Datasets:

Daniel-P-Gonzalez

/

OCD

like 0

# Introduction The *BCL11B* gene encodes a protein which was originally described as chicken ovalbumin upstream promoter transcription factor (COUP-TF)-interacting protein 2 (CTIP2) and radiation induced tumor suppressor gene 1 (*RIT1*). Bcl11b belongs to C₂H₂-zinc finger Krueppel-like proteins, the largest family of transcription factors in eukaryotes. DNA binding is mediated *via* amino acids on the surface of an alpha-helix. Apart from the DNA binding region, Bcl11b possesses domains responsible for transcriptional regulation. The catalogue of proteins and protein complexes known to interact with Bcl11b has grown recently. It includes COUP-TF, the nucleosome re-modeling and histone deacetylation complex (NuRD) and the ubiquitous transcription factor Sp1. Furthermore, recruitment of histone deacetylases (HDAC1 and HDAC2, resp. SIRT1), and the histone methyltransferase SUV39H1 by Bcl11b induces heterochromatin formation and makes it a potent transcriptional repressor. Conversely, Bcl11b interaction with p300 co-activator on the upstream site 1 (US1) of the *IL-2* promoter results in transcriptional activation of *IL-2* expression in activated T-cells. Interestingly, although interaction partners and their binding sequence have been revealed only a few direct target genes of *BCL11B* have been discovered to date. The *P57/KIP2* gene, i.e., a cyclin-dependent kinase inhibitor, is suppressed by Bcl11b. In addition to *P57* and *IL-2* genes, the cancer Osaka thyroid oncogene (Cot) has been recently identified as a direct transcriptional target of Bcl11b. Similar to *IL-2*, the engagement of Bcl11b on the Cot promoter region led to induction of Cot expression and its kinase activity. This caused an augmented phosphorylation of IkappaB kinase which resulted in increased translocation of NF-kappaB to the nucleus and activation of its target genes. The number of Bcl11b-regulated genes was recently extended to include the major cell cycle regulator and tumor suppressor *CDKN1A/p21WAF1* which was demonstrated to be repressed by Bcl11b acting *via* recruiting histone deacetylases and methyltransferases to the *p21WAF1* promoter. The list of biological processes requiring *BCL11B* is constantly expanding. It includes the regulation of T-cell differentiation, normal development of central nervous system (CNS) during embryogenesis, and the maintenance of the latent state of human immunodeficiency virus (HIV) infections. Of note, *BCL11B* which has initially been thought to be of importance to the immune and central nervous systems seems to have a considerably broader impact. The results published within the last years showed the requirement for *BCL11B* in developing skin, where it regulates keratinocyte proliferation and the late differentiation phases determining the process of skin morphogenesis. Moreover, normal tooth development also required *BCL11B* expression and was significantly impaired in *BCL11B*-deficient mice which was accompanied by the decreased expression of ameloblast marker genes and transcription factors driving odontogenesis. The rapidly growing relevance of *BCL11B* for the normal development of different organs and pathogenesis of various diseases requires further investigation of cellular and molecular mechanisms involving Bcl11b. The recently acquired and already established data suggest a critical role of *BCL11B* in three major cellular processes: proliferation, survival and differentiation. The *BCL11B* knockout mouse model revealed the apoptotic phenotype of Bcl11b^−/− thymocytes accompanied by decreased expression of *BCLxL* and *BCL-2* genes. The earlier finding that ectopic expression of *BCL11B* in HeLa cells caused cell cycle retardation inspired the authors to develop a hypothesis of unscheduled proliferation as a primary cause of cell death in Bcl11b-depleted cells. The suppressive influence of accumulated Bcl11b on cell cycle progression was later confirmed in a hematopoietic cell line. However, the mechanism responsible for the reduced proliferation has not been elucidated to date. Moreover, the recently described Bcl11b-mediated transcriptional repression of *P57/KIP2* and *p21WAF1* cyclin-dependent kinase inhibitors responsible for cell cycle restriction should lead to effects opposite to the observed cell cycle retardation. Using a RNA interference approach, we could reproduce the apoptotic phenotype in transformed T cell lines but not in normal mature cells which suggested that apoptosis following Bcl11b depletion is transformation-dependent. These data were confirmed by other reports showing not only reduced survival associated with *BCL11B* knockdown but also impaired response to DNA damage, disabled checkpoint activation and replication stress. These two reports emphasize the anti-apoptotic role of Bcl11b but also uncover its potential function in maintaining genome stability, two features which might contribute to the malignant transformation. The role of *BCL11B* in the pathogenesis of hematological diseases is still a matter of debate. In humans overexpression of *BCL11B* has been linked to lymphoproliferative disorders like the T-cell acute lymphoblastic leukemia (T-ALL), and an acute form of adult T-cell leukemia/lymphoma. Furthermore, *BCL11B* induction correlated with the low differentiation status in head and neck squamous cell carcinoma where Bcl11b co-localized with the cancer stem cell marker BMI-1. In contrast, mouse models of T-cell leukemia revealed frequent homozygous deletions or mutations within the *BCL11B* locus. Moreover, the loss of one allele was identified as a factor predisposing to lymphoma development in p53 (+/−) mice which implicated that *BCL11B* is a haploinsufficient tumor suppressor for thymoma progression in this genetic background and that deletion of one gene copy supports uncontrolled growth. These inconsistencies between animal models and human diseases intensify the dispute on oncogenic or anti- neoplastic properties of *BCL11B*. Here we show that *BCL11B* possesses features which, depending on the expression level and the cellular context, could predispose it to the role of both tumor suppressor and a gene supporting survival of transformed cells. Employing the gene overexpression strategy we demonstrated the relevance of Bcl11b in cell cycle control, chemoresistance and cell death. Our data provide further encouragement for the development of BCL11b-targeting, anti-neoplastic strategies. The attractiveness of such approaches increases together with the accumulating proofs for *BCL11B's* role in different malignant diseases. # Results ## Elevated level of Bcl11b protects transformed T cells from DNA-damage-induced apoptosis but does not interfere with death-receptor pathway The mechanisms of cell death caused by Bcl11b-depletion were already described, but the potential contribution of elevated Bcl11b to cell survival and proliferation has not been explored in detail yet. To address this question, we developed the retroviral-vector-based *BCL11B* overexpression system in two different T cell lines: Jurkat and huT78. The transduction efficiency monitored by measurements of green fluorescence encoded by the reporter gene GFP reached 85%. This resulted in an over 20-fold upregulation of *BCL11B* mRNA expression which reached (huT78) or exceeded (Jurkat) the average expression level measured in primary T-cell leukemia samples. On protein level a 3–5 fold increase of the Bcl11b-specific signal was observed. To ensure an equal and high transgene expression and the reproducibility of further experiments, mock and *BCL11B*-transduced cells were additionally sorted by FACS to eliminate non- modified cells. The elevated Bcl11b level itself did not cause significant difference in cell viability compared to mock-transduced cells as measured by Annexin-V binding assay. In contrast, induction of apoptosis with DNA-damaging agents such as etoposide (ETO) and camptothecin (CAM) was ineffective in Jurkat and huT78 cells expressing high levels of *BCL11B* compared to the parental cell lines. This finding was confirmed for a broad range of ETO and CAM concentrations and for other radiomimetic agents like actinomycin D or dihydroethidium (not shown). We next investigated whether forced expression of *BCL11B* changed only the kinetics of cell death or whether it could also influence the long-term survival upon genotoxic stress. The mock- and *BCL11B*-transduced cells were treated with etoposide for 6h after which the drug was removed by washing. The survival rate was assessed by counting live cells every 48h for 2 weeks. The 6h treatment disabled expansion of mock transduced cells while the cells with elevated Bcl11b restarted proliferation after 4–7 days. The same approach employing camptothecin and actinomycin D validated these findings (not shown). This observation supports the notion that Bcl11b not only altered the kinetics of apoptosis induced by DNA damage but also allowed the survival of cells temporarily exposed to genotoxic stress. To acquire more insight into the Bcl11b-mediated apoptosis resistance, we exposed the cells to the death-receptor ligand TRAIL. Surprisingly, we did not observe any protection against TRAIL treatment in BCL11B-transduced cells neither at low nor at high doses. This indicated that Bcl11b provided selective resistance to genotoxic stress while it had no effect on apoptosis triggered by death-receptors. ## Bcl11b blocks the initiation phase of DNA-damage-induced cell death To explore possible mechanisms of enhanced survival in *BCL11B* overexpressing cells exposed to genotoxic stress we performed a retrograde analysis of the apoptosis cascade. After ETO or CAM treatment the activity of caspase 3 assessed by immunodetection with antibodies raised against the active/cleaved variant was markedly lower in cells transduced with the *BCL11B*-encoding vector. Also the apical caspase 9 acting on the intrinsic DNA-damage-induced apoptotic pathway showed reduced activity in the same conditions. Labeling of the cells with the live-mitochondria-specific dye Mitotracker DeepRed revealed a significant loss of mitochondrial integrity in mock-transduced cells while in *BCL11B*-overexpressing cells the pattern of staining remained unchanged after CAM treatment. Moreover, the activation of Bak-1, a pro-apoptotic member of the Bcl-2 protein family, which initiates the loss of mitochondrial outer membrane potential (MOMP) as a result of DNA damage, was evident in cell expressing endogenous levels of *BCL11B* but only minor in cells with elevated Bcl11b. Sensitization of cells to DNA-damage induced apoptosis with the Bcl-2/BCLxL inhibitor ABT-737 did not overcome the protective effect of Bcl11b at low to intermediate concentrations. However, high doses of the drug induced cell death efficiently in both mock- and *BCL11B*-transduced cells. In contrast to DNA damage, apoptosis initiated by TRAIL treatment proceeded unperturbed regardless of *BCL11B* expression. Both the effector caspase 3 and the apical caspase 8 were efficiently activated in mock- and *BCL11B*-modified cells, in line with previously observed equal Annexin-V binding. We next examined the influence of *BCL11B* expression on induction of DNA breaks resulting from etoposide and camptothecin treatment. The immunostaining for phosphorylation of histone H2AX (γH2AX), an early indicator of the cellular response to DNA damage, showed marked accumulation of γH2AX in cells with endogenous levels of Bcl11b while the cells with elevated Bcl11b showed only weak increase of γH2AX. In order to exclude the detection of apoptosis-related secondary DNA lesions, the measurements were performed before the first signs of apoptosis were detectable ( insets). Alternatively, to prevent the DNA breaks resulting from ongoing apoptosis, the same experiment was performed in the presence of pan-caspase inhibitor Z-VAD. Also under these conditions, *BCL11B*-overexpressing cells acquired less damage upon genotoxic stress (data not shown). Together, these results suggested a role of *BCL11B* in preventing DNA-damaged- induced apoptosis at its early initiation stage. Conversely, the death-receptor triggered apoptosis remained unaffected. ## Bcl11b delays cell cycle progression The analysis of DNA content revealed that in both cellular models induction of *BCL11B* caused slight but significant accumulation of cells at G1 phase of cell cycle. To exclude the possibility that increased number of G1 cells resulted from accelerated cell divisions the cells were treated with a mitosis inhibitor nocodazol. The measurements performed after nocodazol treatment showed that while all mock-transduced cells entered cell cycle and reached either late S or G2/M phase within 12 hours, approximately 20–30% of cells with elevated Bcl11b levels remained at G1. This was further demonstrated by the 3h pulse-labeling of the cells with the nucleotide analog BrdU followed by its immunodetection and FACS measurement. The number of BrdU-negative counts was significantly higher in *BCL11B*-transduced cells compared to cells with wild-type level of the gene. This suggested that forced expression of *BCL11B* caused either G1 arrest or a delayed G1 to S-phase transition in a fraction of cells. The whole-genome expression profiles of mock- and *BCL11B*-transduced Jurkat cells uncovered the regulation of genes known to be involved in cell cycle control (GEO accession number: GSE21382). The list of genes verified by semi- quantitative RT-PCR (sqRT-PCR) is presented in. Our attention focused on genes with previously confirmed function in G1 or G1 to S-phase transition. First, the induction of two cyclin dependent kinase inhibitors *CDKN1C* (p57) and *CDKN2C* (p18) was confirmed on protein level by Western Blotting in both experimental models. While verifying the involvement of the other members of this gene family we found significant accumulation of *CDKN1B* (p27) which resulted either from increased stability or defective degradation since no major increase of p27 mRNA could be measured (data not shown). The later possibility seemed more likely since the transcriptional downregulation of the *SKP2* gene involved in proteasomal degradation of p27 was identified. Moreover, the level of Skp2 declined markedly upon *BCL11B* induction. Interestingly, in Jurkat cells we did not find any differences in pRB phosphorylation status which usually precedes the cell cycle entry. This was confirmed by using phospho-specific antibodies labeled with fluorescent dyes followed by FACS analysis in the other cellular system. The level of retinoblastoma-like protein 1 (*RBL1*, p107) did not vary either (not shown). However, the third member of this gene family *RBL2* (p130), normally undergoing degradation during cell cycle entry, showed significantly increased protein levels. This suggested the possible involvement of pocket proteins and their target genes – E2Fs in cell cycle delay resulting from *BCL11B* induction. This assumption was further strengthened by the 3-fold reduction in the level of the *MYCN* oncogene which is negatively regulated by Rbl2. The decreased *MYCN* transcription was confirmed in *BCL11B*-overexpressing cells in both experimental models by sqRT-PCR and verified on protein level by Western blotting in Jurkat cells. Moreover, the detection of the Myc-n protein by immunofluorescence followed by FACS revealed a negative correlation between the Myc-n signal and the intensity of the reporter gene formerly shown to correspond to Bcl11b levels. This indicated that elevated Bcl11b, apart from inducing p18, p27 and p57 could delay the cell cycle entry by causing accumulation of RBL2 which suppressed *MYCN* expression. The upregulation of the *HBP1* transcription factor (.) known to cooperate with *RBL2* in suppressing the *MYCN* promoter strongly implied the potential involvement of the Bcl11b-p130/Hbp1-Myc-n axis in the observed G1 arrest/delay. In sum, we revealed potential mechanisms of Bcl11b-mediated cell cycle restriction which included activation or induction of p18 and p57 cyclin dependent kinase inhibitors, accumulation of p27 p130 and suppression of *MYCN* oncogene. The increase of cells at G1 phase and decreased fraction of replicating cells could explain the low toxicity of radiomimetic treatment with etoposide and camptothecin. ## Cell cycle restriction and DNA damage resistance caused by *BCL11B* overexpression require its HDAC-interacting N-terminal domain Bcl11b was identified as a part of several multiprotein complexes and a number of Bcl11b domains involved in these interactions have been identified. Therefore, we were interested in identifying the region of Bcl11b relevant for binding complexes/proteins necessary for the DNA-damage resistance and/or cell cycle restriction following its overexpression. We designed, cloned and expressed three different *BCL11B* variants covering the regions known to be important for the formerly established interactions/functions (.). The C-terminally truncated variant one encoded the NuRD interacting N-terminal domain (*BCL11B* ΔC-1), the longer variant 2 contained additionally the SUV39H1/HP1/SIRT1/Sp1 /Tat binding domain consisting of the proline-rich region and two central zinc fingers (*BCL11B* ΔC-2). The third mutant allele contained the whole coding sequence excluding the N-terminal part (*BCL11B* ΔN). The constructs were used for transduction of Jurkat and huT-78 cells and the protein expression and the calculated molecular weight was confirmed by Western blot analysis (.). To test the sub-cellular localization the deletion mutants were sub-cloned in frame into the pEGFP-C1 vector and transfected by nucleofection into Jurkat cells. As shown on, the transfection of EGFP-only vector (mock) caused even distribution of green fluorescence protein within the cells. In contrast, when fused to wild-type *BCL11B* cDNA the EGFP was located in the nucleus showing the spackle-like structures previously described by others. A pattern similar to the wild-type pattern could be detected in cells transfected with *BCL11B* ΔC-2 and *BCL11B* ΔN deletion mutants. Contrary, the N-terminal part (*BCL11B* ΔC-1) localized in the extranuclear compartment. These data indicated that the nuclear localization signal was located downstream of NuRD- interacting N-terminal domain and its deletion caused aberrant non-physiological location of the fusion protein. Next we investigated whether and to which extent the N- and C-terminally truncated *BCL11B* variants preserved the properties of wild-type gene. The induction of DNA damage by the radiomimetic drug etoposide in cells transduced with empty vector, wild-type and truncated *BCL11B* alleles followed by apoptosis assays confirmed the protective activity of wild-type *BCL11B* in both transduced T-cell lines (.). As expected, the aberrantly located *BCL11B*- ΔC-1 variant did not have any influence on DNA-damage induced apoptosis. Conversely, the second C-terminal mutant *BCL11B*- ΔC-2 lacking almost half of the protein including three zinc fingers was as effective in preventing cell death as its wild-type counterpart. The removal of the N-terminal domain (*BCL11B*- ΔN) in turn resulted in the complete loss of anti-apoptotic properties. Since we previously showed that the DNA-damage resistance was accompanied by and presumably associated with cell cycle restriction at the pre-replication phase, we performed the synchronization of the transduced cells at G₂/M using nocodazol (.). DNA staining before nocodazol treatment showed a slightly elevated G₀/G₁ fraction in *BCL11B*- ΔC-2 and wild-type *BCL11B* transduced cells. Eight hours later the accumulation at late S and/or G₂/M phases and essentially no G₀/G₁ was observed in cells transduced with empty, *BCL11B*- ΔC-1 and *BCL11B*- ΔN- containing vectors. However, *BCL11B*- ΔC-2 and wild-type *BCL11B* overexpression caused a significant block of cell cycle at G₀/G₁ which correlated well with the DNA-damage resistance. These results indicated that the anti-apoptotic and cell cycle regulatory activities of BCL11B are strictly dependent on the presence of the N-terminal region, previously reported to interact with protein complexes containing histone deacetylases (HDAC1 and 2). To verify the involvement of HDACs in Bcl11b-mediated cell cycle control and low sensitivity to genotoxic stress we applied the histone deacetylase inhibitor trichostatin A (TSA). At high dose, TSA showed equal toxicity to mock-, *BCL11B*-WT and *BCL11B*-mutant- transduced cells as measured by Annexin V-binding assay (.). The sub-optimal dose of TSA caused generally lower but comparable toxicity ( insets). Besides this, the low doses of TSA re-sensitized the cells transduced with *BCL11B* and *BCL11*B-ΔC-2 mutant to etoposide treatment. Similar results were obtained with a different HDAC class I and II inhibitor SAHA, while a class III inhibitor, blocking SIRT1 activity was ineffective in interfering with Bcl11b mediated protection (data not shown). This provided further evidence for the engagement of HDAC1 and 2 in genotoxic stress resistance and cell cycle restriction observed upon induction of *BCL11B*. Of note, the N-terminal, HDAC-binding region itself was not sufficient for the described effect. Even when physiologically localized in the nucleus *via* addition of the SV40 nuclear localization signal at its C-terminus it did not alter the DNA-damage sensitivity or cell cycle progression (not shown). # Discussion The crucial role of *BCL11B* gene in the regulation of cell cycle and apoptosis has been postulated based on the results obtained from both *in vitro* and *in vivo* studies. However, its potential positive or negative contribution to tumor development was not convincingly settled so far. The growth-promoting properties of truncated *BCL11B* variants isolated from leukemic cells, the losses of heterozygosity at the very early stages of lymphomagenesis and detection of *BCL11B* haploinsufficiency for suppression of mouse thymic lymphomas, support the speculation of *BCL11B* tumor suppressive activity. On the other hand, in the majority of human T-ALLs the expression of *BCL11B* was similar to the level obtained by our overexpression approach in Jurkat and huT-78. Moreover, amplifications of the *BCL11B* locus accompanied by increased mRNA expression and protein were shown in adult T cell leukemia/lymphoma and correlated with the aggressiveness of the disease which indicates a function opposite to tumor suppression. Diverse Bcl11b features can be observed also on the biochemical level. Sharing structural similarities with transcriptional repressors and interacting with chromatin-silencing protein complexes, Bcl11b can act also as a potent transcriptional activator. The list of the genes negatively regulated by Bcl11b contains the known tumor suppressors p57/*KIP2* and p21/*WAF1*, while the activated targets include genes inducing and regulated by NFκB, like Cot kinase or *IL-2*, respectively. These data, supported by the observed elevated expression of *BCL11B* in T cell malignancies, imply its pro-survival and growth-promoting function. The notion was strengthened by findings showing the critical dependence of transformed T cells' survival on *BCL11B* expression. In summary, *BCL11B* seems to function pleiotropic and may lead to opposite effects in different cellular systems. This encouraged us to evaluate the consequences of the increased *BCL11B* levels in transformed cells endogenously expressing the gene. We developed a retrovirus-mediated *BCL11B* overexpression system in two different T cell leukemia cell lines and focused our research on apoptosis and proliferation. We found significantly increased resistance of *BCL11B*-overexpressing cells to genotoxic-stress and cell death which appeared to be blocked at the early stages of the process. Conversely, apoptosis triggered by death receptors remained undisturbed suggesting that the enhanced survival upon genotoxic stress is not due to a general apoptosis failure. Interestingly, global RNA profiling of wildtype and *BCL11B*-overexpressing cells did not reveal any significant differences in the expression of genes known to be engaged in the canonical cell death pathways. The immunodetection of the pro- and anti- apoptotic members of the BH3-only protein family in non- or etoposide treated mock- and *BCL11B*- transduced cells also did not show any major increase of the anti-/pro-apoptotic protein ratio (data not shown). The observed lower number of DNA lesions in *BCL11B*-overexpressing cells upon damage induction suggests limited efficiency of the DNA-damaging drugs as the likely reason for the lower death rate in *BCL11B*-overexpressing cells. In light of the previously described growth-limiting effect of *BCL11B* in *BCL11B*-negative HeLa and hematopoietic progenitor cell FDC-P1lines, a tempting hypothesis is that cell cycle restriction causes the reduced percentage of S-phase cells. Since the DNA-damaging treatment we applied was proven to kill predominantly cells in S-phase the diminished fraction of cells replicating DNA resulting from G1 accumulation would lead to the chemoresistance of the non- replicating cell population. We confirmed this possibility by pre-treating Jurkat and huT-78 cells with the DNA polymerase inhibitor aphidicolin which reduced the rate of G1 to S transition and limited the fraction of S phase cells. When subsequently treated with etoposide, camptothecin or other radiomimetic drugs a fraction of cells blocked at G0/G1 phase cells remained vital while the dead cells originated from S phase (data not shown). The cell cycle analysis and bromodeoxyuridine (BrdU) incorporation assays showed a clear decrease of the S phase fraction in both cell lines upon *BCL11B* overexpression compared to mock treatment. This was due to G1 arrest which was most evident when the cells were synchronized at G2/M with nocodazol which explains the lower sensitivity to radiomimetic drugs. Of note, the G1 arrest was not permanent. When pulse-treated with etoposide or other DNA-damaging drugs, the *BCL11B*-transduced cells restored growth after several days while the mock- transduced cells died. This observation suggests that elevated levels of Bcl11b could trigger a fraction of chemoresistant tumor-initiating cells. Surprisingly, the cell cycle arrest seemed not to be accompanied by the activation of the checkpoint mechanism and the phosphorylation status of the checkpoint proteins Chk1/Chk2 was not significantly altered (not shown). This indicates that although the lack of Bcl11b resulted in checkpoint inactivation, the cell cycle delay caused by *BCL11B* overexpression was mediated by a Chk1/Chk2 checkpoint- independent mechanism. The candidate genes potentially involved in the observed cell growth suppression were identified by the comparison of mRNA profiles of mock- and BCL11B-transduced cells. This analysis revealed upregulated expression of the G1 cycline-dependent kinase inhibitor *CDKN2C* (p18/Ink4c), leading to elevated levels of the encoded protein. The p18 protein binds and inhibits selectively two cyclin dependent kinases, Cdk4 and 6, which are responsible for controlling G1 to S transition. The ectopic expression *CDKN2C* suppresses cell growth in a manner resembling the activation of the tumor suppressor gene pRB. Another representative of the CDKis found to be upregulated in *BCL11B*-transduced cells is *CDKN1C* (p57/Kip2). The protein encoded by this gene is a potent, tight-binding inhibitor of several G1 CDKs, arresting the cells in G1 when overexpressed. The induction of this gene in cells with elevated Bcl11b is especially interesting since the gene has been previously shown to be negatively regulated by Bcl11b *via* associating the NuRD nucleosome remodeling complex. A potential mechanism which could explain the observed transcriptional activation and accumulation of p57 could be based on the “titration” model, in which the accessibility of HDACs necessary for suppression is limited by an excess of Bcl11b. A depleted activity of HDACs as a cause of p57 induction was already identified. Different class I and II HDAC inhibitors showed a potent inducing activity on *CDKN1C*. As shown recently, a different member of the Krueppel-like factor family KLF4 demonstrated a potent anti- apoptotic effect upon HDACi treatment which was mediated in part by direct binding and transcriptional upregulation of p57 followed by inhibition of the stress response phosphorylation pathway. Finally, the role of E2F1 in the transcriptional regulation of p57 must be considered. We report here the activation of the Cdk 4/6 inhibitor p18. As communicated recently, the inhibition of the cyclin dependent kinases by small molecules led to the transcriptional activation of *CDKN1C* which involved direct binding of E2F1 to the promoter. It was suggested that this regulatory loop served to limit the E2F1's death-inducing activity. The third member of the cyclin-dependent kinase inhibitors which accumulated in BCL11B-overexpressing cells was *CDKN1B* (p27/Kip1). It was previously reported that downregulation of Bcl11b led to the transcriptional suppression and decrease of p27 protein, suspected of being one of the cell-death triggering events after *BCL11B* knockdown. This, together with our finding implied a correlation between Bcl11b and p27 levels. However, we did not observe an effect of elevated Bcl11b on p27 transcription indicating the involvement of a post- transcriptional regulation. The level of p27 protein can be modulated *via* multiple posttranscriptional mechanisms. The most potent and the best characterized mechanism of p27 regulation is its degradation preceded by polyubiquitylation by the SCF^SKP2-E3 ubiquitin ligase complex at G1 phase. The involvement of the impaired Skp2-E3-mediated p27 proteolysis in our system is strengthened by the fact that *SKP2* was significantly reduced on mRNA and protein level in cells transduced with *BCL11B*. Support for functional relevance of the decreased Skp2 level following *BCL11B*-transduction was provided by accumulation of another substrate of the Skp2-E3 complex, the retinoblastoma-like protein 2 (*RBL2*/p130) which was not accompanied by any transcriptional activation. The increase of p130 protein was the only change that could be identified within the pocket protein family known to be crucial for G1 to S transition. However, the accumulation of p130 could contribute to cell cycle retardation and apoptosis resistance in *BCL11B*-transduced cells. In neurons, the integrity of the E2F4-p130-HDAC/SUV39H1 complex was shown to be critical for cell survival and overexpression of p130 significantly increased the apoptosis resistance triggered by camptothecin treatment. Furthermore, p130 participated in the suppression of the *MYCN* promoter by interacting with the HBP1 high mobility group (HMG) transcription factor increased transcription of which was observed in *BCL11B*-overexpressing cells. The involvement of the described mechanism in our experimental system was supported by markedly decreased *MYCN* transcription upon BCL11B-transduction. In the cells expressing *BCL11B* at higher levels as visualized by the strong EGFP signal, Myc-N protein was almost undetectable, while the cells with dim green fluorescence showed stronger signals. This observation argues for a strict negative correlation between Bcl11b and Myc-N. Interestingly, inactivation of *MYCN in vivo* led to the activation of p27/Kip1 and p18/Ink4c which indicates that suppression of *MYCN* expression could lead to the effects triggered by *BCL11B* overexpression. However, it seems that changes in the level of Myc-N are not the sole reason for cell cycle abnormalities and cell death resistance. Overexpression of *MYCN* markedly increased proliferation of Jurkat and huT-78 cells expressing endogenous levels of *BCL11B* but only minimally influenced the cell cycle progression in cells transduced with *BCL11B*-encoding vector (data not shown). The notion that parallel activation of multiple pathways seems to lead to the growth suppression and apoptosis resistance observed after BCL11B overexpression is also supported by additional studies on CDKis. Neither the specific knockdown of the particular CDKi nor the simultaneous suppression of two or three CDKis could restore normal growth or apoptosis resistance. In order to get more insight into potential mechanisms of Bcl11b-mediated cell cycle and cell death regulation, we created a series of *BCL11B* deletion mutants lacking the important DNA-binding and protein-interacting domains. Our data strongly indicate that the identified biological consequences of *BCL11B* upregulation were strictly dependent on the presence of the N-terminal HDAC- interacting domain. The removal of the C-terminal half of *BCL11B* cDNA, encoding the three zinc fingers but carrying the N-terminal domain retained the function of wild-type *BCL11B*. Conversely, the deletion of the N-terminal part completely inactivated the mutant protein, although it localized physiologically in the nucleus. Interestingly, the short fragment encoding just the HDAC- interacting region did not show any *BCL11B*-resembling activity, even when localized in the nucleus by addition of viral NLS. Of course one cannot eliminate the improper folding of such a truncated protein as the reason of its inactivation. Taken together, it appears that Bcl11b acts by recruiting some factors to the DNA template through its HDAC-binding domain. In order to provide additional support for this hypothetical mechanism, we treated wt and *BCL11B*-overexpressing cells with the HDAC inhibitor trichostatin A (TSA). Unfortunately, we could not investigate the effects of TSA on cell cycle arrest upon *BCL11B*-transduction because of high toxicity of the drug alone and in combination with the G2/M synchronizing nocodazol. However, we observed a significant re-sensitization of TSA-treated *BCL11B*-overexpressing cells to DNA-damage induced apoptosis, supporting the involvement of HDAC in Bcl11b-mediated cell cycle and apoptosis control. Our results suggest that *BCL11B* may perform opposite functions depending on the cellular context and the expression level. The upregulation of the gene significantly increases the genotoxic stress resistance which could constitute an oncogenic feature providing chemoresistance and allowing the survival of transformed cells. On the other hand, the insensitivity to DNA damage is accompanied by a markedly lower proliferative activity, which is considered to be tumor suppressive. We speculate that in non-transformed tissue *BCL11B* could control the proliferation, ensure genome stability and prevent tumor development and/or apoptosis. This feature qualifies *BCL11B* as a tumor suppressor or a gene maintaining the balance between proliferation and cell death. But in the context of a tumor tissue *BCL11B* could act as potential oncogene being an attractive target for therapeutic approaches. Similar ambivalent modes of action were identified recently for p21/*WAF1* in a mouse model of leukemia. In this setting a prototype tumor suppressor manifested its oncogenic features by restricting cell cycle followed by limited DNA damage and consequently maintaining self-renewal potential of leukemic stem cells. Of note, *BCL11B* was found to be co-expressed with the cancer stem cell marker *BMI-1* in a subset of undifferentiated squamous cell carcinoma cases and cell lines. This indicates that elevated *BCL11B* could play a similar role in tumors expressing it by controlling proliferation and chemoresistance of tumor- initiating cells. The survival assay presented here seems to confirm such a possibility. Jurkat and huT-78 cell forced to overexpress *BCL11B* survived and efficiently restored cell growth upon temporal exposition to genotoxic stress. In conclusion, we believe that the data presented here strengthen the role of *BCL11B* in tumor survival rather than development. Targeting Bcl11b *via* inhibitory approaches might be a promising tool for the treatment of rapidly growing list of cancer types expressing *BCL11B*. Moreover, the identification of HDAC-interacting domain of Bcl11b as a critical mediator of its activity and the finding that BCL11B-mediated apoptosis resistance can be reversed by HDACi may have important implication for therapeutic interventions against *BCL11B*-positive tumors. # Materials and Methods ## Reagents All biochemicals were from Sigma Chemicals (Munich, Germany) unless otherwise specified. Human recombinant TRAIL was purchased from PeproTech (Peprotech Ltd, London, United Kingdom). The Bcl-2/BCLxL inhibitor ABT-737 was provided by Abbott (Abbott Laboratories Il, USA). ## T-cell leukemia samples The 20 mRNAs isolated from T-cell lymphoblastic leukemias were provided by different clinical units. All samples were obtained according to the guidelines for informed consent approved by local Ethical Research Committees. ## Plasmids The constructs used in this study were based on the pMSCV-derived retroviral vector pMIGR encoding internal ribosome entry site followed by EGFP gene. The sequence encoding the full length *BCL11B* was amplified from cDNA prepared from healthy individual and cloned between *Bgl*II and *Eco*RI restriction sites. The deletion mutants of *BCL11B* were constructed using PCR techniques with proof reading polymerase Pfu (Promega, WI, USA) using primers equipped with *Bgl*II and *Eco*RI recognition sequences and start/stop codons where required. All constructs were checked by sequencing to exclude the presence of mutations. The vector carrying the pantropic viral envelope protein pVSV-G was purchased from Clontech (Clontech Laboratories, CA, USA). ## Cell culture, transfection and transduction The human T cell leukemia and lymphoma cell lines Jurkat (DMSZ, Braunschweig, Germany) and huT-78 (ATCC, Rockland, MD) were maintained in RPMI-1640 medium supplemented with 10% fetal calf serum (PanBiotech Berlin, Germany), Glutamax (Invitrogen, CA, USA) and Plasmocin (Lonza Scientific, Switzerland). The retrovirus producing cell line GP2-293 (Clontech Laboratories, CA, USA) was maintained in Dulbecco's modified Eagle's medium (Invitrogen, CA, USA) supplemented as above. For recombinant retrovirus production, non-confluent GP2-293 cells were co- transfected with 10 µg of purified pVSV-G and empty pMIGR plasmid (mock) or pMIGR encoding *BCL11B* variants. The transfection procedure was performed using CalPhos Mammalian Transfection Kit (Clontech Laboratories, CA, USA) according to the manufacturer's instructions. The retrovirus-containing media were collected 48h after transfection and used immediately or stored at 4°C and used within 7 days. To transduce the target cells, the retroviral supernatants were supplemented with 8 µg/ml Polybrene (Sigma Chemicals, Germany) and added to exponentially growing Jurkat and huT-78 cells for 8 hours. The procedure was repeated three times. The efficiency of gene transfer was estimated by FACS detection of EGFP reporter gene. To ensure high proportion of transduced cells, EGFP-positive cells were enriched using high-speed cell sorter FACSAria (BD Biosciences, NJ, USA). Sorted, at least 90% EGFP-positive cells were used for further experiments. ## Apoptosis induction and viability assays To induce DNA-damage, Jurkat and huT-78 cells transduced with the empty- and *BCL11B*-encoding retroviral vectors were treated with etoposide, camptothecin, actinomycin D or dihydroethidium (Sigma, Germany). The influence of histone deacetylase inhibition on the viability of mock- and *BCL11B*-modified cells was verified using increasing concentrations of trichostatin A, nicotinamide (Sigma, Germany) or SAHA (Cayman Chemicals, MI, USA). To investigate the role of histone deacetylases in DNA damage resistance of *BCL11B*-overexpressing cells, the treatment was combined with damage induction using etoposide. Death-receptor pathway was activated by TRAIL (ProImmune Ltd., United Kingdom). Where indicated, incubation with etoposide was combined with the ABT-737 Bcl-2/BCLxL- inhibitor treatment (Abbott Laboratories, Il, USA). The long term survival upon genotoxic stress was initiated by 6 hours pulse treatment with radiomimetic drugs followed by extensive washing and 14 days cell culture accompanied by cell counting every 48 hours. Apoptosis was measured by FACS using Annexin V-APC binding assay (BD Pharmingen, CA, USA). The activation of caspase 3 and 8 was quantified by immunofluorescent detection of the cleaved variants followed by FACS analysis. In brief, mock- and *BCL11B*-transduced Jurkat and huT-78 cells were fixed/permeabilized using Cytofix/Cytoperm kit (BD Pharmingen, CA, USA) and incubated with cleaved caspase 3 (Asp175) or cleaved caspase 8 (Asp391) rabbit monoclonal antibodies and subsequent staining with the secondary goat anti- rabbit IgG F(ab)² fragment stained with APC (Santa Cruz Biotechnology, CA, USA). Activation of caspase 9 was assayed by western blot using rabbit polyclonal antibody recognizing the cleaved/active variant of the protein (Cell Signaling Technology, MA, USA). The integrity of mitochondria was evaluated with the live-mitochondria-specific dye Mitotracker Deep Red FM (Invitrogen, CA, USA) and subsequent FACS measurements. Activation of Bak1 pro- apoptotic protein was tested by immunofluorescence using anti-Bak mouse monoclonal antibody (clone TC-100, Calbiochem, Merck Chemicals, Germany) binding the N-terminal domain exposed upon protein activation and the secondary goat anti-mouse-APC antibody. The determination of the DNA damage induced by radiomimetic drugs was performed by immunofluorescence using AlexaFluor 647-labeled anti-phospho-histone H2A.X antibody (γH2A.X, Cell Signaling, MA, USA) two hours upon treatment initiation. The relative signals obtained in mock- and *BCL11B*-transduced Jurkat and huT-78 cells were measured by FACS. The progress of apoptosis was monitored at the same time by Annexin-V binding assay. To further exclude the secondary, apoptosis related DNA breaks, the same procedure was repeated in the presence of 20 µM pan-caspase inhibitor Z-VAD (Sigma, Germany). ## Cell cycle analysis Cell cycle status has been determined by propidium iodide DNA staining and flow cytometry analyses. Briefly, the cells were washed and resuspended in propidium iodide staining solution containing RNase A (Sigma, Germany) and incubated in the dark for 30 min. Flow cytometric analysis was performed in a FACScalibur instrument (Becton Dickinson, San Jose, CA, USA). Cell cycle analysis was performed using ModFit *LT* (Verity Software House, Topsham, ME, USA). To determine the rate of G1 to S transition, cells were blocked in G2/M phase with 0,5 µM nocodazole and the cell cycle analysis was done at different time-points. In addition, the G1 to S progression was assessed by pulse treatment with the nucleotide analogue BrdU. BrdU incorporation, reflecting the speed of cell cycle progression was determined with a fluorescently labeled antibody and measured by flow cytometry (APC BrdU Flow Kit, BD Pharmingen, Franklin Lakes, NJ, USA). ## RNA isolation, microarray analysis and quantitative RT-PCR Total RNA was isolated from Jurkat cells transduced with empty and *BCL11B* encoding vector with Trizol Reagent (Invitrogen). Affymetrix array analysis was performed using the One-Cycle Target Labelling and Control Reagents, which contain the GeneChip Sample Cleanup Module, and Human Genome U133 Plus 2.0 DNA arrays (Affymetrix, Santa Clara, CA) according to the manufacturer's instructions. The subsequent scanning was performed with the GeneChip Scanner 3000 (Affymetrix). For quantitative RT-PCR (qRT-PCR), RNA was reverse transcribed with MultiScribe reverse transcriptase (Applied Biosysytems) using random hexamers. All primers and probes were synthesized by TibMolBiol (Berlin, Germany). Quantification of *BCL11B* mRNA was performed as described elsewhere. Quantification of genes regulated upon *BCL11B* overexpression was performed in Jurkat and huT-78 cell lines. In brief, PCR amplification was performed in a total volume of 25 µl with 2 µl of cDNA, 25 pmol of each primer and SYBR® Green PCR Master Mix. The list of primer sequences is provided in Supplementary. A non-specific signaling resulting from primer-dimer formation was ruled out by analyzing the dissociation curves and agarose gel electrophoresis. ## Analysis of protein levels Sodium dodecyl sulfate polyacrylamide gel electrophoresis was performed using standard techniques. Equal protein loads were confirmed by Ponceau staining and by immunodetection of ß-actin (Sigma). The Bcl11b protein was detected using antibodies directed against N-terminal, central or C-terminal epitopes (Bethyl Laboratories, TX, USA). The proteins corresponding to the regulated genes confirmed by quantitative RT-PCR were detected with the following antibodies: cleaved Caspase 9 rabbit polyclonal antibody, p57 rabbit polyclonal antibody, p27 rabbit polyclonal antibody, p18 mouse monoclonal antibody, Skp2 rabbit polyclonal antibody, Rb antibody kit, Myc- N rabbit polyclonal antibody (Cell Signaling Technology, MA, USA) and p130 (RBL2) rabbit polyclonal antibody (Santa Cruz Biotechnology, CA, USA). Proteins were visualized by chemiluminescence using rabbit- or mouse-specific Western-*SuperStar*™ Immunodetection system (Applied Biosystems, Life Technologies, CA, USA). The intracellular staining of Myc-N was performed in fixed and permeabilized cells (Cytofix/Cytoperm, BD Pharmingen, Franklin Lakes, NJ, USA) with N-Myc specific rabbit primary antibody and subsequent immunodetection using goat anti- rabbit APC-labeled antibody (Santa Cruz Biotechnology, CA, USA). ## Sub-cellular localization of full-length BCL11B and BCL11B-derived deletion mutants The wild-type *BCL11B* coding sequence or alternatively the truncated variants of the gene were cloned in-frame into pEGFP-C1 plasmid vector (Clontech Laboratories, CA, USA) downstream of the enhanced green fluorescent protein (EGFP). The plasmid vectors were transfected into Jurkat cells using the Nucleofector device (Lonza, Switzerland). After 12–24 hours incubation, cells were fixed and the EGFP expression and localization was determined by fluorescence microscopy. # Supporting Information [^1]: Conceived and designed the experiments: PG VN CAS. Performed the experiments: PG VN M. Delin M. Depke PH. Analyzed the data: PG VN GKP M. Delin M. Depke PH UV. Contributed reagents/materials/analysis tools: M. Depke PH UV. Wrote the paper: PG CAS. Proofreading and corrections of the manuscript: M. Delin M. Depke GKP VN PH UV CAS. [^2]: The authors have declared that no competing interests exist.

# Introduction COVID-19 remains a global public health concern and is especially pernicious in regions with limited public health infrastructure that suffer from inadequate epidemiologic surveillance and delayed implementation of pandemic countermeasures. In the Central Asian states, such as Kazakhstan, substantial underestimations (of \~14-fold) of COVID-19 incidence and associated mortality have led to public distrust and slow uptake of public health measures, including vaccination. To date, the extent of community exposure to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in Kazakhstan is incompletely understood. Thus, as of 7 August 2021 (when analysis presented in this work was completed), the officially reported number of all-time COVID-19 cases in Kazakhstan was 689,402 (626,402 of which were PCR-confirmed, while the rest were diagnosed based on clinical disease manifestations). These figures represented a cumulative prevalence of \~3.7%- a prevalence that appears low given the substantial excess of infections and mortality estimated for Kazakhstan consistent with COVID-19. Disparities between reported cases and true infections occur due to a plethora of factors, including unreported asymptomatic and mild infections, limited access to timely clinical and laboratory confirmation of COVID-19 diagnosis, and false-negative laboratory test results. Case underestimation varies broadly by country and is most pronounced in lower income regions. Similar to its neighbouring Central Asian and Eastern European states, Kazakhstan’s healthcare and vital registration systems have struggled to keep a consistent tally of COVID-19 incidence and mortality. This was especially evident at the onset of the pandemic in Spring-Summer of 2020 when cases were counted solely if COVID-19 was identified as the main cause of hospitalization and/or death, resulting in a significant undercounting of undiagnosed pneumonia cases, which most likely were COVID-19-associated. One way to estimate the proportion of the population with previous exposure to COVID-19 is by using serological surveillance, which has been under-utilized in Kazakhstan and other Central Asian states. Here, we wished to gain insight into the true SARS-CoV-2 exposure rates in the Karaganda district of Kazakhstan, a multi-ethnic region, inhabited predominantly by ethnic Kazakhs (\~60%), and other ethnic groups with diverse Slavic and Eastern European and Central Asian backgrounds. Therefore, we assessed full-length SARS-CoV-2 Spike (S)-specific IgG and IgA titres in a public university-based cohort, representing a diverse array of people with different risks of exposure to COVID-19. # Materials and methods ## Study setting and participant recruitment This study was conducted in conjunction with screening for a clinical trial assessing immunogenicity of the Sputnik-V vaccine (ClinicalTrials.gov \#NCT04871841) based in Karaganda, the capital of Karaganda region situated in Central Kazakhstan. This cohort was chosen for the serologic studies owing to funding availability and perceived feasibility in the context of the readily available resources and active participant recruitment within the infrastructure of the larger clinical trial. Since February 2020 (and to the date when analysis presented in this work was completed), the Karaganda region (population \~1.3M) had over 63,000 (\~5% of regional population) reported COVID-19 infections, placing it behind several other locales including the capital, Nur-Sultan (population \~ 1.0M), which has had a reported all-time COVID-19 prevalence of \>11%. Participant screening occurred in April-May 2021 at a COVID-19 vaccination clinic for university employees at the Karaganda Medical University. The study participants comprised of medical university administrative staff and instructors (61%), clinical laboratory staff (9%), and healthcare practitioners from affiliated teaching hospitals (30%). Consenting, asymptomatic adults, who had not previously received a COVID-19 vaccine, were invited to participate in the study. Exclusion criteria were presence of respiratory symptoms or laboratory-confirmed COVID-19 diagnosis within two weeks prior to the study. Short questionnaires addressing the participants’ demographic background and recent history of COVID-19 exposure were administered. To validate the IgG and IgA assay positivity thresholds, we performed ELISA on pre-pandemic samples, consisting of archived plasma samples (n = 10, 3 men and 7 women, median age (IQR) = 48(34.3–55.5) collected in 2016 as part of clinical studies of colorectal cancer and pertaining to the cancer- free control group in the original study. ## Sample collection and processing Nasopharyngeal swabs were collected following the national guidelines into DNA/RNA shield media (Zymo Research, Irvine, US). Blood (5 ml) was collected by venipuncture into EDTA tubes (Improvacuter, Gel & EDTA.K2, Improve Medical Instruments, Guangzhou, China) both in the pandemic and pre-pandemic studies. Blood plasma was isolated by centrifugation at 2,000 × g for 10 minutes. All samples were stored at -80°C prior to analyses. ## PCR screening for SARS-CoV-2 Total RNA was isolated from nasopharyngeal swabs by magnetic bead-based nucleic acid extraction (RealBest Sorbitus, Vector-Best, Novosibirsk, Russia) and used for SARS-CoV-2 real-time RT-PCR testing by the Real-Best RNA SARS-CoV-2 kit (Vector-Best, Novosibirsk, Russia) targeting the SARS-CoV-2 RdRp and N loci, following the manufacturer’s protocol. ## IgG and IgA assays SARS-CoV-2 S1 IgG and IgA ELISAs were performed using commercially available assays (Euroimmun Medizinische Labordiagnostika AG, Lübeck, Germany) on the Evolis 100 ELISA reader (Bio-Rad) according to the manufacturers’ protocols. Optical density (OD) ratios were calculated as ratio of the OD reading for each sample to the reading of the kit calibrator at 450 nm. In the initial analysis, we used the Euroimmun-recommended OD ratio cutoff values for both IgG and IgA, which are “\<0.8” for Ig-negative samples, “0.8–1.1” for Ig-borderline samples, and “\> = 1.1” for Ig-positive samples. We noted that using the manufacturer’s cutoff values: of all IgG "borderline" participants (n = 9), 7 (77.8%) were IgA+ (IgA OD ratio\> = 1.1), 1 (11.1%) was IgA borderline and 1 (11.1%) was IgA negative, while of all IgA "borderline" participants (n = 6), 2 (30.0%) were IgG+ (IgG OD ratio\> = 1.1), 1 (20.0%) was IgG borderline, and 3 (50.0%) were IgG negative. The mean OD450 ratios of the pre-pandemic samples were 0.3 for both IgG and IgA. Therefore, we empirically assumed that \~99.7% of IgG- and IgA- negative samples would fall within 3 standard deviations of the mean, i.e. within OD450 ratios of 0.45 and 0.51 for IgG and IgA, respectively. Thus, for both IgG and IgA assays, we used the manufacturer-recommended threshold (0.8), which is conservatively above our calculated empiric negative thresholds, and considered all samples with OD ratios \<0.8 and \> = 0.8 as “negative” and "positive", respectively. Using this in-house threshold, we defined the "No Prior COVID" subjects as negative for both IgG and IgA (IgG-, IgA-) and the "Prior COVID" subjects as positive for either or both IgG and/or IgA (IgG+/-, IgA+/-). ## Surrogate SARS-CoV-2 neutralization assays A virus neutralizing assay was performed using a commercially available kit (cPass SARS-CoV-2 Neutralization Antibody Detection Kit, \#L00847-C, GenScript Biotech Co., Nanjing City, China) in a subset of 55 participants. The assay is designed to assess inhibition of the interaction between the recombinant SARS- CoV-2 receptor binding domain (RBD) fragment and the human ACE2 receptor protein (hACE2). Briefly, plasma samples and manufacturer-provided controls were pre- incubated with the horseradish peroxidase (HRP)-conjugated RBD at 37°C for 30 min, and then added to the hACE-2 pre-coated plate for incubation at 37°C for 15 min. After washing and incubation with the tetramethylbenzidine (TMB) substrate, the absorbance of the final solution was measured at 450 nm using the Evolis 100 ELISA reader (Bio-Rad). Quality control was done following the manufacturer’s recommendations. Neutralization % was calculated by subtracting the negative control-normalized absorbance of the samples from 1 and multiplying it by 100%; a manufacturer-recommended cut-off of 30% was used for detectable SARS-CoV-2 neutralization activity. ## Statistical analysis Demographic differences were assessed using the two-sided Mann-Whitney U test for age and BMI, Pearson χ2 for sex, ethnicity and comorbidities and Fisher’s exact test for self-reported history and workplace exposure. The 95% confidence intervals (CI) were calculated using the binomial "exact" method. Correlations among variables were explored using the Spearman rank test and lines of best fit were derived via linear regression. ## Ethics statement All study procedures were approved by the Research Ethics Board of Karaganda Medical University under Protocol \#45 from 06.04.2020. Written informed consent was obtained from all participants. # Results All 100 participants tested negative for SARS-CoV-2 by RT-qPCR at screening. Of 100 participants, 42 (42.0%; 95% CI \[32.2–52.3\]) were S-IgG+. Due to insufficient sample volume, we were unable to include two samples (one IgG+ and one IgG-) in IgA testing, thus out of the 98 participants tested for IgA, 58 (59.2%; 95% CI \[48.8–69.0\]) were S-IgA+. None of the pre-pandemic samples were positive for S-IgG or -IgA. When stratified by the presence/absence of both IgG and IgA, there were 37 (37.8%) IgG+/IgA+, 4 (4.1%) IgG+/IgA-, 21 (21.4%) IgG-/IgA+ and 36 (36.7%) IgG-/IgA- subjects. Cumulatively there were 63.6% (63/99; 95% CI \[53.4–73.1\]) subjects positive for at least one of the antibodies (including one participant with a missing S-IgA test results); these subjects were defined as the "Prior COVID" group. To further characterize the serologic features of the cohort, we compared the levels of SARS-CoV-2 binding antibodies and the capacity of participant plasma to neutralize SARS-CoV-2 RBD-hACE2 interaction *in vitro*. We found that S-IgG and S-IgA levels correlated strongly across the cohort (r = 0.599, p\<0.001). The SARS-CoV-2 neutralization capacity was significantly higher in the Prior COVID group compared to the No Prior COVID group (p\<0.001). In the Prior COVID group, 17 out of 35 tested participants (48.6%) exhibited neutralization exceeding the assay positivity cut-off. SARS-CoV-2 neutralization also significantly correlated with circulating S-reactive IgG in the Prior COVID group. Lastly, we compared the clinical and demographic features of the cohort stratified by the serology-confirmed COVID-19 exposure. The Prior COVID group consisted of 25% more people self-identifying as ethnic Kazakhs (p = 0.022), and more frequently self-reported having had a respiratory illness since March 2020 (p\<0.001), compared to the No prior COVID group. Only in a minority of cases (24%, 8/33) self-reported respiratory illness was confirmed as COVID-19 by PCR and/or serology at the time of sickness; most of the self-reported respiratory illness occurred in March-Aug 2020, and in two cases in February-March 2021. There were no other significant clinical or demographic differences between the Prior COVID and No prior COVID groups. # Discussion Here we present anti-S IgG and IgA-based seroprevalence findings from a cohort of public university employees in Karaganda, Kazakhstan, who were invited to participate in the study prior to receiving their first dose of COVID-19 vaccine. The cohort seropositivity for anti-S IgG and IgA was 41% and 59%, respectively, while 64% of the subjects tested positive for at least one of the antibodies, thus approaching the 67% seroprevalence threshold thought to be required for establishing herd immunity. Consistent with studies of excess infection and death in Kazakhstan, the serologically assessed SARS-CoV-2 exposure in this cohort was 14-15-fold higher than the reported all-time national and regional COVID-19 prevalence. The substantial discrepancy between serology-derived and officially reported COVID-19 exposure estimates is not uncommon across the globe. However, what is most striking about our findings is the unusually high SARS-CoV-2 seroprevalence exceeding the estimates for many other countries, albeit on par with the recent estimates from the neighbouring St. Petersburg, Russia, where antibodies against the receptor binding domain of the SARS-CoV-2 spike were detectable in \~45% of randomly sampled adults. Similarly high SARS-CoV-2 seroprevalence rates have also been observed in the general populations of Brazil (\>76%), Ecuador (\~45%), India and Pakistan (\>52%), and in communities at risk, such as healthcare workers and nursing home residents, across the globe. Consistently, a recent analysis of serology data from a private laboratory network in Kazakhstan obtained for a period of 12 months (July 2020- July 2021) indicated a test positivity rate of 63% for SARS-CoV-2 IgG. At the same time, a recently published household survey conducted across three cities in Kazakhstan between October 2020 and January 2021 found SARS-CoV-2 IgG/IgM-based seroprevalence rates to range from 39–61%. In our cohort, serologically confirmed exposure to COVID-19 was associated with self-documented history of respiratory illness, most of which was dated by the participants to the peak of the first COVID-19 wave in the Spring-Summer of 2020, period during which the country’s healthcare system was overwhelmed and laboratory testing was limited. This timing of self-reported illness suggests that anti-S immunoglobulins remain detectable up to a year after symptomatic COVID-19, consistent with the established long-term persistence of SARS- CoV-2-reactive antibodies. COVID-19 exposure in this cohort was significantly associated with Kazakh ethnicity, consistent with our earlier finding that Kazakh people are more likely have a laboratory-confirmed COVID-19 diagnosis but are less likely to develop severe disease compared to other ethnic groups in Kazakhstan. Somewhat unexpectedly, we did not see any association between the serologically confirmed exposure to COVID-19 and the participants’ professional occupation or any demographic factors. This may be because differences between sub-populations with different COVID-19 exposure risk are overwhelmed by the high seroprevalence in the general population. Alternatively, the risk of infection for healthcare workers may not be elevated relative to the general population because of the adequacy of infection control practices that are in place in healthcare facilities. Given the logistical difficulties with procuring biomedical reagents and limited technological capacity in the setting of Kazakhstan, our choice of the serologic assay in this study was dictated by both assay quality and logistic feasibility. Therefore, we used a commercially available, FDA-approved assay, validated by several research groups, and deployable in a basic clinical lab setting. Furthermore, we chose to use both IgG and IgA based on the evidence of distinct but overlapping temporal patterns seen for these antibodies in COVID-19 patients. Thus, both IgG and IgA appear as early as 2 weeks post-symptom onset (PSO), with IgA increasing up to third week PSO and then dropping, while IgG increases until fourth week PSO, remaining detectable up to 8 months PSO. Finally, we chose not to use IgM, since this antibody is more suitable for detecting acute infection, while in convalescent subjects IgA temporally overlaps IgM, resulting in higher positivity rates. We validated our findings in the current cohort by testing pre-pandemic samples and performing virus neutralization assays. Half of the Prior COVID participants exhibited virus neutralization correlating with S-reactive IgG titres. This finding is consistent with the evidence that SARS-CoV-2 neutralization declines rapidly after COVID-19 disease resolution and \>40% of convalescent subjects show little neutralization activity. Recent studies indicate that people with prior history of COVID-19 have a stronger response to vaccination compared to COVID-19-naive subjects after one vaccine dose. Mass COVID-19 vaccination was launched in Kazakhstan in February 2021, and so far, \~40% of Kazakhstan’s population has received at least one vaccine dose. Considering limited vaccine supply and low vaccination acceptance, our seroprevalence findings therefore could be extended to inform the ongoing vaccination efforts about the existing population-wide anti-SARS-CoV-2 immunity in Kazakhstan. For example, the second dose of COVID-19 immunisation could be reserved for people without prior natural exposure to SARS-CoV-2 but delayed for subjects with prior COVID-19 exposure. Given the small sample constrained to employees of one, albeit large (employing \~3000 staff) organization, our findings should be seen as preliminary “pilot” data on SARS-CoV-2 prevalence in Kazakhstan. Importantly, our analysis was focused on S-reactive immunoglobulins and with increasing vaccination coverage other SARS-CoV-2 antigens should be incorporated into the future seroprevalence surveys across various demographic groups in the region. # Conclusions Continuous epidemiologic surveillance of SARS-CoV-2 exposure is critical for understanding the COVID-19 transmission dynamics and for informing ongoing vaccination efforts and other COVID-19 mitigation strategies. Seroprevalence studies could help fill in the current gaps in COVID-19 case reporting in Kazakhstan. However, large scale seroprevalence studies in resource-limited settings may not be feasible due to limited clinical and laboratory infrastructure- a barrier that could be overcome by using alternative approaches such as analysing blood bank samples, assessing data collected by the private laboratory sector (which has recently dramatically expanded in Kazakhstan), and implementing PCR-based wastewater surveillance. Our seroprevalence study for the first time documents an extremely high rate of SARS-CoV-2 exposure in Karaganda, Kazakhstan. Although pilot in nature, these findings are consistent with high SARS-CoV-2 seroprevalence in other regions of Kazakhstan, corroborating an overall underascertainment of COVID-19 rates across the country. While narrowing the gaps in country-wide infection surveillance necessitates addressing systemic issues related to governance and healthcare delivery and improving vital registration systems, we hope that our findings can inform the regional pandemic response to facilitate a more effective and equitable distribution of resources, such as vaccines. # Supporting information We thank all the study participants and the COVID-19 vaccination clinic staff. We acknowledge that an earlier draft of this manuscript appeared online as a medRxiv preprint. 10.1371/journal.pone.0272008.r001 Decision Letter 0 Acuti Martellucci Cecilia Academic Editor 2022 Cecilia Acuti Martellucci This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 17 Jun 2022 PONE-D-22-14174High seroprevalence of SARS-CoV-2 antibodies in Karaganda, Kazakhstan before the launch of COVID-19 vaccination.PLOS ONE Dear Dr. Yegorov, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Carefully consider the comments of the reviewers, which provide important points for improvement, and also the following: \- Briefly explain the purpose of the pre-pandemic group in the Methods as well; \- Explain which tests were used for which variables (instead of 'as appropriate'); \- Please kindly provide the reference number of the ethical clearance for the larger study within which this serologic survey was performed; \- Results are given for 99 subjects in the Table and for 98 subjects in the Results, please either correct or explain this discrepancy; \- It is not clear whether 55 (methods) or 35 (results) subjects were selected for the neutralization assay, and by which criterion; \- Please only use percentages in the table, as the numbers can be inferred thanks to the total in the top row (you can write "Results are expressed as percentages unless specified otherwise" in the title of the table; consistently, specify that for continuous variables you are providing the mean (and SD?); finally, the tests used should be explained as a note, not within the title of the table. Please submit your revised manuscript by Jul 31 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Cecilia Acuti Martellucci, M.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match.  When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section. 3\. Thank you for stating the following in the Acknowledgments Section of your manuscript:  "The study was funded by the Ministry of Education and Science of the Republic of Kazakhstan (AP09259123) and, in part, by the Nazarbayev University grant \#280720FD1902 to GH. SY was supported, in part, by a M.G. DeGroote Postdoctoral Fellowship." We note that you have provided funding information that is not currently declared in your Funding Statement. However, funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form.  Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows:  "The study was funded by the The study was funded by the Ministry of Education and Science of the Republic of Kazakhstan (AP09259123, <https://www.gov.kz/memleket/entities/edu?lang=en>) and, in part, by the Nazarbayev University (<https://nu.edu.kz/>) grant \#280720FD1902 to GH. SY was supported, in part, by a M.G. DeGroote Postdoctoral Fellowship.  (AP09259123) and, in part, by the Nazarbayev University grant \#280720FD1902 to GH. SY was supported, in part, by a M.G. DeGroote Postdoctoral Fellowship. The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all the data in the study and the lead authors (IK, SY, DB) had final responsibility for the decision to submit manuscript for publication." Please include your amended statements within your cover letter; we will change the online submission form on your behalf. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The paper is well written and is relevant to current situation. Coming from one of the “stans” myself I realize the level of underreporting for such data. This makes it easier for me to understand the situation and importance of this paper. However, I have several concerns: 1\. The history and situation with under reporting in the region is not fully explained. 2\. The cohort chosen cannot be generalized to the whole population - this is mentioned in the paper, but I think it would be good to provide more details about reasons for choosing this particular cohort (availability, connections, etc) 3\. Last, but not least, I felt that there is something missing in conclusion. Specifically, lack of recommendation for further studies and possible actions. It was established that the prevalence of COVID-19 in the region is much higher than officially reported. What can be recommended to the government, how to improve reporting in general, how this underreporting is connected to uptake of vaccinations? If possible this information can be added to the paper to make it more convenient complete. Reviewer \#2: Dear Authors! On acquainting with the presented paper, I found no significant remarks in the Methodology, Results presentation, and the Discussion section. The topic's relevance cannot be considered overestimated, although the tension of the Covid-19 pandemic has significantly decreased. Given the chance of the possible upcoming pandemics, the global scientific community strives to acquire a deeper understanding of all the details of the present pandemic process. In this relation, information from the countries with a comparatively low level of healthcare, such as Kazakhstan (LMICs out of the EU), appears to be particularly valuable. A study on Sputnik V preceded this work, and linked preprints from the MedRxiv Yale were placed into the References. In the methodology narration, the solution to use the manufacturers' cutoff values as being more suitable was relevant, thus indicating the quality of the work done. The dubious moment was the number of pre-pandemic samples (N 10), though it seems that this circumstance did not interfere with the validity of the results obtained. The only I would suggest is to change your title slightly by inserting the brackets to lessen the number of commas: "High seroprevalence of SARS-CoV-2 antibodies in Karaganda (Kazakhstan) before the launch of COVID-19 vaccination" I found this conclusion the most important: "...serologically assessed SARS- CoV-2 exposure in this cohort was 14-15-fold higher than the reported all-time national and regional COVID-19 prevalence; the unusually high SARS-CoV-2 seroprevalence exceeding the estimates for many other countries." These data allow understanding of the shortcomings and pitfalls in the preventive strategies against the pandemic held in Kazakhstan. For instance, the paper showed the extremely low effectiveness of the screening by currently applied means. Albeit all the screened samples were negative, 59.2% of them turned out to be positive for anti-S IgA. I highly appreciated the integrity in presenting your data. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0272008.r002 Author response to Decision Letter 0 28 Jun 2022 Reviewer \#1: The paper is well written and is relevant to current situation. Coming from one of the “stans” myself I realize the level of underreporting for such data. This makes it easier for me to understand the situation and importance of this paper. However, I have several concerns: Authors' response: We are very appreciative of the reviewer’s positive view of our work and thank the reviewer for the insightful comments and questions. Please kindly find our point-by-point responses to the reviewer’s remarks below. 1\. The history and situation with under reporting in the region is not fully explained. Authors' response: Thank you for raising this point. To address this comment, we have now added additional information to the Introduction to describe the situation with regard to COVID-19 under-reporting in Kazakhstan (p3, lines 15-23 and p4, lines 1-3). 2\. The cohort chosen cannot be generalized to the whole population - this is mentioned in the paper, but I think it would be good to provide more details about reasons for choosing this particular cohort (availability, connections, etc) Authors' response: Thank you for this very valid comment. This cohort was chosen for the serologic studies owing to funding availability and perceived feasibility in the context of the readily available resources and active participant recruitment within the infrastructure of the larger clinical trial. We have now added this information to the Materials and Methods (p4, lines 15-18). 3\. Last, but not least, I felt that there is something missing in conclusion. Specifically, lack of recommendation for further studies and possible actions. It was established that the prevalence of COVID-19 in the region is much higher than officially reported. What can be recommended to the government, how to improve reporting in general, how this underreporting is connected to uptake of vaccinations? If possible this information can be added to the paper to make it more convenient complete. Authors' response: Thank you for this suggestion! We have modififed the Discussion/Conclusions (p 12, lines 9-23) to include our recommendations with regard to improving case reporting and pandemic surveillance using serological screening and alternative approaches. Reviewer \#2: Dear Authors! On acquainting with the presented paper, I found no significant remarks in the Methodology, Results presentation, and the Discussion section. The topic's relevance cannot be considered overestimated, although the tension of the Covid-19 pandemic has significantly decreased. Given the chance of the possible upcoming pandemics, the global scientific community strives to acquire a deeper understanding of all the details of the present pandemic process. In this relation, information from the countries with a comparatively low level of healthcare, such as Kazakhstan (LMICs out of the EU), appears to be particularly valuable. A study on Sputnik V preceded this work, and linked preprints from the MedRxiv Yale were placed into the References. In the methodology narration, the solution to use the manufacturers' cutoff values as being more suitable was relevant, thus indicating the quality of the work done. The dubious moment was the number of pre-pandemic samples (N 10), though it seems that this circumstance did not interfere with the validity of the results obtained. The only I would suggest is to change your title slightly by inserting the brackets to lessen the number of commas: "High seroprevalence of SARS-CoV-2 antibodies in Karaganda (Kazakhstan) before the launch of COVID-19 vaccination" I found this conclusion the most important: "...serologically assessed SARS- CoV-2 exposure in this cohort was 14-15-fold higher than the reported all-time national and regional COVID-19 prevalence; the unusually high SARS-CoV-2 seroprevalence exceeding the estimates for many other countries." These data allow understanding of the shortcomings and pitfalls in the preventive strategies against the pandemic held in Kazakhstan. For instance, the paper showed the extremely low effectiveness of the screening by currently applied means. Albeit all the screened samples were negative, 59.2% of them turned out to be positive for anti-S IgA. I highly appreciated the integrity in presenting your data. Authors' response: We thank the reviewer for a very thoughtful and positive review of our work! We agree with the reviewer that the original title of the manuscript appeared somewhat cumbersome, therefore we have adjusted it to: “High SARS-CoV-2 seroprevalence in Karaganda, Kazakhstan before the launch of COVID-19 vaccination.” We would be happy to replace the comma with brackets and ultimately leave this decision to the reviewers' discretion. Editorial Comments: \- Briefly explain the purpose of the pre-pandemic group in the Methods as well; Authors' response: Done, please see p 5, lines 8-11. \- Explain which tests were used for which variables (instead of 'as appropriate'); Authors' response: Done, please see p 7, lines 13-18. \- Please kindly provide the reference number of the ethical clearance for the larger study within which this serologic survey was performed; Authors' response: Done, we have added this information to the "Ethics Statement" (p 7, lines 19-22). \- Results are given for 99 subjects in the Table and for 98 subjects in the Results, please either correct or explain this discrepancy; Authors' response: We apologize for the the lack of clarity here. Two out of 100 participants were excluded from S-IgA testing due to insufficient sample volume. However, because one of these two participants tested S-IgG+, they were still included in the "Prior COVID-19" category" despite the missing S-IgA result. We have now made edits to both the Table (see Table 1 footnote) and the Results (p.8, lines 8-16) to clarify this aspect. \- It is not clear whether 55 (methods) or 35 (results) subjects were selected for the neutralization assay, and by which criterion; Authors' response: We are sorry for the confusion here. The neutralization assays were performed on 55 samples (of which 35 were classified as Prior COVID based on the S-IgG and S-IgA ELISA), as stated in the Methods. We have now clarified this in the Results (p8, lines 7-19). \- Please only use percentages in the table, as the numbers can be inferred thanks to the total in the top row (you can write "Results are expressed as percentages unless specified otherwise" in the title of the table; consistently, specify that for continuous variables you are providing the mean (and SD?); finally, the tests used should be explained as a note, not within the title of the table. Authors' response: Thank you for these suggestions. We have edited the Table as requested by the Editor and among other edits specified in the Table footnote the following: "Continuous and categorical variables are provided as median/interquartile ranges and percentages, respectively." 10.1371/journal.pone.0272008.r003 Decision Letter 1 Acuti Martellucci Cecilia Academic Editor 2022 Cecilia Acuti Martellucci This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 Jul 2022 High SARS-CoV-2 seroprevalence in Karaganda, Kazakhstan before the launch of COVID-19 vaccination. PONE-D-22-14174R1 Dear Dr. Yegorov, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, Cecilia Acuti Martellucci, M.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: (No Response) Reviewer \#2: That's quite an improvement, and I am satisfied with all corrections made. I also agree with your present title. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0272008.r004 Acceptance letter Acuti Martellucci Cecilia Academic Editor 2022 Cecilia Acuti Martellucci This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 18 Jul 2022 PONE-D-22-14174R1 High SARS-CoV-2 seroprevalence in Karaganda, Kazakhstan before the launch of COVID-19 vaccination. Dear Dr. Yegorov: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Cecilia Acuti Martellucci Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction The vast majority of dairy cattle of the Holstein breed display a black and white spotted coat color while a subset of this breed are red and white spotted. The *melanocortin 1 receptor* (*MC1R*) gene determines the basic coat color in cattle and in Holsteins there are four known *MC1R* alleles: Dominant Black (*MC1R*^*D*), Black/Red (*MC1R*^*BR*), the ancestral wild- type allele (*MC1R*^*+*) and Recessive Red (*MC1R*^*e*). The order of dominance is *MC1R*^*D*\>*MC1R*^*BR*\>*MC1R*^{*+ *}\>*MC1R*^*e*. Holsteins homozygous or heterozygous for any combination of the *MC1R*^*+* or *MC1R*^*e* allele are red, while animals heterozygous or homozygous for *MC1R*^*D* are black. Black/Red animals are born red and change to black typically between two to six months of age. The causal mutations for *MC1R*^*D* and *MC1R*^*e* have been identified as constitutive activation and loss of function of *MC1R*, respectively. *MC1R*^*BR* has recently been shown to be a fourth *MC1R* allele through haplotype and linkage analysis with the causal mutation yet unidentified. In Holsteins, *MC1R*^*BR* and *MC1R*^*+* are both rare, such that red vs. black coat color is primarily determined by the segregation of *MC1R*^*D* and *MC1R*^*e*. Holstein breeders have become accustomed to the concept of red coat color being recessive to black and extensively utilize genetic testing of these *MC1R* mutations to mate carriers of *MC1R*^*e* and/or *MC1R*^*+* together to produce red offspring. In 1980, a female Holstein calf (HOCANF3541221, SURINAM SHEIK ROSABEL-RED) was born in Canada that displayed the typical red Holstein coat color phenotype, but from parents that were not thought to carry either of the *MC1R* alleles associated with red color based on pedigree. Genetic testing verified that this red animal was homozygous for *MC1R*^*D* but yet approximately 50% of her progeny were phenotypically red, suggesting the presence of a new dominant form of red coat color, which was termed “Variant Red”. At the time, this name reflected the unknown origin and mode of inheritance of this new genetic cause of red coat color. Here, we refer to this form of red coat color as “Dominant Red” (DR) to more clearly describe the mode of inheritance and differentiate it from the Recessive Red form of coat color caused by *MC1R*. We propose that the *Dominant Red* locus is designated *DR* with two alleles, the derivative allele *DR*^*DR* and the ancestral or wild-type allele *DR*^*+*. The ability of *DR*^*DR* to override the production of black pigment caused by a constitutively activated MC1R (*MC1R*^*D*), the central mechanism of pigment type switching in mammals, indicates that the DR phenotype represents a valuable opportunity to study novel aspects of pigmentation biology. In a previous study the *beta-defensin 103* (*DEFB103*) gene on *Bos taurus* autosome (BTA) 27 was suggested as being associated with DR. However, this analysis was limited to five candidate genes with known roles in pigmentation biology and found only marginal evidence for genetic linkage (LOD = 3.26). Here, we use whole genome linkage and association mapping techniques coupled with whole genome sequencing to show that in fact *DR* is located on BTA3 and is completely associated with a single missense mutation in the *coatomer protein complex*, *subunit alpha* (*COPA*) gene. # Results ## Hair Pigment Analysis Studies of coat color in laboratory mice indicate that the effect of MC1R signaling on hair color represents a switch between the synthesis of two alternative pigment types, red/yellow pheomelanin (caused by loss-of-function *MC1R* mutations) vs. black/brown eumelanin (caused by mutations that constitutively activate MC1R). Hair samples collected from Dominant Black (*DR*^*+*/*DR*^*+*, *MC1R*^*D*/*-*), Dominant Red (*DR*^*DR*/*DR*^*+*, *MC1R*^*D*/*MC1R*^*D*) and Recessive Red (*DR*^*+*/*DR*^*+*; *MC1R*^*e*/*MC1R*^*e*) animals were analyzed for pigment content and composition. Absorbance of solubilized hair pigment at 500 nm (Soluene-350, A500/mg) is an indicator of total melanin. By this criterion, samples from Dominant Black animals had the highest melanin levels, Dominant Red samples were intermediate, and Recessive Red samples had the lowest levels. We used HPLC analysis of melanin degradation products to measure individual pigment types. Pheomelanin content as assessed by 4-amino-3-hydroxyphenylalanine (4-AHP) or thiazole-2,4,5-tricarboxylic acid (TTCA) was approximately 50- to 60-fold and 12-fold higher respectively in Dominant Red and Recessive Red hair samples as compared to Dominant Black. Eumelanin content, as assessed by pyrrole-2,3,5-tricarboxylic acid (PTCA) content, was highest in Dominant Black hair samples; Dominant Red samples exhibited PTCA levels that were approximately 8-fold lower, and Recessive Red levels were 23-fold lower. Although the difference between Dominant Red and Recessive Red PTCA levels (approximately 3-fold) did not achieve statistical significance (*P* = 0.09), when the ratios of 4-AHP/PTCA or TTCA/PTCA were compared between Dominant Red and Recessive Red, the differences were significant (*P*\<0.001). The Soluene-350 A650/A500 ratio, with higher values representing black/brown color and lower values representing yellow/red color were also significantly different for all three groups with Dominant Black samples having the highest value and Dominant Red being intermediate. These results indicate that similarities in hair color phenotype between Dominant Red and Recessive Red reflect similarities in pigment type synthesis, namely, production of pheomelanin rather than eumelanin, although suppression of eumelanin synthesis by Dominant Red is not as effective as in Recessive Red. ## Genetic Mapping Genetic mapping of *DR* was performed using a heterozygous DR bull (HOCANM9626808, MORSAN RED GOLD) as well as 15 DR and 17 non-DR progeny, including their dams, comprising a single half-sib family. Two point linkage analysis between coat color and 2,752 autosomal markers, which were heterozygous in the DR sire, was performed using CRI-MAP 2.503. The highest genome-wide LOD score of 9.6 was found on BTA3. Eight contiguous SNPs on BTA3 had this same LOD score and showed complete linkage with the DR phenotype in this half-sib family. The nearest flanking markers showing recombination to *DR* define a 10.7 Mb region containing the *DR* causal variant by linkage analysis, from BTA-21472-no-rs @ 5,886,143 bp to ARS-BFGL-NGS-20167 @ 16,557,950 bp. The UMD3.1/bosTau6 genome assembly was used for all analyses. A genome-wide association analysis (GWAS) was performed using 95 DR heterozygous animals, 55 non-DR animals closely related to the 95 DR individuals and 500 non- DR animals randomly selected from the population. The 500 non-DR randomly selected animals had previously been genotyped for genomic selection purposes. The DFAM procedure of the PLINK software package was used to perform a family- based association test for disease traits, a method which accommodates the use of unrelated individuals and case/control traits. A single peak on BTA3 was identified as being most highly associated with the DR phenotype. A single SNP (ARS-BFGL-NGS-94819 @ 9,722,400 bp) showed the strongest genome-wide association of-log10(*P*) = 13.3 with all 95 DR animals having at least one copy of the DR associated allele. However, the DR associated allele of ARS-BFGL-NGS-94819 had a frequency of 6.7% in the 500 non-DR animals, indicating that it would result in a high false positive rate if this SNP were used as a genetic marker for DR. We searched for a haplotype that was identical by descent (IBD) in DR animals within the 10.7 Mb region defined by linkage mapping and confirmed by GWAS results in a larger group of animals. The major allele at each SNP across this region in 95 DR animals was considered to represent the allele present on the putative IBD haplotype. The minimal IBD haplotype was defined as the region where all of the 95 DR animals had at least one copy of the major allele. A region satisfying this criterion was identified from ARS-BFGL-NGS-112616 @ 7,906,099 bp to ARS-BFGL-NGS-1429 @ 10,462,387 bp, spanning 2.6 Mb of BTA3. A single individual did not conform to this rule, possibly due to recombination events that we could not resolve. When possible, more than one animal in discordance with this definition was used to define the boundary of the IBD haplotype. ## Whole Genome Sequencing A heterozygous DR individual (HOCANF9845383, BROEDERDALE MALTBY CAMILLA) was selected for whole genome sequencing in order to identify all candidate causal sequence variants within the 2.6 Mb IBD region. This individual was selected based on maximum homozygosity across the IBD region (top row) in order to minimize the number of non-DR associated variants. This means that we selected an animal whose *DR*^*+* chromosome was closely related to its *DR*^*DR* chromosome. At the time of sequencing no *DR*^*DR**/DR*^*DR* homozygotes were yet known. From the whole genome sequencing data 8,485 SNPs were detected within the IBD region, 77 of which cause an amino acid change and 51 of which were heterozygous. Sequence data from the 1000 Bull Genomes Project was used to exclude variants that have been detected in non-DR animals, which further narrowed the list of candidate causal SNPs to 25. Variant coverage, quality scores, and manual inspection of the aligned reads were used to evaluate which SNPs had the most evidence of being true variants. SNP calling criteria was purposefully set at a low threshold to minimize exclusion of true variants. The six candidate causal SNPs with the highest evidence of being true variants were genotyped in a panel of 20 DR heterozygous animals. Three of the SNPs were excluded as false positives as they were not detected by conventional sequencing methods in the same animal used for whole genome sequencing. This suggests that all other remaining candidate causal SNPs with lower quality scores were also likely to be false positives. Two of the SNPs were excluded due to homozygotes of alternative alleles being found in the panel of 20 DR heterozygotes. The one remaining SNP (C\>T @ BTA3:9,479,761 bp) had the highest quality score out of the 25 candidate causal SNPs detected within the entire IBD region and was found to be heterozygous in all 20 DR heterozygotes in the panel. No insertions or deletions that were also heterozygous were detected in the IBD region. ## Analysis of Structural Variants in the IBD Region To determine if a structural variant is associated with the Dominant Red phenotype read depth and deviant mate-pairs within the IBD region were analyzed for insert size, read order and strand orientation. The sequence from the Dominant Red individual was compared to a Dominant Black Holstein reference individual (HOCANM10705608, BRAEDALE GOLDWYN) that had been sequenced to the same average depth. Normalized read depth in windows was analyzed first within both individuals separately using CNVnator and then between individuals, using CNV-seq. The result of the SV analysis is summarized in. Using CNVnator, two regions (BTA3:8,508,750–8,539,500 and BTA3:8,611,500–8,679,000) were called as deletions in both individuals and likely represent assembly/mapping artifacts. In addition, one region (BTA3:8,731,500–8,752,500) was called as a deletion in the reference individual whereas the Dominant Red individual displayed a normal read depth. However, this region does not have support from the CNV-seq analysis or any deviant reads and is likely a false positive. No regions were found to significantly deviate in read depth among individuals using CNV-seq. Analysis of deviant mate-pairs with SVDetect found one putative duplication in Dominant Red (BTA3:9,007,145–9,008,020). This duplication is also supported in the reference individual and may be an assembly artifact. Taken together, no likely candidate causative structural variants within the IBD region specific to the DR individual were identified. ## Causal Variant Validation and Conservation Due to the lack of structural variants and insertions or deletions associated with DR in the IBD region the C\>T SNP at BTA3:9,479,761 bp was considered to be the most likely DR candidate causal variant. A diagnostic test for this SNP was developed and used to genotype 78 known DR heterozygotes. All 78 animals were found to be heterozygous at this SNP. Of 189 randomly selected non-DR animals and 94 non-DR relatives of DR animals, all were homozygous for the wild-type allele. These results demonstrate the complete association of the DR candidate causal variant with DR phenotype. The candidate causal variant at BTA3:9,479,761 bp is located in the coding region of *COPA*, a gene not previously known to have a direct role in pigmentation biology. In the COPA mRNA transcript NM_001105645 this SNP is located at coding position 478 and results in an arginine to cysteine substitution at amino acid position 160 in NP_001099115 (c.478C\>T p.Arg160Cys). Protein sequence alignment to all COPA orthologs present in NCBI databases reveal that an arginine amino acid is completely conserved at this position from mammals to yeast, residing in a highly conserved WD40 repeat motif. Sequence conservation across such a broad range of species suggests evolutionary constraint due to an important function; indeed, the Arg160Cys substitution in human COPA is predicted to be probably damaging or damaging by PolyPhen-2 and SIFT, respectively. The DR candidate causal variant is therefore likely to have a significant effect on the function of this conserved region. Due to the high evolutionary conservation of the amino acid residue affected by the DR candidate causal variant and the fact that no homozygotes for DR had previously been documented, we considered the possibility that the DR variant may be detrimental in the homozygous state. The DR diagnostic test was used to investigate this scenario by genotyping red individuals that were progeny of two DR parents. Out of 14 animals tested to date four individuals were determined to be homozygous for the DR candidate causal variant. All animals appeared healthy indicating that DR homozygotes are viable. ## RNA-Seq Analysis of Dominant Red Skin To gain additional insight into the mechanisms underlying *Dominant Red* gene action, we compared gene expression profiles between skin from Dominant Red (*DR*^*DR*/*DR*^*+*, *MC1R*^*D*/*MC1R*^*D*) and Dominant Black (*DR*^*+*/*DR*^*+*, *MC1R*^*D*/*-*) animals. Skin biopsies from three individuals of each genotype were used as the source of cDNA libraries that were sequenced on an Illumina HiSeq instrument and then aligned against the bovine reference genome (UMD3.1/bosTau6). Differentially expressed genes were identified with DESeq2; among 15,143 genes that were assayed, 111 and 1104 exhibited significantly different levels of expression at a FDR\<0.05 and a FDR\<0.1, respectively. depicts those results with the level of gene expression plotted as a function of the log2 ratio of Dominant Red/Dominant Black; thus, genes overexpressed in Dominant Black compared to Dominant Red skin are on the left side of the plot, and vice versa, with genes exhibiting a significant difference (FDR\<0.1) highlighted in black or red, respectively. Genes previously implicated in melanocyte biology and/or pigment type-switching are annotated (selected genes that do not exhibit a significant difference at FDR\<0.1 are highlighted in blue). Notably, *TYRP1*, *TYR*, and *OCA2*, which encode proteins previously known to be required for and upregulated during eumelanogenesis, are overexpressed in Dominant Black compared to Dominant Red skin. Two additional genes (*PMEL* and *DCT*), also required for and upregulated during eumelanogenesis, are also overexpressed in Dominant Black compared to Dominant Red skin but at an FDR\>0.1. We also considered the three major paracrine ligand-receptor systems involved in melanocyte biology and pigment type- switching: *ASIP* and *POMC—MC1R*, *KITLG—KIT*, *and EDN3—ENDRB*; these genes were either not detected or not differentially expressed, with the exception of *KITLG*, which is overexpressed in Dominant Red compared to Dominant Black skin. Taken together, these results suggest that the biochemical mechanism of pheomelanin production in Dominant Red skin is identical to that caused by MC1R deficiency, and that the phenotype arises due to melanocyte-autonomous alterations in gene expression. We also carried out a functional annotation of differentially expressed genes using the FunNet tool. The most striking result, apparent from a KEGG analysis is that the four pathways most overrepresented in differentially expressed genes are Endocytosis, MAPK signaling, ER protein processing, and Antigen processing and presentation, all of which are increased in Dominant Red compared to Dominant Black skin. These observations are consistent with impairment of protein trafficking due to the *COPA* missense alteration, and upregulation of compensatory pathways. ## Origin of the Dominant Red Mutation The *DR*^*DR* mutation must have occurred on the IBD haplotype described earlier in the germline of either the sire or dam of the DR founder animal ROSABEL. Whole genome genotyping data was available on some of the ancestors of ROSABEL: A PUGET-SOUND SHEIK (sire), PROVIN MTN IVANHOE JEWEL (paternal grand sire), OSBORNDALE IVANHOE (paternal great grand sire) and SEILING ROCKMAN (maternal great grand sire). Using the IBD haplotype data we were able to exclude the sire of ROSABEL as the germline source of the *DR* variant as this individual did not carry the *DR* IBD haplotype. This indicates that the *DR* mutation occurred in either the germline of ROSABEL’s dam or very early in embryonic development of ROSABEL herself. A DNA sample from ROSABEL’s dam was not available, however the maternal great grand sire had previously been genotyped and was found to carry the *DR* IBD haplotype. This individual (HOCANM275932, SEILING ROCKMAN) had at least one copy of the major allele among the 95 Dominant Red heterozygotes at each locus over an expanded 18 Mb region. This individual would be expected to be wild-type for the DR associated mutation in COPA since he was phenotypically black, although a sample was not available for confirmation. Further genotyping of ROSABEL’s maternal ancestors would be needed to conclusively determine which individual contributed the haplotype on which the *DR* variant originated. However, due to the size of the region concordant between SEILING ROCKMAN and the major allele in 95 DR heterozygotes it is likely that this individual contributed the chromosome on which the *DR*^*DR* mutation later occurred. A bull thought to be unrelated to ROSABEL (HOUSAM2141664, MAPLE-VANE SURPRISE- RED) has been reported to transmit a similar dominant red phenotype. A semen sample of this bull was obtained from the USDA National Animal Germplasm Program and genotyped for the *COPA* mutation using the diagnostic test. This individual was in fact heterozygous for the DR-associated missense mutation. Haplotype analysis found that SURPRISE was concordant for the DR associated allele at 19 out of 21 SNPs in the IBD region for which data was available. The same mutation may have occurred independently; alternatively, it is possible that there is an error in the pedigree of SURPRISE, and that the DR-associated allele in SURPRISE originated in ROSABEL. # Discussion Here we provide genetic evidence that a missense mutation in *COPA* causes Dominant Red, a novel dominant red coat color phenotype that recently evolved in the Holstein dairy cattle breed. The identification of the same region of BTA3 as being associated with DR using two different genetic mapping methods indicates that the previous results suggesting *DR* is located on BTA27 and possibly caused by a mutation in *DEFB103* are incorrect. Haplotype analysis narrowed the region containing the DR causal variant to a 2.6 Mb IBD region of BTA3. The sire of the DR founder animal did not carry this IBD haplotype indicating that the causal mutation originated in the maternal germline of the dam or early during embryonic development of the DR founder, ROSABEL. Whole genome sequencing identified six candidate causal variants in the IBD region, five of which were excluded after screening a panel of 20 known DR heterozygotes. The one remaining SNP at BTA3:9,479,761 bp was concordant with expected *DR* genotype in a panel of 78 DR and 283 non-DR animals. This SNP results in an arginine to cysteine substitution at amino acid 160 of COPA, predicted to impair COPA function, and is consistent with differential gene expression detected using RNA-Seq analysis of Dominant Red compared to Dominant Black skin. Taken together with additional results from hair biochemical and RNA-Seq analyses, our results provide strong evidence that a reduction of COPA activity causes a red coat color by interfering with the effects of melanocortin receptor signaling. *COPA* encodes the α-COP subunit of the coat protein I (COPI) seven subunit complex that is involved with intracellular coated vesicle transport. Originally characterized as a critical component for retrograde transport of luminal and membrane proteins from the Golgi to the endoplasmic reticulum and between Golgi compartments, COPI participates in a range of macromolecular complexes with diverse cargoes, including ribonucleoproteins that modulate trafficking of RNA in neuronal cells. Because the pattern of mRNA expression in Dominant Red skin mimics that observed in Recessive Red skin, we speculate that reduced COPA activity caused by the Arg160Cys substitution impairs MC1R signaling via altered localization of its mRNA and/or protein. Like neurons, melanocytes are highly compartmentalized cells, and it is possible that *MC1R* mRNA is normally transported to specific locations in dendrites, a process disrupted by the DR mutation. Alternatively, or in addition, the DR mutation may interfere with MC1R receptor recycling between plasma membrane and endosomal compartments (reviewed in). Previous studies of COPA function have focused mostly on cell biology and biochemical approaches; indeed, heritable variation of *COPA* has not been described previously in mammals, and Dominant Red cattle represent a unique opportunity to study COPI function in an organismal context. Loss-of-function for *COPA*, *COPB*, or *COPB2* in zebrafish causes embryonic lethality, and it therefore seems likely that the Arg160Cys mutation is hypomorphic. An analogous argument has been made for a missense mutation of *Archain 1* (*Arcn1*) in laboratory mice, which causes neurodegeneration and pigmentary dilution; *Arcn1* encodes the delta-subunit of COPI. Pigmentary dilution in the mouse mutant was suggested to involve defects in trafficking of proteins required for melanogenesis; our results suggest the additional possibility that *Arcn1*-related pigmentary changes represent a reduction of MC1R signaling, an idea that could be tested by examination of *Arcn1*; *Mc1r* double mutants. In cattle, both Recessive Red and Dominant Red animals display variation in the intensity of red coloration over the entire body as well as a tendency to darken in the extremities, possibly due to modifier loci or non-genetic effects. However, some have suggested that Dominant Red animals are more often a darker shade of red than Recessive Red animals, consistent with our observation that eumelanin is reduced more in Recessive Red than in Dominant Red. Visually we observed that Dominant Red animals had a tendency to have darker hairs intermixed with lighter hairs while Recessive Red animals displayed a more uniform intra-hair pigmentation color. However, we observed variation in shade of red within both groups of red animals. We also observed that, at birth, Dominant Red animals were indistinguishable from Recessive Red and that any darkening that might occur in some Dominant Red animals typically takes place after four to six months of age. These observations on color are all based on Dominant Red heterozygotes since very few homozygotes are in existence due to the recent occurrence of this mutation. The work reported here provides the basis for careful evaluation of potential additive and epistatic relationships between Dominant Red and Recessive Red in livestock, an opportunity that is usually limited to model organisms. During the preparation of this manuscript it has come to our attention that another group has independently identified the very same COPA mutation as associated with Dominant Red color in Holsteins, providing further support that this missense mutation is causal (Bourneuf E, Otz E, Michot P, Grohs P, Piton C, Deloche M.-C., Cornier M, Delahaye M, Fritz S, Leclerc H, Longin C, Boukadiri A, Saintilan R, Créchet F, Mosca M, Guillaume F, Bouet S, Baur A, Vasilescu A, Genestout L, Allais-Bonnet A, Rocha D, Colle M.-A., Klopp C, Esquerré D, Barbey S, Fayolle G, Danchin-Burge C, Bed'Hom B, Daetwyler H D, Boichard D, Pin D and Capitan A, in preparation). # Conclusions Here we show that a mutation causing an arginine to cysteine substitution at a highly conserved position in the fourth WD40 repeat motif of COPA causes the striking Dominant Red phenotype in Holstein cattle. The mutation affects a key component of the fundamental cellular process of coated vesicle transport with no known adverse effects on other traits. Hair chemical analyses indicate a shift from eumelanin to pheomelanin production in Dominant Red individuals, similar to what is seen in the Recessive Red phenotype caused by a loss-of- function mutation in *MC1R*. RNA-Seq analyses indicate that Dominant Red skin samples have decreased expression of genes necessary for eumelanin synthesis (*TYRP1*, *TYR*, and *OCA2*). Expression of these genes is upregulated by MC1R signaling, which suggests that altered COPA activity in Dominant Red skin samples impairs eumelanin synthesis by selectively inhibiting MC1R signaling. The COPA mutation associated with Dominant Red provides an opportunity to study the potential relationship between coated vesicle transport and MC1R signaling. # Materials and Methods ## Genetic nomenclature In cattle as well as in many other domestic animals, variation at *MC1R* was originally recognized and described as the “Extension” (*E*) locus. In accord with the HUGO Gene Nomenclature Committee Recommendations, we use *MC1R* as the gene name instead of *E*, with superscripts *D*, *BR*, +, and *e* to refer to alleles. As described here, we provide very strong evidence that *Dominant Red* represents allelic variation in *COPA*; however, since there are no known *COPA* alleles in other animals yet, we refer to the ancestral and derivative *Dominant Red* alleles as *DR*^*DR* and *DR*^*+*, respectively. ## Animals Holstein cattle hair samples were obtained from dairy farmers in Canada and the USA unless stated otherwise. Where relevant and available, individuals are identified by their breed association registration number and name to facilitate lookup of additional information on breed association websites. Animals used for chemical hair pigment and RNA-Seq analyses were selected based on pedigree information so as to maximize within-group genotype similarity. Specifically, none of the animals used came from backgrounds that include *MC1R*^*BR* or *MC1R*^*+* (which are both rare in the Holstein breed), all Dominant Red and Dominant Black animals had no record of Recessive Red in their background, and all Recessive Red animals had no record of Dominant Red in their background. Additionally, all Dominant Red animals had one black and one red parent; thus, the phenotypic designations (and inferred genotypes) are: Dominant Black (*DR*^*+*/*DR*^*+*; *MC1R*^*D*/*MC1R*^*D*), Dominant Red (*DR*^*DR*/*DR*^*+*; *MC1R*^*D*/*MC1R*^*D*), and Recessive Red (*DR*^*+*/*DR*^*+*; *MC1R*^*e*/*MC1R*^*e*). This study utilized tissues and/or DNA obtained from domestic animals maintained by their owners. All samples were collected following best practice husbandry and veterinary procedures. No laboratory animals or laboratory animal research subject to IACUC regulations were used. ## Pigment Analysis Hair was collected from the paralumbar fossa region of females approximately one year of age. Hair was cut immediately adjacent to the skin and wrapped in aluminum foil until analysis. Hair samples (15–20 mg) were homogenized with a Ten-Broeck homogenizer at a concentration of 10 mg/ml and 100 μl aliquots were subjected to Soluene-350 solubilization, alkaline hydrogen peroxide oxidation, and hydroiodic acid hydrolysis. ## Genotyping The sire, progeny and respective dams of the half-sib family used for linkage analysis were genotyped on the BovineLD Genotyping BeadChip (Illumina). Animals used for the GWAS were genotyped on the BovineSNP50 Genotyping BeadChip (Illumina) or genotypes were imputed to this level from a lower density SNP panel. Additional animals were targeted for genotyping through the established genomic selection system with assistance from Canadian Dairy Network, Holstein Association of Canada, and Holstein Association USA. Owners provided hair samples directly to commercial genotyping labs. A subset of samples was also sent by the owners to the researcher’s lab for follow up genotyping. These samples were composed of either hair or blood and were isolated using the Gentra Puregene kit (Qiagen). Animals that could be both Recessive Red and/or Dominant Red based on pedigree analysis were excluded from all analyses unless a negative genetic test for Recessive Red was obtained. ## Sequencing SOLiD mate pair libraries were prepared from a Dominant Red individual (HOCANF9845383, BROEDERDALE MALTBY CAMILLA) according to the SOLiD System Mate- paired Library Preparation protocol of the Applied Biosystems SOLiD System (ABI). The libraries were sequenced on two flowchips using the SOLiD 5500xl platform. The initial libraries had a large number of duplicate reads so a paired-end library was then prepared and sequenced on the same machine. The resulting data was of good quality. The sequence data have been uploaded to the NCBI sequence read archive (SRA: SRS834811). Short reads were aligned in color space against the UMD3.1/bosTau6 genome assembly using BWA v0.5.9. After mapping, the duplicates within each BAM file were marked using Picard v1.54 (<http://broadinstitute.github.io/picard>). Because the first two runs were using the same mate-pair library, the BAM files were merged and duplicates were marked again for removal. Variants (SNPs and indels) were called using SAMtools v0.1.18 mpileup with no quality filtering in order to not exclude any true positives with low quality scores. Genotypes from the 50K SNP chip were used to validate sequencing results. Genotypes were concordant between variant calling results and existing genotype data for all SNPs present in the IBD region that are also on the 50K SNP chip, 12 homozygous non-reference SNPs and 10 heterozygous SNPs. SNPs and indels were annotated with predicted functional consequences using NGS- SNP. Overlapping genes, transcripts, proteins, protein domains, and variants were included among the annotations, along with any known pathways or phenotypes linked to the genes in cattle or to their orthologues in humans. Sequence alignments were constructed between missense SNP-altered proteins their orthologues and scored using the SIFT algorithm to predict the functional significance of protein substitutions. Variants were classified as “known” if the non-reference allele was present in the dbSNP database and “novel” otherwise. The source databases used by NGS-SNP during annotation include Ensembl release 68, dbSNP build 133, Entrez Gene, and UniProt release 2012_09. ## Detection of Structural Variations In order to detect putative causative structural variations within the IBD region the two SOLiD mate-pair libraries for the Dominant Red individual was mapped against the UMD3.1/bosTau6 reference assembly. For comparison, two SOLiD mate-pair libraries from a Holstein bull (HOCANM10705608, BRAEDALE GOLDWYN) were downloaded from the NCBI sequence read archive (SRA: SRR592656, SRR592657) and used as a reference since this individual does not have the Dominant Red phenotype. The reads from both individuals were mapped with MOSAIK v2.2 using a hash size of 14 (-hs 14), allowing four mismatches per read (-mm 4), an alignment candidate threshold of 20 (-act 20) and otherwise default settings. Non-unique alignments were excluded and duplicated read pairs were removed with the Picard v1.123 (<http://broadinstitute.github.io/picard>) package’s MarkDuplicates utility before subsequent analyses. Putative copy number variants (CNVs) were analyzed across BTA3 but only called within the IBD region. First the program CNV-seq was used to compare read depth between the Dominant Red and reference individual in overlapping windows across the chromosome. Based on the default significance threshold of *P*\<0.001 and log₂-fold change threshold of ±0.6, CNV-seq automatically calculated a suitable window size of 2,979 bp. In order to call a significant CNV at least two consecutive windows needed to be significant. Second, the program CNVnator was used to call putative CNVs based on read depth within both individuals separately in sliding windows of 750 bp across chromosome three. CNVnator combines nearby windows that display a similar CNV signal and calculates statistics for the resulting regions. A significance threshold of *P*\<0.001 was used for the CNVnator analysis. In addition to the read depth based analysis, the program SVDetect was used to detect putative SVs based on deviant read mate-pairs. A read mate-pair is called deviant if it has an abnormal insert size, read order orientation or read strand orientation. This method makes it possible to detect copy number balanced variants such as inversions and translocations in addition to duplication and deletion events that affect read depth. SVDetect clusters reads into links that corresponds to a potential SV and classifies it based on how the mate-pairs deviates. Only high quality alignments were used by excluding reads with a mapping quality less than 20. For SVDetect, default settings were used to detect the abnormal mate-pairs. Deviant reads were detected and clustered into links separately for the Dominant Red and reference individual. The putative SVs were selected in the Dominant Red individual from links that were supported by at least five reads and had a final score of one (highest confidence). Any links crossing an assembly gap were removed. These candidates were then compared with the reference individual and overlapping links of the same type were deemed as assembly/alignment artifacts or not associated with the Dominant Red phenotype. ## Diagnostic Test A diagnostic test was developed for the causal Dominant Red SNP (COPA c.478C\>T p.Arg160Cys) at BTA3:9,479,761 bp using a Custom TaqMan SNP Genotyping Assays (ABI). Primers and probes were Forward: 5’-TCAGAAGACCTGGTCGTGTCA-3’, WT allele: 5’-VIC-CAAACGCGCACAGTC-NFQ-3’, DR allele: 5’-FAM-CCAAACGCACACAGTC-NFQ-3’ and Reverse: 5’-TGGGCAGCTCACCAGAAATATC-3’. Genotyping was performed according to the standard protocol using TaqMan Genotyping Master Mix (ABI). ## RNA-Seq Punch biopsies were performed to obtain skin samples from six individuals (three Dominant Red, three Dominant Black), and then stored in RNALater at -80°C. Total RNA was isolated using RNeasy Fibrous Tissue Mini Kit (Qiagen), assessed for integrity with an Agilent Bioanalyzer instrument, and cDNA libraries then prepared with the TruSeq RNA Sample Preparation Kit v2 (Illumina). Individual libraries were multiplexed as six per lane and sequenced as paired-end 50 base pair reads on an Illumina HiSeq 2000 instrument at the Genome Sequencing Laboratory of the HudsonAlpha Institute. RNA-Seq reads obtained for each sample were aligned against the bovine reference genome (UMD3.1/bosTau6) with TopHat2 software, using genomic sequence and transcript annotations obtained from Ensembl (release 72). On average ≈34 million paired-end reads were obtained for each sample, out of which ≈66% were mapped individually and ≈62% were mapped as concordant pairs against the reference transcriptome. Gene counts were computed from read alignments with GenomicRanges and GenomicFeatures packages from Bioconductor (release 2.14), and then used to test for differential expression between Dominant Red and Dominant Black skin samples with DESeq2 package from Bioconductor. DESeq2 relies on negative binomial generalized linear models to determine whether the number of counts for a transcript or gene is significantly different across a range of experimental conditions. # Supporting Information Major contributors of samples were made by Morsan Farms (Canada), Clint Broeders (Canada), Cody Orton (USA), Jason Kearns (USA) and Matt Dorshorst (USA). Sample collection was coordinated by Canadian Dairy Network, Holstein Association of Canada, and Holstein Association USA. GP thanks Jackson Mah for sample analysis at the University of Alberta, and Michelle Miller and Yee Ying Lock at Delta Genomics (Edmonton, AB) for the sequencing of HOCANF9845383 BROEDERDALE MALTBY CAMILLA. The CriMap 2.503 software update was courtesy of Jill Maddox and Ian Evans. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BD GSB LA. Performed the experiments: BD CH SI KW. Analyzed the data: BD CH XL MSA CJR SI KW PS GSB LA. Contributed reagents/materials/analysis tools: BVD GP PS. Wrote the paper: BD GSB LA. Conceived the study: BD LA.

# Introduction Combination antiretroviral therapy (cART) for human immunodeficiency virus (HIV)-1 infection typically comprises a ‘backbone’ – two nucleoside analogue reverse transcriptase inhibitors (NRTI) – and a third drug – either a non- nucleoside reverse transcriptase inhibitor (NNRTI), a protease inhibitor (PI) or an integrase strand-transfer inhibitor (INSTI). The United States Department of Health and Human Services (DHHS) guidelines form a major basis for HIV public policy in resource-rich settings. As of February 2013, an NRTI backbone of tenofovir-emtricitabine/lamivudine, with either efavirenz (NNRTI), raltegravir (INSTI), or ritonavir-boosted atazanavir or darunavir (PI) comprises ‘Preferred’ initial therapy. When to commence cART remains guided by the clinical stage and CD4 lymphocyte count; pre-treatment plasma HIV viral load was removed as an indication for starting cART in 2007. While such recommendations arise from serial evaluation of individual studies by expert bodies, a systematic review of outcomes across multiple studies may reveal characteristics associated with success/failure, and so inform drug development, future study design, treatment guidelines and ultimately patient care. Of the available meta-analyses of initial cART, most focus upon specific comparisons, or earlier studies. None has evaluated outcomes beyond 48 weeks or by the regimen type (‘Preferred’ vs. ‘Alternative’). Much data (some of it unpublished) have been generated since the last broad-ranging analysis of initial cART efficacy, and an updated comprehensive assessment of initial cART efficacy and its associations is warranted. # Methods This systematic review evaluated all prospective studies of initial cART in adults reported through December 31, 2012. The primary outcome measure was antiviral efficacy, defined as undetectable (study-defined) plasma HIV viral load reported on an intention-to-treat (study-defined) basis. Substitution of any initial drug was regarded as treatment failure. Secondary outcomes were efficacy at weeks 48, 96 and 144, change in efficacy post-week 48, and premature cessation of initial cART. Subgroup analyses of efficacy were performed within pre-treatment HIV viral load strata (≥ or \<100,000 copies/mL plasma) and by use of a ‘Preferred’ or ‘Alternative’ regimen as per the February 2013 edition of the DHHS guidelines. Additionally, we aimed to identify characteristics associated with heterogeneity of summarised efficacy, and premature treatment cessation due to participant decision, adverse events or virological failure. ## Study protocol and eligibility criteria Conduct of this study was in accordance with the PRISMA Statement. The protocol/analysis plan are available from the editors or authors upon request. This review aimed to include all studies of initial cART, subject to strict eligibility criteria to ensure data quality. Included studies: were conducted in consenting, treatment-naïve, HIV-1-infected adults; were a prospective cohort or randomised trial; reported efficacy data; and had a minimum of 48 weeks' follow- up. Comparative trials of initial cART were assessed only for the duration that the original randomisation was preserved. We excluded studies of: retrospective or cross-sectional design; cART regimens categorised as not to be offered at any time in key treatment guidelines from 1996 through 2013, ; multiple/variable regimens within a single study arm; and directly-observed therapy. However, treatment groups with fixed, unspecified, dual-NRTI backbones and a common third drug were permitted. Studies of novel, class-sparing regimens were considered for inclusion on an individual basis after review of the efficacy data (relative to standard triple-drug therapy based on a dual-NRTI backbone) by at least two authors. Boosting-dose ritonavir was not regarded an antiretroviral drug (but included in pill counts/dosing requirements). Apart from excluding any study presenting \<48 weeks of efficacy data, eligibility criteria were the same as our previous systematic review of pre-2008 studies. ## Data sources and search strategy The search period was January 1, 2008 to December 31, 2012. Electronic databases searched were: MEDLINE; Cochrane Central Register of Controlled Trials; United States National Institutes of Health clinical trials registry; and the International Standard Randomised Controlled Trial Numbers registry. For each of these, the search strategy was: ‘(“drug”) AND (HIV OR antiretroviral) AND (cohort OR randomised trial)’, where “drug” was the generic or pre-approval code name of an antiretroviral drug. No language restriction was applied. Abstracts from the following key scientific meetings between 2008 and 2012 were also searched: Conference on Retroviruses and Opportunistic Infections; International AIDS Society; Interscience Conference on Antimicrobial Agents and Chemotherapy; and the International Congress on Drug Therapy in HIV. Product labels and medical reviews of antiretroviral drugs published by the United States Food and Drug Administration and the European Medicines Agency between January 1, 1996 and December 31, 2012 were reviewed manually. The above data were combined with all eligible 1995–2008 studies included in our earlier systematic review. All were manually reviewed in duplicate by an author (FJL) for eligibility before being combined with results of the latest search. Discrepancies were discussed with a second author (AC). Study synopses accessible on the following pharmaceutical company websites up to December 31, 2012 were reviewed: Abbott; Boehringer-Ingelheim; Bristol-Myers Squibb; Gilead Sciences; GlaxoSmithKline; and Roche. Other antiretroviral manufacturers did not have such websites. All manufacturers were approached for relevant data missing from the above sources; data were provided by Bristol-Myers Squibb (August 15, 2012), Gilead Sciences (August 10, 2012), MSD (July 31, 2012), and ViiV Healthcare (November 29, 2012). The PRISMA flow diagram is depicted in. ## Data extraction Study arms were regarded as individual groups. For every group, characteristics collected were: year of commencement; participant numbers; latest reporting format; study characteristics; eligibility criteria; participant/disease characteristics; treatment characteristics; efficacy; and rates of premature treatment cessation. Efficacy data were collected cumulatively through study weeks 48, 96 and 144. Dosing requirements were assumed to be in accordance with the product label at the time of study initiation or subsequent approval, unless otherwise stated. Efficacy data reported using the ‘time to loss of virological response’ (TLOVR) algorithm were collected preferentially. The newer ‘Snapshot’ algorithm was not considered, due to the small numbers and reported similarity of results to TLOVR. All plasma viral load assays were considered equivalent – when more than one threshold was reported (e.g. both \<400 and \<50 copies/mL plasma), the lowest was used. Data were abstracted manually in triplicate by one author (FJL) using a standardised data collection form. Ethnicity was recorded as ‘white’, ‘black’ or ‘other’, defined as previously. Cause of premature treatment cessation was categorised as one of: adverse event, participant decision (withdrawal, loss to follow-up), virological failure, and other. No imputation was made for missing data (marked as unavailable). Completed forms and the resulting electronic database were audited by a second author (AC) to ensure data quality. ## Summary measures and data synthesis The study arm was the unit of analysis, with clustering by trial accounted for. Principal summary measures were the proportion of participants per group and the mean (for continuous variables, including proportions). For summary measures across study arms, outcomes were expressed as a mean percentage, weighted for group size. Standard error of the mean was used to adjust for group size differences and heterogeneity was quantified using the *I*² statistic. Adjusted absolute difference in outcomes within categorical variables was expressed as a percentage coefficient. Differences between means were compared using Student's *t*-test. A multivariable, random-effects, linear regression approach was used to explore sources of heterogeneity in measures of efficacy and treatment cessation. Variables pre-selected for testing included all study, participant, disease and treatment characteristics as well as adverse events. Year of study commencement was excluded from the primary predictor analysis, because of its likely relationship to other variables (drug potency/tolerability, pill/dose counts, availability of genotypic testing, emphasis on maximum pill adherence). Fixed NRTI backbones and third drug classes were compared, not individual drugs. This was done to retain analytical power and sensitivity across subgroup analyses. Furthermore, comparative prospective trials are powered for non-inferiority, and no clear superiority has hitherto been demonstrated for individual third drugs within a given class for initial cART – excepting boosted vs. unboosted PI. Unboosted PIs, (including unboosted atazanavir) were considered here as a separate class. Backwards, stepwise selection was used: only those statistically significant (*p*≤0.05) in univariable analysis were further assessed in multivariable models. Additionally, where covariate data were missing for \>20% of individuals within a group, that group was excluded for assessment of that covariate. For stratified antiviral efficacy (pre-treatment viral load; regimen type) tests for interaction were conducted first. Only variables with statistically significant interactions with the stratified outcome were further interrogated using the aforementioned linear regression approach. The following pre-planned sensitivity analyses were performed for the predictor linear regression models: (1) including only recent data by excluding studies commenced prior to 2005; (2) excluding cohorts and abstracts to correct for potential differences in study/report quality; and (3) including year of commencement as a variable. Four post-hoc analyses were performed: (1) descriptive efficacy/premature cessation data for groups using tenofovir- emtricitabine and abacavir-lamivudine, both overall and stratified by pre- treatment plasma viral load ≥ and \<100,000 copies/mL; (2) a variant of the primary linear regression model for overall efficacy conducted after exclusion of all abacavir-lamivudine groups without HLA-B\*5701 screening; (3) descriptive efficacy/premature cessation data and a linear regression model for overall efficacy when considering raltegravir separately from other INSTI drugs; and (4) descriptive efficacy/premature cessation data and a linear regression model for overall efficacy when considering unboosted atazanavir separately from other unboosted PIs (**Tables S1-S6 in : Supplementary post-hoc stratified analyses**). All analyses were performed using STATA, Version 11 (StataCorp LP, College Station, Texas, USA). The *meta* suite of commands were used for combining data across trial arms. # Results ## Study selection The search (2008-2012) yielded 2,272 studies (2,008 publications, 306 abstracts), with 42 duplicates; 45 studies met the eligibility criteria. Following review and addition of pre-2008 studies (removal of duplicates and one pre-2008 cohort with only 24 weeks of follow-up), 114 studies (103 publications, 11 abstracts) were included. ## Study and participant characteristics Of 114 included studies, 97 (85%) were randomised trials and 17 (15%) prospective cohorts, encompassing 216 treatment groups with 40,124 participants (median 112 participants/group; interquartile range 63 to 200). This represents 73 new groups (32 randomised trials, 3 cohorts, 17,057 participants) since our earlier review. Participant and treatment characteristics are shown in ; these were similar in terms of NRTI backbone and third drug class, demographics and disease stage for each analysis population (data not shown). ## Overall efficacy: all studies Mean overall efficacy was 60% (SD 16) after a mean follow-up of 82 weeks (SD 38) with greater efficacy in more recent studies (**,**). Collected data were highly heterogeneous (*I*² = 96%). Study phase, intention-to-treat analysis method, genotype/CD4 eligibility restrictions, NRTI backbone, third drug class, daily pill and dose counts were identified as the primary sources of efficacy heterogeneity on univariable analysis. In the multivariable analysis, higher efficacy was associated with: NRTI backbone (favouring tenofovir-emtricitabine, *p*\<0.001); third drug class (favouring INSTI, *p*\<0.001); and studies using intention-to-treat algorithm (favouring ‘missing equals failure’, *p*\<0.001). This model accounted for 56% of the variance (*r²*) in reported efficacy. Tenofovir-emtricitabine was associated with higher efficacy than abacavir- lamivudine (coefficient 7.6%; 95% confidence interval \[CI\] 2.6 to 12.7; *p* = 0.003). Only two groups, both 96 weeks in duration, used pre-treatment HLA-B\*5701 screening before administering abacavir; overall weighted efficacy of these abacavir groups was 55% (SD 13), vs. 63% (SD 6.4) for the 24 groups without HLA-B\*5701 screening. Amongst third drug classes, INSTI use was associated with higher efficacy compared to either an NNRTI (coefficient 11.9%; 95%CI 4.6 to 19.2; *p* = 0.002) or a boosted PI (coefficient 12.7%; 95%CI 4.6 to 19.2; *p* = 0.001). There was no difference in efficacy of NNRTI compared to boosted PI regimens. ## Change in efficacy by follow-up duration Efficacy at 96 and 144 weeks was reported for 85 (39%, 19,959 participants) and 25 (12%, 8,330 participants) groups, respectively. For these groups, mean efficacy at 48 weeks was 66% (SD 16), declining to 60% (SD 16) at week 96, then 52% (SD 18) at week 144; the mean fall in efficacy was 8.3% (SD 3.9) between weeks 48 and 96, with a further 6.3% (SD 5.3) decline between weeks 96 and 144. Type of NRTI backbone (favouring tenofovir-emtricitabine, *p*\<0.001), third drug class (favouring INSTI, *p*\<0.001) and efficacy analysis method (favouring ‘missing equals failure’, *p* = 0.001), eligibility by pre-treatment resistance genotype (favouring restriction, *p* = 0.007) and CD4 lymphocyte count (favouring no restriction, *p* = 0.001) were each associated on multivariable analysis with greater efficacy at week 48. At week 96, greater efficacy was associated with: phase 2 studies (vs. phase 3 or 4, *p*\<0.001) and no eligibility restriction for CD4 count (*p*\<0.001). The *r²* values were 66% and 39% for weeks 48 and 96, respectively. Due to insufficient data, a stable multivariable model could not be generated for efficacy through week 144. For those studies reporting efficacy data through to week 96, a multivariable analysis for the decline in efficacy between weeks 48 and 96 was performed. Lesser decline was associated with phase 2 studies (*p* = 0.002), placebo use (*p*\<0.001), and NRTI backbone (*p* = 0.002), but there was no significant difference between use of tenofovir-emtricitabine and abacavir-lamivudine. ## Efficacy stratified by regimen type: February 2013 DHHS guidelines A total of 63 groups (29%, 14,233 participants) were allocated either a 2013 DHHS ‘Preferred’ regimen (27 groups, 13%, 5,677 participants) or an ‘Alternative’ regimen (39 groups, 18%, 9,305 participants). Amongst the ‘Preferred’ regimens, 14 groups (2,729 participants) initiated efavirenz: efficacy 72% (SD 7.6) over 108 weeks (SD 40); nine groups (1,776 participants) initiated atazanavir/ritonavir: efficacy 75% (SD 7.3) over 87 weeks (SD 42); one group (343 participants) initiated darunavir/ritonavir: efficacy 79% over 96 weeks; and three groups (829 participants) initiated raltegravir: efficacy 82% (SD 7.9) over 99 weeks (SD 59). ‘Preferred’ regimen efficacy was 75% (SD 7.9) over a mean follow up of 99 weeks (SD 41), compared to 66% (SD 6.6) over 82 weeks (SD 35) with ‘Alternative’ regimens (difference 10%; 95%CI 7.6 to 15.4; *p*\<0.001). This difference persisted when stratified by high (difference 10%; 95%CI 3.0 to 17.0; *p* = 0.006) and low (difference 16%; 95%CI 7.9 to 23.7; *p*\<0.001) pre- treatment viral load strata. The initial regimen type selected was found to significantly interact with the relationship of third drug class (*p* = 0.040), dosing requirements (*p* = 0.024), previous AIDS events (*p* = 0.003), and pre-treatment CD4 lymphocyte count (*p* = 0.049) on efficacy. On regression analysis, the pre- treatment CD4 count was associated with efficacy within ‘Preferred’ regimens (coefficient \<0.1%; 95%CI 0.0 to 0.1; *p* = 0.035); that is, efficacy was \<0.1% greater for every one cell increase in the absolute CD4 count. Amongst ‘Alternative’ regimens, previous AIDS events were associated with lesser efficacy (coefficient −0.4%; 95%CI −0.6 to −0.1; *p* = 0.011), INSTI use with greater efficacy (vs. NNRTI; coefficient 15.3%; 95%CI 8.7 to 21.9; *p*\<0.001), and a requirement for fasting when taking medications was associated with lesser efficacy (vs. no dosing requirement; coefficient −10.6%; 95%CI −17.7 to −3.4; *p* = 0.005). ## Efficacy stratified by pre-treatment viral load For high (≥100,000 copies/mL) and low (\<100,000 copies/mL) viral load strata, 98 groups (45%, 22,878 participants, mean 81 weeks' follow-up \[SD 36\]) had efficacy data available. Mean efficacy was 62% (SD 15) and 70% (SD 15), respectively (difference 8.4%; 95%CI 6.0 to 10.9; *p*\<0.001). Stratification by pre-treatment viral load showed no significant interaction with the relationship of any pre-selected variable upon efficacy (*p*\>0.05), and univariable-multivariable analysis was not pursued. Despite this lack of interaction, ‘Preferred’ regimens (February 2013 DHHS guidelines) maintained greater efficacy at both higher and lower viral load strata; efficacy within these strata were 73% (SD 12) and 82% (SD 12), respectively (difference 9.1%; 95%CI 4.3 to 14.0; *p* = 0.001). ## Premature treatment cessation: all studies An average of 25% (SD 11) of participants prematurely ceased initial cART. The most common reason was participant decision (11%, SD 6.6) followed by adverse events (8.1%, SD 5.9); virological failure was less common (3.5%, SD 4.0). Total premature cessation was lower with more recent studies. Most cessations (20%, SD 9.0) occurring through week 48, 9.2% (SD 11.0) ceasing between weeks 48 and 96, and a further 4.9% (SD 2.7) ceasing between weeks 96 and 144. On multivariable analysis adjusting for study, participant, disease and treatment characteristics and adverse events, industry-only sponsorship was associated with lower cessation rates due to participant decision, compared to industry-supported academic (coefficient −2.6%; 95%CI −4.8 to −0.5; *p* = 0.015) and academic-only sponsorships (coefficient −4.0%; 95%CI −6.3 to −1.6; *p* = 0.001). Similar associations existed between sponsorship and cessation due to adverse events. Phase 2 studies were associated with lower rates of cessation due to adverse events than phase 3 or 4 studies (coefficients −2.6% and −4.0%, respectively; *p*\<0.015). Neither NRTI backbone nor third drug class were associated with cessation attributed either to adverse events or participant decision. ## Studies commenced 2005 and later The pre-planned sensitivity analysis of the most recent studies (commenced 2005 and later) included 82 groups. Of these, 76 (93%) were randomised trials and 6 (7%) were prospective cohorts, encompassing 16,795 participants (median 118 participants/group; interquartile range 47 to 170). Participant characteristics were similar to that of the primary analysis (mean age 37 years \[SD 1.9\]; mean CD4 count 238 cells/µL \[SD 80\]; mean HIV-1 plasma viral load log₁₀ 4.9 \[0.3\]). Mean overall efficacy was higher than for all included studies: 71% (SD9.6) over a mean follow-up of 86 weeks (SD 37). Efficacy at 96 and 144 weeks was reported for 37 (45%, 9,961 participants) and 12 (15%, 3,486 participants) groups, respectively. Mean efficacy at 48 weeks was 77% (SD 8.6), 70% (SD 11) at 96 weeks, and 66% (SD 6.3) at week 144. Total mean premature treatment cessation was 21% (SD 8.9). Like the primary analysis, participant decision (8.8%, SD 5.4) was the most common reason, followed by adverse events (6.5%, SD 4.0) and virological failure (4.3%, SD 4.3). Multivariable analysis of studies commenced 2005 or later resulted in similar findings to the primary analysis. Tenofovir-emtricitabine continued to be associated with higher efficacy than abacavir-lamivudine (coefficient 8.8%; 95%CI 3.9 to 13.8; *p* = 0.001), as was INSTI over NNRTI (coefficient 5.5%; 95%CI 0.1 to 11.0; *p* = 0.050). ## Other sensitivity analyses The separate analyses conducted after excluding cohorts and abstracts, and then including the year of commencement as a variable both gave similar results to the primary analysis (data not shown). # Discussion For studies initiated from 1995 through to 2010, the overall mean efficacy of initial cART was low (60% over 82 weeks), although it has risen substantially for more recent studies, suggesting it could rise further with studies commenced post-2010 (not available for this analysis before locking of the final database). Efficacy was higher (75% over 99 weeks) with current ‘Preferred’ cART regimens. There is ongoing loss of efficacy through the second and third years of initial cART, mainly because of participant decision or adverse events. Across all analyses, tenofovir-emtricitabine was more effective than abacavir- lamivudine, and INSTI more effective than NNRTI. As the sensitivity analyses demonstrated, this also applied to the more recent studies of initial cART. Participants with pre-treatment viral load \<100,000 copies/mL plasma had significantly better antiviral outcomes on initial cART, a finding persisting within ‘Preferred’ regimens, and similar in magnitude to that favouring February 2013 DHHS ‘Preferred’ over ‘Alternative’ regimens. Initial cART should aim to induce and maintain long-term virological control. For a disease that presently requires life-long treatment, this analysis demonstrates suboptimal efficacy and durability even with currently ‘Preferred’ initial regimens, corroborating similar findings of the Antiretroviral Therapy Cohort Collaboration. In practice, individuals failing one regimen are likely to be switched to a different, effective combination. Therefore, the present data reflect success or failure of initial cART only, not the absolute failure of all treatment, which we have not examined. Nevertheless, our data show for the first time that almost 15% of participants failed cART between weeks 48 and 144 of therapy. If this trend were ongoing, the majority of initial regimens would fail within 10 years of commencement. As the pace of new antiretroviral development slows, treatment options may become limited as patients are progressively exposed to switching of cART regimens, even in resource-rich settings. Participant decision was the most common identifiable cause of failure, ahead of adverse events or virological failure. This was the case both for studies of ‘Preferred’ regimens, and more recent studies (post-2005). The underlying reasons for this decision, or whether participants re-started cART outside a study, were not routinely reported. Uncovering and systematically documenting reasons for premature cART cessation and subsequent outcomes will be necessary if participants are to be retained more successfully on treatment. Industry- sponsored studies were associated with lower rates of premature cessation due to participant decision or adverse events. It is possible that the cost-free medication and more frequent clinical follow-up commonly mandated within such studies may have attenuated cessation rates. Treatment cessation due to virological failure was consistently low. This was the case even in the earliest cART studies, possibly because the limited treatment options available at the outset of cART prevented drug switches even in the event of study-defined virological failure. Most treatment failure/cessation occurred early, within the first 48 weeks, with incremental declines of \<10% per year at weeks 96 and 144. It may seem self- evident that the longer an individual remains on a single regimen the lower the likelihood of failure, but this may explain why at week 96, only study design characteristics were associated with efficacy, rather than treatment or disease characteristics. There were, however, considerably fewer study groups with data for 96 weeks of follow-up. Unsurprisingly, of the intention-to-treat strategies used to calculate efficacy, the ‘missing equals failure’ algorithm was associated with greater treatment success, as it has the fewest imputations regarding failure. This raises the possibility that overall efficacy may have been even lower if other, more stringent algorithms had been used exclusively. However, the ‘missing equals failure’ approach arguably more closely reflects actual clinical practice. The four key international guidelines use the CD4 lymphocyte count as a key marker for when to start cART, with none recommending pre-treatment viral load as an indication. Viral load was removed from DHHS recommendations in 2007, because of data indicating that risk of AIDS or death in individuals receiving cART with pre-treatment CD4 counts ≥350 cells/µL was \<2% regardless of viral load. Our findings that: (1) the high-vs.-low viral load and ‘Preferred’-vs.-‘Alternative’ regimen efficacy differences were similar (8.4% vs. 10%, respectively); and (2) ‘Preferred’ regimens had greater efficacy at pre-treatment viral loads \<100,000 copies/mL, are therefore striking, since it suggests that the pre-treatment viral load exerts an independent effect upon efficacy, even as guidelines place a primary emphasis on regimen selection. An implication of this finding is that the presence or absence of pre-treatment viral loads ≥100,000 copies/mL may influence the efficacy ultimately reported in studies. In particular, it may help explain the lesser efficacy of abacavir- lamivudine compared to tenofovir-emtricitabine – a finding noted in the ACTG A5202 study. In the existing pre-planned multivariable analyses, neither median viral load, nor the proportion of participants with a pre-treatment viral load ≥ or \<100,000 copies/mL was significantly associated with efficacy in the overall multivariable analysis. Within the pre-planned subgroup analyses, there were also no significant interactions between: (1) the regimen type and pre-treatment viral load strata; or (2) the viral load-stratified efficacy and treatment characteristics. However, a post-hoc analysis of our data comparing tenofovir- emtricitabine with abacavir-lamivudine mirrors the gaps in efficacy seen between ‘Preferred’ and ‘Alternative’ regimens (**Table S1 in : Supplementary post-hoc stratified analyses**). At viral loads ≥100,000 copies/mL, overall efficacy with tenofovir-emtricitabine was higher than abacavir-lamivudine (71% \[SD 10\] vs. 59% \[SD 12\], respectively; difference 11%; 95%CI 3.6 to 17.7; *p* = 0.004), with a similar difference at viral loads \<100,000 copies/mL (79% \[SD 12\] vs. 67% \[SD 14\], respectively; difference 14%; 95%CI 6.5 to 20.8; *p*\<0.001). The mean follow-up periods were 83 (SD 37) and 74 (SD 38) weeks, respectively. The descriptive findings, both pre-planned and post-hoc, suggest that the overall superior efficacy of tenofovir-emtricitabine (and ‘Preferred’ regimens) is independent of the plasma viral load and further implies that for a given drug or combination, efficacy at lower viral loads is better than at higher viral loads. Not only does this validate current ‘Preferred’ regimens, it argues for future guidelines recommending cART initiation when the plasma viral load rises towards 100,000 copies/mL. Comparable results have been reported recently, albeit limited to 48 weeks’ follow-up. The ACTG A5202 study showed a higher risk of study-defined *virological* failure with abacavir-lamivudine for viral loads ≥100,000 copies/mL at interim analysis (resulting in the unblinding of that stratum), rather than all-cause, intention-to-treat failure. As we did not have premature cessation data stratified by pre-treatment viral load, the results are not directly comparable. While viral load and regimen type did not significantly interact to influence efficacy, precluding multivariable analyses of these subgroups, it is worth noting that the high-low threshold of 100,000 copies/mL (log₁₀5.0) reported in studies is arbitrary. A meta-analysis of efficacy data from individual participants may reveal a clinically relevant association between gradations of viral load and long-term efficacy. The superiority of tenofovir-emtricitabine over abacavir-lamivudine, although statistically significant on primary analysis, remains confounded by one major issue – that of abacavir-related hypersensitivity. The association between HLA-B\*5701 and abacavir-related hypersensitivity, first reported in 2002, is well-described. It would have been advantageous to have more efficacy data with pre-treatment HLA-B\*5701 screening. However, due to limited availability, testing for HLA-B\*5701 did not become the standard of care until its inclusion in DHHS guidelines from 2007. By that point, 24 of the 26 groups using abacavir in our review had already commenced, leaving only two studies (total 227 participants) which utilised HLA-B\*5701 screening (efficacy 55% \[SD 13\] over 96 weeks). Frequency of abacavir-related hypersensitivity is estimated at between 2% and 9%, with some ethnic variation. A meta-analysis of 5,332 patients exposed to abacavir reported a mean incidence of 4% (range 3% to 6%). Hypersensitivity does contribute to the lesser efficacy of abacavir-lamivudine vs. tenofovir- emtricitabine in our primary analysis, but within the limitations of the source data its relative contribution to higher treatment failure cannot be quantified. Given the insoluble nature of the missing data, one approach to addressing this problem is to infer that the adjusted efficacy difference of 10% between tenofovir-emtricitabine and abacavir-lamivudine is partially due to hypersensitivity (assuming a mean incidence of 4%), with other factors responsible for the balance. For instance, amongst post-2005 studies, the approval of single-tablet, fixed-dose preparations is a possible reason for patients electing to switch away from abacavir mid-clinical trial. During the period of analysis (up to 2010), all such preparations contained tenofovir- emtricitabine as the NRTI backbone. The pairing of tenofovir with emtricitabine, rather than lamivudine, may also contribute to the efficacy difference. Although emtricitabine and lamivudine have closely related chemical structures, and are regarded as interchangeable by the DHHS and the World Health Organization, emtricitabine has a longer intracellular half-life than lamivudine. The alternative approach is a separate, post-hoc, multivariable analysis excluding the 24 abacavir-lamivudine groups (5,289 participants) that did not use HLA-B\*5701 screening. Such an analysis was performed (**Table S2 in : Supplementary post-hoc stratified analyses**); in this model, the comparative difference in efficacy (vs. tenofovir-emtricitabine) was non-significant (coefficient −14.4%; 95%CI −30.2 to 2.5; *p* = 0.175), although the large negative coefficient favoured tenofovir-emtricitabine. In one of the two included abacavir groups (from study CNA109586), hypersensitivity accounted for more premature cessation in the abacavir arm than the comparator tenofovir- emtricitabine arm (6% vs. \<1%, respectively), despite HLA-B\*5701 screening. With only the two HLA-B\*5701-screened groups included, this post-hoc analysis is under-powered to compare abacavir-lamivudine with other NRTI backbones. However, as all other groups were left unchanged from the original primary analysis, the other predictive associations (third drug class, intention-to- treat analysis method) were preserved. The disadvantage of removing the majority of abacavir-lamivudine groups is that it also removes a large amount of data for the concomitant drug classes from our study. If further studies were then to be excluded on the basis either of currently non-recommended regimens or a non-standard practice, it would also necessitate the exclusion of older cART regimens (e.g. zidovudine-lamivudine), and similarly, the many earlier studies commenced before pre-treatment resistance genotyping became the recommended standard of care in 2006. With the removal of such a large number of groups from the analysis, the benefits of ecological analysis may be lost, and with it, several comparisons between NRTI backbones and third drug classes. This systematic review only evaluated entire third drug classes. While this may limit the specificity of the comparisons, assessing individual drugs will result in a loss of analytical power. Our approach is vindicated by long-term prospective data. The adjusted efficacy difference between INSTI and NNRTI classes is consistent with the 10% lower efficacy of efavirenz vs. raltegravir (both ‘Preferred’ agents), as well as with recently-published five-year outcomes of the 004 and STARTMRK studies, and the short-term results of the SPRING-1 study comparing efavirenz with dolutegravir. Examining pooled data can effectively portend long-term prospective results, displaying the value of systematic review. The favourable efficacy of INSTI- over NNRTI-based initial cART has been noted elsewhere. Our analysis supports this finding, but extends it to include a treatment advantage over boosted PI therapy, and updates earlier meta-analyses conducted prior to INSTI availability. In order to investigate whether the efficacy results for INSTI-based regimens were inflated by the presence of short-term efficacy data for elvitegravir/cobicistat and dolutegravir, a stratified post-hoc analysis was performed separating raltegravir (5 groups, 1,246 participants) from the other INSTI drugs (4 groups, 904 participants). No studies from the primary analysis were excluded. Efficacy overall and at weeks 49 and 96 were similar for raltegravir and non-raltegravir INSTIs (**Table S3 in : Supplementary post-hoc stratified analyses**), although follow-up was shorter for the latter (83 weeks \[SD 51\] vs. 56 weeks \[SD 21\]). When multivariable analysis was performed using raltegravir and non-raltegravir INSTIs as separate categorical variables, the results were similar to those of the primary analysis (**Table S4 in : Supplementary post-hoc stratified analyses**). Both raltegravir (coefficient 10.9%; 95%CI 1.2 to 20.6; *p* = 0.028) and non-raltegravir INSTIs (coefficient 13.0%; 95%CI 2.6 to 23.4; *p* = 0.014) were superior third drug options compared to the reference (NNRTIs). These results imply similar efficacy within the INSTI class, a position since adopted by the DHHS in October 2013, when all three INSTIs became listed as ‘Preferred’ options. In the primary analysis, unboosted PIs (regarded as a separate third drug class) were inferior to NNRTIs, while boosted PIs and NNRTIs were similar. The DHHS guidelines do not sanction the use of any unboosted PI, with the exception of unboosted atazanavir, which was used by five groups (932 participants) in our analysis. Unboosted atazanavir was reported as non-inferior when compared to efavirenz and atazanavir/ritonavir in the AI424-034 and ARIES studies, respectively,. However, intention-to-treat efficacy in AI424-034 was low (32% at 48 weeks), while the ARIES study used 36 weeks of atazanavir/ritonavir as an induction therapy prior to switching to unboosted atazanavir. Thus whilst atazanavir/ritonavir is a DHHS ‘Preferred’ third drug, unboosted atazanavir is listed as a ‘less satisfactory’ treatment option, one that is neither ‘Preferred’ nor ‘Alternative’. We conducted a post-hoc analysis considering unboosted atazanavir as a separate third drug class. The results of this analysis were similar to the primary analysis (**Table S5 in : Supplementary post-hoc stratified analyses**). Unboosted atazanavir remained significantly associated with lesser efficacy, as did the other unboosted PIs. The stratified descriptive data supports this; compared to the overall efficacy of unboosted PIs in the primary analysis (42% \[SD 11\]), the efficacy of unboosted atazanavir and other unboosted PIs were similar −40% (SD 12) and 42% (SD 11), respectively (**Table S6 in : Supplementary post-hoc stratified analyses**). In contrast, the efficacy of atazanavir/ritonavir-based regimens was 72% (SD 8). This supports the findings of the ACTG A5175 study, where the unboosted atazanavir arm was stopped early because of inferior efficacy. The pharmacokinetics of unboosted atazanavir is another reason for caution, as its use effectively precludes combination with tenofovir-containing NRTI backbones, as the latter appears to reduce the bioavailability of atazanavir significantly. In the groups examined in this systematic review, unboosted atazanavir was paired with NRTI backbones selected from older, less effective agents (didanosine, stavudine, zidovudine), which may further attenuate its usefulness in clinical practice. Despite the advent of fixed-dose combinations, neither daily pill count nor doses significantly predicted efficacy, even amongst ‘Preferred’ or ‘Alternative’ regimens. However, such fixed-dose combinations are a relatively recent development (only three groups had a daily pill count of one), so there were insufficient data to address this question adequately. From our results, however, it appears that while a one-pill-per-day regimen is convenient, it is not necessarily superior. This remains an inference, as importantly, 44% of studies (weighted) used placebo controls, inflating the daily pill count. We did not analyse ‘Preferred’ and ‘Alternative’ cART regimens by guidelines other than of the DHHS (e.g. World Health Organisation, International Antiviral Society – USA, European AIDS Clinical Society). International ART guidelines have all been derived from the same pool of publicly reported data. Such duplicated analyses are likely to show results similar to the ones we obtained for DHHS-specified cART regimens, i.e. that use of a ‘Preferred’ cART regimen (however defined) resulted in higher efficacy than with ‘Alternative’ regimens. Similarly, despite our updated search being restricted to between 2008 and 2012, an additional sensitivity analysis for post-2008 studies was not performed. This would have effectively duplicated the ‘Preferred’-vs.-‘Alternative’ regimen analysis, while arbitrarily excluding studies of currently ‘Preferred’ (and therefore relevant) regimens that commenced pre-2008. There are several limitations to our study. Chief of these is that the base unit of analysis in our methodology was the treatment group. Hence ours was an ecological analysis, using mean aggregate data, rather than individual participant data. Also, other, more commonly-reported meta-analyses of cART aim to combine similarly-designed, randomised comparisons of efficacy/failure to achieve data homogeneity. Such analyses therefore have a different hypothesis, i.e. they evaluate the comparison, whereas a study such as ours describes overall efficacy and failure in each group exposed to the treatment drug(s). By including a wide variety of treatments and settings, our data set is more heterogeneous; a meta-regression approach permits exploration of potential sources of heterogeneity in the estimation of efficacy. This would not be possible with a ‘typical’ meta-analysis of similar studies. However, because broad associations are identified, it is not possible to infer causality, and it would be incorrect to deduce the impact of a particular cART regimen on efficacy in participants of a certain age, gender, race, location or clinical status. However, as standardised outcome measures (virological efficacy and failure) have been applied to a large weighting of participants, it is best suited to representing the likely outcomes for populations with a similar profile, reporting associations that may advise public policy, like the DHHS and other international antiretroviral guidelines, and identifying topics for future study. As most study groups were of predominantly white race, this may limit applicability to currently resource-limited settings. A multivariable approach was used to interrogate heterogeneity within the efficacy data. With this approach, bias due to missing data may allow some clinically relevant sources of heterogeneity to be missed whilst other, non- relevant sources might reach statistical significance. The large number of groups examined reduces, but does not eliminate, the risk of such bias. The inability of our study to assess the effect of HLA-B\*5701 screening on the relative efficacy of abacavir-lamivudine is a key example of this, and one which necessitated post-hoc analysis in an attempt at characterisation. The failure of some covariates to reach significance may also highlight which data are currently poorly reported in studies. We were able to source some non-published data from pharmaceutical sponsors, but very little from academic sponsors. For each subgroup analysis, population numbers changed, so comparisons between subgroups can only be inferred. For most studies, randomisation was not stratified by pre-treatment viral load. It is likely that variables not reported by viral loads strata (previous AIDS events, CD4 lymphocyte count, adherence, co-infection with viral hepatitis) can partly explain the differences seen between high and low viral load strata. This limitation does not affect other subgroup analyses. Finally, HIV viral load was used as the primary outcome measure, rather than a clinical endpoint. Our data identifies pre-treatment HIV-1 viral load as a determinant of efficacy, which should prompt a re-examination of its place in treatment guidelines, and also suggests that HIV-infected adults should initiate cART for plasma viral loads rising towards 100,000 copies/mL. Whether this difference is driven by lesser antiviral potency at high viral loads is unknown because of missing data, but requires further investigation. One possible option would be a prospective study comparing triple- and quadruple-drug combinations as initial therapy in high viral loads. The medium-term efficacy of initial cART alone remains unsatisfactory, and prospective studies require longer follow-up and better reporting of adverse events and reasons for participant-initiated treatment cessation. As an ecological analysis of pooled outcomes, our findings are associations rather than causes, best applied to populations with demographics similar to our source data. Future analyses performed using data from individual participants are needed to allow specific causality to be inferred for the trends identified in this study. # Supporting Information The authors acknowledge the contributions of Bristol Myers-Squibb, Gilead Sciences, MSD, ViiV Healthcare, Rebekah Puls and Kathy Petoumenos in providing unpublished pre-treatment, eligibility and outcome data. [^1]: The authors have read the journals' policy and have the following conflicts to declare: Frederick J. Lee currently receives research funding from the National Health and Medical Research Council of Australia (Grant Number 1017991), and has received a Clinical Immunology Society Fellowship, travel sponsorships from the Australasian Society of Clinical Immunology and Allergy and the Australasian Society for HIV Medicine, conference sponsorship from MSD and Gilead Sciences, and consultancy fees from MSD and AstraZeneca. Andrew Carr has received research funding from Baxter, Gilead Sciences, MSD, Pfizer and ViiV Healthcare, consultancy fees from Gilead Sciences, MSD, and ViiV Healthcare, lecture and travel sponsorships from Bristol-Myers Squibb, Gilead Sciences, Janssen, MSD, and ViiV Healthcare, and has served on advisory boards for Gilead Sciences, MSD, and ViiV Healthcare. Janaki Ami declares no conflict of interest. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: FJL JA AC. Performed the experiments: FJL JA AC. Analyzed the data: FJL JA AC. Contributed reagents/materials/analysis tools: FJL JA AC. Wrote the paper: FJL JA AC.

# Introduction Soil is considered to be a major reservoir of *Bukholderia pseudomallei*, the causative agent of melioidosis in humans and animals. Factors responsible for the occurrence of *B*. *pseudomallei* in endemic areas are poorly understood, although studies have suggested several properties related to soil that may influence the distribution of the organism. Physicochemical properties such as temperature, pH, soil water contents and sunlight have been shown to influence the survival of *B*. *pseudomallei* in soil under laboratory conditions. In addition, the intensity of rainfall, season, changes in landscape, soil type, farm management and human activities were suggested to influence the occurrence of melioidosis in endemic areas. Defining the pattern of *B*. *pseudomallei* distribution may help in assessing the risk of melioidosis infection. In Malaysia, the prevalence of the organism in soil is unknown however the overall seroprevalence of melioidosis in livestock was reported to be 5.7% with the reactor rates in sheep and goats found to be 13.6% and 2.6% respectively during a 10-year study period. Sampling of soil from the endemic region of Australia found a prevalence of 14%. Contact with contaminated environmental reservoirs appeared to be an important risk factor for infection among animals. Soil movement, heavy rainfall resulting in water run-off and/or aerosolization of *B*. *pseudomallei* during high velocity wind were suggested to be risk factors for infection. The presence of some basic cations, macro and microelements, organic matter and the cation exchange capacity of soil were found to be potential properties of soil that could influence microbial growth in general. However, information on the effects of these substances on the presence of *B*. *pseudomallei* in soils of the endemic areas is limited. This study investigates the physicochemical properties of soil that may influence the presence of *B*. *pseudomallei* in small ruminant farms in Malaysia. # Materials and Methods ## Study area Malaysia is a country located in the Southeast Asian region and is comprised of two main parts, the West Malaysia (Peninsular Malaysia) and East Malaysia (Sabah and Sarawak on the Borneo Island). The Peninsular Malaysia is made up of 11 states and two Federal territories (Wilayah Persekutuan and Putrajaya) while the East is comprised of Sabah and Sarawak states and the Federal territory of Labuan all located on the Borneo Island. The two parts of Malaysia are separated by the South China Sea. Peninsular Malaysia covers an area of 131,598 square kilometers sharing common borders with Thailand in the north and Singapore in the south. The 11 states in Peninsular Malaysia include Johor, Kedah, Kelantan, Melaka, Negeri Sembilan, Pahang, Perak, Perlis, Pulau Pinang, Selangor and Terengganu. The country lies entirely in the equatorial zone and is situated in the northern latitude between 1° and 6° N and the eastern longitude from 100° to 103°E. The climate of Peninsular Malaysia is influenced by two monsoon seasons, the northeast monsoon, which starts in November and ends in March and the southwest monsoon, which starts in May and ends in September. The northeast monsoon is usually characterized by prolonged heavy rainfall in northern and eastern regions of the Peninsular Malaysia causing severe floods in low-lying areas especially the east coast states of Kelantan, Terengganu, Pahang and east Johor. On the other hand, the southwest monsoon is characterized by drier conditions with less amount of rainfall throughout the Peninsular. According to the Malaysian Meteorological Department, Peninsular Malaysia has an average rainfall of 2,400 mm with hot and humid weather throughout the year. ## Study design and study population Goat and sheep farms were selected from four states in Peninsular Malaysia and visited for sample collection. The states from which farms were selected included Negeri Sembilan, Pahang, Perak and Selangor states. Letters requesting for the farmers to participate in the study were sent to the individual farmers before commencement of the study. Only those that indicated their willingness and agreed to participate were visited. At each of these farms, the farmers were interviewed using a structured closed-ended questionnaire. The farmers gave written consents for the use of the questionnaire results. In fact, The questionnaire was designed to capture information on farm demography, farm size, small ruminants population size, breeds, farm management, animal health status, source of water and presence of water bodies around farm, soil conditions such as soil pH, applications of fertilizers, herbicides or lime and soil-related activities as well as occurrence of adverse climatic events such as drought, erosion, landslide, water logging etc. around farm. ## Sample collection and bacterial isolation A total of 60 small ruminants farms were selected from a database of farms utilized for disease surveillance and monitoring by the Department of Veterinary Services (DVS) and the Veterinary Research Institute (VRI) Ipoh, Malaysia. A detailed description of the database was published by Musa et al.. The farms visited were categorized based on management practice as intensive (21), semi- intensive (30) and extensive (9) farms. For the intensively managed farms, samples were taken within the farm premises especially near the sources of the animals’ feed and water. In farms practicing semi-intensive and extensive systems, soil samples were taken within the farm premise near sources of animals’ feed and water and from areas where the animals graze. The soil sampling procedure was performed as described by Chantratita *et al*.. About 500ߞ700g of soil was taken at depth of 30cm from the surface. Triplicate soil samples were collected from three different spots. One of the sampling spots in each farm was located near the animals’ source of food one from their source of water and the third at least 100m away from the first two and roughly formed the third point of a triangle marked on the ground. A total of 180 soil samples (three from each farm) were collected for this work. Samples were sealed, labeled and transported to laboratory for processing. The sampling utensils were cleaned and disinfected between each use. The soil samples were processed according to the standard method procedure by Limmathurotsakul et al with slight modification. Briefly, 100g of soil sample was mixed with 100ml of distilled water, shaken for 1minute and allowed to sediment overnight. Then 100μl of supernatant was plated on Ashdown‘s agar, incubated at 37°C and observed for growth. Additionally, an enrichment process where 1ml of the supernatant was added to 9ml of threonine basal salt solution supplemented with 50mg/l polymixin B as enrichment broth was carried out. Both broths and agar plates were incubated at 37°C and observed for 48 hours. Suspected isolates were screened using catalase and oxidase tests, colony appearance and growth on Ashdown’s agar and presumptive isolates were stored in glycerol/ brain heart infusion (BHI) medium at -20°C until confirmation using PCR as described by Brook et al.. ## Identification of *Burkholderia pseudomallei* Suspected colonies of culturable *B*. *pseudomallei* were initially screened on the bases of colony appearance, growth on Ashdown’s agar, catalase and oxidase tests. The characteristic colony morphological descriptions as purple, flat, dry and wrinkled by Chantrantita et al. were utilized as guide in the screening. Colonies of suspected to be *B*. *pseudomallei* were further screened using API 20NE kits according to manufacturer’s instructions. Presumptive isolates that were positive using the API 20NE were confirmed using polymerase chain reaction (PCR) amplification. The PCR protocol was as described by Brook et al., using PPM3 and PPM4 primers selected from the 16S rRNA region of *B*. *pseudomallei*. The sequences were from position 452 to 472 (PPM3-forward primer) `5' AATCATTCTGGCT AATACCCG 3'` and position 1023 to 1042 (PPM4-reverse primer) `5' CGGTTCTCTTTCGAGCTCG 3'`. The DNA templates were prepared from the suspected isolates using boiling method. The positive control was a DNA of previously confirmed *B*. *pseudomallei* isolated from kidney of rabbit, which died of melioidosis. The negative control used was sterile distilled water. The amplification process was carried out using a thermalcycler (MyCyler^®, Bio Rad, US). The PCR amplification consisted of 30 cycles of 1 minute at 94°C, 30 seconds at 54°C and 2 minutes at 72°C, with a final extension step of 10 minutes at 72°C. Products were visualized by electrophoresis on a 1.0% agarose gel stained with 0.1% ethidium bromide. ## Determination of physicochemical properties of samples Before analyses was performed to determine the soil physicochemical properties, soil samples were divided into two groups (based on the PCR finding); samples positive for *B*. *pseudomallei* (n = 32) and samples negative for *B*. *pseudomallei* (n = 28). Samples from each farm were pooled by mixing 200g from two of the triplicate samples obtained from the farm. Determination of water contents was carried out according the procedure described by Forster. Briefly, this involved measuring the weights of moist soil samples after which the samples were dried and re-weighed to obtain the percentage weight loss. Each sample from the two groups was separately air-dried, ground and sieved. Soil textures were analyzed using mechanical method as described by Miller and Miller. Organic matter contents were determined using the loss on ignition (LOI) method as described by Salehi et al., where samples were dried in the oven at 105°C overnight then cooled in a desiccator. Then 200mg of each of the samples were weighed and combusted at 550°C for 2hours in a muffle furnace (Model 1400 Furnace, Humboldt, Germany). After combustion, the samples were cooled in a desiccator and weighed again. An estimation of soil organic matter percentage was calculated as the difference between the oven-dried weights and the combusted weights. The pH values were measured at room temperature using a pH meter after samples were dissolved in equal amounts of distilled water. The trace elements that included iron, copper, manganese and zinc as well as basic cations that included sodium, magnesium, calcium and potassium were determined using atomic absorption spectrophotometer (PerkinElmer, Inc., Shelton, CT, USA) according to the manufacturer’s instructions and published guides for soil samples. The cation exchange capacities (CEC), measured in CentMol/l (CMol-¹l) of samples were determined using auto-analyzer (QuickChem^® 8000 Series, Flow Injection Analysis System, Lachat Instruments, Loveland, USA) in accordance with the manufacturer’s guide. Carbon, nitrogen and sulfur contents were analyzed using CNS elemental analyzer (TruMac CNS Analyzer, Leco, US) according to the manufacturer’s instruction. ## Data Analysis Data were entered into Microsoft Excel and analyzed using JMP^® for Mac (version 9.0.1 SAS Institute inc., Cary, NC, USA). Mean values of physicochemical properties were compared between *B*. *pseudomallei* positive and negative samples using independent T-tests with p-values \<0.05 considered to be significant. A Chi-square test for association was performed to examine relationship between presence of *B*. *pseudomallei* in soil and farm management type and the type small ruminant kept in the farm. A multivariable logistic regression analysis was carried out using backward stepwise method in which soil parameter with univariable level of significance p\<0.25 were selected for inclusion in the base model, and variables were excluded if the p-value was \>0.05 and did not meaningfully alter the point estimates of the remaining variables. Exploratory correlation analysis shows significant collinearity between the predictor variables namely CEC, organic matter, carbon and nitrogen contents and so were excluded from the analysis one at a time. The outcome variable used in the analysis was the outcome of PCR confirmation of *B*. *pseudomallei* (positive/negative) isolation from soil sample. Chi-square test for independence was used to assess association between the independent variables. The overall goodness-of-fit of the model to the data was examined using the Hosmer-Lemeshow test. # Results ## Bacterial isolation shows the results of *B*. *pseudomallei* isolation from soil samples collected from small ruminant farms from the four states in this study. Screening of the suspected *B*. *pseudomallei* isolates using the API 20NE kits found samples from 33 farms to be positive while those from 27 farms were negative. However, PCR amplifications of these samples confirmed 32 were positive while 28 negative. One sample each from Pahang and Selangor, which initially gave negative results using the API 20NE kit yielded positive results using PCR amplification. In the same vein, a sample from Negeri Sembilan and two samples from Perak that were positive using the API 20NE kits was negative using PCR amplification. The results from the confirmatory PCR were used for further analysis of the data in this study. We found no significant association between the proportions of isolation of *B*. *pseudomallei* from soil samples to the type of management systems (intensive, semi-intensive and extensive) and the type of animals kept in the farm (goats only, sheep only and mixed) *(χ*^*2* = 2.41, df = 2, p = 0.29 and *χ*^*2* = 0.10, df = 2, p = 0.94, respectively). ## Univariable analysis of physicochemical parameters of soil samples The descriptive statistics and mean percentages of the physicochemical parameters of *B*. *pseudomallei* positive and negative soil samples from the study farms are presented in. In terms of soil texture, there were significant differences (p\<0.05) between the means of clay and silt contents of *B*. *pseudomallei* positive compared to those of the negative samples. Similarly, comparisons of the means of organic matter, water, clay, iron, carbon and nitrogen contents as well as the cation exchange capacities between *B*. *pseudomallei* -negative and positive soil samples using T-test showed significant differences (p\<0.05) between the two groups. ## Multivariable Logistic Regression The multivariable logistic regression of the soil properties influencing the presence of *B*. *pseudomallei* in soil samples is presented in. The Hosmer- Lemeshow goodness of fit test showed that the model significantly fitted the data (χ^*2* = 3.67, df = 8, p = 0.88). The final model revealed that only three of the factors tested remained significant (p\<0.05) as predictors of the presence of *B*. *pseudomallei* in the soil samples when the effects of other variables were accounted for. We found that when compared with *B*. *pseudomallei*-negative soil sample, a positive soil sample was significantly more likely to have higher: 1. iron content, whereby the odds of isolating *B*. *pseudomallei* from soil sample will increase by a factor multiplied by 1.01 for every unit increase the soil iron content; 2. water content, whereby the odds of isolating *B*. *pseudomallei* from soil sample will increase by a factor multiplied by 1.28 for every percentage increase in soil water content, 3. clay content whereby odds of isolating *B*. *pseudomallei* from soil sample will increase by a factor multiplied by 1.54 for every unit increase in clay content of the soil. Other soil physicochemical properties included in the final model were soil pH and its copper contents that were not significant but were considered biologically meaningful and important in contributing to the explanation of the variation seen in the study. # Discussion Soil texture, the relative proportion of sand, silt and clay contents of the soil, have been suggested to be essential for understanding the physical, physicochemical and biological properties of soil. In this study, we screened several physicochemical properties using the univariable analysis and identified seven that significantly influenced the presence of *B*. *pseudomallei*. However, when these variables were considered together in a multivariable logistic regression, only five remained in the final model with three associations found to be significant while the other two were approaching significance. We believe that the power of the study may not be adequate to detect small differences in association of the latter two variables because our sample size was not large. It is also possible that the method used to detect the presence of the bacteria in this study (ie culture and identification) prior to PCR is less sensitive than direct PCR amplication from soil samples.. Our finding on the increased likelihood of isolating *B*. *pseudomallei* when soil has high contents of clay agrees with report from previous researches in the endemic regions of Australia. Clay soil tends to support survival of *B*. *pseudomallei* because clay has excellent water and nutrient retention capabilities due to its large surface area and chemical activity. Our finding also agrees with other studies reporting isolation of *B*. *pseudomallei* from rice fields that are predominantly clayey in texture and have the tendency for waterlogging. In general, clay is known to support bacteria due to its small pore size, enabling this structure to withhold nutrients and water. It has also been reported that electrostatic interactions of clay particles with bacterial extracellular polysaccharides popularly known as “clay hutches” aid the survival of *B*. *pseudomallei*. Clay contents of soils in general have been reported to influence other soil parameters such as the “protected biomass” availability of substrates for activity of microorganisms present in soil, soil porosity and ecology and microenvironment of soil microorganisms. The soil sand content was not significant in our final model. Reports have been inconsistent on the findings about this variable as some study have found that soil with high sand contents is less likely to support the persistence of *B*. *pseudomallei* due to low water and nutrient contents while others have recorded that, *B*. *pseudomallei* has been isolated from soil samples with high percentage sand contents and poor nutrient contents. We believe that complex interaction exists between parameters in the soil beyond the soil texture alone and more work need to be done to arrive at conclusion about the organism’s niche preferences for perpetuation. Of the four trace elements that included iron (Fe), copper (Cu), zinc (Zn) and manganese (Mn) investigated in the study, only Fe contents was found to be significant, consistent with the recent finding published by Wang-ngarm et al.. Iron in soil exists in ferrous (Fe²⁺) and ferric (Fe³⁺) forms where the ferrous being forms available for plant and microbial activity. The availability of the iron in these forms is in turn affected by other soil properties such as pH, aeration and organic matter contents and plants adaptation. The iron oxides and hydroxides in soil are responsible for the yellowish and reddish color in soil. These colors reflect the high iron contents which has been found to be associated with the occurrence of *B*. *pseudomallei*. Our finding is supported by Draper et al. who found strong association between occurrence of *B*. *pseudomallei* in water with higher iron contents. In addition, laboratory study where iron (in form FeSO₄) was added to the growth medium of *B*. *pseudomallei* showed an increased biofilm formation which may lead to improved resilience of *B*. *pseudomallei* against environmental inhibitory factors. In the same vein, increased levels of intracellular iron has been shown to enhance the invasion efficiency, replication and intracellular survival of this organism. Furthermore, iron (Ferric ion) regulate the expression of respiratory enzymes involved in *B*. *pseudomallei*’s biological processes therefore enhancing their survival. The acquisition of these iron was suggested to be regulated by siderophore biosynthesis in *B*. *pseudomallei*. Successful competition for iron has been reported to be an essential aspect of *B*. *pseudomallei* ability to establish infection in a host. Iron is also utilized by the agent to form superoxide, FeSOD (iron superoxide dismutase) which protects it against the “host” immune system thereby enhancing its survivability in host’s system. Our findings on iron contradicts those of Baker et al. who reported that the iron contents of soil samples positive for *B*. *pseudomallei* were significantly lower when compared with samples that were negative. This inconsistency may be due to the way data were analysed. In our study, data were analyzed using a multivariable logistic regression, which consider putative variables that may influence the presence of *B*. *pseudomallei* together in a model, while the data from Baker et al. were analyzed using univariable analysis, which consider the variables individually thus may not have accounted for the effects of other variables in the study. Alternatively, the disparity may also be due to the presence of other unexplained interacting factors that might not have been adequately captured in our study. Complex interactions amongst several factors have been suggested to be the major determinants of *B*. *pseudomallei* and melioidosis in the endemic regions. Copper exists in soil in silicate minerals or carbonates forms which are largely unavailable for plant and microbial activities. The form available is cation (Cu²⁺) found usually on surfaces of clay minerals or in association with organic matter where the availability is in turn influenced by soil pH and organic matter contents.We found a significant difference between copper contents of samples that were negative for *B*. *pseudomallei* to those positive samples in the univariable analysis, consistent with that found by Baker et al.. However the variable did not remain significant when other variables were accounted for. The information on the exact role played by soil copper contents on survivability of *B*. *pseudomallei* in the environment could not be obtained from literature to fully understand the effects of copper on the agent. However, copper has been found to slow down biofilm formation in other Gram-negative bacteria such as *Legionella pneumophila* in water due to its toxic or inhibitory effect on bacteria. Our finding that water contents in samples affects the likelihood of the organism’s survival in those soil samples is expected as water is essential in the maintenance of ecological balance in soil microorganisms, as availability of nutrient and the integrity of bacterial membranes are dependent on the soil solution which in turn depends of soil water. Water in soil is maintained in network of pores where gravitational water circulates in macropores and capillary water in the micropores. Consequently, the water in micropore offers favorable environment for bacteria by protecting them from desiccation and from exogenous toxic water-soluble substances. To highlight the importance of soil water contents on the survival of *B*. *pseudomallei*, a laboratory investigations by Tong et al and by Chen et al have both shown that water in combination with pH and temperature were the three major ecological factors affecting survival of *B*. *pseudomallei* in soil. Pumpuang et al., showed that the organism can survive in distilled water for more than 16 years which underscore the importance of water in survival and maintenance of this organism even in the absence of other nutrients. The mean percentage water contents of the positive soil samples in our study was about 11% and this is observed to be well within the range of positive *B*. *psseudomallei* sites in Thailand with 9–18% moisture contents. However, *B*. *pseudomallei* may survive even in the environment with lower moisture than those aforementioned such as the desert. Soil pH was included in the logistic regression model as one of the predictors for the occurrence of *B*. *pseudomallei* in soil even though the association was not significant. Laboratory investigations by Tong et al and another by Chen et al have both shown that pH, together with water and temperature were the three major ecological factors affecting survival of *B*. *pseudomallei* in soil under laboratory conditions. Under field conditions, lower pH has been found in other endemic regions such Thailand and Australia to promote the presence of *B*. *pseudomallei*. The bacteria has been observed to prefer more acidic medium for its growth and survival and according to Dejsirilert et al., acidic medium probably enhanced the production of acid phosphatase in the bacteria which in turn aid its survival in the environment. However, earlier study in soils from Malaysia isolated *B*. *pseudomallei* from soils of varying pH values between 2.8 and 7.4 indicating the organism’s preference for more acidic to neutral medium. Our previous findings on the decreasing risk of melioidosis in farms that have treated their soil with lime further support this finding. In conclusion, this study found that iron, water and clay contents of soil samples were the three most important factors influencing the occurrence of *B*. *pseudomallei* in farm environment in Peninsular Malaysia. Higher iron, clay and water contents of soil samples appear to support the presence of *B*. *pseudomallei* in the soil. The copper contents of soil also appear to meaningfully decrease the likelihood of the pathogen survival. This study revealed information on the physicochemical properties of soil that may influence the occurrence of *B*. *pseudomallei* in soil in Malaysia. However, the sample size available for this study is limited therefore may not have uncovered other factors that could have played significant role in the survival of *B*. *pseudomallei* in the farm environment. # Supporting Information We thank all state and district veterinary officers for their contribution and technical assistance throughout the study. We acknowledge the assistance of Haji Jamil Omar of Soil Research Laboratory, Department of Land Management, Faculty of Agriculture, UPM for the analysis of the soil parameters. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** HIM LH ZS ZZ SAA. **Formal analysis:** HIM ZS LH ZZ SAA. **Investigation:** HIM LH. **Methodology:** HIM ZS LH CP ZZ SAA. **Project administration:** LH. **Resources:** HIM LH ZS. **Software:** HIM LH. **Supervision:** LH ZS ZZ CP SAA. **Writing – original draft:** HIM LH ZS ZZ SAA. **Writing – review & editing:** HIM LH. [^3]: ‡ These authors also contributed equally to this work.

# Introduction According to UNAIDS report 2019, there are approximately 1.7 million children less than 15 years old living with HIV worldwide and 160000 children became newly infected. Maternal-to-child transmission (MTCT) accounts for more than 90% of pediatric HIV infection. In the era of antiretroviral therapy (ART), the disease progression has been controlled and the life expectancy of HIV-infected children has greatly increased. However, with an expanded life span, HIV- infected children have high prevalence of metabolic abnormalities such as dyslipidemia, lipodystrophy and insulin resistance, which potentially increases the risk of early cardiovascular atherosclerosis, kidney failure, and diabetes in this population. Several factors including HIV, host immune response and ART contribute to these metabolic complications. Previous study demonstrated that prolonged exposure to protease inhibitors (PIs) in HIV-infected children was associated with lipid abnormalities and lipodystrophy. The use of ART and persistent inflammation in children living with HIV were shown to be associated with increased carotid intima media thickness, which is a marker of cardiovascular disease (CVD). In HIV-infected children, increased use of ART also reported to cause dyslipidemia and insulin resistance. Thus, metabolic abnormalities in perinatally HIV-infected children are of great concern since ART is given early in life. However, even before the introduction of ART alterations in lipid metabolism and sensitivity to insulin were reported during HIV infection. A recent study showed increased cardiovascular abnormalities in HIV-infected children without ART(treatment- naïve). Since existing clinical indicators of HIV are not infallible, observing the development of HIV infection in children and monitoring their responses to ART using NMR metabolic profiling will be germane in the identification of novel biological markers. Several studies had focused on the impact of ART on metabolism in perinatally HIV-infected children, metabolic complications due to HIV in treatment-naïve HIV-infected children is less understood. It is thus imperative to monitor metabolic state of children with or without ART. Metabolomics is an unbiased method to identity and quantify metabolites present in the biological fluids and tissues, providing the metabolic fingerprint of an organism. Metabolomics helps to determine metabolic alterations during various diseases like coronary heart disease, colon cancer and hepatitis and has been used to discover novel clinical biomarkers and therapeutic targets. The NMR spectroscopy was first used by Hewer et al. to profile metabolites in the serum of chronically HIV-infected adults. They demonstrated a distinction in the sera among treatment-naïve, ART-suppressed (ART-experienced and virologically suppressed) HIV-infected subjects and HIV negative controls. Since then, other studies have used NMR and MS to demystify metabolic response to HIV in adults. Moreover, it is also necessary to comprehensively study HIV-infected children plasma metabolic profile as they have remarkable difference in HIV- disease progression with higher viremia compared to HIV-infected adults and also have metabolic abnormalities. In this study, we performed untargeted metabolic profiling of treatment-naïve and ART-suppressed perinatally HIV-infected children and compared them to age-matched uninfected controls by using 1H NMR spectroscopy. # Material and methods ## Study participants This pilot cross-sectional study included 30 perinatally HIV-infected children and 12 age-matched uninfected controls who had been recruited in June 2016 from the south part of India. All the children were less than 15 years old. Perinatal infection was confirmed by the documentation of HIV infection in children within 1 year of life along with documentation of maternal HIV infection. Among HIV- infected children, 15 children were on ART for more than 6 months, whereas the other 15 children were not on ART (treatment-naïve). Those on ART had undetectable plasma viral load (VL \< 50 copies/ml) and 60% of them were receiving a combination of Zidovudine (ZDV), Lamivudine (3TC) plus Nevirapine (NVP), 20% were receiving Atazanavir (ATV), 3TC plus Efavirenz (EFV), 13% were receiving Stavudine (d4T), 3TC plus NVP and the remaining 7% were receiving ZDV, 3TC plus Lopinavir/Ritonavir (LPV/r). Uninfected controls were matched with the HIV-infected children for age and sex. Children with age greater than 15, with co-infections such as tuberculosis or viral hepatitis and metabolic abnormalities were excluded from this study. All study participants belonged to same ethnicity, geographical area and had similar dietary habits. This study was approved by Institutional Review Board (IRB) of St. Johns Medical College & Hospital at Bangalore, Jawaharlal Nehru University (JNU) at New Delhi and King George’s Medical University (KGMU) Lucknow, and was carried out in accordance with the approved guidelines. The biosafety approval was obtained from Institutional Biosafety Committee (IBSC) of JNU for handling plasma samples. Informed written consent was obtained from parents/guardian of all the children who participated in our study. The characteristics of subjects are listed in. ### Plasma viral load and CD4 count Plasma HIV-1 viral load (VL) was measured using Abbott Real Time HIV-1 assay with a lower limit of detection of 50 copies of RNA/ml (Abbott Molecular Inc., Des Plaines, IL, USA). CD4 T cell count was measured using FC500™ flow cytometer (Beckman Coulter, Fullerton, California, USA). ### Sample collection Peripheral blood was collected into vacutainer blood collection tubes containing EDTA and whole blood was centrifuged to isolate plasma. Plasma samples were stored at -80°C for NMR metabolic profiling. ## Sample preparation and NMR spectroscopy Plasma samples were thawed on ice and centrifuged at 10,000g for 10 min to remove particulates. 70 μL of supernatant was then taken and immediately added to 140 μL of 0.1M potassium phosphate buffer (composition: 0.2M of K₂HPO₄ and KH₂PO₄, 10% 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS), D₂O, pH 7.4). The mixture was transferred to an NMR tubes (outer diameter: 3.0 mm) for NMR analysis. NMR experiments were performed at 298K on a Bruker 500.13 MHz AVANCE III spectrometer equipped with 5mm triple-resonance z-gradient cryoprobe (CPTCI 1H–31P/13C/2D Z-GRD). TopSpin, version 3.5 (Bruker Corporation), was used for spectrometer control. Temperature was calibrated using a 100% d4-methanol sample. ¹H Carr–Purcell–Meiboom–Gill (CPMG) echo sequence was used to record NMR spectra. A total CPMG delay of 300 ms was used with an echo time of 200 μs was used based on experimental optimization. NMR spectra were obtained with 32 scans and a relaxation delay of 4 sec. Free induction decays (FIDs) 10,000 Hz spectral width and 3.33 sec acquisition time were collected for each sample. ## NMR data processing NMR data processing was done using Topspin, version 3.5. Automatic matching, tuning and shimming of each data were done. They were Fourier transformed after zero-filling to varying points and application of a line-broadening factor of 0.30 Hz. Consequently, NMR spectra were phased and automatically baseline corrected. Spectra were referenced to DSS at 0 ppm and spectral region between 4.5 and 4.7 ppm was set to zero to remove water resonance suppression effect. Peaks for each samples were picked and each NMR peak list and intensity data were then imputed into Metaboanalyst 3.0. This program groups peaks based on their ppm values using a moving window of 0.03 ppm and a step of 0.015 ppm and peaks within the same group were aligned to their median ppm. Peaks appearing in less than half of the samples were excluded from downstream analysis. ## Statistical analysis, metabolite identification and pathway analysis MetaboAnalyst 3.0 (<http://www.metaboanalyst.ca/>) was used for statistical analyses and NMR spectral annotations. This is a web server with numerous statistical and machine learning algorithms for comprehensive metabolomics data analysis, visualization, and interpretation. Processed raw data from TopSpin were log transformed, normalized by sample median and auto-scaled by mean- centering and dividing by the standard deviation of each variable for better, reliable and accurate biological inference before univariate and multivariate data analysis in Metaboanalyst 3.0. Supervised and unsupervised multivariate statistical analysis including principal component analysis (PCA), partial least squares-discriminant analysis (PLS-DA) and orthogonal partial least squares-discriminant analysis (OPLS-DA) were performed to discriminate between our study groups. Metabolites responsible for difference in the metabolic profiling were obtained from variable importance in projection plot (VIP) of threshold 1.0 in PLS-DA. Both PLS-DA and OPLS-DA models were validated for predictability and PLSDA: R², accuracy, Q² and OPLS-DA: R²Y and Q² were generated to confirm whether the difference between our study groups was statistically significant. One-way analysis of variance (ANOVA) was employed to analyze the significance of differential metabolites across our study groups. Corresponding metabolites for differential peaks were identified from NMR spectra databases which include Human Metabolome Data Bank (HMDB) and complex mixture analysis (COLMAR) query web server as listed in. These metabolites were checked with two-dimensional (2D) homonuclear ¹H-¹H J-resolved experiments and compared with previously published one and two dimentional values. The representative 500 MHz 1H-1H J-Resolved (JRes) spectrum is shown in. Metabolic pathway analysis was carried out using MetaboAnalyst 3.0, Kyoto Encyclopedia of genes and genomes (KEGG) and HMDB pathway. # Results ## Multivariate analysis distinguishes plasma metabolic profile among treatment-naïve, ART-suppressed perinatally HIV-infected children and uninfected controls A representative one dimensional ¹H-NMR spectra of plasma from treatment-naïve, ART-suppressed perinatally HIV-infected child and uninfected control is depicted in. To determine whether it was possible to differentiate HIV-infected children from uninfected controls on the basis of the NMR spectra, we carried out unsupervised multivariate analysis such as PCA and supervised multivariate analysis such as PLS-DA and OPLS-DA. PCA score plot showed that treatment-naïve HIV-infected children could be distinguished from ART-suppressed HIV-infected children and uninfected controls. Whereas PLS-DA score plot revealed excellent separation among treatment-naïve, ART-suppressed perinatally HIV-infected children and controls. On validation by permutation tests, we obtained PLS-DA model with accuracy 0.972 (R² = 0.990 and Q² = 0.919) and p = 0.02 and OPLS-DA with R²Y = 0.979 and Q² = 0.917. For biological data, a model with R² = 0.7 and Q² = 0.4 is considered to be good. ## Perturbed levels of metabolites in HIV-infected children despite suppressive therapy To identify the metabolites with the highest degree of differences in the metabolomes among treatment-naïve, ART-suppressed HIV-infected children and uninfected controls, we used PLS-DA to evaluate variable importance in projection (VIP) scores. A heatmap showing relative intensity of some metabolites as identified by VIP scores. We identified 66 metabolites to be significantly altered in untreated HIV-infected children when compared to uninfected controls. Out of these, the levels of 38 metabolites were significantly high in untreated HIV-infected children compared to controls, while 28 metabolites were significantly low in untreated HIV-infected children compared to controls. Insulin resistance is one of the major metabolic disorders seen in patients living with HIV. Within altered metabolites, the levels of lactate, glucose, phosphoenol pyruvic acid and TCA cycle metabolites such as succinic acid, oxoglutaric acid, and oxaloacetic acid were significantly higher in the plasma of treatment-naïve HIV-infected children compared to uninfected controls. Compared to controls, the plasma levels of some other metabolites like propionic acid, acetate, acetoacetate, myoinositol and trimethylamine-N-oxide (TMAO) were significantly elevated whereas 2-ketobutyric acid and choline were siginificantly lowered in treatment-naïve HIV-infected children. treatment-naive HIV-infected children also showed abnormalities at the aminoacid levels with significantly lower plasma level of glycine, sarcosine, serine, tyrosine, valine and higher plasma level of alanine, creatine, glutamine, leucine, methionine and taurine compared to controls. In treated children the plasma levels of glucose, propionic acid, 2-ketobutyric acid, succinic acid, acetic acid, acetoacetic acid, TMAO, glycine, sarcosine, tyrosine, valine, alanine, creatine, leucine, methionine, and taurine were restored to the levels found in uninfected controls. However, the plasma levels of lactate, phosphoenolpyruvic acid, oxoglutaric acid, oxaloacetic acid, myoinositol and glutamine remained elevated and the plasma levels of choline and serine remained lowered in treated HIV-infected children suggesting the failure of ART in restoring the levels of these metabolites. ## Metabolic pathway analysis Metabolites that were significantly altered in treatment-naïve and ART- suppressed perinatally HIV-infected children when compared to controls, were subjected to pathway analysis. We found 22 metabolic pathways among treatment- naïve, ART-suppressed perinatally HIV-infected children and uninfected controls that were significantly modulated. The prominent changes were observed in amino acid metabolism, TCA cycle, glycolysis pathway. # Discussion In this study, we profiled the plasma metabolome of treatment-naïve and ART- suppressed HIV-infected children, and compared them with that of uninfected controls using NMR spectroscopy. Our data suggests perturbed levels of metabolites in treatment-naïve perinatally HIV-infected children compared to uninfected controls. The ART appears to normalize majority of metabolites in perinatally HIV-infected children under treatment; some metabolites like lactate, phosphoenolpyruvic acid, oxoglutaric acid, oxaloacetic acid, myoinositol and glutamine remain elevated either due to previous exposure to HIV or ART itself. Altered metabolites due to perturbed metabolic process might cause metabolic disorders like insulin resistance, hyperlipidemia, obesity and high blood pressure. It was beyond the scope of clinical routine practice and the study scope to measure lipid levels, insulin resistance etc. However, none of the HIV- infected children in this study had high BMI. Previous studies have shown a direct link between perturbed glycolytic pathway and chronic inflammation. Metabolic profiling of plasma of HIV-infected adults had shown higher levels of glucose and lactic acid, which is suggestive of perturbed glycolytic pathway. In our study, the increase in the level of glucose, lactic acid and phosphoenolpyuvic acid among treatment-naïve perinatally HIV-infected children is an indication of perturbed glycolysis and supports the previous findings in HIV-infected adults. In addition, high level of glucose or hyperglycemia has been known to promote HIV pathogenesis. This was supported by a study which showed that the high glucose increases the expression of CXCR4 in T cells thereby enhancing the entry of HIV into T cells. Glucose metabolism has also been reported to play a major role in interleukin-1β (IL-1β) production and thus high glucose level during chronic HIV infection results in the upregulation of IL-1β level in the plasma of HIV-infected subjects. A study by Mikulak et al showed that IL-1β enhances HIV entry and its persistence in human podocytes. Interestingly, the use of ART does not seem to have any effect on the perturbed glycolytic pathway. In ART-suppressed HIV-infected children, the glucose level is restored to that of uninfected controls, however, the levels of lactate and phosphoenolpyruvic acid remain elevated despite the suppressive therapy. Hyperlactatemia has been reported in HIV-infected children receiving antiretroviral treatment as well as in ART-exposed HIV uninfected children born to HIV-infected mothers. High plasma lactate concentration is directly associated with insulin resistance and increased hepatic gluconeogenesis in type-2 diabetic subjects. Our results of high levels of lactate in treated and untreated perinatally HIV-infected children are in agreement with the previous finding and suggests that perinatally HIV-infected children are at high risk of developing insulin resistance and diabetes even during suppressive treatment. High level of plasma succinic acid is associated with inflammation as shown in case of cancer. It induces primary inflammatory cytokine, IL-1β, through HIF-1α signaling. Also, when dendritic cells (DCs) were simultaneously primed with both succinate and antigen, T cell activation was increased, as shown by elevated TNF-α and IFN-γ production from these cells. Thus, succinate is an important metabolite that can act as an inflammatory signature. Increased levels of succinic acid in treatment-naïve perinatally HIV-infection children reflect elevated inflammation that appears to normalize during treatment. We observed elevated levels of plasma TMAO and propionic acid, a gut-microbiome dependent metabolite, in untreated HIV-infected children. Gut microbiota has a significant role in metabolism of host. Gut microbiota is associated with numerous metabolic conditions including diabetes, obesity, and CVD. Choline which comes from dietary phosphatidylcholine is metabolized by gut microbes to trimethylamine (TMA), which further oxidizes to TMAO in liver. Wang et al. showed that the metabolites of dietary phosphatidylcholine promote cardiovascular disease (CVD). Plasma level of TMAO has been shown to be associated with CVD in general population and carotid artery atherosclerosis progression in HIV-infected adults. This suggests that perinatally HIV-infected children are at the risk of developing CVD. Further, propionic acid, that is synthesized by the gut microbes during complex carbohydrates fermentation in the colon is taken up to the circulation where it further metabolizes in the liver and serves as an energy source. Propionic acid regulates glucose homeostasis and is antiobesogenic. However, elevated levels of propionic acid were reported in obese individuals and also in obese diabetes mice, indicating that propionic acid might may pathogenic roles via mechanisms yet to be elucidated. The increase in TMAO and propionic acid level in untreated HIV-infected children might be attributed to the gut dysbiosis as we observed in our previous study. Elevated Myo-inositol among untreated HIV-infected children is similar to earlier reports where they showed myo-inositol to be elevated in left frontal brain of HIV-infected children. Increased myo-inositol in both gray and white matter and also in basal ganglia regions were reported in chronically HIV- infected adults. Myo-inositol is a marker of glial cells and elevated myo- inositol indicates gliosis, a condition of proliferation or hypertrophy of glial cells in response to damage to the central nervous system (CNS). Increased myo- inositol in both treatment-naïve and ART-suppressed HIV-infected children suggests the presence of neuroinflammation and thus might potentially contribute to neurocognitive disorders despite successful virologic control by ART. Previous study has also demonstrated neuroinflammation and cognitive impairment in treated HIV-infected adults. Increased glutamine among untreated HIV-infected children could reflect CD4 T cell skewing and activation and as shown earlier in mice. This was further confirmed by Wang et. al, where glutamine catabolism was shown to be tightly coupled with the biosynthesis of polyamines that are essential for T cell proliferation. A study by Ziegler et al reported decreased levels of plasma amino acids except glutamate in HIV-infected youth. Our study also observed decreased valine, aspartatic acid, glycine, tyrosine, serine and sarcosine levels in the HIV infected treatment-naïve group. This might be due to intestinal malabsorption as observed in HIV-infected individuals. Intestinal epithelial apoptosis and increased epithelial barrier disruption during chronic HIV infection might reduce amino acids absorption. In addition to nutritional deficit, the change in plasma amino acids levels is also due to alteration in aminoacid metabolic pathway. In conclusion, our NMR-based metabolomics with an array of robust multivariate statistical techniques have unraveled distinct metabolic phenotypes and pathways among treatment-naïve, ART-suppressed perinatally HIV-infected children and uninfected controls. Untreated and ART-suppressed HIV-infected children had perturbed glycolysis, TCA cycle, aminoacid metabolism. Developing therapeutic stratergies targeting metabolic abnormalities and restoring healthy gut microbiota may be beneficial for preventing diabetes, cardiovascular disease or other associated complication in perinatally HIV-infected children. # Supporting information We would like to thank ICGEN Core Funds and DBT for NMR facility at ICGEB, New Delhi, India. U.K. is thankful to DBT-SRF fellowship and also Fogarty International Center (FIC) of the National Institutes of Health and the National Institute Of Diabetes And Digestive And Kidney Diseases (NIDDK) under grant \#3D43TW009345-08S1 awarded to the Northern Pacific Global Health Fellows Program for support. B.F.O thank ICGEB for the award of Arturo Falaschi ICGEB pre-doctoral fellowship. R.T. is thankful to UGC-Faculty Recharge Programme (FRP). [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Molecular Biology Research Group, Department of Science Technology, The Federal Polytechnic Ado-Ekiti, Ado Ekiti, Nigeria

# Introduction Age-related macular degeneration (AMD) is a common multifactorial and heterogeneous disorder, characterized by progressive degeneration of the central region of the retina (macula). Pigmentary abnormalities of the retinal pigment epithelium (RPE) and extracellular deposits (drusen) under the retina are among the early-stage manifestations of AMD. As the condition progresses, extensive atrophy of the RPE and outer retina (geographic atrophy or dry AMD) or abnormal vessel growth underneath the macula (exudative or wet AMD) are common advanced- stage manifestations. AMD affects 30–50 million individuals worldwide and is a leading cause of legal blindness among older individuals in developed countries. Although the precise etiology of AMD remains elusive, genetic studies have provided significant insights into the molecular basis of AMD. Several genes encoding proteins involved in the complement pathway have been shown to be associated with susceptibility to AMD, including the complement factor H gene (*CFH*) on chromosome 1q32, two neighboring genes, complement component 2 (*C2*) and complement factor B (*CFB*) on 6p21, the complement factor I gene (*CFI*) on 4q25, and the complement component 3 gene (*C3*) on 19p13. These findings strongly implicate aberrant regulation and/or activation of the complement pathway in the mechanism of susceptibility to AMD. In addition to the association with complement pathway genes, AMD has been convincingly shown to be associated with two adjacent genes on 10q26 (age-related maculopathy susceptibility 2 \[*ARMS2*\] and high-temperature requirement factor H \[*HTRA1*\]), which together account for nearly half of the heritability of AMD. AMD susceptibility loci have been primarily discovered in populations of European descent, of which only the association of *CFH* – and the *ARMS2*/*HTRA1* loci, have been convincingly validated in Asian populations. We recently reported a significant association of wet AMD in a Japanese population with the same susceptibility variant near *CFI* as that observed in individuals of European descent, indicating that, along with *CFH* and *ARMS2*/*HTRA1*, *CFI* is a susceptibility locus of AMD that transcends racial boundaries. However, studies have also revealed the existence of genetic heterogeneity in AMD susceptibility at the *C3* locus between populations of European and Asian descent. A nonsynonymous coding variant in *C3*, rs2230199 (R102G), was consistently found to be associated with AMD in Caucasian populations, , but not in Asians. Furthermore, the allelic frequency of the R102G variant is absent in Japanese and rare (\<1%) in Chinese populations, according to the data from the International HapMap Project and published studies, while risk allele frequency is almost 20% in individuals of European descent. It has been proposed that genetic effects of disease-associated variants are similar across racial boundaries regardless of their widely divergent allelic frequency between different populations. However, it has also been documented that genetic heterogeneity of disease susceptibility between ethnic groups is common in complex diseases, and thus, disease-associated variants present in populations of European descent might not be applicable to Asian populations because of underlying genetic heterogeneity. Indeed, two recent studies have suggested a role for common intronic variants of the *C3* locus in susceptibility to wet AMD in Japanese and Chinese populations, implying that more common *C3* variants are associated with the disease in Asians. Here we genotyped 13 tag single nucleotide polymorphisms (SNPs) that capture the majority of common variations in the *C3* locus and tested for associations between these SNPs and wet AMD in a Japanese population comprising 420 case subjects and 197 controls. # Materials and Methods ## Ethics Statement The study protocol was approved by the Institutional Review Board at Kobe University Graduate School of Medicine and performed in accordance with the Declaration of Helsinki. Written informed consent was obtained from all subjects before participation in this study. ## Study participants All cases and controls included in this study were Japanese individuals recruited from the Department of Ophthalmology at Kobe University Hospital in Kobe, Japan. The demographic details of the study population are shown in. All cases and control subjects underwent comprehensive ophthalmic examination, including visual acuity measurement, slit-lamp examination, and dilated funduscopic examination. Fundus findings in each eye were classified according to the clinical age-related maculopathy staging system (CARMS) as previously described. All of our case subjects had wet AMD and associated manifestations such as nondrusenoid pigment epithelial detachment, serous or hemorrhagic retinal detachment, and subretinal or sub-RPE hemorrhages and fibrosis; they were categorized as having CARMS stage 5. The controls were individuals aged 56 years or older and were defined as cases without macular degeneration and changes, such as drusen or pigment abnormalities. Thus, controls were categorized as having CARMS stage 1 on the basis of comprehensive ophthalmic examinations. ## Genotyping Genomic DNA was extracted from peripheral blood using standard methodology. Genotyping was performed using the TaqMan® SNP Genotyping Assays (Applied Biosystems, Foster City, CA) on a StepOnePlus™ Real-Time PCR System (Applied Biosystems) in accordance with the manufacturer's recommendations. ## SNP selection To comprehensively yet efficiently screen *C3* sequences for common genetic variations, tag SNPs were selected from the HapMap Project database for the Japanese in Tokyo (JPT) population using the tag selection tool. Thirteen tag SNPs were selected for genotyping, which captured 29 of 34 SNPs in the *C3* locus exhibiting a minor frequency greater than 10% with a mean r² value of 0.986. ## Statistical analysis Allelic associations were evaluated for each SNP by chi-square tests on 2×2 contingency tables using the software package PLINK v1.00 (<http://pngu.mgh.harvard.edu/purcell/plink/>). The odds ratio (OR) and corresponding 95% confidence interval (CI) were calculated relative to the major allele. In addition to obtaining nominal *P* values, corrected empirical *P* values for multiple testing were generated by 10,000 permutation tests using the Max (T) permutation procedure implemented in PLINK. We also applied a Bonferroni correction, where nominal *P*-values were multiplied by 13 (the number of SNPs tested for association). To adjust for age and gender differences between the case and control subjects, logistic regression analysis was performed using SNPStats (<http://bioinfo.iconcologia.net/SNPStats>), with age and gender controlled as covariates. Age and gender were included in this model as a continuous covariate measured in years and a categorical covariate, respectively. Deviations from the Hardy–Weinberg equilibrium were tested using the exact test implemented in PLINK. Haploview software was used to assess linkage disequilibrium (LD) patterns and haplotype association statistics. Haplotype blocks were determined using the solid spine of LD algorithm with a minimum D' of 0.8. To correct for multiple testing in the haplotype association analysis, 10,000 permutations were run using this software. An omnibus (or global) test of the haplotype association was performed with PLINK. To determine whether a single variant could explain an entire omnibus haplotype association, conditional haplotype-based likelihood ratio tests implemented in PLINK were conducted. The haplotype association was assessed further using sliding window analyses of four adjacent SNPs across the *C3* region. For this analysis, sliding windows of overlapping haplotypes were tested in sequence. For example, SNPs rs2250656, rs2230205, rs11569429, and rs11672613 were treated as a single haplotype, followed by SNPs rs2230205, rs11569429, rs11672613, and rs428453. The significance values were evaluated on the basis of omnibus test *P* values. The sliding window analyses were conducted using the PLINK software. The FASTSNP program (<http://fastsnp.ibms.sinica.edu.tw/pages/input_CandidateGeneSearch.jsp>) was used to predict the function of a SNP of interest. To examine a genetic effect detected here in the context of three validated AMD- risk loci for Asians (the A69S variant \[rs10490924\] in *ARMS2*, and the I62V variant \[rs800292\] and Y402H variant \[rs1061170\] in *CFH* –), we conducted logistic regression analyses with the R statistical analysis package (<http://www.r-project.org/>). For each locus, the genetic model of best fit was determined before genotypes were coded according to additive, dominant, and recessive models. Akaike Information Criterion (AIC) was used to select the model of best fit. The best models for each locus were then combined into multilocus models, and an effect of the *C3* variant after controlling for *ARMS2* A69S, *CFH* I62V, and *CFH* Y402H was estimated. Furthermore, we compared two logistic regression models (the full model including all four variants versus a reduced model in which the C3 variant was omitted) by using a likelihood ratio test and calculating AIC values. To determine epistatic effects between *C3* rs2241394 and *CFH* I62V or Y402H, pairwise interaction analysis was performed using the epistasis option in PLINK. # Results None of the 13 SNPs reported in the present study showed significant deviation from the Hardy–Weinberg equilibrium in both the case and control subjects (*P*\>0.05). Marker information, allelic frequencies, and summary statistics for all evaluated SNPs are shown in. In single-SNP analyses, two of the 13 SNPs showed nominally significant associations with wet AMD (rs2241394, nominal *P* = 8.32×10⁻⁴; rs428453, nominal *P* = 0.0240), of which only rs2241394 withstood multiple test corrections (corrected empirical *P* = 0.0102; Bonferroni-corrected *P* = 0.0108). The minor allele C of rs2241394 was associated with protection against the disease, with a frequency of 0.052 in disease cases and 0.104 in controls (per allele OR = 0.48 \[95% CI = 0.31–0.74\];). In a dominant genetic model, OR for individuals carrying at least one copy of the protective allele was 0.45 (95% CI = 0.28–0.72; *P* = 7.81×10⁻⁴). Inclusion of age and gender as covariates in the logistic regression model did not substantially change the significance of the association (age- and gender-adjusted OR = 0.48 \[95% CI = 0.30–0.75\], *P* = 0.0016, additive model; age- and gender-adjusted OR = 0.44 \[95% CI = 0.27–0.72\], *P* = 0.0012, dominant model). The pairwise LD structure was constructed with the 13 SNPs genotyped. Five haplotype blocks were defined, and association with the disease was restricted to block 4 where the disease-associated SNP rs2241394 resided as demonstrated by the significant omnibus result (omnibus *P* = 0.00367 at 2 degrees of freedom). Only one haplotype in block 4 was found to be significantly associated with the disease, with a haplotype frequency of 0.052 in affected individuals and 0.104 in controls (*P* = 8.0×10⁻⁴; OR = 0.48 \[95% CI = 0.31–0.74\];). This association remained statistically significant after correction for multiple testing (permutation *P* = 0.011). The disease-associated haplotype was completely described by the protective allele C of rs2241394, and a conditional haplotype-based likelihood ratio test revealed that the significant omnibus haplotype association detected in haplotype block 4 disappeared when it was estimated to be conditional on rs2241394 (omnibus *P* = 0.85), confirming that rs2241394 is responsible for the haplotype association detected in this LD block. To further assess haplotype associations, we conducted a sliding window analysis of four adjacent SNPs across the *C3* region. Significant associations were observed only around rs2241394, and the strongest association was found when four variants—rs2241393, rs2241394, rs1389623, and rs7951—were included together (omnibus *P* = 9.81×10⁻⁴). To examine the possibility that the disease-associated SNP rs2241394 might be correlated with untyped SNPs, we investigated the LD structure across the genomic region extending approximately 200 kb upstream and downstream of the *C3* locus. Genotype data were retrieved from the 1000 Genome Project (August 2010 release) and International HapMap (release 24) JPT+CHB datasets, and correlations (as defined by r² values) were examined. In this genomic region, we found 594 SNPs but did not identify any SNPs that are highly correlated with rs2241394 (all pairwise r²\<0.45). Next, we examined the genetic effect of rs2241394 in the context of three validated AMD-risk loci for Asians (*ARMS2* A69S, *CFH* I62V –, and *CFH* Y402H ). Using unconditional logistic regression, the genetic model of best fit for *C3* rs2241394, *ARMS2* A69S, *CFH* I62V, and *CFH* Y402H was determined and genotypes were coded according to additive, dominant, and recessive models. On the basis of AIC values, *ARMS2* A69S, and *CFH* I62V had the best fit under an additive model, and *C3* rs2241394 and *CFH* Y402H had the best fit under a dominant model. The best models were then combined into multilocus logistic regression models. After including the effects of *CFH* I62V, *CFH* Y402H, and *ARMS2* A69S, *C3* rs2241394 retained significant association (model 1;). Furthermore, we found that the model including all four variants–*C3* rs2241394, *ARMS2* A69S, *CFH* I62V, and *CFH* Y402H–fit significantly better than the model without *C3* rs2241394 (likelihood ratio test χ² = 10.32, *P* = 0.00132, model 1 vs. model 2; AIC = 692.0 and 700.3 for model 1 and 2, respectively;). Finally, we conducted pairwise interaction analysis to evaluate potential epistatic effects between *C3* rs2241394 and *CFH* I62V or *CFH* Y402H, because the proteins encoded by these loci biologically interact in the complement pathway. However, we did not find any evidence of epistasis between rs2241394 and *CFH* variants (all *P*\>0.05). # Discussion We genotyped 13 tag SNPs that capture the majority of common genetic variations in the *C3* locus and found statistically significant evidence for association between an intronic *C3* variant (rs2241394) and wet AMD in a Japanese population (*P* = 8.32×10⁻⁴). Haplotype analyses identified the LD block where rs2241394 resides as being the only significant locus, and haplotype association was completely explained by rs2241394. Logistic regression analysis showed that the effect of rs2241394 is independent of the established associations of *ARMS2* A69S, *CFH* I62V, and *CFH* Y402H, and that the model including these three established loci plus *C3* rs2241394 provides a better fit than the model without rs2241394. Although the proteins encoded by *C3* and *CFH* are involved in the same biological pathway, we found no evidence of epistasis between rs2241394 and the two *CFH* variants. Complement has emerged as an important element in AMD pathology, because of the identification of various complement-related molecules in drusen and nearby RPE. In addition, recent successes in the identification of genetic susceptibility loci for AMD have revealed several molecules involved in the complement pathway, including CFH, CFB, C2, CFI, and C3. Furthermore, systemic complement activation was observed in AMD patients – and nutritional supplementation with zinc was shown to delay the progression of AMD, an effect likely mediated by an inhibitory effect of zinc on complement activity. C3 is a central component of all three pathways of complement activation: the alternative, classical, and mannose-binding lectin pathways, all of which lead to the cleavage of C3 into biologically active C3a and C3b fragments. Notably, an animal study has shown that C3 deficiency prevented the formation of choroidal neovascularization induced by the rupture of Bruch's membrane with laser photocoagulation in eyes of *C3*^−/−mice, indicating that C3 is a key factor in the development of choroidal neovascularization. A nonsynonymous coding *C3* variant, rs2230199 (R102G), is strongly associated with AMD in populations of European descent, and this variant is presumed to be the most likely causal variant responsible for the disease association based on mechanistic functional evidence. However, the association of R102G has not been reported in Asian populations, and allele frequencies of R102G vary widely among different ethnicities. For example, the risk allele is absent in Japanese and rare (\<1%) in Chinese populations, while the corresponding rate in Caucasians is 20%. In the present study, we have found that a more common SNP of *C3*, rs2241394, is associated with AMD in a Japanese population. This association has not been documented by any previous genetic studies of AMD in European populations. These findings suggest that the susceptibility conferred by the R102G variant does not transcend ethnic lines and that there may be a significant difference in disease susceptibility loci in the *C3* region of populations of European and Asian descent. Notably, rs2241394 has previously been reported in a Japanese population to be associated with polypoidal choroidal vasculopathy, a major subphenotype of wet AMD in East Asian populations, and the direction of association was consistent with our findings. However, suggestive evidence for association of rs2250656 with wet AMD previously reported in a Chinese cohort was not detected in the present study. We sought further evidence from a recent genome-wide association study of wet AMD in Japanese populations ; however, the arrays used in this study (Illumina HumanHap610-Quad BeadChip and Illumina HumanHap550v3 Beadchip) did not suit rs2241394. The *C3* variant rs2241394 is an intronic SNP, and there is currently no evidence supporting its functional relevance. Using the FASTSNP program , we investigated potential functions of rs2241394. According to the analysis, this SNP was identified as lying in an intronic enhancer region created by a “C→G” change at rs2241394 that may lead to the creation of a binding site for the transcriptional factor *GATA-1*. Therefore, this SNP may have a functional relevance to disease risk for Japanese populations in the absence of surrounding 1000 Genome Project and HapMap SNPs that are highly correlated with rs2241394; however, fine-mapping and resequencing efforts are required to identify any potential as yet unidentified variants of more functional relevance. In conclusion, we report a significant association between wet AMD and a common noncoding *C3* variant in a Japanese population. Our study provides evidence that *C3* is a common AMD-associated locus that transcends racial boundaries and provides an impetus for more detailed genetic characterization of the *C3* locus in Asian populations. [^1]: Conceived and designed the experiments: SY NK AM WM SK SH YT AN. Performed the experiments: SY NK AM WM SK. Analyzed the data: SY NK AM WM SK SH YT AN. Contributed reagents/materials/analysis tools: SY NK AM WM SK SH YT AN. Wrote the paper: SY NK. Critical revision of the article: AM WM SK SH YT AN. [^2]: The authors have declared that no competing interests exist.

# Introduction Alterations in arterial structure and function occur with advancing age in healthy individuals, and the aging-induced decrease in endothelial function contributes to the increases in arterial stiffness. Several studies have shown that arterial stiffness is lower in physically active individuals compared with sedentary individuals. Furthermore, aerobic exercise training reduces arterial stiffness, which increases with advancing age. Thus, regular aerobic exercise prevents or reduces arterial stiffness. Nitric oxide (NO), which is produced from L-arginine by endothelial NO synthase (eNOS) in endothelial cells, contributes to the underlying mechanism of this effect of exercise. NO causes vasodilation and inhibits the development of arteriosclerosis and atherosclerosis. Aging impairs arterial eNOS protein and mRNA expression, however, eNOS expression levels are increased by endurance exercise training in aged rats. Moreover, in middle-aged and older woman, moderate regular exercise training elevates plasma NOx levels with reduction of blood pressure. Thus, aging impairs NO bioavailability, and it may result in increased arterial stiffness. Recently, apelin has been detected in several tissues, such as white adipose tissue, kidney, heart, and vessel. Apelin is initially synthesized as preproapelin, which consists of 77 amino-acid residues. Following enzymatic cleavage, the C-terminus is released into the circulation as the biologically active fragment, apelin. Apelin acts via APJ receptor in the expressing endothelial cells. Clinical evidence suggests that plasma apelin levels are generally lower in patients with cardiovascular diseases, such as heart failure and hypertension. In animal studies, apelin administration decreased blood pressure in normal and hypertensive rats. This effect was blocked in the presence of a NOS inhibitor, suggesting that apelin causes vasodilation through a mechanism that involves NO. In the APJ-deficient mice, the hypotensive effect to apelin was blocked with the downregulating eNOS phosphorylation in the endothelial cells. Therefore, reduced plasma apelin levels may be associated with reductions of NO bioavailability in older adults, and this may in turn result in increased arterial stiffness. Additionally, in patients with type 2 diabetes mellitus, regular exercise training elevates circulating apelin levels, and higher levels of physical activity caused larger increases in apelin levels than lower levels of activity. Furthermore, a study in hypertensive rats showed that exercise training promotes mRNA expression and tissue concentration of apelin in the aorta. However, the association between aerobic exercise training-induced changes in arterial stiffness and circulating apelin level in healthy middle-aged and older adults remains unclear. We hypothesized that aerobic exercise training would elevate plasma apelin levels along with plasma NOx levels in middle-aged and older adults, and that this increase in apelin may be participated in endurance exercise training- induced reduction of arterial stiffness. To test our hypothesis, we measured plasma apelin levels, nitrite/nitrate (NOx) concentrations, and arterial stiffness in middle-aged and older adults using a randomized controlled exercise intervention trial. # Methods ## Subjects Thirty-four healthy middle-aged and older subjects (total, *n* = 34, 67.0±1.3 years; male, *n* = 14, 70.4±1.6 years; female, *n* = 20, 64.7±1.8 years) volunteered to participate in this study. Subjects were recruited via advertisement from a local community health center and a community recreation center. All volunteers provided written informed consent before participating in the study, which was approved by the Ethics Committee of Ritsumeikan University and was conducted in accordance with the Declaration of Helsinki. Subjects with taking medication such as anti-hyperlipidemic, anti-hypertensive, or anti- hyperglycemic, and a history of stroke, diabetes, hypertension, hyperlipidemia, cardiac disease, chronic renal failure, and mental disorder were excluded from the study, and the subjects in this study hardly drank alcohol. Subjects were randomly divided into two groups: the training group (*n* = 18 \[male  =  7/female  = 11\], 66.4±2.1 years) and the control group (*n* = 16 \[male  =  7/female  =  9\], 67.8±1.5 years). ## Experimental design For all subjects, VO_{2 peak}, body weight, body fat, height, resting systolic blood pressure (SBP), resting diastolic blood pressure (DBP), resting heart rate (HR), resting plasma NOx concentrations, resting plasma apelin concentrations, and serum concentrations of total cholesterol, HDL cholesterol, and triglycerides were measured at the beginning and end of the experiment. Carotidβ-stiffness was examined as an index of arterial stiffness. Before subjects were tested, they sat quietly for 30 min. Resting brachial SBP, DBP, and HR were measured in duplicate in the supine position at rest, using a vascular testing device (OMRON COLIN Co., Tokyo, Japan). At the beginning and end of the study period, fasting blood samples were drawn following at least 48 hours of rest after the last exercise-training session. All subjects were instructed not to eat or drink fluids other than water for at least 12 h prior to blood sampling. In addition, we checked to be sure that participants did not intake any dietary sources of NOx over the 24 h prior to testing in both groups, since NOx can be affected by diet. Thus we were able to rule out both acute effects from the most recent bout of exercise and oral sources of NOx other than NO. Serum and plasma samples were immediately centrifuged (1500×*g*, 15 min, 4°C). Blood samples were stored at −80°C until use. Room temperature was maintained at 24°C throughout the experiment. ## Exercise intervention Aerobic exercise-training program consisted of cycling on a leg ergometer (828E Monark cycle ergometer, Stockholm, Sweden) for 55 min, 3 days/week, for 8 weeks. Each exercise session consisted of a 5-min warm-up period at 40% peak oxygen uptake (VO_2peak), followed by 45 min of cycling at a resistance that elicited 60–70% VO_2peak, and ended with a 5-min cool-down period at 40% VO_2peak. The exercise training program was conducted in the group at AM9:00–11:00 after breakfast. Exercise compliance was carefully monitored by direct supervision. Additionally, the sedentary control subjects were encouraged to maintain the activities of daily living and might not be changed during the 8-week experiment period. The subjects in both groups were encouraged to maintain the food intake and might not be different from usual during the experiment period. ## Measurement of VO_2peak VO_2peak was measured during breath-by-breath assessment using an incremental cycle exercise test on a cycle ergometer (MINATO, AE-310SRD, Osaka, Japan). Incremental cycle exercise began at a work rate of 60 W (30–90 W) for men and 30 W (0–60 W) for women, and power output was increased by 15 W•min⁻¹ until the subjects could not maintain a fixed pedaling frequency of 60 rpm. The subjects were encouraged during the ergometer test to exercise at maximum intensity. Heart rate and rating of perceived exertion (RPE) were monitored minute by minute during the exercise. RPE was obtained using the modified Borg scale. The highest 30-second averaged value of VO₂ during the exercise test was designated as VO_2peak if three out of four of the following criteria were met: (I) plateau in VO₂ with an increase in external work, (II) maximal respiratory exchange ratio ≥1.1, (III) maximal heart rate ≥90% of the age-predicted maximum (208–0.7× age;), and (IV) RPE ≥18. ## Measurement of the carotid β-stiffness index Carotidβ-stiffness was examined as an indicator of arterial stiffness. A combination of ultrasound imaging of the pulsatile common carotid artery and simultaneous applanation of tonometrically obtained arterial pressure from the contralateral carotid artery allowed noninvasive determination of arterial compliance. The carotid artery diameter was measured from images obtained using an ultrasound system equipped with a high-resolution linear array transducer. A longitudinal image of the cephalic portion of the common carotid artery was acquired 1–2 cm proximal to the carotid bulb. All image analyses were performed by the same investigator. Pressure waveforms and amplitudes were obtained from the common carotid artery using a pencil-shaped probe with a high-fidelity strain gauge transducer (SPT-301; Millar Instruments; Houston, TX;). Because baseline blood pressure levels are dependent on hold-down pressure, the pressure signal obtained via tonometry was calibrated by equating the carotid mean arterial blood pressure and DBP to brachial artery values,. The carotid β-stiffness index was calculated using the equation \[ln(P1/P0)\]/\[(D1–D0)/D0\], where D1 and D0 are the maximum (systolic) and minimum (diastolic) diameters, and P1 and P0 are the highest (systolic) and lowest (diastolic) blood pressures, respectively. ## Measurement of plasma NOx concentrations NOx concentrations in the plasma were measured by the Griess assay (R&D Systems, Minneapolis, MN), according to the manufacturer's protocol. All samples were assayed in duplicate. Optical density at 540 nm was qualified using a microplate reader (xMark microplate spectrophotometer; Bio-Rad Laboratories, Hercules, CA). Samples were converted into concentration by a linear fit of the log–log plot of the standard curve. ## Measurement of plasma apelin concentrations Apelin concentrations in the plasma were measured by an enzyme-linked immunosorbent assay (ELISA; Phoenix Pharmaceuticals inc., Burlingame, CA, USA), according to the manufacturer's protocol. Apelin-12, which was used in the present study, has cross-reactivity with apelin-13 and apelin- 36. All samples were assayed in duplicate. Optical density at 450 nm was qualified using a microplate reader (xMark microplate spectrophotometer; Bio-Rad Laboratories). Samples were converted into concentration by a linear fit of the log–log plot of the standard curve. ## Measurements of serum cholesterol and triglyceride levels Fasting serum concentrations of total cholesterol, HDL cholesterol, and triglyceride levels were determined using standard enzymatic techniques. ## Statistical analysis Values are expressed as the means ± SE. Differences between groups and two time points were assessed by two-way repeated-measure ANOVA, followed by a Fisher's post hoc test that was applied when a measurement was significantly different. The unpaired Student t-tests were used to compare differences in percent change from baseline in carotid β-stiffness, plasma NOx concentrations, and plasma apelin concentrations between training and control groups. Relationships between plasma apelin concentrations and carotid β-stiffness, plasma NOx concentrations were determined using the Pearson correlation coefficient. *P*\<0.05 was defined as statistically significant. All statistical analyses were performed using StatView (5.0, SAS Institute, Tokyo, Japan). # Results ## A comparison of baseline in the training and control groups Before the exercise training intervention, there was no significant difference in VO_2peak between the training and control groups, however, there was a significant interaction of groups and time in VO_2peak (*P* = 0.046). Specifically, in the training group, VO_2peak was significantly increased after the exercise training intervention. Although there was no significant difference in carotid β-stiffness between the training and control groups before the exercise training, there was a significant interaction between groups and time on carotid β-stiffness (*P* = 0.038). After the exercise training intervention, the carotid β-stiffness significantly decreased. In addition, the percent change of carotid β-stiffness was also significantly higher in the training group compared to control group. However, no significant changes were noted between before and after the exercise training intervention, or between groups, in age, height, body weight, BMI, HR, SBP, DBP, total cholesterol levels, HDL cholesterol levels, or triglyceride levels. ## Comparison of plasma apelin and NOx concentrations between the training and control groups Before exercise training intervention, there was no significant difference in plasma apelin and NOx concentrations between the training and control groups. There was a significant interaction between groups and time on plasma NOx concentrations (*P*\<0.0001). Specifically, after exercise training, plasma NOx concentrations were significantly increased in the training group as compared to the control group. Additionally, the percent change in plasma NOx concentrations was significantly higher in the training group than in the control group (-A). A significant interaction among groups and time was seen in plasma apelin concentrations (*P*\<0.0001). After exercise training intervention, plasma apelin concentrations were significantly increased in the training group as compared to the control group. In addition, the percent change of plasma apelin concentrations was also significantly higher in the training group compared to control group (-B). ## The correlation between plasma apelin concentrations and carotid β-stiffness, plasma NOx In the training group, the percent change of plasma apelin concentrations negatively correlated with the percent change of β-stiffness (r = −0.508, *P* = 0.032, -A). Furthermore, the percent change of plasma apelin concentrations positively correlated with the percent change of plasma NOx concentrations in the training group (r = 0.494, *P* = 0.037, -B). # Discussion This study investigated the effects of regular aerobic exercise on apelin concentrations in middle-aged and older adults before and after 8-week aerobic exercise training. After the exercise training intervention, arterial stiffness decreased, concomitantly, plasma apelin levels elevated along with plasma NOx levels. By contrast, there were no significant changes in these parameters in the sedentary control group. Furthermore, the effect of training on carotid β-stiffness was negatively correlated with the effect on plasma apelin levels. Thus, an elevation of plasma apelin levels is associated with a concomitant decrease in arterial stiffness through aerobic exercise training in middle-aged and older adults. Additionally, the effect of training on plasma NOx levels was positively correlated with the effect on plasma apelin levels. Several studies have suggested that increases in NO production resulting from endurance exercise training may be a causal factor in reduction of arterial stiffness risks, in both humans and animals. Therefore, these results suggest that arterial NO bioavailability via apelin may be participated in variation of arterial stiffness in middle-aged and older adults. In a previous study, circulating apelin levels were increased by 6 months of aerobic training consisting of walking, treadmill running, cycling, or calisthenics at 60–70% of maximal heart rate for 60 min, 4 days/week, whereas 8-types of resistance training at 60–80% of one repetition maximum, 4 days/week, did not change apelin levels in patients with type 2 diabetes mellitus. Patients with higher physical activity had higher plasma apelin levels than less physically physical active patients. In animal studies using hypertensive model rats, 9-week swimming training normalized mRNA expression and tissue concentration of apelin in aorta, along with apelin plasma levels. Previously, however, the effect of regular aerobic exercise on apelin concentration in middle-aged and older adults had remained unclear. This study demonstrated that 8-week aerobic exercise training, consisting of cycling on a leg ergometer at 60–70% VO_2peak for 45 min, 3 days/week, elevated plasma apelin levels in middle-aged and older adults. Thus, aerobic exercise training may be an effective way to elevate plasma apelin levels and reduce arterial stiffness in both healthy and at-risk subjects. This study demonstrated elevated NOx plasma levels and decreased arterial stiffness after aerobic exercise training in middle-aged and older adults. In a previous study, middle-aged and older women who performed aerobic exercise training exhibited elevated plasma NOx levels and reduced blood pressure. Aging leads to increased risk of arterial stiffness; this risk is associated with attenuation of NO generation. However, an animal study revealed that eNOS protein and mRNA expression levels in the aortas of older rats were ameliorated by regular exercise training. In this study, plasma apelin levels were increased by regular exercise training. Moreover, plasma NOx levels were associated with plasma apelin levels. In another study, administration of apelin induced NO production and promoted eNOS mRNA expression in the isolated rat aortic tissue, but did not alter iNOS expression. Furthermore, eNOS phosphorylation of isolated endothelial cells in mice was accelerated by the treatment with apelin. Apelin- induced eNOS activation is mediated by the activation of phosphatidylinositol-3 kinase (PI3K)/Akt signaling pathway in endothelial cells, resulting in increased NO production. Apelin treatment increased eNOS phosphorylation in the aorta, but in isolated endothelial cells from APJ-deficient mice, increased eNOS phosphorylation in response to apelin was inhibited. Additionally, in the left internal mammary artery of patients with stable coronary artery disease, exercise training raised eNOS expression and phosphorylation in association with change of Akt phosphorylation. Apelin is involved in regulation of eNOS gene expression, and contributes to NO production in the endothelial cells of aorta. Thus, the elevation in circulating apelin levels induced by regular exercise training may participate in the acceleration of NO generation in middle-aged and older adults. We demonstrated that exercise training in middle-aged and older adults elevated apelin plasma concentration. However, the source of the exercise training–induced increase in apelin plasma levels is unclear. Zhang et al. demonstrated that aortic, myocardial, and plasma apelin concentrations were all increased by exercise training in hypertensive rats, concomitant with elevation of the apelin mRNA expression level in aorta and heart. Therefore, aorta and heart tissues are one possible source of the exercise training–induced apelin production. However, apelin has also been detected in several tissues, including white adipose tissue and kidney. Further studies should investigate the source of exercise training–induced increase in apelin plasma levels. After the aerobic exercise training intervention, plasma apelin levels increased, and concomitantly, plasma NOx levels were increased. The mechanism underlying these effects is unclear. Hypoxic inducible factor 1–α, bone morphogenetic protein receptor 2, insulin, tumor necrosis factor–α, and mechanical stress are all potential inducers of apelin production. However, it is unclear whether the levels of these inducers are influenced by exercise training, and whether they contribute to elevated apelin production in this context. Further studies are needed to examine the effects of regular exercise training on these inducers in endothelial cells. Although the older increase the risk of cardiovascular disease, the present study recruited healthy middle-aged and older subjects. Therefore, it should focus on the effect of exercise training on apelin production in the elderly patient. In conclusion, we investigated the effects of regular aerobic exercise on plasma apelin concentrations in middle-aged and older adults before and after 8-week aerobic exercise training. After the exercise training intervention, plasma apelin levels increased along with plasma NOx levels, whereas arterial stiffness decreased. Additionally, the plasma apelin level was negatively correlated with carotid β-stiffness, and also was positively correlated with the plasma NOx level. Thus, the increase in plasma apelin levels may partly contribute to the improvement in arterial stiffness and NO bioavailability resulting from aerobic exercise training in middle-aged and older adults. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MI. Performed the experiments: S. Fujie K. Sato EM NH K. Sanada MI. Analyzed the data: S. Fujie. Contributed reagents/materials/analysis tools: S. Fujita TH. Wrote the paper: K. Sato MI.

# Introduction Gait asymmetry is one of the most common conditions following neurological impairments, such as a stroke, Parkinson’s disease (PD), or multiple sclerosis (MS). Deficits in temporal and spatial gait symmetry produce gait abnormalities, such as reductions in walking speed and balance and unbalanced propulsions between the legs. Thus, restoring spatiotemporal gait symmetry is an important goal in gait rehabilitation to minimize challenges that face many neurologic populations. In an effort to alleviate spatial asymmetric gait, split-belt treadmill (SBT) walking training has been tested, where the different belt speeds exaggerate patients’ step length asymmetry. For instance, when the different split-belt speeds exaggerate stroke patients’ step length asymmetry, they start to adapt their gait pattern by increasing their shorter step length on the faster belt. When the two belts are returned to the same speed, the gait adaptation to SBT walking has been shown to produce aftereffects of newly adapted symmetric gait patterns, which can lead to temporal corrections in spatiotemporal gait symmetries. However, the aftereffects in response to the mechanical perturbations by SBT are short-lived (i.e., poor retention of the adapted gait patterns), which results in a limited transfer of newly learned motor patterns to post-training movements. In motor learning, the interaction between explicit and implicit learning processes is related to consolidating motor memories, and there has been considerable interest in the role of implicit processes in motor adaptation and retention. An implicit learning process generally refers to acquiring a certain motor pattern without voluntary or conscious movements. As an alternative training intervention to modulate gait symmetry in a non-conscious manner, we have previously proposed a novel visual perturbation paradigm in which subjects adapt their gait patterns in response to visual feedback distortion (VFD) of their step length symmetry. For this, we distorted the visual feedback of subjects’ gait step symmetry so that subjects perceived their gait asymmetrically. The subjects were not informed that the feedback was being distorted, and they were not given any experimental tasks where they had to consciously perform an action during the trials. We found that a gradual distortion of visual feedback systematically modulated gait step length away from symmetry and the adapted gait pattern (aftereffects) remained even after the visual feedback was removed. The subjects did not notice distortion or modulation of their symmetric gait pattern. In addition, our prior study on healthy subjects compared the magnitude of aftereffects of step length asymmetry acquired using implicit VFD versus SBT walking, and we discovered that subjects who were trained with VFD retained aftereffects significantly longer compared to the SBT walking trial. These results suggest that distorted visual information in an unconscious manner can create sensory-prediction errors and provide a greater degree of implicit motor memory, which is thought to be responsible for the longer retention of newly learned motor patterns. SBT walking is a well-characterized gait learning intervention via mechanical perturbation and can induce sizeable changes in spatiotemporal gait symmetry. One way to improve the efficacy of symmetric gait adaptation and retention would be to integrate the visual distortion and the mechanical perturbations into a single training session. Gait control may employ distinct adaptive learning processes during walking under distinct perturbations, and the different perturbation modalities could potentially have complementary effects on gait symmetry adaptation and retention. For visual perturbations, while walking on a split-belt treadmill, visual bars that represent the right and the left step lengths are displayed on a computer screen, and only one of the bars is distorted without subjects’ knowledge of the distortion. The mismatch between the predicted and actual visual information of gait symmetry may induce asymmetric gait patterns with a longer retention effect. Thus, we aimed to investigate the potential benefits of concurrent gait training that combined VFD with SBT walking in healthy individuals. We evaluated the short-term retention (aftereffects) of adapted spatial gait symmetry from the implicit VFD+SBT combining strategy and compared them to those with SBT-only walking. We hypothesized that the concurrent gait adaptation to implicit VFD+SBT perturbations in an unconscious manner produces complementary benefits of facilitating the adaptation process and retaining adapted gait patterns longer than the single mechanical perturbation strategy. We also compared the retention from the implicit VFD+SBT walking trials with the SBT walking with conscious correction trials to confirm that unconscious processes associated with motor adaptation can have longer-lasting retention effects than conscious (explicit) processes. A positive outcome of this study would indicate that implicit visual perturbations through VFD can be useful for gait rehabilitation when used in conjunction with SBT walking because motor learning has a better retraining effect. # Materials and methods ## A. Subjects Thirty healthy adult volunteers, 20–50 years with no visual or physical disability, participated in this study. The subjects’ average age was 22.6 ± 6.2 years, their average weight was 71.5 ± 12.1 Kgs, and their average height was 174.7 ± 9.5 cm. Subjects were given a detailed explanation of the study procedure by the researchers before providing their written informed consent to participate in this study. All subjects were informed of the physical fatigue they would experience during the experiment and were accustomed to walking on a treadmill. All protocols (14-EF-023) were approved by the Institutional Review Board of California Baptist University to confirm the study meets national and international guidelines for research on humans. ## B. Experimental setup All walking experiments were performed using a split-belt treadmill (Woodway USA, Waukesha, WI) consisting of two separate belts, each with its own motor that permitted the speed of each belt to be controlled independently. The treadmill was equipped with supporting handrails to ensure the safety of subjects. All subjects were given a 5–10 minute habituation period to adjust to walking on the treadmill. Two motion-capturing markers were attached to the back of the subject’s shoes. These markers were seen by an optometric motion capture system (OPTOTRACK 3D Investigator, Northern Digital Inc., Canada) that locates the positions of the markers. The retrieved data was then sent to a PC in real- time using a program developed with LabVIEW (National Instruments Corp., TX) to graphically represent the subjects’ step length measurements on a computer screen. For visual feedback, real-time visual feedback of step length was provided to subjects on a 48-inch LCD display placed directly in front of the treadmill (about 2 meters away). The display shows two vertical bar graphs representing the step length of each leg. The step length is defined as the distance between the two legs, expressed as the vertical bar’s height. Thus, during the swing phase of a leg, the height of the corresponding vertical bar initially had no height and increased in real time. The height of the vertical bar freezes when a heel strike occurs, and it resumes to change when that leg reenters the next swing phase and passes the contralateral stance foot (i.e., when the distance is zero). Therefore, subjects observe the maximum stride length of the right and the left leg side by side at every heel strike. We thoroughly explained the meaning of the bar to the subjects. During the visual feedback distortion trial, we distorted the length of only the right bar (representing the right step length) in 8% decrements from 100% down to 60%. For example, a -8% distortion of step length changed the bar height from 100% to 92%. Thus, subjects would perceive their right step length to be 8% shorter than the actual length during the visual distortion. Each distortion level lasted for 60 seconds for the first 7-minute adaptation period. After the adaptation phase, the visual feedback was removed, and the subjects continued walking for the remaining 15-minute post-adaptation period. Our post-test questionnaire confirmed that the subjects did not notice the distortion. ## C. Experimental protocol The experiments were designed a) to primarily investigate the effects of visual feedback distortion within an unconscious manner on the retention of the new gait pattern (asymmetric gait pattern) with the SBT walking trial and b) to see if there was an advantage of using the implicit visual distortion strategy versus the explicit conscious correction strategy in terms of the retention of the newly learned gait pattern. 30 subjects participated in two different trials at least several days apart: 1) a trial with SBT walking only (***SBT-only***) and 2) a trial with SBT walking combined with implicit visual feedback distortion (***implicit VFD+SBT***). Of the 30 subjects, 12 subjects also participated in a conscious correction trial that combined SBT walking with veridical visual feedback (***conscious VF+SBT***). In this additional trial, the subjects consciously (voluntarily) corrected their asymmetric gait pattern under SBT walking perturbation. Due to the mechanical perturbation by SBT, the two vertical bars would look asymmetric, and the subjects were then asked to match the height of the bars over the entire adaptation period. The order of trials was randomized between the SBT-only and combined trials. As for the combined trials, the implicit VFD+SBT trial was performed at intervals of several days before the conscious VF+SBT trial. summarizes the three experimental conditions tested in this study. Each experimental trial consisted of a 5-minute control and 25-minute main sessions. For the first control trial, the subjects walked comfortably while the treadmill belts were moving at the speed of 2 mph (0.894 m/s), whereas during the control trial on their second visit, they were asked to look at visual feedback (no distortion applied) on the screen while walking. For the 25-minute main trial, there were three distinct sections: the baseline period (3 minutes), the adaptation period (7 minutes), and the post-adaptation period (15 minutes). The treadmill’s speed during the pre-adaptation phase was a consistent speed of 2 mph for the 3 minutes. During the adaptation period for all three conditions (*SBT-only*, *implicit VFD+SBT*, and *conscious VF+SBT*), the speed of the left leg stayed at 2 mph while the right leg speed was increased from 2 to 3 mph (1.341 m/s) by 0.2 mph (0.0894 m/s) every minute. The speed of the right belt reached 3 mph after 4 minutes and stayed at 3 mph for another 3 minutes until the post-adaptation phase started. During the 15-minute post-adaptation period, the speeds of the two belts returned to the same speed (2 mph), and the visual feedback was also removed. The subjects continued to walk until the end of the trial. During the experimental session, we measured the changes in gait symmetry and assessed the retention of asymmetric gait patterns during the post- adaptation period. For the *implicit VFD+SBT* trial, the split-belt speed ratio change was identical to the *SBT-only* trial. Subjects were asked to keep their gaze on the screen and walk comfortably with no further specific instructions. For the initial pre-adaptation period, visual feedback without any distortion was shown to the subjects. During the adaptation period, only the right step length’s visual feedback was distorted by -8% every minute until a total of 60% distortion was achieved (in this way, visual feedback in the implicit VFD+SBT condition was distorted in a manner to exaggerate the typical step length asymmetry by SBT walking). Following the adaptation period, the visual feedback disappeared from the screen, the speeds of the two belts returned to the same speed, and subjects continued to walk for the remaining 15 minutes of post- adaptation. This was considered an implicit condition because the subjects were given visual feedback without knowing that the visual feedback of their step length was being distorted, and there was no experimental task they had to perform consciously during trials. They also did not notice the asymmetric step pattern caused by the visual distortion. In the *conscious VF+SBT* trial, the split-belt speed ratio change was identical to the *SBT-only* trial. However, subjects were asked to consciously match the heights of the visual feedback bars to induce greater asymmetric adaptation during the adaptation period. ## D. Data analysis The primary measure used in this study was step length symmetry. To compute the step length symmetry, the ratio (%) between the left step length and the right step length was calculated for each gait cycle by using the following formula: 100×(right step length−left step length)/(0.5×(right step length + left step length)). Positive symmetry ratios mean that the right step lengths are longer than the left ones, and negative symmetry ratios indicate that the right step lengths are shorter than the left ones. Among the 30 subjects, three did not exhibit typical changes in step length symmetry in the implicit VFD+SBT walking. This was most likely because they tried to voluntarily change their gait steps by matching the heights of the two visual bars, even though it was an implicit (unconscious) condition. Thus, these three subjects’ data were excluded from the data analysis. To analyze the change in step length symmetry during the adaptation period (7 minutes) and the post-adaptation period (15 minutes) over time, the step length symmetries of each step over 30-second intervals were averaged. Then, the group means and SDs of the step length symmetry were calculated across all the subjects. In our experiments, healthy subjects showed some variation (2.6% on average) in step length symmetry ratio during the baseline walking. Thus, to minimize the variability of gait symmetry among subjects, the baseline step length symmetry (the symmetry value measured in the last two minutes of the baseline period) was subtracted from step length symmetry data. The main goal of the present study was to compare the amount of adaptation and short-term retention (aftereffects) between the *implicit VFD+SBT* trial and the *SBT-only* trial, as well as between the *conscious VF+SBT* trial and the *SBT- only trial*. We did statistical analysis for the 7-minute and 15-minute post- adaptation periods separately. Two-way repeated measures ANOVA was performed to see if different experimental trials had a significant effect on the changes in step length symmetry. The two factors used for this analysis were the experimental conditions (different training strategies) and the level of the split-belt ratio or the visual distortion (for the analysis during the adaptation period) or the time (for the analysis during the post-adaptation period). In these analyses, Mauchly’s test of sphericity was used to test the assumption of sphericity, and the degrees of freedom were adjusted accordingly using the Greenhouse-Geisser correction before calculating the *p*-value. All statistical tests were performed at a significance level of *p* \< 0.05. The *p*-values are displayed in the main text unless depicted in Figs. Additionally, we performed a paired t-test to examine at which level of perturbation (or at which time) the two training strategies showed a significant difference in step length symmetry. # Results shows an example of changes in step length symmetry obtained from a subject as a function of time for the implicit VFD+SBT and SBT-only trials. Each point represents the mean step length symmetry (%) of strides averaged over 30-second intervals. We observed a downward trend in the step length symmetry in the 7-minute adaptation period following the first 3-minute baseline period. If the speed ratio of the two belts had not been gradually adjusted, the subject’s step length symmetry would have approached the baseline (zero step length symmetry ratio) over time. However, in our experiment, the subject showed asymmetric gait throughout the adaptation period, as the two-belt speed ratio increased every minute. In the implicit VFD+SBT trial, the visual bar of the right step length was implicitly distorted shorter than the actual step length (the subject was unaware of the manipulation), and we observed reduced asymmetric gait patterns (closer to the baseline symmetry) during the adaptation period compared to the SBT-only trial. This indicates that the subject showed visuomotor adaptation (implicitly adapted to the asymmetric visual information) by producing longer right strides than during the SBT-only trial. During the following 15-minute post-adaptation period, the subject showed aftereffects of the adapted asymmetric pattern immediately after the speed of the two belts became the same, and the step length symmetry steadily decreased (moved toward the baseline). However, the rate of de-adaptation appeared to be different between the two trials. While the step length symmetry rapidly dropped towards the baseline in the SBT-only trial, the implicit VFD-SBT trial seemed to retain the asymmetric pattern longer. shows an example of changes in step length symmetry obtained from a different subject for the conscious VF+SBT and SBT-only trials. During the adaptation period, the subject was told to match the height of the visual bars. If the subject had matched the bar perfectly, the symmetry ratio would have been very close to the baseline (zero), which was not the case for this subject. We observed that the step length symmetry rapidly decreased towards the baseline during the post-adaptation period in both trials. According to the data, we found only 16 subjects (adapted group) who showed visuomotor adaptation in response to implicit VFD while walking on a split-belt treadmill and 11 subjects (non-adapted group) who did not show visuomotor adaptation. The adapted group showed reduced gait asymmetry than that measured in the SBT-only trial during the perturbation period. On the other hand, the non-adapted group was clearly different from the adapted group in that their gait symmetry changes were similar to the typical changes shown in the SBT-only trial. shows the group means and variabilities in step length symmetry during the SBT-only trial versus the implicit VFD+SBT trial from the adapted group (16 subjects). The trend shown in was consistently observed in the group results. We first observed the extent of how the two different groups changed their step length symmetry over the course of the adaption period. Statistical analysis using two–way repeated measures ANOVA on the step length symmetry revealed that there was a significant effect of trials on the changes in step length symmetry during the adaptation period (*F(1*, *15) = 45*.*758 with the Greenhouse-Geisser correction*, *p \< 0*.*001*, *partial η*^*2* *= 0*.*753)*. There was also a significant interaction between the trial and the perturbation level (*F(4*.*299*, *64*.*485) = 0*.*931 with the Greenhouse-Geisser correction*, *p \< 0*.*001*, *partial η*^*2* *= 0*.*058*). Using paired t-tests, we then examined the differences in changes in step length symmetry between time points for the two trials. There were significant differences over the entire adaptation period *(p = 0*.*0021*, *0*.*0016*, *0*.*0000*, *0*.*0007*, *0*.*0038*, *0*.*0054*, *0*.*0069*, *0*.*0122*, *0*.*0133*, *0*.*0163*, *0*.*0229*, *0*.*0055*, *0*.*0209*, *0*.*0124*, *respectively)*. During the post-adaptation period, we observed that the implicit VFD+SBT group deadapted much slower than the SBT-only group. Also, two-way repeated measures ANOVA results showed a significant effect of trials (*F(1*,*15) = 8*.*086* with the Greenhouse-Geisser correction, *p\<0*.*05*, η² = 0.350) and significant interaction between trial and time on the rate of deadaptation (*F(7*.*49*,*112*.*356) = 3*.*550 with the Greenhouse-Geisser correction*, *p\<0*.*001*, *η*^*2* *= 0*.*191*). This shows that the retention of aftereffects changed differently between the SBT-only and the implicit VFD+SBT trials during the post-adaptation period. Paired t-test results showed the differences in the group size of aftereffects over the entire post-adaptation period except for the first six minutes. *(p = 0*.*0052*, *0*.*0141*, *0*.*0101*, *0*.*0014*, *0*.*0053*, *0*.*0026*, *0*.*0052*, *0*.*0111*, *0*.*0027*, *0*.*0020*, *0*.*0022*, *respectively)*. shows the results from the non-adapted group (11 subjects) who did not show visuomotor adaptation in response to the implicit VFD while walking on a split- belt treadmill. Statistical analysis using two–way repeated measures ANOVA on the step length symmetry revealed no statistically significant differences in step length symmetry changes over time in both the adaptation and the post- adaptation periods. shows the group means of step length symmetry changes between the conscious VF+SBT trial versus the SBT-only trial. Only 12 subjects participated in the conscious VF+SBT trial, where they were instructed to match the veridical visual feedback bars during the adaptation period. We compared the difference in the retention of aftereffects of adapted symmetric gait patterns between the two conditions. The results of the two-way repeated measures ANOVA revealed a significant main effect of the trial on step length symmetry during the adaptation period (*F(1*,*11) = 5*.*695 with the Greenhouse-Geisser correction*, *p \< 0*.*05*, *η*^*2* *= 0*.*341*), but no significant difference in the retention of aftereffects during the post-adaptation period. During the adaptation period, paired t-test results showed significant differences in changes in step length symmetry most of the time *(p = 0*.*0223*, *0*.*0134*, *0*.*0023*, *0*.*0244*, *0*.*0199*, *0*.*0279*, *0*.*0103*, *0*.*0333*, *respectively)*. shows the group means of retention (de-adaptation) rate changes during the post- adaptation period for the three trials. For computing the group means of the SBT-only trial, we did not include the data from the non-adapted group who did not show visuomotor adaptation in response to implicit VFD. However, if any subject in the non-adapted group participated in the conscious VF+SBT trial, we include their data in the SBT-only trial group mean for comparison purposes. Thus, this figure shows the group means values averaged across 12 subjects, 16 subjects, and 19 subjects for the conscious VF+SBT, implicit VFD+SBT, and SBT- only trials. Also, the step length symmetry data was normalized by an initial aftereffect value. Compared to the implicit VFD+SBT trial, the short-term retention of adapted step length symmetry was less pronounced in both the SBT- only and the conscious VF+SBT trials, indicating that the adaptation was stored less prominently. To investigate differences in the retention rate among the three experimental conditions in more detail, one-way repeated measures ANOVAs were used to examine the retention changes in the adapted symmetric gait pattern during the post-adaptation period. This test showed a significant main effect, and *post-hoc* analyses were then performed using the Bonferroni correction for pairwise comparisons among the three conditions (see marked time points). # Discussions In this study, we evaluated the potential benefits of concurrent gait training that combined a visual feedback distortion (VFD) paradigm with split-belt treadmill (SBT) walking in a single gait training by investigating the robustness of aftereffects of a learned gait symmetric pattern on a short-term basis. Understanding how gait adaptation and learning processes operate to make long-lasting changes in motor learning is crucial in gait rehabilitation. Our main finding was that there was an advantage of the implicit manipulation of visual feedback of gait symmetry in an unconscious manner for the longer-lasting effect of a newly learned gait pattern. However, not all subjects responded to the implicit VFD similarly. For the subjects who did not show visuomotor adaptation in response to the implicit VFD, the longer retention of aftereffects did not manifest. ## Effects of implicit visual feedback distortion on split-belt treadmill walking adaptation In the implicit VFD+SBT trial, subjects walked on split-belts moving at different speeds while looking at the implicitly imposed distortion of visual feedback of their step length. The right step was initially shorter than the left one due to the asymmetric speed perturbation during the adaptation period. The right step length’s visual bar was only gradually distorted by being shorter than the actual step length. The subjects were given no instructions other than to walk comfortably while looking at the screen. We observed that subjects spontaneously adapted to the asymmetric visual information by producing longer right strides than those in response to the SBT-only trial. As shown in, this made gait asymmetry closer to the baseline during the adaptation period compared to the SBT-only trial. Significant differences were also found in the retention of aftereffects between the two conditions during the post-adaptation period. The results of this study demonstrated that SBT walking with the implicitly distorted visual feedback in an unconscious manner retained the acquired asymmetrical gait pattern longer than the SBT walking only, suggesting that sensory-prediction error unconsciously driven by the VFD influences retaining aftereffects longer. We found, however, that 11 subjects among our subjects did not show visuomotor adaptation in response to the implicit VFD. Interestingly, there was no difference with the SBT-only trial in gait symmetry changes in both the adaptation period and the de-adaptation periods in this group. Although a further investigation should be conducted to ascertain why some subjects do not adapt in response to the VFD, from our previous VFD experimentation experience, visuomotor adaptation did not tend to occur when subjects seemed distracted during tests or often turned away from the screen. If a subject had not faithfully paid attention to the visual information displayed on the screen, the results would have been the same as the SBT-only trial, as shown in our results. This study did not measure whether the subjects were constantly looking at the screen, and it was not easy to quantitively measure how much they engaged in the trials. Since the implicit VFD+SBT trial was based on an environment without the subjects’ intention during the performance, we realized the importance of motivating subjects to engage in a trial. When applying our proposed method to clinical applications in the future, more research should be conducted on characterizing the experimental environment in which subjects can more consistently produce visuomotor adaptation under unconscious conditions. ## Comparison of implicit visual distortion vs. conscious correction on retention of aftereffects Gait adaptation appears to accelerate when training interventions involve subjects’ conscious (voluntary) correction in an explicit manner. Conscious repetition of a step asymmetry can cause a use-dependent bias toward aftereffects. While encouraging voluntary (consciously) corrections during training is beneficial in quickly forming new motor patterns, it has been suggested that unconscious (automatic) processes involved in motor adaptation can lead to long-lasting retention effects or reduce the amount of forgetting over time compared to conscious (explicit) processes. To understand this better, we compared the implicit visual feedback condition results to those from the conscious correction condition by assessing the rate of aftereffects from both conditions. During the conscious correction condition (conscious VF+SBT), subjects walked under the SBT-induced perturbation while consciously correcting their asymmetric gait patterns. To aid them in accomplishing this task, the veridical visual feedback displayed the bars representing the right and the left step lengths. Subjects were then instructed to match the height of the two bars on display with each step. As shown in, as the subjects consciously corrected the asymmetric gait pattern induced by SBT walking, their step length symmetry stayed close to the baseline (toward symmetric gait). This indicates that the subjects had to produce longer right and left steps during the adaptation period. In other words, the subjects had to experience a larger magnitude of gait asymmetry than in the implicit VFD+SBT trial. Nonetheless, the aftereffects of the acquired asymmetric gait pattern still tended to decay quickly, and there was no significant difference in the retention rate compared to the SBT-only trial. When we looked at the deadaptation rate among the implicit VFD+SBT, conscious VF+SBT, and SBT-only groups, our results revealed that the aftereffects in the implicit VFD+SBT trial stayed significantly longer than in two other trials. In recent years, several studies by Wood et al. investigated asymmetric gait adaptation using a similar visual feedback distortion paradigm and showed that sensory-prediction error via distorted visual feedback did not improve storage of learning when subjects’ conscious movements were involved in a training task. Interestingly, although the experimental conditions were not entirely the same, our conscious trial results seem to be consistent with their results: the conscious VF+SBT trials where subjects consciously corrected their steps while matching the bars did not improve the retention rate compared to the SBT-only trial. This may suggest that the adapted motor behavior could be more prominently stored (at least on a short-term basis) when the visuomotor adaptation process via the implicit VFD is decoupled with consciously repetitive movements. ## Potential benefits of engaging the implicit (unconscious) motor learning process in gait training Effective interventions need to facilitate a more efficient motor adaptation and ensure more persistent changes in newly learned motor patterns. Various researchers have suggested that the mechanism underlying the retention benefit of learning is the reliance on the unconscious (automatic) mode of a learning process that may accompany dual-task (e.g., distraction task), random practice, and visual perturbation. The reason why implicit learning allows learned motor behaviors to remain longer has not yet been clarified. We speculate that the brain areas underlying implicit learning develop earlier in the period of growth and are, therefore, more effective at retaining motor learning. When subjects were instructed to explicitly think about their movements, whether a given visual feedback was distorted or veridical, different brain systems could be engaged than those involved during implicit learning. However, the precise role of the implicit process in the adaptation process needs to be further elucidated through future studies. Split-belt walking is known to alter gait patterns by employing both explicit and implicit processes. During split-belt walking, however, subjects are more likely to perceive the speed difference of the split- belt, which may require a slight conscious effort and may reduce the ability to retain aftereffects longer. The SBT walking paradigm can be easily utilized to facilitate adaptation, whereas the implicit VFD paradigm within an unconscious (attention-independent) manner can have longer-lasting effects on a newly learned walking pattern. Indeed, the results of this study corroborate the effects of implicit VFD on the retention of symmetric gait patterns acquired with SBT walking. Our results demonstrated the potential advantages of combining different types of perturbation (implicit visual distortion and mechanical perturbation) in improving retention through an additive effect of each perturbation. Overall, our results support the idea that the gait changes spontaneously induced by visual distortion result in an augmented implicit learning mode, and its effect can still be sustainable when the visual distortion is implemented with split-belt treadmill walking. ## Limitations and further directions This study evaluated the aftereffects (de-adaptation) rate in integrating implicit visual feedback distortion into SBT walking and conscious correction with SBT walking. However, our study is not without limitations. The current study focused only on the spatial aspects of walking, and the temporal elements were not explored. Second, regarding the limitations of this study’s experimental protocol, the post-adaptation period was only 15 minutes, and the speed of the slower belt was fixed at 2 mph (0.894 m/s). Certainly, the longer the period, the better understanding of the longer-term effect. If subjects were tested in a preferable walking speed environment, they might have shown a different amount of motor adaptation. Third, during SBT perturbation, only the speed of the right belt was increased regardless of the subjects’ leg dominance (preferential use of one leg over another). Gait asymmetric adaptation may vary slightly from individual to individual depending on leg dominance. Fourth, although the subjects were instructed to walk while continuously looking at the VFD, we could not quantitively measure how attentive they were to the instructions. Therefore, when a subject did not show visuomotor adaptation, we cannot confirm whether it was a tendency inherent in an individual or was just because the subject was not fully attentive during the trials. Future studies will be needed to answer these questions. Finally, in our study, there may be various independent experimental variables, such as whether there is visual feedback during SBT walking, whether distortion is applied to visual feedback, and whether explicit instructions are given to subjects. Future research should require a more rigorous and better design for an experimental protocol to explicitly demonstrate the effects of each independent variable involved in our experiments. In conclusion, this study examined how combining two different forms of error- driven adaptation trials would affect storing a learned motor pattern (retention of a learned pattern). We found that SBT walking with the distorted visual feedback in an unconscious manner retained the acquired asymmetrical gait pattern longer than SBT walking only. Additional gait pattern changes made consciously during SBT walking showed little effect on increasing retention. A better rehabilitation program can be considered as one that enables longer retention of learned motor patterns than others. In this respect, our results suggest that integrating visual feedback distortion into split-belt treadmill rehabilitation could potentially enhance the outcome of gait therapy with longer retention by providing an implicit mode of learning. For better rehabilitation strategies, some form of visual feedback is often incorporated into training programs to enhance patients’ participation, and the method of using visual feedback distortion in an unconscious manner could be an effective rehabilitation program. However, further study should seek treatment strategies that can induce more consistent gait adaptation even under unconscious training conditions and determine the long-term functional effects of the visual feedback distortion paradigm on the overground transfer of an adapted step length symmetry. Further research quantifying the extent of interaction of explicit and implicit control of gait is also needed to fully understand the mechanisms underlying the rate of de-adaptation. # Supporting information The authors thank the College of Engineering at California Baptist University and appreciate valuable inputs from Elijah Brown on this study. [^1]: The authors have declared that no competing interests exist.

# Introduction Peripheral inputs contribute to kinesthesia, the sense of joint position and movements, and blocking peripheral primary afferents impairs the perception of limb position and movements. More directly, artificial activation of muscle tendons or surface cutaneous receptors induces a powerful illusion of movement. Human studies have shown that peripheral deafferented patients showed error in hand movements without visual feedback. In monkeys that have had the dorsal root transected at the level of the cervical spinal cord, precision grip is severely impaired. Thus, positional information arising from inputs of peripheral afferents is critical to achieve accurate and dexterous movements of the hands and arms of primates. Microneurographical recordings from humans or single fiber recordings from animals have been conducted to examine the responses of peripheral afferents to mechanical stimuli. Individual sensory receptors were found to respond to movements of the hand or arm. However, recording from a single afferent neuron has limitations for examining the sensory processing from an ensemble of peripheral receptors during the dynamic movements. To understand the neural processing of natural movements, it is requisite to simultaneously record the activity of a population of peripheral afferents and to investigate the computation required for those multiple receptors to represent the kinesthesia. Recent advances in multichannel recordings allow the simultaneous detection of the activity of neuronal ensembles at nerve bundles or dorsal root ganglions (DRGs). Analysis of the activity of peripheral afferent populations in cat lumbar DRGs allowed the reconstruction of the kinematic state of the leg accurately using the linear regression model. The results of these studies demonstrate that populations of sensory receptors contain rich information that represents various kinematics of each joint in the hindlimb during movements in two dimensional space. However, it remains unclear whether linear models can be used to encode complex and dexterous movements in primates from DRG recordings. To elucidate the encoding of forelimb position and movements by a population of DRG neurons, we performed multichannel recordings from the DRGs in the lower cervical segments of anesthetized monkeys. We applied the sparse linear regression (SLiR) algorithm to encode the forelimb joint kinematics from activities of DRG neuronal ensembles. The SLiR effectively and automatically selected appropriate feature sets from thousands of parameters to attain an improved generalization of performance over that obtained from other ordinary linear regression models. By selecting the optimal ensemble from recorded units, the analysis may reveal the underlying physiology in encoding of joint kinematics. We classified the recorded units into two groups, putative muscle and cutaneous units, by the response property to peripheral mechanical stimulations. We then analyzed neuronal populations selected by the SLiR, and examined the contribution of the two groups to the encoding of joint kinematics. # Materials and Methods Two adult male monkeys (Macaca mulatta, *body weight; 4.6 and 9.6 kg, respectively*) were used in this study. The experiments were approved by the animal experimental committee of the National Institute of Natural Sciences (Approved Nos.: 09A196, 10A203, 11A168) and were performed in accordance with the Weatherall report, “The use of non-human primates in research”. Before the experiments, the animals were housed individually on a 12-hour light/dark cycle and provided a rubber toy. They were not food and water deprived. ## Preparation The monkeys were sedated using a mixture of xylazine (0.4 mg/kg; Bayer Health Care, Monheim, Germany) and ketamine (5 mg/kg; Daiichi Sankyo, Tokyo, Japan), and then anesthetized with isoflurane (exhaled level; 1–2%) and nitrous oxide gas (1–2%). Atropine (0.1 mg/kg; Mitsubishi Tanabe Pharma, Osaka, Japan) and dexamethasone (0.15–0.3 mg/kg; Banyu, Tokyo, Japan) were administered intramuscularly immediately after the anesthesia. The animals were paralyzed using pancuronium bromide (Mioblock; 0.2 mg/kg/h; Schering-Plough Corporation, Kenilworth, NJ), and artificial pneumothorax was introduced and artificial respiration was provided. Expiratory CO₂ levels were monitored continuously and maintained within the physiological range (3.3–4.2%). Blood pressure was maintained above 80 mmHg. The depth of anesthesia was monitored continuously by checking the stability of blood pressure, heart rate and lack of pupillary reflex. After shaving the hair on the back and the left forelimb, a partial laminectomy was performed to expose the DRGs at the C7 and C8 segments. A lateral mass of C5–Th1 segments was dissected. Two multi-electrode arrays (Blackrock Microsystems, Salt Lake City, UT) were inserted with a high-velocity inserter through the dura into the C7–C8 DRGs on the left side. Reference wires were placed into the back muscles. After surgery, the monkeys were suspended in a spinal frame and radiant heat was used to maintain body temperature near 37°C. After finishing the recording session, the animals were deeply anesthetized using pentobarbital sodium (100 mg/kg, intravenous injection) and perfused transcardially with 0.1 M phosphate buffered saline and 10% formalin (Nacalai tesque, Kyoto, Japan), and the placement of the electrode arrays into the DRGs was confirmed. ## Neural Recording and Spike Detection The implanted arrays consisted of 48 platinized**-**tip silicon electrodes (100–1,000 kΩ at 1 kHz), arranged in a square grid (400 µm on center), 1 mm in length, and in a 5×10- configuration. The size of the array covered a DRG of 2–3 mm in diameter and 4 mm in length. The electrode arrays were connected to a 128-channel amplifier (Cerebus; Blackrock Microsystems) with a gain of 1000, and signals from each electrode were sampled at 30 kHz. Filtered waves (250–7500 Hz) above the amplitude threshold, which is 5 times the estimated background noise based on the median of the absolute value of the bandpass filtered signals, were extracted from 0.33 ms before to 0.73 ms after threshold crossing. Spikes with similar features on the principal component analysis (PCA) projection were grouped into clusters by semi-automatic spike sorting methods (Offline sorter; Plexon, Dallas, TX). If an interval between two consecutive units was less than 1 ms, we used first spikes for analysis of unit activity, even though these constituted a small population. Although a portion of the units (14.3% in Monkey 1, 8.7% in Monkey 2) responded to stimulation of distinct areas located far from one another, and were considered to be mixtures of more than one neuron, most of the units were the activity of one neuron. In some analyses, we excluded the contamination of multi neuronal activity by setting the amplitude threshold as larger values in each channel. Inter-spike intervals of units isolated by the sorting method were more than 1 ms, which implied no contamination from other neurons. Neuronal firing rates for each unit were computed at 5-ms bins, corresponding to the sampling rates of the motion capture system. ## Motion Capture Movements of the upper limb, from shoulder to fingers, were recorded using reflective markers tracked with an optical motion capture system (Eagle-4 Digital RealTime System; Motion Analysis, Santa Rosa, CA) and synchronized with the neural recordings. The system used 9 or 6 infrared cameras operating at 200 frames/s to track the position of multiple reflective markers (6-mm-diameter spheroids) with submillimeter accuracy in Monkey 1 and 2, respectively. A total of 7 markers were attached to the dorsal side of the forelimb using mild adhesive, and a comprehensive catalog of 3 anatomically defined upper extremity joint angles (elbow, wrist, finger) were analyzed. In particular, Euler angles were used to represent relative joint rotations. Elbow joint angles were calculated from two vectors (one from marker 2 to marker 1 and the other from marker 2 to marker 4 in Monkey 1; one from marker 2 to marker 1 and the other from marker 2 to marker 3 in Monkey 2). Wrist joint angles were calculated from two vectors (one from marker 4 to marker 2, a cross product of a vector from marker 4 to marker 6, and the other from marker 4 to marker 5). Finger joint angles were calculated from two vectors (one from marker 6 to marker 7, a cross product of a vector from marker 4 to marker 6, and the other from marker 4 to marker 5). The trajectory of an endpoint of the limb was obtained by subtracting the position of marker 1 from that of marker 7 in the Cartesian coordinate system (from caudal to cranial (CC), from distal to proximal (DP), and from ventral to dorsal (VD)). To reduce noises from various sources, temporal changes in the joint angles and the position of the endpoint of the limb were smoothed using a 5 Hz cutoff frequency in a low-pass digital filter. For convenience, we refer to the first and second time derivatives of joint angles and the position of the endpoint as ‘velocity’ and ‘acceleration’, respectively. ## Mechanical Stimulations The left forelimb, from forearm to digits, was moved manually. In single joint movements, passive movements were applied to one of the 3 forearm joints: the elbow, wrist, and metacarpophalangeal (MCP) joints of digit 5 (D5). These joints were repeatedly moved from the neutral position (elbow joint angle 90°, wrist 90°, finger 90°) toward the extension or flexion direction and back to the original position. Monkey 2 also underwent pronation. One trial took 5 s in Monkey 1 and 3 s in Monkey 2, from 2 s and 1 s before the stimulation onset, respectively. One session was composed of 26–30 trial repetitions. To investigate more complex, compound movements, the left forearm of Monkey 1 was moved sequentially and arbitrarily by an experimenter at 7 forearm joints, including the elbow, wrist, and MCP joints of the 5 digits for 10 min. Compound movements of the forearm were performed in 2 sessions. Tactile stimulations were applied 29–30 times to the skin surface of the forelimb at various sites with a paint brush. ## Linear Regression Joint angle, velocity, and acceleration were modeled as a weighted linear combination of neuronal activity using a multidimensional linear regression as follows:where; y*_j*(*t*) is a vector of kinematic variables *j* (joint angle, velocity, acceleration of elbow, wrist, finger) at time index *t*. x*_k*(*t+lδ*) is an input vector of unit *k* at time index *t* and time-lag *lδ* (*δ* = 5 ms). w*_j,k,l* is a vector of weights on unit *k* at time-lag *lδ*, and b*_j* is a vector of bias terms to y*_j*. The units that showed no more than one spike in the training data sets were omitted before the regression analysis. Because external stimulation was the cause of any afferent activity, time-lag was set at future, positive values. According to the goodness of reconstruction of the kinematic data from the neural activity as a function of the input-signal duration, we set the duration of signals at 100 ms (maximum *l* = 20). When we changed the length of the time window, the plateau level in the accuracy of estimation of joint kinematics was achieved at 100 ms. If we consider the conduction velocity of afferent nerves to be more than 10 m/s and their length to be ∼30 cm, 100 ms is presumed to be sufficient. This is probably because good prediction of the encoding of joint kinematics for 3-dimensional movements required sufficient amounts of DRG activity, but the firing frequency of individual DRG neurons was quite low. A set of linear weights were estimated by the linear regression method from a set of training input (unit activity) and output (kinematic variables) data. For comparison with the SLiR method, we also adopted a regularized linear regression method. The regularization parameter was determined from the data using a Bayesian estimation method. After low-pass filtration of the acquired reconstruction at 5 Hz, we compared the predicted kinematics to the observed kinematics for validation of the model. ## Sparse Linear Regression (SLiR) An SLiR algorithm used in other studies, was applied to analyze the population coding of DRG neurons. The SLiR effectively and automatically selected appropriate feature sets and pruned less important signals from thousands of explanatory variables to attain a better generalization performance compared to the regularized linear model. This is because having too many parameters relative to the limited number of training data sets is known to lead to poor generalization performance (over-fitting problem). We used a Bayesian SLiR algorithm that introduced sparse conditions only for the unit dimension, and not for the temporal dimension (see 35). This method estimated the weight and the automatic relevance determination (ARD) parameters, which represented how much the weight contributes to the reconstruction. Based on the values of the ARD parameters, relevant features were selected automatically and irrelevant features were discarded. ## Data Analysis In regression analyses of single joint movements, an SLiR model was generated for each of the kinematic parameters investigated (joint angle, velocity, and acceleration). The models generated in the training data sets were tested against a test data set. The training data sets were composed of 40 randomly selected trials (20 extension movement trials and 20 flexion movement trials) for each joint movement (a total of 120 trials). The test data sets were composed of 6 randomly selected trials (3 extension trials and 3 flexion trials) in each joint movement (total of 18 trials). As the detection of some reflective markers were missed in Monkey 2, data of wrist flexion, pronation, and digit 5 joint extension and flexion movements were used in wrist joint encoding, and data of wrist flexion and digit 5 joint extension and flexion movements were used in digit 5 joint encoding. In regression analyses of compound movements, continuously recorded data were partitioned into 20 trials (one trial for 25 s data). Among the 20 trials, 17 randomly selected trials were used for the training data sets, and the remaining 3 trials for the test data sets. To assess model generalization, we used data from 3 different movement blocks (single joint movements and compound movement sessions 1 and 2) of Monkey 1. The test data sets were built from the different movement blocks of the training data sets. For example, a model was constructed from compound movement session 1, and the model was tested using a data set from session 2. In characterizations of units selected by the SLiR, we surveyed the training and test data sizes to evaluate the test performance. When the test performance was evaluated using a limited number of data samples, there were trade-offs between the test performance and accuracy of the evaluation. The test performance increased as the training data size increased, while the accuracy of the test performance evaluation increased with increasing test data size. Therefore, we evaluated the balance between the training and the test data sizes by changing the sizes of the training and test data sets. To do this we constructed data sets from combinations of the different movement blocks of Monkey 1, using 150 5-s trials from the single joint movements and 100 5-s partial trials from each session of compound movements to construct pooled data totaling 350 trials. To quantify the generalization and survey the training and test data sizes, we applied simultaneously the SLiR to data composed of 9 joint kinematics data (angle, velocity and acceleration of each of 3 joints). In other cases, we applied individually the SLiR to the kinematics data of each joint. Prediction accuracy was evaluated as the correlation coefficient between the observed kinematics of test data sets and the predictions from the model. The root mean squared error (RMSE) between the observed kinematics and the predictions from the model was also calculated to assess the prediction accuracy in some analyses. To assess the model, 5 pairs of training and test data sets were generated, and a 5-fold cross-validation was performed in all the analyses. In control analyses of model prediction, we created surrogate training and test data sets in which temporal firing profiles of individual neurons were shuffled independently across different trials and tested subsequently for their prediction of each kinematic parameter. To infer the importance of individual neurons in reconstruction, we defined two indices: (i) the corresponding weight value determined through the regularized linear regression analysis with population data and normalized by the power of the unit activity in the training data sets, and (ii) the correlation coefficient between the observed kinematics and the predictions derived from single unit activity and corresponding weight values. In encoding of the joint kinematics by an individual class of sensory neurons, the number of putative muscle units used in the model training was matched to that of the putative cutaneous units to compare the contribution of the putative muscle and cutaneous units. Using random selection, we produced 20 data sets of identical numbers of putative muscle and cutaneous units. We then fit the SLiR using individual data sets and generated a new prediction. ## Statistical Analysis The data were analyzed using a two-way analysis of variance (ANOVA), the non- directional unpaired- or paired Student’s t-test, with Bonferroni correction if necessary. An alpha level of significance was set at 0.05 for all statistical tests. Data are expressed as the mean ± standard deviation (mean ± S.D.) or the mean ± standard error (mean ± S.E.). We found 95% confidence intervals for proportions based on the inverse of the appropriate cumulative Beta distribution. We used Matlab (Mathworks, Natick, MA) for the statistical analysis. # Results Recordings were obtained from the DRGs at the C7 and C8 segments with two multi- electrode arrays. A total of 112 units (39 from C7, 73 from C8) were discriminated from 43 channels in Monkey 1 and 92 units (38 from C7, 54 from C8) were discriminated from 44 channels in Monkey 2. ## Reconstruction of the Forelimb Joint Kinematics from Activities of the DRG Neuronal Ensembles using the SLiR Model Population recordings during passive movements showed that the temporal discharge patterns of individual isolated units were correlated with temporal changes in the joint angles and that the temporal changes in the firing of each unit varied among the isolated units. To examine whether neuronal ensembles in the DRGs convey rich information about joint kinematics, we applied the SLiR model to the encoding of kinematic variables from the activities of all the single and multiple units. As the activation of peripheral afferents was induced by external stimulation, we considered that the peripheral afferents carried information concerning limb position immediately before their firing. Therefore, the kinematic variables were defined as a weighted sum of neural firing for the upcoming 100 ms (here grouped into 20, 5-ms bins) in the SLiR. shows the result of the encoding of elbow joint angle, velocity, and acceleration in a test data set composed of multiple single joint movements, from the activities of a neuronal ensemble. The SLiR provided accurate predictions of the joint kinematics. The prediction performance of the test data sets (test performance) from the actual neural firing pattern was much better than that from the shuffled data (paired student’s t-test; p\<0.0001;). Next, we applied the SLiR to neuronal activities recorded during compound movements of the forelimb to encode temporal changes of the joint kinematics. illustrates the results of encoding of the elbow joint angles, velocity, and acceleration from ensemble activities in a test data set of compound movements. Even in complicated, three-dimensional movements, prediction of the joint kinematics was accurate. The test performance from the actual neural firing pattern was much better than that from the shuffled data (paired Student’s t-test; elbow and wrist; p\<0.0001, finger; p\<0.05;). In the compound movements, a more accurate prediction of the joint kinematics was obtained the greater the distance the joint was from the end of the limb. In both the single joint movements and the compound movements, the prediction accuracies of angle and velocity were higher than that of acceleration at the three joints. These results demonstrate that neuronal ensembles in the DRG conveyed rich information about joint kinematics, especially for angle and velocity. Finally, we tested whether kinematics of the limb endpoint in the Cartesian coordinate system were encoded by the DRG neuronal activity using the SLiR. Examples of encoded kinematics of the endpoint are shown in and in single joint movements and compound movements, respectively. The SLiR also provided accurate predictions of kinematics of the limb endpoint in both movements. The test performance from the actual neural firing pattern was much better than that from the shuffled data (paired student’s t-test; p\<0.001;). These results demonstrate that the activity of neuronal ensembles in the DRGs also carried rich information about the kinematics of the limb endpoint. ## Improved Generalization in Reconstruction of the Forelimb Joint Kinematics using the SLiR The SLiR reduced the number of inputs used in the prediction. Note that the numbers of selected units were higher in the prediction of angle and velocity than that of acceleration (paired Student’s t-test with Bonferroni correction (n = 3); p\<0.0001), which were correlated with the prediction accuracies of the respective kinematics. In any case, the SLiR accurately encoded temporal changes in the joint kinematics from activities of the DRG neuronal ensembles using reduced numbers of units. To validate the importance of the units selected by the SLiR, we compared the extent of generalization between the SLiR and the regularized linear regression, which used all of the recorded units to encode the joint kinematics. We used data from 3 different movement blocks (single joint movements and compound movement sessions 1 and 2) of Monkey 1. To assess the generalization of the encoder, an encoder was tested using data sets from the different movement blocks. When an encoder was constructed from compound movement session 2 and the performance of the encoder was tested using data sets from compound movement session 1, the SLiR predicted the joint kinematics more accurately than did the regularized linear regression. When training data sets were built from approximately 85% of the total data, model performance of the SLiR was statistically better than that of the regularized linear regression model in 5 combinations of training and test data sets (paired Student’s t-test; p\<0.05). Thus, the SLiR extracted a limited number of units from the DRG ensemble to encode the joint kinematics with an improved generalization performance. ## Extraction of Appropriate Units to Reconstruct Forelimb Joint Kinematics using the SLiR Based on a high encoding accuracy with a limited number of units selected by the SLiR, we examined which units were selected from the total sets of recorded neurons. In the following analysis, we used units which were isolated using the more strict sorting rule to reduce the possibility of contamination of any other neuronal activity into isolated single neurons (69 units in Monkey 1 and 74 units in Monkey 2). Using this data set, superior generalization performance of the SLiR was also achieved. We first surveyed the training and test data sizes to evaluate the test performance. We analyzed the entire data sets from Monkey 1, because we obtained a larger amount of data sets from this animal than the other. The relationship between the size of the training data sets and the prediction accuracy indicated that the performance achieved a plateau at 140 trials (corresponding to recording for 700 s) per training data set. At the data size, the superior generalization performance of the SLiR was also confirmed. Using 140 trials as the size of the training data set, we calculated two indices based on univariate statistics, (i) weight value normalized to the power of the unit activity and (ii) the correlation coefficient between the observation and the prediction by individual single neuron activity. show the distributions of the respective indices for individual single neurons. In both the normalized weight values and the correlation coefficient, units selected by the SLiR were statistically larger than those of the pruned units (Student’s t-test; p\<0.0001). The SLiR selected mainly units that contributed to the reconstruction of joint kinematics or determined the outline of temporal changes of these kinematics. Note that some of the selected units had low scores in the univariate analysis, indicating that they contributed less to the reconstruction individually, but might still play important roles as part of the ensemble. Finally, we examined what DRG neurons were selected by the SLiR. We classified the recorded units into two groups based on their response property to the peripheral stimulations. We first determined whether units responded to the peripheral stimulation by experimenters and then confirmed the presence of statistical significance (p\<0.05) between before and during the stimulation using the Wilcoxon signed-rank test. If units responded to a soft brush, they were classified as the putative cutaneous units, while units that responded to passive movements only were classified as the putative muscle units. When units did not fire sufficiently for us to categorize them into either of two classes, they were classified as unidentified units. The numbers of units belonging to putative muscle units, putative cutaneous units and unidentified units were 26, 14 and 29, respectively, in Monkey 1, and 41, 22 and 11, respectively, in Monkey 2. The composition of modalities selected by the SLiR for encoding of joint kinematics using 140 training data sets are shown in. Compared to the composition of the originally recorded units, putative muscle units comprised the majority of the selected units (original, 37.7%; selection, 58.0% (50.0 to 65.6 as confidence intervals for proportions)). The SLiR selected a majority of the putative muscle units that were recorded (66.9% (58.4 to 74.4)). These results suggest that putative muscle units contributed much to the encoding of joint kinematics. Although the proportion of putative cutaneous units was reduced following the SLiR selection (21.4% (13.5 to 32.4)), the selected units included a number of putative cutaneous units, suggesting that they also contributed substantially to the encoding of joint kinematics. ## Contribution of the Putative Muscle Units and the Putative Cutaneous Units to Encoding of Joint Kinematics To confirm that the putative muscle units or the putative cutaneous units individually encode joint kinematics, we applied the SLiR to encoding of joint kinematics from neuronal activities of each category. Two-way ANOVA revealed significant main effects of category (*F*_{(2, 243)} = 8.2, *p* = 0.0004) and kinematics (*F*_{(8, 243)} = 5.1, *p*\<0.00001) with no significant interaction (*F*_{(16, 243)} = 0.76, *p* = 0.73). A post- hoc test showed that the prediction accuracy for encoding of elbow joint kinematics by the putative muscle units activity was higher than that from the putative cutaneous units (paired Student’s t-test with Bonferroni correction (n = 3); p\<0.05). To examine the contribution of the putative cutaneous units to encoding of joint kinematics, we compared the prediction accuracy by the SLiR using neuronal activities of both the putative muscle and the putative cutaneous units to that of the putative muscle units alone. By adding the neuronal activity of the putative cutaneous units to those of the putative muscle units, the SLiR selected both the putative muscle and cutaneous units and the prediction accuracy in all the kinematics was significantly improved (paired Student’s t-test with Bonferroni correction (n = 3); p\<0.05). These results suggest that, while a great deal of elbow joint kinematic information was conveyed by the putative muscle units, the putative cutaneous units provided the central nervous system with joint kinematic information that the putative muscle units may not code, such as subtle forelimb movements accompanied with skin deformation. In encoding of wrist and finger joint kinematics, the prediction accuracy from the putative cutaneous units was similar to that from the putative muscle units or superior in encoding of finger velocity (paired Student’s t-test with Bonferroni correction (n = 3); p\<0.05). Comparison of the prediction accuracy by the SLiR using neuronal activities of both the putative muscle and the putative cutaneous units to that of the putative muscle units alone also indicated contribution of the putative cutaneous units to encoding of some joint kinematics (paired Student’s t-test with Bonferroni correction (n = 3); p\<0.05) . Furthermore, in the prediction of joint kinematics by the SLiR using neuronal activities of both the putative muscle and the putative cutaneous units, the SLiR selected both of them. These results suggest that, in wrist and finger joints, the putative cutaneous units provided the central nervous system with joint kinematic information as did the putative muscle units. # Discussion The results of the present study demonstrate that temporal changes in angle, velocity, and acceleration of various forelimb joints can be reconstructed from the population activities recorded from cervical DRGs in monkeys using the SLiR. The analysis provided improved generalization performance in the reconstruction of various joint kinematics from a subset of somatosensory neural activity. The analysis elucidated that, not only the putative muscle units, but a number of the putative cutaneous units also contributed to reconstruction of joint kinematics. The result was confirmed by reconstruction of joint kinematics from activity of individual classes of units. The putative muscle units contributed greatly to encoding of elbow joint kinematics, but the putative cutaneous units provided additional information regarding the joint kinematics, especially at wrist and finger joints. ## Multichannel Recording from the Cervical DRGs The activity of peripheral afferents have been recorded by single fiber recordings in human and animals, in which isolated afferent nerves were recorded by inserting a single microelectrode into the peripheral nerve. These studies identified the response properties of each recorded afferent fiber to various external stimulations or voluntary movements. However, to produce kinesthesia, the sensory information is thought to be assembled in the central nervous system. The coding of positions and movements of the ankle joint by populations of sensory neurons was examined by collecting a number of separate microneurographical recordings during similar movements. Although these studies discriminated movement directions in two dimensional space at a single joint of a leg from the collected recording data, application of the same methodology to the prediction of more complicated natural movements of hand and arm is difficult. For the first time, we recorded simultaneously from approximately 100 sensory neurons using multi-electrode arrays from the cervical DRGs of non-human primates. High variability in the ensemble recording enabled us to calculate various joint kinematics in three dimensional space, especially of compound multi-joint movements. Furthermore, while the microneurographical recording is technically difficult to be performed on subjects who are moving their limbs, we recorded neuronal activity during dynamic movements of the entire forelimb. Thus, the recoding of sensory neural activity using multi-electrode arrays from the DRGs in monkeys affords a new experimental paradigm to explore sensory coding by peripheral afferents as one possible alternative for microneurography. A series of studies examined the encoding of cat hindlimb movements from ensembles of lumbar DRG neurons, calculating position and velocity of the foot in two dimensional space from up to 100 discriminated sensory neurons. We applied a similar recording technique to the DRG neurons at the cervical segments of monkeys, but the forelimb movements in three dimensional space were more complicated than those in the cat studies. Since prediction of movements in three dimensional space from neural activity is more difficult than that in two dimensional space, the prediction accuracy was slightly lower than in the encoding of cat hindlimb movements. On the other hand, prediction of forelimb joint angle or trajectory of the limb endpoint in reaching and grasping movements in three dimensional space from ensemble recordings of the motor cortex provide correlation coefficient values around 0.7, which is similar to the prediction accuracy in our results. Thus, we succeeded in reconstructing elbow, wrist, and finger joints kinematics during movements in three dimensional space using DRG population activity. ## Encoding using SLiR We demonstrated encoding of forelimb joint kinematics from multichannel recordings of DRG neurons. The elbow and wrist joint kinematics were estimated with higher accuracy than that of the finger. It is possible that higher correlation between the elbow and wrist motions accounts for the higher accuracy in estimation of kinematics of these joints. When the joint movements were correlated, it is difficult to evaluate prediction of kinematics at each joint independently. However, the elbow and wrist motions were not more correlated with each other (0.13 and 0.09 in session 1 and 2, respectively) than other combinations (elbow and finger; 0.33 and 0.32, respectively, wrist and finger; −0.26 and −0.20, respectively). Furthermore, a principal component analysis of joint movement trajectory showed that three joint motions were almost independent in compound movement session 1. In compound movement session 2, their motions were coupled, but the degree of freedom for their motions were three. Thus, we conclude that the result of encoding properties of each joint was not dependent on their movement profile, but due to the discharge pattern of recorded ensembles. The forearm trajectory of the monkey was predicted previously with high accuracy from activities of subsets of motor cortical neurons obtained by multichannel recordings using the SLiR. The authors showed that the SLiR outperformed that of the conventional Wiener filter, which is consistent with our analysis. Their and our present studies suggest that the SLiR selected important units as an ensemble. The ensemble selected by the SLiR was different from units ranked by the conventional univariate analysis, which contained units whose activities were less correlated to joint kinematics as individual units. Although some units with weaker correlation showed high contribution to the prediction of joint kinematics, these units did not contribute to the whole structure of the joint kinematics but may be relevant to local changes within them. The improved generalization performance of the sparse method has also been accounted for by the correlation structures among ensemble components from the analysis of fMRI data. Reduction of input parameters without any deterioration of coding performance gives a benefit to BMI systems that require less computational burden due to a limited hardware. Arm trajectory can be predicted from the activity of a small number of neurons in the primary motor cortex of primates. The optimal number of neurons to achieve high model performance has been determined by neuron dropping analysis,. In this analysis, the linear models were generated from the remaining units after randomly removing a single unit until one neuron was left. The procedure was repeated many times for each ensemble to obtain an optimal ensemble size. One problem is that the method requires significant computational resource and has less practical use in the situation in which composition of chronically recorded units was variable during the experimental period. The SLiR automatically selected important units from the entire recoded population without affecting model performance through machine learning. Thus, the SLiR is a practical method in analyzing data containing an enormous number of recorded neurons. ## Encoding of Joint Kinematics by the Putative Muscle Units and the Putative Cutaneous Units We classified recorded units into two classes based on their response properties to the peripheral stimulations. Identification of the unit modalities and receptive fields of approximately one hundred units individually as in single afferent recordings is very time consuming. Therefore, we did not apply the identification method to the multichannel recording. Instead, we applied the mechanical stimulation to most of the surface skin of forelimb with a paint brush, and analyzed the response property of each unit after recording. Although the latter offline analysis was not so strict to identify fine receptive field, we considered that the units that responded to tactile stimulation were classified as putative cutaneous units. These units were assumed to be equivalent to cutaneous receptors. The rest of the units that responded to joint movement were classified as putative muscle units, thought to contain muscle spindle, tendon or joint receptors, while muscle spindles may be more commonly encountered. The putative muscle units tended to be more sensitive to passive movements than the putative cutaneous units. Chronic implantation of array electrodes into the DRGs may allow us to identify unit modalities and receptive field definitively. Muscle receptors are generally considered to be the major source of movement perception. The notion partially stems from the evidence that tendon vibration induces a strong illusion of position and movements,. In this study, we showed that the putative muscle units were the major components that encoded the joint kinematics and test performance of elbow joint kinematics prediction was higher from the putative muscle units than the putative cutaneous units. The richness of kinematic information encoded by muscle spindle receptors may be of great value in human psychophysical studies. Nevertheless, cutaneous receptors also encoded additional important positional information. Human experiments showed that specific activation of cutaneous receptors by local stretch and electrical stimulation induced movement illusion. Simultaneous skin stretch and tendon vibration increased the illusory movement compared with vibration only. Thus, the present results that both the putative cutaneous and muscle units contribute to encode joint kinematics are consistent with the results in psychophysical experiments. Ensembles of cortical neurons in the primary somatosensory cortex in macaque monkeys have provided accurate information about limb kinematics. Cortical neurons in the primary somatosensory cortex are divided mainly into two classes in their receptive fields, cutaneous and proprioceptive receptive fields. Weber and colleagues showed that two classes of cortical neurons contributed equally to encoding of kinematics of the limb endpoint. In the present study, both the putative cutaneous and muscle units also contributed to encoding of joint kinematics at the peripheral level. These results suggest that information about limb kinematics coded by both cutaneous and proprioceptive afferents is conveyed to the cortical neurons with each modality in the primary somatosensory cortex, respectively. Position sense may be different between distal and proximal regions, especially in the forelimb. The extent of contribution of the putative cutaneous units to encoding of joint kinematics tended to be higher in distal forelimb region than proximal one. This may be explained by the physical structure of forelimb muscles. As finger muscles, such as the extensor digitorum muscle, control multiple digits, tendon vibration to the finger extensor induces simultaneous illusory movements of these digits. Additional skin stretch to the tendon vibration focuses the sensation of movement to the stretched digit. These results suggested that cutaneous receptors contributed to the positional sense of fingers to a substantial extent. It is well known that two-point discrimination threshold is lower in distal body parts than proximal ones. One explanation of this is that distal parts of the human forelimbs are much more densely innervated by mechanoreceptors than the proximal parts. On the other hand, the number of muscle spindles acting on the joints decreases toward the end of the limbs. Relative high density of peripheral mechanosensory receptors and low density of spindles in the distal forelimb compared to the proximal one may increase the contribution of cutaneous receptor activation to distal kinesthesia compared to those in proximal parts of the body. In addition, results of the present study showed that the putative cutaneous units encoded joint kinematics of distal forelimbs to a greater extent, which suggests that single distal cutaneous units are more important in sensing body position and movements than proximal cutaneous units. These features may constitute high sensitivity to skin deformation in the distal body parts for dexterous manipulation. We thank T. Sakatani for technical support in operating the motion capture system. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: TU KS MK TI. Performed the experiments: TU KS TI. Analyzed the data: TU. Contributed reagents/materials/analysis tools: MS. Wrote the paper: TU KS MS YN MK TI.

# Introduction P2X receptors, ion channels activated by extracellular ATP and permeable to Na⁺, K⁺ and Ca²⁺, are involved in a large array of physiological functions including presynaptic modulation, , synaptic transmission, smooth muscle contraction, inflammation, , cancer, intestinal motility, taste, pain, and the regulation of immune and cardiovascular, responses, and therefore, are potential therapeutic targets of many diseases. Today, seven genes have been identified to encode distinct P2X subunit isoforms, denoted P2X1 to P2X7, with functional P2X receptors formed as homotrimers or heterotrimers of those subunit isoforms. All P2X subunits share a common topology, characterized with a large, glycosylated, disulphide-rich extracellular (EC) loop, two transmembrane (TM) domains and intracellular N- and C- termini. ATP binds to extracellular subunit interfaces and evokes a conformational change of P2X receptor, leading to the opening of non-selective channel pore formed by TM domains. Notably, the seven homomeric P2X receptors differ considerably in their biophysical and pharmacological properties, and the formation of heteromeric P2X receptors further confers the divergence in channel properties. Determination of ATP-recognition (AR) is a prerequisite for better understanding physiological and pharmacological functions of these homo- or hetero-trimeric P2X receptors. The EC loop of P2X receptors has ∼280 amino acids, containing no known consensus sequences for agonist binding identified in other ATP-sensitive protein. Experiments on chimeric receptors have narrowed down the region of the ATP-binding sites from K67 to K313 (rat P2**X**4 receptor numbering). Using alanine-scanning mutagenesis, further studies identified eight residues having the potential to participate in the formation of the ATP binding site: K67, K69, F185, T186, N293, F294, R295 and K313 (rat P2**X**4 numbering). The positively charged lysines coordinate the binding of the negatively charged triphosphate of ATP, while aromatic phenylalanine residues could coordinate the binding of the ATP adenine ring. The open structure of zebra fish P2**X**4 (zfP2**X**4) receptors partially confirmed this predicted AR mode of P2X receptors. Besides the AR mode described above (we called it AR1), at least two other AR modes may take place when ABP is exposed to a certain concentration of ATP. Jiang *et al.* have located ATP-binding sites in rat P2X2 (rP2X2) receptor using an engineered affinity labeling approach. They reported two previously unidentified residues N140 and L186 (rP2X2 numbering) from two adjacent subunits separated by about 18 Å in the rP2X2 homology model at the resting state, suggesting the existence of at least two distinct AR modes. One shows similar features to AR1, while the other mode exhibits distinct characters. Additionally, numerous *in silico* docking studies supported the presence of a third AR mode, in which the adenine ring of ATP is deeply buried into the hydrophobic region at the interface of rigid lower body domains from two subunits. All of these observations imply that ABP recognizes ATP through different modes. However, why only AR1 can trigger P2**X**4 channel activation and how the architecture of ABP favors the occurrence of AR1 rather than other AR modes still remain uncertain. Further investigation on the underlying mechanism of the AR determining will help to understand the ligand recognition of these ATP-gated trimeric ion channels. Taking advantage of the resolved crystal structures of zfP2**X**4, here we then study the structural determinant of AR of P2**X**4 receptors using a diversity of computational approaches, including *in silico* docking, molecular dynamics (MD) simulations, normal mode analysis (NMA), conformational sampling and binding free energy (*Δ*G_bind) calculations. Our results reveal a potential role of inherent dynamics of head domain in AR determining of P2**X**4 receptors. # Results ## Three Distinct ATP-recognition Modes of zfP2X4 Receptor: AR1, 2 and 3 The closed structure of zfP2**X**4 receptor reveals a large ATP-binding pocket (ABP) at the interfaces of two subunits. ABP is formed by the head and upper body domains of one subunit, and the lower body and dorsal fin domains of another subunit. It is widely accepted that the conspicuous cluster of basic and polar residues, R298, K316, K70, K72 and N296, constitute a pocket specific for recognizing the triphosphate moiety of ATP. However, as revealed by closed structure of zfP2**X**4, there are at least three sites in ABP, S1, S2 and S3, can dock the adenine ring of ATP. We established those three recognition modes using *in silico* docking, AR1, 2 and 3, to show initial interactions between ATP and those three sites. In our modes, AR1 had similar features with the open crystal structure of zfP2**X**4, in which the adenine ring made contacts with site S1 (consisting of L191, K70, K72, I232 and T186). For AR2, the adenine ring of ATP is deeply buried into the interface of rigid lower body domains and contacts with I94, F297, Q97 and Y295, a recognition mode proposed by numerous computational studies. AR3 showed similar features revealed by covalent binding of NCS-ATP to engineered cysteine residues in the putative ATP-binding sites of rP2X2, where the adenine ring of ATP contacts with site S3 (W167, I173, L170, D145 and E171). We further characterized these three distinct AR modes through additional computational approaches. It has been reported that ATP may bend into ‘U’ or ‘V’-shaped structure with the *β*- and *γ*-phosphates folded towards the adenine ring, as observed in ATP-bound opening zfP2**X**4 and class II aminoacyl transfer RNA synthetases. Our *in silico* docking studies showed that the bound ATP could adopt bent conformations when adenine ring of ATP made contacts with any one of those three sites despite the differences in the bending angle, suggesting that sites S1, 2 and 3, together with the charged region, are capable to fit bended ATP. Additionally, as revealed by combining the docking and *Δ*G_bind calculations, AR2 induced a little more free energy release than AR1 did (*Δ*G_bind = −33.6±4.1, −43.2±3.9 and −28.8±4.4 kcal/mol for AR1, AR2 and AR3, respectively), indicating higher ATP-affinities of AR2 than AR1, at least in molecular simulations. Thus, the *Δ*G_bind analysis does not support the dominant presence of AR1. Additionally, ATP was also docked into the closed structure of zfP2**X**4 that has been previously sampled and averaged on short timescale (2-ns) MD simulations, and an induced fit docking strategy was also performed on this closed structure. *Δ*G_bind calculations indicated that those two docking strategies still did not support the dominant presence of AR1, implying that more dynamic features in ABP are responsible for the dominance of AR1 over AR2 and 3. ## Inherent Dynamics of Head Domain is Able to Change the Preference of ABP for AR1, 2 and 3 As revealed by crystal structures of zfP2**X**4, both closed and open structures, head domain exhibits higher B-factor values than other domains involved in forming ABP, such as the lower body domain and dorsal fin domain. Previous studies have demonstrated the critical role of head dynamics in channel activations. Here, taking account of its role in ABP formation, we proposed that the dynamics of head domain would affect the recognition of ATP. To test this hypothesis, we extensively investigated the relationship between the inherent dynamics of head domain and AR of P2X receptors using MD, NMA and *in silico* docking. As revealed by low frequency modes of NMA, head domain displayed three possible dynamic modes: a downward motion, a rotation around the central axis of itself, and a rotation around the central axis of protein. MD simulations confirmed a more notable conformational fluctuation of head domain than that of body, left flipper, right flipper and dorsal fin domains in the absence of ATP. Moreover, a spontaneous downward motion of head domain was observed in MD simulations, indicating the preference of zfP2**X**4 to the downward motion of head domain even without ATP binding. In open structure of zfP2**X**4, ATP binding induces slightly downward motions of the head domain and subsequently closing of ATP-binding jaw, confirming the idea that the downward motion may be the inherent dynamics of head domains. To further test the contribution of head domain dynamics to determination of AR in P2**X**4, six representative conformations were sampled by getting the ‘average’ conformations around 0, 12, 24, 48, 100, and 250 ns upon the initiation of spontaneous downward motion of head domain during MD simulations. ATP was docked into the sampled conformations of zfP2**X**4, and various ATP- poses were then characterized and plotted according to their total docking scores and *vdW* scores of ATP contacting with L191 (L191_vdW) and I232 (I232_vdW) (see, each colorful dot represents an ATP pose). Poses with lower values of docking score (\<−7), and lower values of L191_vdW and I232_vdW scores (\<−1) (region III) were regarded as rational AR1 mode in which the adenine ring of ATP makes a contact with L191, I232 and nearby hydrophobic residues while the triphosphate moiety contacts with clustered basic residues. On the other hand, lower values of docking score (\<−7), and higher values of L191_vdW and I232_vdW scores (\>−1) (region I) indicated that the adenine ring of ATP makes no contacts with these two amino acid, reflecting rational AR2 or 3 modes. Along with the motion of head domain, the top ATP poses with the lowest docking score within region I (AR2 and 3) and III (AR1) were obviously altered (within the highlighted circles). For initial conformation of zfP2**X**4 (0 ns), the top ATP poses distributed both in region I and III showed comparable lowest docking scores. However, the top ATP poses only occurred within region III instead of region I along with the downward motion of head domain, indicating that this motion favors the AR1’s occurrence rather than AR2 and 3. These data suggested that inherent conformational fluctuations of head domain, especially the downward motion, can alter the preference of ABP for AR1, 2 and 3. A superposition of closed structure and equilibrated representative structure after 300-ns MD simulations showed that the downward motion of head domain was followed by a marked change in the orientation of R143 side chain, a leftward movement of loop_139–146 (a region within the head domains) and a rightward movement of loop_169–183 (a region covalently linking with the head domains). The closing movement between these two loop regions dramatically changed the shape of site S3, which prevented ATP from contacting with site S3. It also brought the basic residues K70, K72 and R298 as well as *β*1 and *β*12 close to each other, which significantly narrowed entrance of ATP into site S2 and prevented contacts between the adenine ring of ATP and site S2. On the contrary, the downward motion of R143 and rightward movement of loop_139–146 facilitated the contacts between ATP and site S1 via a mechanism of partially trapping adenine ring of ATP into the narrowed cleft formed by loop_139–146 and dorsal fin domain. Meanwhile, the altered orientations of K70, K72 and R298 would favored coulombic interactions between triphosphate moiety of ATP and basic groups of those three residues. These allosteric changes resulted in an increased preference of ABP for AR1 and prevent ATP from interacting with sites S2 and S3. ## AR1 Instead of AR2 and 3 Favors the Downward Motion of Head Domain and Downstream Allosteric Changes Associated with the Channel Activation of P2X4 Receptors We also found that AR1 rather than AR2 and 3 can promote the inherent downward motion of head domain. A combination of 10-ns MD simulations of and PCA analysis of snapshots from simulations revealed that AR1-induced a downward movement of the head domain, upward motion of dorsal fin domain and a subsequent closing of ATP-binding site jaw, consistent with previous observations. Meanwhile, it has been well established that those allosteric changes are associated with downstream opening-related allosteric changes, such as the radial expansions of extracellular vestibule and the final iris-like channel opening. The potential energy calculations on resting and open states of zfP2**X**4 receptors revealed that AR1-induced allosteric changes led to energy releases during conformational transition between the resting and open states. In contrast, MD simulations together with PCA analysis showed that AR2 and 3 can only induce outward and upward movements of head domain, respectively. Therefore, AR2 and AR3 would not trigger downstream allosteric changes. These observations provided a conceivable explanation for why only AR1 rather than AR2 and 3 is able to trigger channel activation of P2**X**4 receptors. The *Δ*G_bind together with the downstream allosteric changes-induced energy releases may enable AR1 to efficiently overcome the energy barrier for the channel gating. Meanwhile, based on the observation that conformational fluctuations of head domain, especially the downward motion, greatly enhance the preference of ABP for AR1, we proposed that the downward motion of head domain facilitated by AR1 may further increase the preference of ABP for AR1. It looks like an induced- fit/positive feedback mechanism when AR1 occurs. Indeed, when we compare conformations of resting sate and the open state as well as their corresponding *Δ*G_bind values of ATP, AR1-induced downward movement of the head domain, upward motion of dorsal fin domain and the closing of ATP-binding site jaw in open state led to significant increase in free energy release than that in resting state (*Δ*G_bind of ATP = −33.6±4.1 and −46.79±7.7 kcal/mol, for resting and open states, respectively). This increased free energy release upon ATP-binding may attribute to the dynamics of head domain together with dorsal fin upward movement-induced partial trapping of the adenine ring of ATP into site S1 of ABP. Taken together, the contact between ATP and site S1 (AR1) promoted downward motion of head domain correlated well with conformational changes of ABP and subsequently increased preference of ABP for AR1. Once ATP comes in contact with site S1, it will initiate an ‘induced-fit’/positive-feedback mechanism, with the increased free energy release and the downstream allosteric changes-induced energy releases acting as the driving force. In contrast, AR2 and 3 are not capable of facilitating the downward motion of head domain and therefore, would not evoke such positive feedback. ## The Allosteric Changes of Head Domain Induced by AR1 Attenuate the Preference of ABP for AR2 and 3 Beside the positive-feedback mechanism of AR1 on itself, the possibility that AR1 exerted a negative effect of AR1 on the occurrence of AR2 and 3 is also examined here. Taken into consideration of the involvement of head domain in forming S2 and S3, AR1-induced dynamics of head domain may also change the preference of ABP for AR2 and 3 accompanying the AR1’s occurrence. To test this idea, ATP was further docked into ABP of open crystal structure of zfP2**X**4, a state representing the stable conformation of AR1-induced alterations both in head domain and ABP. Interestingly, the top ATP poses with very low values of docking scores (\<−12) and low values of L191_vdW and I232_vdW scores can be easily observed in region III, confirming the ‘induced-fit’ mechanism of AR1 on itself. In contrast, as revealed by the approaching zero scores of I94_vdW in all the ATP poses , no poses possessing features of AR2 occurred during *in silico* docking. This result indicated that AR1-induced conformation changes of ABP preclude the contact between ATP and site S2. The I173_vdW scores revealed that a few poses possessing features of AR3 can be also observed in molecular docking (blue dots). However, AR3-induced free energy release was much less than that of AR1 (*Δ*G_bind = −46.79±7.7 and −27.52±5.04 kcal/mol, for AR1 and 3 in open state, respectively). These results suggested that AR1-induced allosteric change of head domain would decrease the preference of ABP for AR2 and 3. The positive feedback of AR1 on itself together with the negative effect it exerted on AR2 and 3 will dramatically improve the ability of ABP for AR1 recognition. Notably, along with the downward motion of head domain, the closing movement of loop_139–146 and loop_169–183 and structural rearrangements of K70, K72, R298, R143 and D145 enable ABP in the open structure to discriminate AR1 from other AR modes. Consistent with the MD data, our results support the pivotal role of the conformational changes of head domain in AR determinants. ## Discussion and Conclusions Agonist-recognition is a fundamental question for ligand-gated ion channels. Although the newly resolved crystal structures of zfP2**X**4 uncover exact AR mode (AR1) of these trimeric ion channels, it remains unclear why ABP of P2X receptor favors AR1 while other AR modes proposed by both experimental and computational approaches cannot trigger the activation of those receptors. Understanding the underlying mechanism of AR will greatly contribute to the development of novel antagonists/allosteric modulators of P2X receptors. Using a diversity of computational approaches, we suggested that the following factors may contribute to the preference of ABP for AR1 and abilities of AR1 rather AR2 and 3 in P2**X**4 receptors gating: *First,* site S1 is located on the surface of P2X receptor while S2 and 3 are partially buried by head domain and body domain, thus, the probability of ATP contacting with site S1 is a little higher than that of S2 and S3. *Secondly,* although the binding free energy releases of three distinct AR modes at the very beginning of ATP contacting with resting P2**X**4 receptors were comparable, the increased free energy release, springing from AR1-mediated promoted downward motion head domain, together with the downstream allosteric changes-induced significantly increased energy releases, which would enable AR1 to overcome the energy barrier required for channel gating. In contrast, AR2 and 3 would preclude the downward motion of head domain and the downstream allosteric changes correlated with channel gating. *Third,* beside the positive feedback of AR1 on itself, AR1’s occurrences-induced allosteric changes of ABP is able to preclude the occurrence of AR2 and 3. This idea is supported by the observation that no AR2 poses was observed during docking of ATP into open conformation of zfP2**X**4. All of these contribute to the preference of ABP for AR1 over AR2 and 3 during channel gating. The comparisons between closed and open structures, and the equilibrated averaged structure after MD simulations would not only facilitate studying ligand-recognition of P2**X**4 receptors, but also will provide some clues to understanding of channel gating mechanism. The movements observed in MD simulations exhibited high a similarity to bound-ATP induced conformational changes of ABP in open crystal structure. For instance, the structural rearrangements of K70, K68, R289 and R143, and closing movements of loop_139–146 and loop_169–183. However, the rightward movement of loop_169–183 and the down movement of head domain during MD simulations were more evident than those of ATP-bound open structure ( and). On the contrary, the upward movement of head domain in MD data was not as obvious as that of open structure. These observations confirmed the crucial role of inherent dynamics of head domain in both AR and channel gating of P2**X**4 receptors. However, the complex dynamics in the process of channel gating well beyond the inherent dynamics of head domain, during which bound-ATP evoked coordinated movements of multiple domains, especially for the head and dorsal fin domains, are crucial for this process. Our findings also provide new strategies for developing specific blockers/allosteric modulators by preventing the dynamics of head domain associated both with AR and channels activation of P2**X**4 receptors. Three types of new molecules could interrupt P2**X**4 activation based on our findings: *First,* we can *de novo* design high affinity small molecules containing both the acidic group to interact with K70, 72, K316, K193, R298 and the heterocyclic group to finely match the shape of site 2 or 3. These small molecules can compete with ATP and meanwhile act as allosteric modulator to prevent the dynamics of head domain. The second type of molecules are designed to fill up the cleft between the head domain and the dorsal fin domain. This strategy has also been proposed by Jiang *et al.* as inhibitor binds to this position can prevent the closing of ATP-binding site jaw, an allosteric change essential for channel activation. Here we predict that small molecules at this position would block the inherent dynamics of head domain. As a result, more ATP would contact with sites S2 and S3. Additionally, ATP contacting with site S1 induced positive-feedback can also be inhibited by these small molecules. Therefore, molecules that filling up the cleft between the head domain and the dorsal fin domain would impede the probability of AR1’ occurrence at the very beginning of ATP contacting with P2**X**4. *Third*, we can design *de novo* small molecules with a bulky size to occupy both the whole ABP and the cleft between head and dorsal fin, which can both compete with ATP and prevent the dynamics of head domain as well as the closing of ATP-binding site jaw. In summary, our study brings new insights into the mechanism of ATP recognition. The crucial role of head domain dynamics in channel activation has been well established, here we discuss its potential role in determining AR of P2**X**4 receptors. Our findings provide strategies in designing new blockers that prevent the dynamics of head domain associated with both AR and channel activation of P2**X**4 receptors. # Methods ## *In silico* Docking The docking program Glide was applied to dock ATP into the potential binding site of closed (PDB entry code: 3H9V) and open (PDB entry code: 4DW1) zfP2**X**4 as our previous description. Briefly, for Glide docking, the zfP2**X**4 structure was preprocessed using the protein preparation and refinement components. Then the grid for the protein was defined as an enclosing cubic box within 30 Å of centroid of selected reside (K70) in ABP of zfP2**X**4, where sites S1, S2 and S3 were all included. Conformations of ATP were generated by LigPrep. For docking runs, the extra precision (XP) docking mode was selected. All of the procedures including protein preparation, refinement, grid generation, and docking were performed using the default parameters except for the parameters for ATP poses output. During *in silico* docking, at most 100000 poses per docking run were selected, among of which top 300 poses per conformation of ATP were performed post-docking minimization. The threshold for rejecting minimized pose is 0.5 kcal/mol. At most 100 poses per conformation of ATP will be finally wrote out. ATP-per residue interaction scores were also measured during *in silico* docking. The docking scores and ATP-residue interaction scores were summarized, sorted and then plotted by Maestro (<https://www.schrodinger.com/productpage/14/12/37/>). Induced fit docking was performed by Glide³⁸. The residues within 15 Å of ligand pose were refined. ## MM-GBSA Binding Free Energy Calculation Binding free energy (*Δ*G_bind) calculations were performed by Prime MM-GBSA. The pose files obtained from Glide were the source of structures, where ligands and receptors were properly prepared beforehand. Residues that have any atoms within the 15 Å of the ATP processed are included in the flexible region. The movement of the flexible residues were not constrained with a harmonic potential. The binding energy is calculated according to the equation: *ΔG*_bind* = G*_complex*−*(*G*_ligand*+G*<sub>receptor</sub >). ## Molecular Dynamics Simulations According to our previous description, MD simulations were performed by using the program Desmond 3.0 with constant number of particles, constant pressure and temperature, and periodic boundary conditions, which uses a particular “neutral territory” method called the midpoint method to efficiently exploit a high degree of computational parallelism. Briefly, the closed structure of zfP2**X**4 (PDB entry code: 3H9V) or closed structure of zfP2**X**4 (PDB entry code: 3H9V) complexes with pre-docked ATP were used as the starting structures for MD simulations. A large dimyristoylphosphatidylcholine bilayer was constructed to generate a suitable membrane system where the TM domain of the zfP2**X**4 could be embedded. The protein/dimyristoylphosphatidylcholine system was then solvated in a bath of simple point charge water molecules. Counter ions Na⁺ were subsequently added to compensate for the net negative charge of the system. A default OPLS_2005 force field was employed for the protein or protein-ligand complex. To maintain the system at a constant temperature of 300 K, the Berendsen algorithm was applied to couple protein and other molecules separately with a coupling time of 0.1 ps. All of the bond lengths including hydrogen atoms were constrained by the Linear Constraint Solver algorithm. Electrostatic interactions between charged groups at a distance less than 9 Å were calculated explicitly. Long range electrostatic interactions were calculated using the smoothed particle mesh Ewald method. All of the MD simulations were run on the DAWNING TC2600, with 200 AMD Opteron™ 8374HE CPUs). All MD simulations were repeated in at least two independent runs. ## Normal Mode Analysis and PCA Analysis According to our previous description, the atomic coordinates for the crystal structure of zfP2**X**4.1 (PDB entry code: 3H9V) with proper preparation and minimization beforehand was used as the starting structure in a series of computational simulations and calculations. NMA was conducted using the web server developed by Delarue et al. (<http://lorentz.immstr.pasteur.fr/nomad- ref.php>). During the NMA simulations, the single-parameter Hookean potential, a simplified all-atom potential, was used:where *d_ij* is the distance between two atoms *i* and *j*, *d*⁰*_ij* is the distance between the atoms in the three-dimensional structure, *c* is the spring constant of the Hookean potential (assumed to be the same for all interacting pairs) and *R_c* is an arbitrary cut-off. In this study, *R_c* was set to be 10 Å. PCA analysis were applied by using the program ProDy 1.1. PCA modes of zfP2**X**4 dynamics were obtained by essential dynamics analysis (EDA) of snapshots of MD simulations of zfP2**X**4 receptors that ATP has been pre-docked into sites S1, S2 or S3. This measurement was performed using the default Prody parameters. 5 modes were generated in PCA analysis by Prody. We thank Dr. Tian-Le Xu for his permission to use DAWNING TC2600. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: Y. Yu. Analyzed the data: Y. Yu XYL JW. Wrote the paper: Y. Yu XYC Y. Yang. Supervised Y. Tian: XYL. Performed MD simulations, NMA analysis, and docking: LDH YZF YT JW WSZ PC. MM-GBSA binding free energy calculations and induced fit docking: YL WCZ.

# Introduction Antimicrobial proteins were first described in early 1960s with the identification of three basic proteins with bactericidal activities from a lysosomal fraction of guinea-pig polymorphonuclear leukocytes. Based on the molecular masses and antimicrobial activities, these proteins were later named as antimicrobial peptides (AMPs). The AMPs, or host defense peptides, are a diverse group of molecules which play an important role in the innate immune response and have been isolated from bacteria, insects and other invertebrates, birds, fish, amphibians, plants, and mammals. Based on the amino acid composition and structure, AMPs are divided into several subgroups such as anionic peptides, linear cationic α-helical peptides, cationic peptides enriched for specific amino acids, anionic and cationic peptides that contain cysteine and form disulfide bonds, and anionic and cationic peptide fragments of larger proteins. The majority of AMPs are known to interact with the lipid bilayer, leading to formation of ion channels, transmembrane pores, and membrane rupture resulting in microbial cell lysis, although AMPs can also interact with other microbial intracellular targets. Human granulysin as well as porcine and bovine NK-lysin, have also been described as AMPs. Granulysin and NK-lysin are found in the cytosolic granules of cytotoxic T-lymphocytes and NK cells and have been found to exhibit activity against a variety of Gram positive and Gram negative bacteria, viruses, fungi, parasites and tumor cells. Both granulysin and NK-lysin are structurally related to the saposin-like protein (SAPLIP) family of lipid binding proteins. Although most mammalian species, including humans and swine, have only a single antimicrobial protein gene (granulysin or NK-lysin), the cattle genome is known to contain four functional NK-lysin genes, *NK1*, *NK2A*, *NK2B* and *NK2C*. Gene expression analysis has revealed that while *NK1* and *NK2A* were highly expressed in the intestinal Peyer’s patches, while *NK2C* was highly expressed in the lungs. It has also been previously reported that the potent antimicrobial activity of bovine NK-lysin is associated with helix 2 through helix 3 regions . Bovine respiratory disease complex (BRDC), also known as shipping fever, is the most common infectious disease affecting older weanling, stocker, or feeder calves causing extensive economic losses to the North American beef and dairy cattle industry. BRDC develop as a complex interaction between environmental stress factors, host factors, and multiple viral and bacterial infectious agents. Various viral pathogens such as bovine herpes virus-1, bovine viral diarrhea virus, bovine respiratory syncytial virus, bovine parainfluenza-3, bovine coronavirus and bovine adenoviruses have been implicated to predispose calves to secondary bacterial infections. The most important bacterial pathogens associated with BRDC are *Mannheimia haemolytica*, *Pasteurella multocida*, *Histophilus somni*, and *Mycoplasma bovis*. Each of these bacteria are commensals in the nasopharynx of cattle. *H*. *somni* is a Gram negative bacterium and is a commensal of the upper respiratory as well as genitourinary tracts of cattle. *H*. *somni* can cause a wide variety of disorders in cattle in addition to respiratory disease, including septicemia, thrombotic meningoencephalitis, arthritis, myocarditis, and abortions. Antimicrobial activity of bovine NK-lysin derived peptides against BRDC pathogens *Mannheimia haemolytica*, and *Pasteurella multocida* has been reported. It has also been previously reported that compared to *M*. *haemolytica* isolates, *P*. *multocida* isolates were less susceptible to bovine NK-lysin mediated killing. *M*. *haemolytica* isolates were susceptible for NK2A and NK2C, but they were resistant to NK1 and NK2B. However, *P*. *multocida* isolates were susceptibility to NK1 and NK2A, they were resistant to NK2B and NK2C. Since these two closely related pathogens showed differential sensitivity to bovine NK-lysins, it was of our interest to assess the sensitivity of the other most important BRDC associated pathogen *H*. *somni* to bovine NK-lysins. Therefore, all four bovine NK-lysin derived peptides corresponding to the functional region of helices 2 and 3 were synthesized and their antimicrobial activity on several bovine pneumonic isolates of *H*. *somni* were examined. Ultrastructural changes to the cell membrane of *H*. *somni* following incubation with NK2A peptide were also examined. # Materials and methods ## NK-lysin peptide synthesis As previously described, four 30 amino acids long linear peptides corresponding to the functional regions of bovine NK-lysin (helices 2 and 3) of NK1 \[VIIHVTSKVCSKMGLWSILCNQMMKKYLNR\], NK2A \[TVIEVASKMCSKMRLLKGLCKSITKRFLRR\], NK2B \[TVIEAASKVCGKMGPLKGLCKSITKRFLRR\] and NK2C \[TVIEEASKVCSKMRLLKGLCKSIMKKFLRT\]) were synthesized by Peptide 2.0 Inc (Chantilly, VA), with over 95% purity. Lyophilized peptides (1 mg aliquot) were stored at -20°C, dissolved in phosphate-buffered saline (PBS, pH 7.4) and stored in aliquots again at -20°C until used. ## Bacterial isolates and culture conditions Bovine pathogenic *Histophilus somni* pneumonia isolate (2336) was kindly provided by Dr. Lynette B. Corbeil at the Department of Pathology, School of Medicine, San Diego, CA and three *H*. *somni* isolates (21, 22, and 91) which were originally isolated from pneumonic cattle lungs were kindly provided by the Veterinary Diagnostic Laboratory, College of Veterinary Medicine, Iowa State University. *Histophilus somni* isolates were maintained as frozen stocks (-80°C) in Columbia broth with 10% glycerol. All *H*. *somni* isolates were grown individually in trypticase soy agar supplemented with 5% defibrinated sheep blood (TSA II™, Becton, and Dickinson Co., Sparks, MD) at 37°C in a humidified atmosphere of 7.5% CO₂ for 16 to 48 hrs. ## Antimicrobial killing assay Antimicrobial killing assay was performed as described previously, but with minor modifications. Briefly, *H*. *somni* isolates grown on TSA sheep blood agar plates were collected using a cotton swab and diluted in Columbia broth to an optical density at 600 nm of 0.8 (\~1 × 10⁹ colony forming units per milliliter (CFUs/ml)). The bacterial suspension was further diluted in PBS to obtain \~1–5 × 10⁶ CFU/ml. One hundred microliter aliquots of bacterial suspension (\~1–5 × 10⁵ CFUs) were placed in a non-tissue culture treated flat-bottom 96-well plate (Becton Dickinson) and gently mixed with 20 μl of each of the diluted NK-lysin peptides (2, 5, 10, 20, and 30 μM final concentration) or PBS. The plate was covered with a lid and incubated at 37°C in a humidified atmosphere of 7.5% CO₂ with constant shaking (\~50 rpm) for 60 min. For quantitative counting, bacterial samples were then serially diluted in PBS, a 100 μl aliquot of each dilution was spread on TSA sheep blood agar plates and incubated at 37°C for up to 2 days. Bacterial colonies were enumerated for each dilution. Antimicrobial killing assays were repeated at least twice with three replicates for each peptide. ## Bacterial viability staining To easily and reliably distinguish dead bacteria from live bacteria following incubation with NK-lysin peptides, *H*. *somni* was stained using LIVE/DEAD *Bac*Light bacterial viability kit as described by the manufacturer, but with some modifications (Cat no. L13152; ThermoFisher Scientific, Carlsbad, CA). Briefly, \~1 × 10⁸ cfu of *H*. *somni* suspension in 100 μl PBS was placed in a 96-well plate and 20 μM final concentration of each of the NK-lysin peptide was added and samples were incubated at 37°C in a humidified atmosphere of 7.5% CO₂ with constant shaking for 30 min. Untreated and ethanol killed *H*. *somni* were used as live and dead controls, respectively. Fifty microliters of Syto 9 and 50 μl of propidium iodide (approximate final concentrations of Syto 9 and propidium iodide were 6 μM and 30 μM, respectively) were added to each well and incubated at 37°C for additional 15 min. The viability of *H*. *somni* was then determined by confocal laser-scanning microscopy and flow cytometry (green (live) versus red (dead) bacteria). ## Confocal laser-scanning microscopy Approximately 150 μl of bacterial suspensions, previously incubated with 20 μM NK-lysin peptides or PBS along with Syto 9 and propidium iodide, were adhered on to glass slides using Shandon cytospin 2 and coversliped using *Bac*Light mounting medium (ThermoFisher). Live and dead bacteria were then visualized using Nikon A1R+ Confocal System microscope (Nikon Instruments, Melville, NY). Syto 9 and propidium iodide were excited at 488 nm and 561 nm solid state laser beam and emission signals 500–550 nm (green) 570–620 nm (red) were recorded, respectively. Images were collected using proprietary NIS-Elements Advanced Research software. Calibration was created for both dyes using the software and sequentially collected frames of individual channels were merged and saved as TIFF files. Images were obtained with plan Apo 60× objective lens (oiled) at numerical aperture 1.4. Final figures were prepared using Adobe Photoshop Elements 11. ## Flow cytometry Two-color flow cytometric analyses was performed using a BD LSRII flow cytometer (BD Biosciences). *Histophilus somni* were visualized in forward and side light scatter and electronic gates were set to contain single bacterial cells. Single (Syto 9 or propidium iodide) and double fluorescence dye labeled live and dead bacteria were included to further optimize acquisition gates and compensation for each fluorochrome. Both Syto 9 and propidium iodide were excited at 488 nm laser beam and the emission signals were detected using a 530/30 nm and 575/25 nm long-pass filters, respectively. Approximately, 10,000 events were collected for data analysis and relative live/dead bacterial changes were determined using FlowJo software (FlowJo LLC, Ashland, OR). ## NK-lysin bacterial binding assay To determine the binding of NK-lysin peptides with bacteria, *H*. *somni* was incubated with Cyanine dye 5 (Cy5)-conjugated NK2A and analyzed by flow cytometry and confocal microscopy. Since NK2A showed the highest antimicrobial activity, NK2A peptide was chosen for Cy5 labeling. Peptide-dye conjugation was performed with Lightning-Link^® Cy5 rapid conjugation system as described by the manufacturer (Innova BioSciences, Cambridge, UK). Approximately, 1 × 10⁸ CFU of *H*. *somni* suspension in 100 μl PBS was placed in a 96-well plate and incubated with either 20 μM final concentration of NK2A-Cy5 conjugate or similar amount of unconjugated Cy5 dye for 30 min at 37°C in a humidified atmosphere of 7.5% CO₂ with constant shaking. Samples were then directly assessed by flow cytometry and confocal microscopy as describer earlier, but with Cy5 excitation (633 nm) and emission (660/20 nm) filters. In order to visualize NK-lysin binding to *H*. *somni* by confocal microscopy, coverslips were mounted using ProLong^® Gold antifade mount medium with DAPI (4\`, 6\`-diamidino-2-phenylindole). ## Propidium iodide uptake assay To determine kinetics of NK-lysin induced damage to *H*. *somni* cell membrane, propidium iodide (PI) uptake assay was performed. Briefly, \~1 × 10⁸ cfu of *H*. *somni* suspension in 100 μl PBS was placed in a 96-well opaque plate and mixed with 5 μl of PI (final concentration 3 μM). Samples were then incubated at 37°C in a humidified atmosphere of 7.5% CO₂ with constant shaking for 5 min. After pre-staining *H*. *somni* with PI, different concentrations of NK2A peptide (1, 2, 5, and 20 μM final concentrations), 1% (v/v) Triton X-100, or kanamycin (50 μg/ml) were added into each well. At selected time points, fluorescent intensity of each sample was measured using a fluorescent microplate reader with excitation and emission at 530 nm and 620 nm, respectively. Ethanol-killed and Trion X 100 treated *H*. *somni* were used as positive controls and untreated and kanamycin treated *H*. *somni* were used as negative controls. ## Transmission electron microscopy Approximately 2 × 10⁹ CFUs of *H*. *somni* in 100 μl of PBS was transferred into a 96-well plate and incubated with 30 μM of NK2A peptide or PBS at 37°C in a humidified atmosphere of 7.5% CO₂ with constant shaking for 60 min. The bacterial suspension was mixed with an equal volume of 3% glutaraldehyde in 0.1 M cacodylate buffer (pH 7.4) and processed for electron microscopy essentially as described previously. Briefly, after fixation, bacterial pellets were rinsed in cacodylate buffer, postfixed in 1% osmium tetroxide, dehydrated in alcohols, and embedded in epoxy resin. Ultrathin sections of the bacterial pellets were cut and stained with uranyl acetate and lead citrate. Sections were examined with a FEI Tecnai G2 Biotwin (FEI Co., Hillboro, OR) transmission electron microscope and images were captured with Advanced Microscopy Technologies (AMT Inc., Danvers, MN) imaging camera. ## Statistical analysis Mean number of viable bacteria (CFUs/ml) and standard errors of means were determined. Student’s *t*-test was used to compare percentages of live and dead *H*. *somni* in control and NK-lysin treated samples. The term significant indicates *P* value of less than 0.05. # Results ## Antimicrobial activity of bovine NK-lysin peptides on *H*. *somni* The antimicrobial activity of bovine NK-lysin peptides against both Gram- positive, as well as Gram-negative bacteria, have been previously reported. Since *H*. *somni* is one of the primary bacterial pathogens involved in BRDC, a study was conducted to evaluate whether bovine NK-lysin peptides show antimicrobial activity against this bacterium, similar to that which has been reported earlier for two other BRDC causing bacterial pathogens, *M*. *haemolytica* and *P*. *multocida*. To compare relative antibacterial activity of bovine NK-lysins on *H*. *somni*, we synthesized previously reported 30-mer peptides corresponding to the functional region of bovine NK-lysin helices 2 and 3 of NK1, NK2A, NK2B and NK2C. The antimicrobial activity of the peptides on *H*. *somni* were assessed using multiple techniques, including an antimicrobial killing assay, a Live/Dead *Bac*light bacterial viability assay with confocal microscopy and flow cytometry, as well as transmission electron microscopy. Bovine NK-lysin peptides killed *H*. *somni* isolates in a dose-dependent manner, with up to 100% of the bacteria killed during a 60 min incubation period. At higher concentrations (10–30 μM), all four peptides showed higher antibacterial activity against all four bovine *H*. *somni* isolates. However at lower peptide concentrations (2 and 5 μM), reduced antibacterial activity as well as variations in sensitivity among *H*. *somni* isolates was observed. Control *H*. *somni* isolates which were not incubated with any of the peptides but with PBS did not show detectable changes in the bacterial numbers (CFUs/ml) during 60 min incubation period (data not shown). ## Bovine NK-lysin peptides cause *H*. *somni* cell membrane damage Although the precise mechanism of antimicrobial activity of antimicrobial peptides (AMPs) is yet to be fully understood, previous studies report that AMPs can damage outer and inner bacterial membranes. *H*. *somni* isolates were sensitive to NK-lysin-derived peptides as evidenced by reduction in bacterial numbers (CFUs/ml) in killing assays however, it was not clear whether the antibacterial activity of NK-lysin peptides on *H*. *somni* were bactericidal or bacteriostatic. Therefore to identify the mechanism of action, we performed Live/Dead bacterial viability staining after the incubation of *H*. *somni* with peptides. Syto 9 is a cell permeant green-fluorescent dye which can efficiently bind with both prokaryotic and eukaryotic DNA. Unlike Syto 9, propidium iodide is a cell non-permeant red-fluorescent dye and only stain DNA in the cells with compromised cell membranes. Therefore, we used Live/Dead bacterial viability staining kit containing both Syto 9 and propidium iodide to differentiate live from dead bacteria after incubation with NK-lysin peptides. As expected, majority of bacteria in control sample stained with Syto 9 (green) and in killed bacteria stained with propidium iodide (red). These findings confirmed that the cell membranes of live bacteria were intact. The damage to *H*. *somni* cell membranes was clearly visible as bacteria treated with NK-lysin peptides stained mostly with propidium iodide. Unlike with NK2B and NK2C, clumping of dead *H*. *somni* were observed following incubation with NK1 and NK2A. Concurrently, similarly incubated *H*. *somni* samples were also analyzed by flow cytometry. Since dead bacteria tended to clump, samples were thoroughly mixed before flow cytometry and electronic gates were positioned for single cell discrimination. Again as expected, most of the bacteria in the control sample were viable with intact cell membranes and all the bacteria treated with ethanol were found dead with compromised cell membranes. Over 93% of *H*. *somni*, when incubated with NK-lysin peptides, were positive for propidium iodide staining or in the dead bacterial region further confirming the damage to the bacterial cell membranes. Taken together, both confocal microscopic and flow cytometric analysis suggest that all four NK-lysin peptides caused damage to *H*. *somni* cell membranes and are bactericidal in nature. It has been previously reported that most antimicrobial proteins (AMPs) interact with lipid bilayer, however some AMPs can interact with other intracellular targets such as nucleic acids. For example, Buforin II, an AMP isolated from the stomach tissue of Asian toad *Bufo bufo garagrizans*, penetrates the bacterial membranes (without damaging the cell membrane) and enters the cytosol, then binds with bacterial DNA and RNA. Both our confocal microscopic and flow cytometry analysis (Figs) confirmed that bovine NK-lysins caused damage to *H*. *somni* cell membranes, but we wished to further evaluate whether NK-lysin peptides can also interact with other bacterial targets, specifically with *H*. *somni* DNA. Therefore, *H*. *somni* was incubated with NK2A-Cy5 conjugate (for 30 min) and observed the bacteria under confocal microscope after staining bacterial DNA with DAPI. Unconjugated or free Cy5 dye did not bind with bacteria but their nuclear materials were clearly stained with DAPI. Although NK2A-Cy5 staining was visible with *H*. *somni*, we did not observe any co-localization of NK2A-Cy5 conjugate with DAPI suggesting that NK2A did not bind with DNA, but interacted with bacterial cell membrane. In order to examine the kinetics of NK-lysin-induced *H*. *somni* cell membrane damage, we performed propidium iodide (PI) uptake assay in a time-dependent manner following incubation of bacteria with different concentrations of NK2A. As early as 1 min post-incubation, higher PI fluorescence emission was observed with higher concentrations of NK2A (5 and 20 μM) and PI uptake continuously increased during the 30 min incubation period. Lower, but time-dependent increases in PI uptake was also observed with lower concentrations of NK2A (1 and 2 μM). As expected, higher PI uptake was observed with ethanol-killed and Trion X-100 treated samples while low-to-no PI uptake was observed with NK2A untreated and kanamycin treated samples. To visualize ultrastructural damage to cell membranes, we prepared control and NK2A treated *H*. *somni* samples for transmission electron microscopic assessment. Both inner and outer membranes of control *H*. *somni* were intact and their cytosol was filled with electron dense material suggesting these bacteria are alive and healthy. In contrast, NK2A treatment led to extensively damaged *H*. *somni* cell membranes and their cytosol appeared to be clear indicating cytosolic contents leakage. Although the damage to the bacterial inner membranes was very clear (arrows), *H*. *somni* were also found to contain single to multiple protrusions (letter “a”) of the outer membranes. Using six electron micrographic images (taken at 11,000 magnification), we calculated percent viable and dead *H*. *somni* in both control and NK2A treated samples and statistically analyzed the data using student’s *t*-test. Compared to control samples (\~5% dead bacteria), significantly higher dead *H*. *somni* (\~45% dead bacteria with membrane damage, protrusions and clear zones of cytosol) in NK2A treated samples, further confirming the antimicrobial activity of NK2A. # Discussion Both *H*. *haemolytica* and *P*. *multocida* belong to the *Pasteurellaceae* family and share many common properties, yet their relative sensitivity to NK- lysin peptides were different. It has been previously reported that unlike *M*. *haemolytica* isolates, *P*. *multocida* isolates were less susceptible to antimicrobial activity of all four peptides. Although *P*. *multocida* isolates showed moderate sensitive to antimicrobial activity of NK1 and NK2A peptides, *M*. *haemolytica* isolates showed enhanced sensitive to NK2A and NK2C peptides. Similar to *M*. *haemolytica* isolates, NK2A and NK2C peptides consistently displayed the highest antimicrobial activity against *H*. *somni* isolates. Ten micromole concentrations of NK2A and NK2C peptides were able to completely eliminate viability of all *H*. *somni* isolates within 60 min incubation period. However, unlike with *P*. *multocida* and *M*. *haemolytica* isolates, NK1 and NK2B peptides also displayed substantial antibacterial activity against all four *H*. *somni* isolates, but at relatively higher peptide concentrations. Amino acid sequence analysis revealed that NK2A, NK2B and NK2C (146 residues) showed higher identity to each other in comparison to NK1 (142 resides). However, all four 30-mer long synthetic NK-lysin peptides corresponding to the functional region of helices 2 and 3 showed similar hydrophobicity (40–43%) and basic residues (20–30%) such as arginine, lysine, and histidine. All four peptides also contain two conserved cysteine residues at 10^th (helix 2) and 20^th (helix 3) position. It has been previously reported that secondary structure of NK2A, NK2B and NK2C peptides were very similar while NK1 peptide differed structurally with a lower degree of helicity. Despite these differences, NK1 peptide showed the highest antimicrobial activity against *E*. *coli* and *S*. *aureus* while NK2A and NK2C peptides showed the highest antimicrobial activity against *M*. *haemolytica* and *H*. *somni* isolates (this study). Given the similarities in amino acid residues and secondary (helical) structures among bovine NK-lysin peptides, differences in antibacterial activity among bovine NK-lysin peptides on additional bacterial pathogens should be examined in future studies. Antimicrobial peptides (AMPs) are a diverse group of molecules which play an important role in the host innate immune response. AMPs such as granulysin and NK-lysin are well-known to kill both Gram-positive and–negative bacteria, parasites and fungi. Propidium iodide is widely used in flow cytometry and fluorescent microscopy to differentiate live from dead cells as it cannot cross the intact membrane of live cells (to stain DNA). In our propidium iodide uptake assay, both NK2A untreated and kanamycin treated samples did not show any increase in PI fluorescence suggesting a lack of *H*. *somni* cell membrane damage. However, time- as well as concentration-dependent increased in PI fluorescence or PI uptake in NK2A treated samples was observed strongly suggesting NK2A-induced *H*. *somni* cell membrane damage. Furthermore, incubation of *H*. *somni* with Cy5 directly conjugated NK2A peptide confirmed a lack of association of bovine NK-lysin with bacterial DNA, but rather showed an interaction with bacterial cell membranes. Although our confocal microscopic images confirmed the extensive damage to *H*. *somni* cell membranes following incubation with NK-lysin peptides as treated bacteria were positive for propidium iodide staining, the resolution of the confocal images were not adequate to identify ultrastructural changes to the bacterial cell membranes. Therefore, we assessed ultrastructural morphological changes of control and NK- lysin treated *H*. *somni* by transmission electron microscopy. As has been observed with other bacterial pathogens, electron micrographs confirmed the rupture of *H*. *somni* inner membranes when incubated with NK2A. It has been previously reported that NK-lysin induced damage to *P*. *multocida* cell membranes with cytosolic content leakage, but without membrane-enclosed bodies. Such membrane-free coagulated content leakage was not very visible with *H*. *somni* upon incubation with NK-lysin. Rather most of the cytosolic contents were observed to be attached to the bacterial outer membrane or released as outer membrane enclosed vesicles/bodies. Similar single to multiple outer membrane protrusions development have been previously reported with *Mycobacterium tuberculosis* when incubated with human granulysin. Although there are many *H*. *somni* vaccines (bacterins) currently available in the market, they show limited efficacy due a variety of reasons, such as, antigenicity of vaccine strains may be different from *H*. *somni* isolates currently circulating in the field. A subunit vaccine comprised of a recombinantly expressed DR2 domain of ibpA or multiple antigens expressed using a genome-based reverse vaccinology technique with a field *H*. *somni* isolate showed enhanced protection in cattle against subsequent challenge with *H*. *somni*. In this study, we showed that all four NK-lysin peptides are highly effective against four different *H*. *somni* isolates, but at 10 μM or higher concentrations. Therefore, combination of NK-lysin peptides, particularly NK2A, with *H*. *somni* recombinant or subunit vaccines might provide enhanced protection in cattle against *H*. *somni* infections. However, one must consider the cost associated with AMP production particularly for large animal trials, as well as the possibility of development of resistance to AMPs by microorganisms. However, such resistance to AMPs among microbes has been shown to be very limited. Furthermore, although several AMPs are in clinical trials, the FDA has yet to approve any AMPs for medical usage. Therefore, it will be interesting to determine whether bovine NK-lysin in general, and NK2A in particular, can be used as a treatment or preventive to reduce the occurrence of BRDC in calves. # Conclusions Using several techniques such as bacterial killing assay, Live/Dead bacterial staining, followed by confocal laser-scanning microscopic and flow cytometry analyses have revealed that bovine NK-lysin-derived peptides are bactericidal to *H*. *somni*. Moreover, electron microscopic analyses revealed that NK-lysin- derived peptide (NK2A) caused extensive damage to *H*. *somni* inner and outer membrane, cytosolic content leakage, as well as protrusions of outer membranes. Based on our results and those observed previously, now we can conclude that bovine NK-lysin-derived peptide NK2A is effective against the bacteria agents involved in BRDC such as *M*. *haemolytica*, *P*. *multocida*, and *H*. *somni*. The authors would like to thank Tracy Porter, Judith Stasko, Adrienne Shircliff, Sam Humphrey Brad Chriswell, and Virginia Montgomery at the NADC for their excellent technical support. [^1]: The authors have declared that no competing interests exist.

# Introduction MicroRNAs (miRNAs) are small noncoding RNAs comprising 20–25 nucleotides, and miRNAs regulate gene expression by inducing mRNA degradation and inhibiting mRNA transcription into proteins. More than 2,500 miRNAs have been identified previously. It has been reported that the changes in miRNA expression observed in blood are associated with human diseases such as cancer, cardiovascular disease, and Alzheimer’s disease. Because early detection and treatment of cancer leads to improved survival rates, screening tests are conducted for each type of cancer. However, testing for multiple types of cancer is a physical and financial burden for patients. MiRNA-based liquid biopsy is attracting attention since it is a less invasive method for cancer screening. Previous studies have reported that the expression levels of some miRNAs can be used to distinguish between individuals with and without cancer. Furthermore, the development of technologies for comprehensive data acquisition, such as next-generation sequencing (NGS) and microarrays, has made it possible to obtain the expression profiles of more than 2,500 different miRNAs in a single measurement. Through the identification of cancer-specific markers, cancer and noncancer were able to be distinguished with an area under the receiver operating characteristic (ROC) curve of 0.92–0.98. MiRNA biomarkers are reported to be useful not only in distinguishing the presence or absence of cancer but also in binary classification of certain cancer types, with 95% accuracy for bladder cancer and more than 95% accuracy for six other cancer types. MiRNA biomarkers were also shown to be useful in distinguishing between ovarian cancer and noncancer with 84% accuracy and in distinguishing between ovarian cancer and four nonovarian cancer types with more than 80% accuracy. It has been suggested that a screening test that simultaneously classifies multiple cancer types can be achieved by obtaining a comprehensive miRNA expression profile. In previous studies on the discovery of cancer-specific markers, miRNAs that showed large fold changes in expression between a certain cancer type and other cancer types were identified as cancer-specific miRNAs. Although studies have identified cancer-specific markers, the results regarding the changes in the expression levels of these miRNAs are controversial. We speculate that the low reproducibility of the results regarding the changes in the expression levels of the same miRNAs is attributable to differences in the data acquisition methods used among the studies. Since sample collection is typically conducted using different operational methods across each facility, the type of blood collection tubes and the temperature and timing of storage of whole blood and plasma samples are potentially variable between facilities. The inconsistencies in the results regarding cancer-specific marker among studies are thought to be related to these differences in preanalytical conditions. Studies on the type of anticoagulant in blood collection tubes have revealed different expression levels of miRNAs, including hsa-miR-191-5p, hsa-miR-320a, hsa-miR-21-5p and hsa- miR-451a; hsa-miR-21 and hsa-miR-29b; and hsa-miR-16, in samples obtained from ethylenediaminetetraacetic acid (EDTA)-treated plasma and citrate-treated plasma obtained from the same source. Although these studies focused on a limited number of miRNAs, studies on the storage conditions used for whole blood and plasma have reported that some miRNAs show small changes in expression over 24 hours in whole blood and over 14 days in frozen plasma. However, these studies on preanalytical conditions have investigated the expression changes in specific sets of miRNAs, and findings on the effects of different anticoagulants used in blood collection tubes and sample storage conditions on comprehensive miRNA expression profiles are still limited. To establish preanalytical conditions for obtaining highly reproducible data, we investigated the effects of different anticoagulants and storage conditions on miRNA expression profiles, which could potentially contribute to the variability in miRNA expression levels among samples collected at multiple facilities. MiRNA expression profiles acquired from blood samples using NGS were analyzed according to the following parameters: the library concentration, human genome identification rate, ratio of unique sequences and difference in expression profiles. First, we discovered highly reproducible preanalytical conditions by analyzing miRNAs extracted from healthy volunteer samples. Then, using these preanalytical conditions, we collected samples from cancer patients at multiple independent facilities, measured the miRNA expression level, and evaluated the expression distribution range among facilities. # Materials and methods ## Blood sample collection from healthy volunteers Blood samples were collected from 18 healthy volunteers employed by DeNA Co., Ltd. and DeNA Life Science, Inc. Participants were healthy adults who had been examined by a physician and had no diseases. Whole blood was collected by venipuncture into anticoagulant-containing tubes at time of the physical examination. To investigate the effect of whole-blood storage conditions, samples were refrigerated for one hour, six hours, one day, two days and three days after collection. To investigate the effect of storage periods, samples were refrigerated within one hour after collection. To investigate the effects of anticoagulant type, blood samples were collected from each donor using tubes containing EDTA-2Na, EDTA-2K, sodium fluoride and sodium citrate. Plasma separation was performed within two hours after collection. To investigate the effects of storage periods, blood samples were collected using EDTA-2Na tubes. All samples were mixed immediately by inverting the tube 10 times after sample collection. Before participating in this study, the participants completed a written informed consent form. This research was approved by the ethics committee of DeNA Life Science, Inc. (approval number: 20180228_1) and was conducted according to the guidelines of the Declaration of Helsinki. ## Blood sample collection from cancer-free donors in the biobank Blood samples were obtained from 60 cancer-free donors and stored in the biobank of the National Cancer Center (NCC) Institute for Cancer Control. Cancer-free donors were defined as donors who were not diagnosed with cancer during cancer screening performed by the NCC. Whole blood was collected by venipuncture into EDTA-2Na-containing tubes before treatments. All samples were mixed immediately by inverting the tube 10 times after sample collection. The collected samples were refrigerated within 30 minutes. Before participating in this study, the participants completed a written informed consent form. This research was approved by the NCC Review Board (approval number: 2018–200) and was conducted according to the guidelines of the Declaration of Helsinki. ## Blood sample collection from cancer patients Blood samples were collected from 60 cancer patients. Whole blood was collected by venipuncture into EDTA-2Na-containing tubes before treatments. All samples were mixed immediately by inverting the tube 10 times after sample collection. The collected samples were stored at 4°C for 30 minutes. From facility A (n = 20), samples from esophageal (11), gastric (5), and pancreatic (4) cancer patients were collected. From facility B (n = 20), samples from esophageal (5), gastric (6), and pancreatic (9) cancer patients were collected. From facility C (n = 20), samples from esophageal (1), gastric (10), and pancreatic (9) cancer patients were collected. All samples were randomly collected without any selection of patients with specific cancer types. Before participating in this study, the participants completed a written informed consent form. This research was approved by Juntendo University Hospital, NCC, and the Osaka International Cancer Center Review Board (approval number: 2020111, 2018–200 and 20043–2) and was conducted according to the guidelines of the Declaration of Helsinki. ## Plasma sample preparation During the collection of samples from healthy volunteers, the samples were centrifuged once at 1,200 ×g at 4°C for 10 minutes. During the collection of samples from cancer-free patients, the samples were centrifuged once at 1,830 ×g at 4°C for 10 minutes within 12 hours of venipuncture. During the collection of samples from cancer patients, samples were centrifuged once at 1,500 ×g at 4°C for 10 minutes within 12 hours of venipuncture. The supernatant was immediately removed without aspiration of the buffy coat and then cryopreserved in RNase- free cryotubes. There was no significant difference in the amount of miRNA recovered, and the correlation of miRNA profiles was more than 0.99 when the centrifugation conditions were selected between 1,200 x g and 1,900 x g. ## RNA purification MiRNAs were automatically extracted from 300 μL of plasma using the Maxwell RSC miRNA Plasma and Serum Kit (Promega, Madison, US) with the Maxprep Liquid Handler (Promega) and Maxwell RSC (Promega) equipment. Purifications were performed according to the manufacturer’s instruction manual. The RNA concentration was measured by a QuantiFluor RNA System (Promega) with a GloMax Explorer (Promega). ## Reverse transcription and sequencing data acquisition Complementary DNA (cDNA) libraries were synthesized from 5 μL of RNA eluate with 22 polymerase chain reaction (PCR) cycles using the QIAseq miRNA Library Kit (QIAGEN, Venlo, NL) with the Biomek i5 Automated Liquid Handling Workstation (Beckman Coulter, Brea, US). The concentration of each amplification product was measured with a QuantiFluor ONE ds DNA System (Promega). cDNA libraries were merged in one tube at a final concentration of 1.0–1.5 pmol/L. Small RNA sequencing data were obtained using NextSeq 550Dx with the NextSeq 500/550 High Output Kit v2.5, 75 cycles (Illumina, San Diego, US) and the NextSeq Phix Control Kit (Illumina). An automated program was set up for each instrument according to the manufacturer’s instruction manual, and all the samples were processed using the same program. ## MiRNA expression analysis The raw sequencing data were processed using fastp (version 0.19.6) to remove data regarding the amplification adapter sequence. The unique molecular identifier (UMI) sequences and the probe adapter were trimmed using umi_tools (version 1.0.1). Mapping was performed using Bowtie (1.3.0), and hg19 was used as the reference human genome. The following options were applied: -n0, -v0, -l18, -m1, and—best. After sorting with SAMtools (version 1.7), an index was created and deduplicated with umi_tools. Then, featureCounts (version 2.0.0) was used to obtain the expression count levels of each miRNA with reference to miRBase (version 20). To estimate RNA quality from the sequence reads, the UMI ratio was calculated from the total number of UMIs and the total reads obtained from each sample. The human genome identification rate (identification rate) was a number obtained by dividing total number of quality filtered reads processed by fastp (Assigned values) by total number of quality filtered reads processed by fastp (Reads_processed). These parameters are referred to in the column descriptions in. MiRNA expression counts are shown in. ## Statistical analysis We calculated the means and standard deviations of the miRNA expression counts, generated scatter plots, and performed one-way analysis of variance (ANOVA) with a post hoc Tukey test or Dunnett’s test using the statistical analysis software R (version 3.6.3). One-way ANOVA with the Tukey test was applied to compare the miRNA levels during the investigation of the effects of different blood collection tubes among facilities. One-way ANOVA with Dunnett’s test was applied to compare the miRNA levels during the investigation of time-dependent differences. The fold change depicted in the scatter plot was calculated from the miRNA expression counts. The data points on the solid line indicate miRNAs with equal expression levels, and the points above or below the dashed line indicate that miRNAs with expression levels that were 2-fold higher or 1/2-fold lower, respectively. # Results ## Study design In this study, we explored a standardized protocol for the collection of samples from clinical cancer patients from multiple facilities. First, the preanalytical conditions were investigated after miRNA expression profiles were acquired from plasma samples obtained from healthy volunteers. Then, we evaluated the use of the established preanalytical conditions as a method for collecting samples from clinical cancer patients. The differences between samples subjected to different preanalytical conditions, including different types of blood collection tubes, were evaluated, and the analysis included assessments of the stability of the samples. During the discovery stage, preanalytical conditions were examined using plasma samples from healthy individuals. During the investigation of the effects of anticoagulants, blood samples from the same donor were collected into EDTA-2Na, EDTA-2K, sodium fluoride (NaF) and sodium citrate (SC) blood collection tubes, which are used in a typical clinical laboratory. The storage stability of refrigerated or frozen samples was evaluated in both whole-blood samples and plasma samples. Whole-blood samples were stored for one hour, six hours (room temperature), one day, two days, and three days at 4°C. Plasma samples were stored at 4°C, -20°C or -80°C. The storage periods used for samples maintained at 4°C and -20°C were one hour, one day, three days, five days, seven days and 30 days. The storage periods for samples maintained at -80°C were three, four and five years. During the discovery stage, we analyzed the following factors to identify the optimal preanalytical conditions. First, the library concentration was measured to examine the effect on biochemical processes, such as RNA extraction and reverse transcription. Second, the identification rate was calculated to examine the effect on the purity of sequences identified as originating from the human genome in the extracted RNA fragments. Third, the UMI ratio was calculated to examine the effect on the uniqueness of the origin RNA located in the library DNA. Finally, the expression levels of erythrocyte- derived miRNAs and platelet-derived miRNAs were compared to examine hemolysis and platelet contamination. During the validation stage, the selected preanalytical conditions were assessed using 60 samples from clinical cancer patients. Samples were obtained from 20 patients each from facilities A, B and C. The library concentrations, identification rate and UMI ratios were measured. Then, the adaptability of the selected preanalytical conditions was assessed as follows: (i) whether each measurement was significantly different between facilities and (ii) whether measurements obtained from one facility fit within the standard deviation of the results obtained from another facility. ## Evaluation of different blood collection tubes The impacts of anticoagulants were examined using EDTA-2Na, EDTA-2K, NaF and SC tubes. Regarding the library concentration, although there was no significant difference resulting from the use of EDTA-2Na and EDTA-2K tubes, the use of NaF and SC tubes resulted in lower library concentrations than the use of EDTA tubes. Regarding the identification rate, the use of EDTA-2Na, EDTA-2K and NaF tubes did not result in difference, but the use of SC tubes resulted in a lower identification rate. Regarding the UMI ratio, the use of EDTA-2K tubes resulted in the highest UMI ratios, but the UMI ratios did not differ following the use of EDTA-2Na and NaF tubes. There were significant differences resulting from the use of EDTA tubes and SC tubes. It was observed that EDTA-2K and EDTA-2Na exhibited similar properties, while NaF and SC exhibited different properties from EDTA. To determine the differences in the expression of miRNAs resulting from the use of various blood collection tubes, a profile comparison was performed in three ways: EDTA-2Na vs. EDTA-2K tubes, EDTA-2Na vs. NaF tubes and EDTA-2Na vs. SC tubes. The numbers of miRNAs that exhibited increased levels of more than 2-fold between samples obtained in EDTA-2Na and EDTA-2K tubes, EDTA-2Na and NaF tubes and EDTA-2Na and SC tubes were 2, 25, and 52, respectively. The numbers of miRNAs that exhibited decreased levels of less than 1/2-fold between samples obtained in EDTA-2Na and EDTA-2K tubes, EDTA-2Na and NaF tubes and EDTA-2Na and SC tubes were 1, 2, and 10, respectively. No common features were observed among the miRNAs with decreased expression levels. The miRNAs that exhibited increased expression in samples obtained in NaF and SC tubes were examined, and 24 miRNAs were identified. shows the list of these miRNAs ranked by their expression levels. Hsa-let-7b-5p, hsa-miR-451a, hsa-miR-144, hsa-miR-486, hsa-miR-363, hsa- miR-96 and hsa-miR-4732-5p have been reported as erythrocyte-derived miRNAs. The fold change in the levels of erythrocyte-derived hsa-miR-451a between samples obtained from EDTA-2Na and those obtained from NaF tubes and EDTA-2Na and SC tubes was more than 6-fold, with significant differences between samples obtained from EDTA-2Na and NaF tubes or SC tubes. In addition, the expression levels of platelet-derived hsa-miR-126-3p were compared to determine the degree of platelet contamination, but a significant difference was not observed. In this study, we focused on hsa-miR-126-3p as a representative marker to investigate the expression of platelet-derived miRNAs. The expression levels of other platelet-derived miRNAs, such as hsa-miR-223-3p, hsa-let-7f-5p and hsa- miR-16-5p, were investigated, and there were no differences in expression between samples obtained in the different tubes. To examine differences in RNA concentrations, blood samples were collected from five donors in EDTA-2Na, EDTA-2K, NaF, and citrate blood collection tubes. Samples obtained in EDTA-2Na and EDTA-2K tubes exhibited the highest RNA concentrations, which were not significantly different among them, but the RNA concentrations in samples obtained from NaF and citrate samples were lower. In addition, differences among samples subjected to different storage periods were examined. Blood samples were collected from five donors in EDTA tubes at 1 hour, 6 hours (RT), 1 day, 2 days, and 3 days at 4°C. There was no significant difference in the concentration of RNA between samples extracted at 1 hour after collection and those collected at each other time point. From the results of the differences in samples obtained from different blood collection tubes, it was observed that EDTA-2Na and EDTA-2K tubes had similar properties. However, the properties of NaF and SC tubes were different from those of EDTA tubes, suggesting that hemolysis was promoted in these tubes. ## Stability of whole blood over storage periods of up to three days at 4°C The storage stability of whole-blood samples was investigated. To examine the time-dependent stability of samples subjected to various storage periods, blood samples stored for one hour after blood collection were used as the baseline condition and compared with blood samples stored for 1 day, 2 days, and 3 days. Plasma samples were prepared after these storage periods, and various measurements were performed. In addition, plasma samples prepared after 6 hours of storage at room temperature were examined for reference information. The library concentration was decreased in a time-dependent manner in all donor samples. Significant differences were observed after one day. The identification rate was decreased in a time-dependent manner in all donor samples. Significant differences were observed after two days. However, the UMI ratio showed no time- dependent differences in any donor samples. From these results, it was observed that the quality of whole-blood samples did not change until one day after blood collection. To investigate the time-dependent change in miRNA expression, miRNA expression profiles were compared between samples stored for one hour and samples stored for various periods. The numbers of miRNAs with levels that increased more than 2-fold between one hour and six hours, one day, two days, and three days were 0, 5, 14, and 38, respectively. Thirteen of the 38 miRNAs that exhibited levels that were increased in samples stored for three days were also found to have increased levels in samples stored for two days. shows the mean expression levels of these 13 miRNAs. The miRNA with the highest expression was hsa- miR-451a, followed by hsa-miR-144-3p. These miRNAs have been reported as erythrocyte-derived miRNAs. The levels of hsa-miR-451a and hsa-miR-144-3p were found to increase by more than five- and sixfold, respectively, from one hour to three days. The expression level of hsa-miR-451a significantly increased from one hour to 3 days of storage. However, no significant difference was observed in the levels of hsa-miR-126-3p, suggesting that platelet destruction did not progress during storage. The expression levels of the platelet-derived miRNAs hsa-223-3p, hsa-let-7f-5p and hsa-miR-16-5p did not differ during the storage periods. In addition, to examine markers reflecting sample degradation, miRNA expression profiles in samples stored for each storage period were compared with miRNA expression profiles in samples stored for one hour after blood collection. The numbers of miRNAs with expression levels that decreased less than 1/2-fold between one hour and 6 hours, 1 day, 2 days, and 3 days were 0, 0, 4, and 8, respectively. The expression levels of hsa-miR-197-3p, hsa-miR-485-3p, and hsa- miR-130b-3p were observed to decrease in a time-dependent manner. It was suggested that these miRNAs could be useful as markers for monitoring the degradation of whole-blood samples. We found that hemolysis gradually increased during the storage period, and miRNAs that could be used for monitoring blood degradation were identified. From the results described above, it was found that the quality of whole-blood samples could be maintained if samples were stored within 6 hours at room temperature and within one day at 4°C. ## Stability of plasma over storage periods of up to 30 days at 4°C Depending on the laboratory conditions, plasma samples are sometimes preserved in cold storage before freezer storage. Thus, the stability of plasma samples stored at 4°C was investigated. To examine the time-dependent stability of samples during the storage period, plasma samples stored for one hour after separation from whole blood were used as the baseline condition and compared with samples stored for 1 day, 3 days, 5 days, 7 days and 30 days. Although the library concentration, identification rate and UMI ratio decreased in a time- dependent manner in all donor samples, significant differences among samples stored for the various storage periods were not observed. To investigate the time-dependent change in expression, miRNA expression profiles were compared between samples subjected to one hour of storage and samples stored for other periods. The numbers of miRNAs with levels that increased more than 2-fold between samples stored for one hour and those stored for one day, three days, five days, seven days and 30 days were 0, 0, 3, 0 and 3, respectively. However, the numbers of miRNAs with levels that decreased less than 1/2-fold between one hour and one day, three days, five days, seven days and 30 days were 0, 1, 1, 3 and 2, respectively. No common features were observed in either the miRNAs with increased expression miRNAs or those with decreased expression. Overall, the miRNA expression profiles did not show any remarkable changes after 30 days at 4°C. ## Stability of plasma over storage periods of up to 30 days at -20°C Plasma samples are routinely stored in a freezer until measurement experiments. Thus, the stability of plasma samples at -20°C was investigated. To examine the time-dependent stability of samples during the storage period, plasma samples stored for one hour after separation from whole blood were used as the baseline condition and compared with samples stored for 1 day, 3 days, 5 days, 7 days and 30 days. The library concentration, identification rate and UMI ratio did not significantly differ among samples subjected to the various storage periods. To investigate the time-dependent change in expression, miRNA expression profiles were compared between samples subjected to one hour of storage and samples stored for other periods. The numbers of miRNAs with levels that increased more than 2-fold between one hour and one day, three days, five days, seven days and 30 days were 1, 0, 1, 1 and 1, respectively. However, the numbers of miRNAs with levels that decreased less than 1/2-fold between one hour and one day, three days, five days, seven days and 30 days were 0, 1, 0, 1 and 1, respectively. No common features were observed among miRNAs with increased expression and those with decreased expression. Overall, the miRNA expression profiles did not show any remarkable changes after 30 days at -20°C. ## Stability of plasma samples from a biobank over storage periods of up to five years at -80°C Samples that have been stored for years in a biobank at -80°C are often used in the development of diagnostic tests using clinical samples. Thus, we investigated the stability of plasma samples obtained from a biobank. To examine the time-dependent degradation of samples over various storage periods, plasma samples stored for 3 years were used as the baseline condition and compared with samples stored for four and five years. The library concentration did not differ between samples stored for 3 years and samples stored for 4 years, but a significant difference was observed between samples stored for 3 years and samples stored for 5 years. The identification rate and UMI ratio did not significantly differ among samples subjected to the various storage periods. To investigate the time-dependent change in miRNA expression, miRNA expression profiles were compared between samples stored for three years and samples stored for four and five years. The numbers of miRNAs with levels that increased more than 2-fold between three years and after four years and five years were 0 and 1, respectively. However, the numbers of miRNAs with levels that decreased less than 1/2-fold between three years and after four years and five years were 0 and 0, respectively. Although a significant difference was shown in the library concentration between samples stored for 3 years and samples stored for 5 years, the miRNA expression profiles differed only slightly between samples stored for these periods. To examine the differences in the levels of hemolysis that occurred over different storage periods, the expression of hsa-miR-451a was measured and showed no significant differences between periods. During the examination of platelet contamination, the expression of hsa-miR-126-3p was measured, but no significant differences in expression were observed between periods. It was observed that the quality of the 5-year-old samples was maintained. From these results, we found that remarkable quality degradation was not observed in frozen samples stored for years in the biobank. ## Assessment of selected preanalytical conditions using clinical cancer samples We selected the preanalytical conditions based on the results obtained using healthy samples as follows. Blood samples collected using EDTA blood collection tubes were refrigerated within 1 hour, plasma separation was performed within 12 hours, and the obtained plasma samples were frozen at -80°C. Prior to the large- scale collection of clinical samples, the developed preanalytical conditions were validated using clinical samples collected on a small scale. For this purpose, a total of 60 cancer samples were collected from three facilities (20 samples per facility). First, the samples obtained from cancer patients were evaluated for equivalence among the three facilities in terms of the library concentration, identification rate and UMI ratio. Significant differences were not observed in these evaluations among facilities. This finding suggested that the properties of the samples were equivalent among different facilities. Next, to examine the differences in the levels of hemolysis in samples obtained from different facilities, the expression of hsa-miR-451a was compared. The fold changes in the levels of this miRNA in samples collected at facility A vs. facility B, samples collected at facility B vs. facility C and samples collected at facility A vs. facility C were 1.8-fold, 1.2-fold and 2.3-fold, respectively. Significant differences were observed between samples obtained in facilities A and C. However, the mean expression of hsa-miR-451a in samples collected at facility A was within a standard deviation of the mean expression of hsa- miR-451a in samples collected at facilities B and C. The same was true in comparisons of samples obtained from each individual facility with samples obtained from the other two. The variation in hemolysis levels between samples obtained from different facilities did not exceed the variation observed in samples obtained within facilities. In addition, to evaluate the differences in the levels of platelet contamination that occurred in samples obtained from different facilities, hsa-miR-126-3p expression was measured. There were no significant differences in hsa-miR-126-3p expression among samples obtained from facilities A, B and C. The expression levels of platelet-derived miRNAs hsa- miR-223-3p, hsa-let-7f-5p and hsa-miR-16-5p were not different among samples obtained from the different facilities. From these assessments it was revealed that major differences in the library concentration, identification rate, UMI ratio or platelet contamination were not observed among samples obtained from the three facilities. Although significant differences in hemolysis levels were observed among samples obtained from the different facilities, those differences were not based on the specific features of the facilities. The selected conditions were thought to demonstrate robustness in the sample collection methods used among facilities. We found that the optimal preanalytical conditions were established for the collection of samples from cancer patients in multiple facilities. # Discussion To determine the optimal preanalytical conditions for acquiring highly reproducible data, we investigated the effects of anticoagulants and sample storage conditions on miRNA expression profiles using plasma samples from healthy volunteer. Variances in the miRNA concentration, library concentration, human genome identification rate, ratio of unique sequences and expression profiles were compared between samples subjected to each condition. Then, the optimal preanalytical conditions for acquiring highly reproducible data were established. The selected preanalytical conditions included the use of EDTA as the anticoagulant, whole blood storage at room temperature for 6 hours or 4°C for 24 hours, and plasma storage at 4°C or -20°C for within 30 days or -80°C for within 5 years. Next, a protocol was formulated with these preanalytical conditions, and samples obtained from cancer patients were collected from three independent facilities. We found no significant differences in the distribution range of measurements among samples obtained from the different facilities. The preanalytical conditions identified in this study were shown to be practical for developing protocols to collect samples from multiple facilities. Plasma and serum are used at approximately the same frequency for cancer liquid biopsy. A review article of studies on cancer-specific miRNAs reported that of 154 cancer-specific miRNAs, 42% were discovered using plasma, 57% using serum, and 1% using both. To our knowledge, there is no consensus on whether plasma or serum should be used in discovery studies for cancer-specific miRNA markers. Regarding serum preparation, differences in miRNA expression profiles between samples obtained from different operators are assumed since there are factors that depend on the manipulation technique, such as the times used for clot formation and the intensity of inversion mixing. However, regarding plasma preparation, there are thought to be fewer technique-dependent variations because only centrifugation is needed after blood collection to obtain plasma samples. For these reasons, we selected plasma so that a robust measurement system could be established by minimizing factors that depend on manual procedures when samples are collected across multiple facilities. We investigated time-dependent differences in miRNA expression profiles when whole blood was refrigerated for 3 days, and we found that the expression of erythrocyte-derived miRNA (hsa-miR-451a) increased proportionately in a time- dependent manner. In contrast, the levels of hsa-miR-197-3p, hsa-miR-485-3p and hsa-miR-130b-5b were found to decrease in a time-dependent manner. Regarding the storage of plasma samples, it is known that the expression levels of hsa-miR-16 and hsa-miR-21 decrease in a time-dependent manner to approximately 60% and 70% over 100 days at 4°C. Regarding the storage of serum samples, it was reported that miRNA concentrations showed only a slight difference between 2 and 4 years of storage at -20°C, whereas miRNA concentrations decreased to approximately 1/2 after 6 years and to 1/4 after 10 years. However, to the best of our knowledge, there are no miRNAs that exhibited a time-dependent decrease in expression in whole-blood samples, and this is the first report showing that hsa-miR-197-3p, hsa-miR-485-3p and hsa-miR-130b-5b are potential markers for tracking sample degradation in whole blood. The expression of erythroid-derived hsa-miR-451a was more than 2-fold higher in plasma obtained from NaF and SC tubes than in plasma obtained from EDTA tubes, and 7 of the 25 miRNAs that were highly expressed in plasma from both NaF and SC tubes were erythroid-derived. The degree of hemolysis was higher when blood was collected in NaF or SC blood collection tubes than in EDTA blood collection tubes. We speculated the following on the basis of these results. It has been reported that NaF reduces the amount of linolenic acid, a component of erythrocyte cell membranes, that is incorporated into the cell membrane to 1/3 or less, suggesting that intracellular miRNA leaks from the damaged cell membrane compartment. Although we could not find reports on the effects of SC on erythrocyte membranes, it was suspected that if the SC blood collection tubes contain 3.2% SC pH 5.5 solution, which is equivalent to 1/10 the volume of blood, the erythrocyte membrane could be damaged by the pH gradient when the first drops of blood mix with the SC. Therefore, we speculated that plasma collected in NaF and SC tubes would have higher levels of hemolysis than plasma collected in EDTA tubes. Considering the results of this study, which showed higher expression of several erythroid-derived miRNAs in plasma collected from NaF and SC tubes and the effects of NaF and SC on erythrocyte cell membranes, we recommend the use of EDTA as an anticoagulant to reduce hemolysis levels. The analysis of the expression of erythrocyte-derived hsa-miRNA-451a across facilities revealed that the expression distribution range among facilities was less than the expression distribution range among samples obtained at the same facility. These results demonstrate the potential for standardizing preanalytical conditions across facilities to establish a robust miRNA assay system. However, significant differences were observed between samples obtained from facility A and facility C. This finding also suggests that samples obtained from the same facility could show significantly different degrees of hemolysis among individuals. Therefore, when samples are collected from a number of facilities, it is assumed that a certain number of samples with a high degree of hemolysis could be included. A method to standardize the hemolysis level using the differential expression of hsa-miR-451a and hsa-miR-23a-3p has also been reported. To generate a classification model for multiple cancer types, precise miRNA expression profiles are needed. We speculate that these miRNAs could be applied as a quality criterion for data acquisition in the future. We examined the effect of the type of anticoagulant used for plasma sample preparation and the conditions of sample storage on the comprehensive miRNA expression profile. We concluded the following: (i) EDTA should be used as an anticoagulant in blood collection tubes; (ii) whole blood should be left at room temperature prior to plasma preparation within 6 hours and (iii) refrigerated within 24 hours; and plasma should be stored at (iv) 4°C or -20°C within 30 days and at (v) -80°C within 5 years. These findings are worth considering during determinations of the type of anticoagulant and storage conditions for samples because they contribute to reducing the variation in sample quality among facilities. # Supporting information We would like to thank the participants in this study for their involvement. We would like to thank Hana Takahashi, Hiroko Oka, Hiroyuki Hamada and Yuko Yamakage for performing NGS data acquisition; Hidetaka Nakamoto, Hitomi Fukushima, Keisuke Inagaki and Kiyohide Ishikura for the contribution of recruiting healthy volunteers. [^1]: The authors have declared that no competing interests exist.

# Introduction Depression is a mood disorder characterized by both emotional and cognitive symptoms. Despite the intense research in the field, the neurobiology of depression remains elusive, however emerging evidences place the glutamatergic system as central to the neurobiology and treatment of the mood disorders. Our understanding of the etiology of the disease is limited to a list of risk factors where genetic predisposition and environmental risk factors, such as stressful life events, are thought to interact. If the stress-induced mechanisms of depression differs from the endogenous (genetic) factors is not clear. Social isolation is known to be a strong stressor for both rodents and humans. In humans, social isolation is associated with higher risk of mental health problems, such as depressive and anxiety disorders and increased risk of mortality. In rodents, social isolation induces anxiety- and depressive-like behaviors, aggression and memory impairments, whereas social interactions has been shown to be protective against stress-induced changes. Recent work shows how neurobiological factors affected by stress can be studied to understand resilience to stress, an important way to understand psychiatric disorders Here we studied the effect of social isolation on the hippocampal glutamatergic system of a selectively bred strain of rats: the Flinders Sensitive Line (FSL). The FSL rat has been extensively used as a model for depression since it exhibits several depressive-like responses in behavioral tests. The strain was originally created to be supersensitive to acetylcholine esterase, and it is possible that the increases sensitivity to cholinergic agents is associated with the increased sensitivity to stress displayed by these animals, but it should be noted that FSL rats also have deficiencies in many other systems, including dopaminergic and glutamatergic transmission. Our lab have associated the depressed behavior in FSL rats with alterations in the hippocampal glutamatergic system, including impairments in synaptic transmission and plasticity (LTP), down-regulation of the glial glutamate transporter GLAST, and decreased levels of the NMDA receptor co-agonist D-serine. These identified hallmarks in the FSL rat make them an optimal model to investigate the potential converging endogenous and stress-induced mechanisms of depression. Endurance exercise has been show to protect the brain from the stress, and previous work has shown that running partly reduces the depressive like behavior in FSL rats. Thus, we used voluntary running as an intervention that reduces depressive symptoms and found that running differentially affects the mechanisms affected by stress but not the endogenous mechanisms. # Methods The *Flinders Sensitive Line* (FSL) is a selectively bred rat line derived from Sprague-Dawley rats. The animals used in this study were bred at the Karolinska Institute and the Ethical committee for animal research in Stockholm, Sweden approved all animal experiments. The animals had free access to food and water and housed in a controlled environment of 12-h light/dark cycle. 10–12 weeks old male Sprague Dawley (SD) and FSL rats were divided in three groups: group-housed (GH; 3–4 rats per cage), individually housed (SI) and individually housed with free access to a running wheel (34 cm in diameter) for five weeks (SI+R). Running behavior was recorded every tenth minute by a computer-based data system with customized software and animal run on average 1250 ± 146 m per day. Note that in the SI group the running wheel was present and accessible to explore during the whole period but blocked for running, in order to not have differences in environmental enrichment between the two groups. After 5 weeks, animals were anesthetized with isoflurane and decapitated. One of the hippocampi was immediately frozen down to -80°C and kept for posterior molecular analysis. The second hemisphere was cut into 400 μm hippocampal slices and LTP induction protocol was performed in CA1 area as previously described. ## Electrophysiology Briefly, slices were incubated in an interface chamber containing artificial cerebrospinal fluid (aCSF) (in mM): 130 NaCl; 3.5 KCl; 1.25 NaH2PO4, 24 NaHCO3, 2 CaCl; 1 MgCl and 10 glucose, pH 7.4) for at least 2 h and then transferred to the recording chamber. Field excitatory post-synaptic potentials (fEPSP) at Schaffer collateral (SC)-CA1 synapses were elicited at 10-s intervals with a bipolar concentric electrode (FHC, ME, USA) and a extracellular recording pipette (filled with regular aCSF) placed in the stratum radiatum. Input-output (I/O) curves were obtained. The stimulus intensity was set to approximately 60% of the intensity that triggered population spikes and was determined empirically for each cell. For measuring long-term plasticity (LTP) in the CA1 region, stimuli were applied every 60 s for at least 20 min before LTP was induced using three trains of high frequency stimulation (HFS; 100 pulses at 100 Hz applied at 20-s intervals). Synaptic strength was monitored for 60 min and calculated using the initial rising slope phase of the fEPSP. The data was normalised with respect to the mean fEPSP slope that was recorded during the last 20 min of the baseline period. I/O curves were constructed using the Prism 5 program (GraphPad software, Inc., USA) following software instructions: first the X values (Intensity) of the I/O curve data were transformed to log form and the Y values (response: fEPSP slope) normalized. Data was fit to a sigmoid curve defined in Prism and the best-fits values of the curves from the different experimental groups were compared statistically using the F test, which compares the difference in sum-of-squares with the difference you would expect by chance. The result is expressed as the F ratio, from which a P value is calculated. ## Western blotting Rats were anesthetized with isoflurane, sacrificed by decapitation and the brains were quickly removed to dissect the hippocampi. The tissue was immediately sonicated in MAPK-buffer (containing Triton-X, SDS, Tris-HCl, NaCl, EDTA and H2O) with two types of protease inhibitors (1:100 from Sigma \#P8340 and 1:10 from Roche \#04693124001) and one phosphatase inhibitor (1:7 from Roche \#04 906 837 001). The samples were centrifuged at 13.200 rpm during 10 minutes at 4°C and supernatant was kept for protein analyses. BCA colorimetric method was used to determine the total amount of protein obtained. Equal amounts of protein (30 μg) were loaded onto a NuPAGE 4–12% Bis-Tris gel, Novex (Life Technologies, Glasgow, UK) and transferred to PVDF membranes (0.45μm) Immubolon- FL (Millipore, Temecula, CA, USA). Detection was based on a fluorescent secondary antibody that was visualized using the LICOR (Lincoln, NE, USA) Odyssey infrared fluorescence detection system. The data were quantified using ImageJ software (NIH, Bethesda, MD, USA) normalizing the values with **β-**actin. The primary antibodies used were at the following concentrations: GluA1 (1:100; Millipore, Temecula, CA, USA), GluA2 (1:1000; Millipore, Temecula, CA, USA), GluN1 (1:500; SYSY, Gottingen, Germany), GluN2A (1:1000; Tocris Bioscience, Bristol, UK), GluN2B (1:1000; Abcam, Cambridge, UK), GLAST (1:5000; human EAAT1) (Abcam, Cambridge, UK) and GLT-1 (1:5000; human EAAT2) (Abcam, Cambridge, UK) and **β-**actin (1:10000; Millipore, Temecula, CA, USA). ## Enantioselective liquid chromatography—fluorescence detection Tissue samples were prepared by sonicating fresh frozen hippocampi in 500 ml of 0.1M NHClO4. Each standard/sample was neutralized with an equal amount of 0.1M NaOH, and 9 ml of the neutralized standard/sample was derivatized using 9 ml of the o-phthalaldehyde/N-isobutyryl-L-cysteine mixture. The enantioselective chromatography experiments were performed using a Shiseido capcell pak MG C18 column (Analis, Namur, Belgium). The compounds were eluted by gradient elution with a mobile phase that was delivered at a rate of 0.17ml/min. The gradient elution was performed using mobile phase A (0.025M phosphate solution, pH 9) and mobile phase B (methanol:water 60:40). For fluorescence detection, a RF-10Axl Shimadzu spectrofluorometric detector was modified by introducing a 2-ml semi- microcell (Shimadzu, Duisburg, Germany). The derivatives were measured with excitation and emission wavelengths of 340 and 450 nm, respectively. The integration computer programme, Clarity (DataApex, Antec, Zoeterwoude, the Netherlands), was used to integrate the chromatograms. A scientist blinded to the rat strain performed all analyses. ## Statistical analysis Statistical analysis was performed using the Prism 5 program (GraphPad software, Inc., USA). Statistical significance was tested using the unpaired two-tailed Student’s *t*-test and one-way ANOVA where applicable. # Results Basal synaptic transmission from hippocampal slices from FSL and SD rats that had been groups housed (GH), socially isolated (SI) or socially isolated with access to a running wheel (SI+R) was measured using extracellular field recordings to generated input-output (I/O) curves to assess the differences in the responses to stimuli of a given intensity (from 10–70 μA). In FSL rats, the I/O curves from the three groups were calculated independently and followed a sigmoidal function (r2 = 0.53, no difference between the groups), but while the GH and SI+R groups share the same curve (P = 0.74), the one from the SI group was significantly left shifted (\*\*\*P\< 0.0001), indicating an increase in basal synaptic transmission. No significant changes were observed in any of the groups of SD rats. We then showed that social isolation reduces long-term potentiation (LTP) in the FSL rats and, remarkably, running can prevent this reduction. At 45–55 min post- induction, potentiation in the GH and SI+R group was 121± 13% and 125 ± 6% respectively, versus 111 ± 8% in the SI group (\*P \< 0.05 vs. GH and SI+R; one- way ANOVA followed by Bonferroni’s multiple comparisons test;). As we previously reported (13), CA1-LTP was significantly decreased in FSL rats compared to SD rats (124.2± 8% vs. 148 ± 9%; \*\*\*P\<0.001; one-way ANOVA followed by Bonferroni’s multiple comparisons test;). Importantly, we did not observe any effect of social isolation or running on plasticity (LTP) in the SD rats confirming that the effect synaptic transmission and plasticity that we observe in the FSL rats upon social isolation is associated with their increased vulnerability to stress. To understand if the changes in I/O-curve and LTP were associated with presynaptic changes we studied PPF in slices from FSL rats, applying two consecutive pulses with the same intensity but at different intervals (from 50–750 ms). We found a reduction in the PPF ratio only at the 50 ms interpulse interval in the SI group (\*P\<0.05; one-way ANOVA followed by Bonferroni’s multiple comparisons test;), which was again reversed in the SI+R group. Previously, we reported that the reduced LTP in FSL rats compared to SD rats was partially due to decreased levels of D-serine, a NMDA co-agonist known to enhance synaptic plasticity. We therefore hypothesized that D-serine might be involved in the effect of social isolation on LTP. However, using enantioselective chromatography, we found similar D-serine levels in hippocampal homogenates from the three groups. To further explore the mechanisms responsible for the changes in synaptic transmission and plasticity due to social isolation, and the effect of voluntary running, we analyzed the expression levels of several proteins involved in glutamatergic transmission. We did not find any alterations in any of the NMDA-R subunits (GluN1, GluN2A, GluN2B; data not shown). However, we found that social isolation induces a significant decrease in the protein expression level of the AMPA receptor subunit GluA2 (53.3 ± 5.6% of GH; \*\*\*P\<0.0005 vs. GH; one-way ANOVA followed by Bonferroni’s multiple comparisons test; left), with no effect on the GluA1 subunit ( right). Interestingly, running partially rescued the decrease in GluA2 protein levels (72.7 ± 6.5% of GH; \*P\<0.05 vs. GH; one-way ANOVA followed by Bonferroni’s multiple comparisons test) ( left). Next, we observed that social isolation decreased the levels of the glial fibrillary acidic protein, (GFAP) and that this protein remains decreased in the running group (65± 6.6% and 67.1± 5.2% of GH, respectively; \*\*\*P\<0.0005; one-way ANOVA followed by Bonferroni’s multiple comparisons test). This is in agreement with findings from stress-induced depression and opposite to the increase in GFAP levels observed in the FSL rats compared to the SD rats. Giving these two opposing results we consider that more work is needed to interpret and understand the functional consequences of changes in GFAP levels. In addition, we also observed that the levels of the glial Glutamate Aspartate Transporter (GLAST), already reduced in FSL rats, showed a non-significant decreasing trend in the SI rats (87.1 ± 4.3% of GH; P = 0.08; one-way ANOVA followed by Bonferroni’s multiple comparisons test) that was further decreased by running (78.5 ± 4.2% of GH; \*P\<0.05 vs. GH; one-way ANOVA followed by Bonferroni’s multiple comparisons test) ( left). In contrast, the protein expression levels of the other glial glutamate transporter (i.e., GLT-1) was significantly decreased in the SI group (79.2 ± 4.2% of GH; \*\*P\<0.005; one-way ANOVA followed by Bonferroni’s multiple comparisons test) and fully restored by running (#P\<0.05 vs. SI; one-way ANOVA followed by Bonferroni’s multiple comparisons test) ( right). GLT-1 is not affected in the FSL model compared to SD rats but a decreased level of this transporter has been reported in the chronic mild stress model of depression. # Discussion The selectively bred FSL rat strain is extensively used as an animal model for depression, and one of its characteristic behavior is increased sensitivity to stress. One neurobiological mechanism associated with its depressed behavior is increased glutamate transmission and reduced plasticity in the hippocampus, conveyed by reduced levels of the glutamate transporter GLAST and reduced levels of D-serine. Here we show that social isolation further enhances glutamate transmission and decreases synaptic plasticity in the FSL rats by targeting a different set of proteins involved in the glutamatergic system: GLT-1 and GluA2. Interestingly, we demonstrate that five weeks of voluntary running protects the brain against social isolation but does not affect the mechanisms already affected in the FSL rat. A reduction in synaptic plasticity in the hippocampus is one of the most robustly proven neurobiological consequences of stress, whereas the change in basal glutamate transmission has been shown to increase, decrease or not change at all, depending on animal strain, stress protocol, age of the animals etc.. In our experiments, social isolation induced a potentiation of glutamate transmission (as assessed by a shift in the I/O curve). This increase is compatible with the observed reduction of the glutamate transporter GLT-1, since reduced levels of glutamate uptake will directly lead to increased levels of extracellular glutamate and a concomitant increase in glutamate transmission. A reduction of GluA2 seems more paradoxical, however, AMPA receptors lacking the GluA2 subunit have significantly different properties compared to GluA2 containing AMPA receptors, one of the most important being that they are permeable to calcium. Indeed, it has been shown that activation of GluA2-lacking receptors induces a retrograde signal that enhances release probability at the presynaptic terminal. Thus, a reduction of GluA2 could lead to increased glutamate transmission through an increase in release probability. We have previously reported that the decreased hippocampal synaptic plasticity in the FSL compared to SD rats is due to decreased D-serine levels, a NMDA co- agonist known to enhance synaptic plasticity. However, the effect of social isolation on synaptic plasticity is independent of D-serine, since we did not observed any differences in D-serine levels between these groups. Interestingly, we observed that although running restores LTP to the level of group-housed FSL, it does not increase LTP to the level of SD rats, suggesting that running does not affect LTP in the CA1 *per se*; rather it is protective from the deleterious effect of social isolation on LTP, maybe through the restoration of the synaptic proteins, GLT-1 and GluA2. On the other hand, social isolation has been shown to reduce hippocampal neurogenesis and one study in mice showed that running increases LTP in the DG by increasing neurogenesis in this area. In this sense, we cannot exclude the possibility that the beneficial effect of voluntary running on SC-CA1 synapses may be mediated by increased hippocampal neurogenesis. In fact, since the hippocampus is a feed-forward network it is possible that the increase in number of young dentate granule cells after voluntary running can be accompanied by functional changes in the CA1 area. In fact, the antidepressant effect of both selective serotonin reuptake drugs and voluntary running has been suggested to occur through neurogenesis in the DG. The astrocytic proteins GLAST, GLT-1 and GFAP are differentially regulated by social isolation and running. This fact illustrates the complex effect of physical activity and stress on the brain (for review, see). Both treatments have a direct effect on the HPA axis- and sustained physical exercise can reduce the response to stress on the HPA axis on several parameters, including reducing the levels of peripheral glucocorticoids. Paradoxically though, physical activity also activates the HPA axis functioning as a stressor. Finally, running also triggers other processes, such as oxidative stress, metabolic rate and blood flow, which affect the brain independently of the HPA axis illustrating the multiple and complex mechanisms underlying both the benefits and caveats of running. The fact that social isolation and running have differential effect on specific astrocytic proteins in our model highlights this complexity and places the astrocytes as main actors in the mechanisms underlying the stress and running response in the brain. In summary, our results show that in the FSL model of depression running counteracts stress-induced mechanisms of depression, but not endogenous mechanisms. Since most experimental depression models are stress-induced, these findings explain why physical exercise has consistently given good results in animal experiments, while effects are less clear in patients. They are also in line with the recommendation that endurance activity is a good therapy for stress-induced depression, and that running can increase resilience to depression in individuals with a genetic predisposition for depression by acting on compensating mechanisms. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** ML SB. **Formal analysis:** AVE IS MGG TF. **Funding acquisition:** EA ML. **Investigation:** MG TF EA LG. **Methodology:** MGG EA LG AVE IS TF. **Project administration:** ML. **Resources:** AVE SB ML. **Supervision:** SB ML. **Visualization:** TF MGG. **Writing – original draft:** MGG ML. **Writing – review & editing:** MGG ML EA AVE IS TF LG SB.

# Introduction Recent evidence from cognitive social neuroscience has accelerated our understanding of intricate social brain functions, including processes involving the perception of others and their apparent behavior. However, relatively little research has been conducted to evaluate agency and its role in intentional bias. Moreover, there is limited evidence regarding how the intentional brain can be differentiated from the visual brain. For example, some configural cues such as contingent movement of geometrical patterns trigger an agency or animacy detectors in the brain that can partially explain intentional agents such as other people's minds. We hypothesize that the specific intentional brain function of estimating others' mental states based on agency attribution is an extended version of the visual brain. This extension involves recruiting higher brain regions found in the temporo-parietal cortices like the superior temporal sulcus (STS). The social braininvolves consciousness of one's own and others' mental states, intentions, attitudes, beliefs and motives and, therefore, is closely related to the theory of mind (ToM) and intentional agents. The ToM requires the ability to estimate the intentional states of others. Estimating another's state of mind involves modeling the other person's intention, possibly by agency attribution and one's own past experience. Current social neuroscience studies suggest that the superior temporal sulcus (STS) and medial prefrontal cortex (MPFC) are likely essential components of the social brain region involved in intentional tasks. In order to examine this issue, we developed simple animations that manipulated intentional bias (higher- and lower-intention involved animations) by representing geometrical shapes as opposed to complex verbal or visual tasks. In their seminal research, Heider and Simmel (1944) and Michotte (1963) used simple moving geometrical patterns as intention-involving agents in a local environment (i.e., a house having walls and a door). In Heider and Simmel's classic experiment, observers were asked to interpret a moving-picture film in which three geometrical figures (i.e., a large triangle (“T”), a small triangle (“t”) and a circle (“c”)) moved in various directions. A rectangle (“house”) with a wall section that opened and closed as a door was also shown. In their original film sequence, the animation was as follows. When the door opened, “t” and “c” moved into the “house.” Then, “T” moved into the “house” and shut the door. Next, “T” and “t” fought and “T” won. Finally, “t” and “c” broke through the door and ran away from the house. This work suggests that moving shapes can simulate the actions of living beings and, therefore, can represent agents performing actions. Accordingly the moving shapes are perceived to have goals and to possess qualities of an intentional mind. Therefore, the moving shapes are likely observed as if they represent the intentional states of others. In his theory of interpersonal relations, Heider proposed that individuals perceive and create explanations for the behavior of other's, a process he called “attribution”. Researchers have documented that higher-order cognition involving concepts such as causality and agency can be elicited by observing interactions, but not by observing the independent random movements of simple geometrical objects. If animations could possibly evoke mental state attributions based on intention, we propose that attributions of a mental state can be applied to animated objects. If this supposition is true, it would suggest that the neural substrate associated with understanding intentional events would include the same substrate (i.e., the STS) that becomes active when watching an interactive animated object in cooperation with other regions. To date, however, there have been few empirical studies to investigate why and how these attributions are affected by animations containing objects with lower- or higher-intentional involvement. In mentalization studies in which the ability to estimate another's mind is required, the observer must infer and model the intentions of another person. In this type of paradigm, the observer models the behavior of the other person prospectively by using attributions that are represented as animated dots or cartoons. For example, Baron-Cohen et al. (1994) found a rCB (regional cerebral blood flow) increase in the orbitofrontal cortex of the right hemisphere during the TOM task. Abel, Happe, and Frith, using two triangles moving around the screen in one of three ways (ToM-like, in a goal-directed way, or randomly), compared the attribution of the mental state in autistic children having less TOM than that of normal children, finding that the former used mentalizing (ToM- like) descriptions less often than the latter did. In another study, Schultz et al. presented short animations to participants in which two moving disks appeared to be either interacting or moving independently from each other. Using fMRI, they found that activation in the STS increased in proportion to the degree of correlation between the motion of two disks, and that an increase in correlation increased the amount of interactivity and animacy the observers attributed to the two disks. Perception of animacy also influences interactive behavior. Recent fMRI studies using non-Heider & Simmel patterns showed that the STS is also activated by simple moving objects whose interactions appear causal or intentional and that the STS is involved in the representation of observed intentional actions. Saxe et al. presented a real movie of a human walking into a room with or without occlusion (e.g., bookcase), finding that the walking figure activated the right posterior STS, which appears to be sensitive to the relationship between the observed motion and local environment. They further hypothesized that the right posterior STS is involved in the representation of observed intentional actions. In a study using PET, Castelli, Happe and Frith presented participants with a silent, computer-generated animation involving two simple geometric shapes (e.g., triangles) that resembled Heider and Simmel patterns. They found that the STS, MPFC, and temporal regions, including the fusiform gyrus, temporal pole, and occipital gyrus, were activated. The investigators argued that these animations strongly evoked mental state attributions based on intentions and hypothesized that the ability to make inferences about another's mental state evolved from the ability to make inferences about another's apparent behavior. Their findings suggest that controlling the degree of intention from high to low evoked by animations that vary in attribution appears to be critical in this type of research. They had six adult participants observe an animation that involved two moving triangles that manipulated the degree of intention from high to low in three ways: 1) ToM-like, corresponding to high intention; 2) goal- directed, corresponding to intermediate intention; and 3) randomly, corresponding to low-intention intention. These stimuli could therefore be graded from random movements to goal-directed actions, and finally to complex intentional states. The primary goal of the current work was to evaluate the degree to which intentional bias could result in greater STS activation and less MPFC activation. Similar animations were used such that objects always stayed within the same local region. However, animations differed in terms of their movements. Specifically, some animations were designed to give a graded impression of either intentional-oriented interactions or mechanical-oriented movements. In other words, a primary aim of our study was to describe how the social brain is influenced by animations that evoke high intention relative to less or no intention. We sought to replicate and extend the findings of Castelli et al. using a larger sample and event-related technology, and by grading stimuli based upon random movements, goal-directed actions, and complex intentional states. # Methods ## Participants Twelve healthy, right-handed participants (4 males and 8 females; mean age = 25.2) and fifteen separate participants (11 males and 4 females, mean age = 25.8) were recruited for the fMRI experiment and preliminary rating study, respectively. All had normal or corrected-to-normal vision, and were screened for the presence of current or past neurological and psychiatric disorder. ## Ethics Statement The experiment was conducted in accordance with the guidelines of the ethical committees of the Brain Activity Imaging Center (ATR, Kyoto, Japan) and of Kyoto University. All individuals voluntarily participated in the study and provided their written, informed consent prior to study participation. ## Procedure The animations used in the study was modeled on that of Heider and Simmel. depicts examples of the five-second animations (moving from left to right) used. Two or three triangles of different colors (blue, pink, and green) moved around on a black background. These triangles corresponded to the “t,” “c,” and “T” stimuli used in the Heider and Simmel animation. Additionally, the animation had a “house” with a gap on its side wall. The upper panel of shows a high-intention-involved animation (rated 5.77 and corresponding to condition i = 1 in ; see movies for details). In our preliminary study (see below), one participant reported a ToM-like story corresponding to the high-intention-involved animation as follows: “When the door of the ‘house’ opened, the blue and pink triangles moved in. Then, a green triangle moved in. Green and pink fought and green won. Blue and pink broke out of the ‘house’ and ran away. Based on this script, the two triangles were chased and persecuted by the green triangle and each triangl moved in an interactive way. The lower panel of depicts a low-intention-involved animation (rated 1.79 and corresponding to condition r = 1 in ; see movie file in detail). In our preliminary study (see below), a typical response to a story corresponding to one of the low-intention-involved animations as follows: “Triangles moved merely randomly or drifting without interaction”. By varying the motion path of the triangles, 25 different pairs consisting of one high- and one low-intentionality animation were designed for a total of 50 animations. Interactive motion (two triangles chased and persecuted by the third triangle) was varied by the experimente. In order to test the effect of the number of objects, we used three triangles in all but six pair in which the green triangle did not appear. The animations were created and encoded using Adobe Flash CS3 (30 flames per second, 320×320 pixel). ## Preliminary study In the preliminary behavioral study, 15 participants rated each animation based on an intentionality score. The intentional score was rated using a Likert-type scale of 1 to 7 (1: not at all intentional; 7: highly intentional). Next we selected 25 “high” and 25 “low” intentionality animations. Observers were asked to rate intentionality between the blue object and the other objects based on their mutual actions. Pairs of high- and low-intention animations were created. Their paths of motion are shown in. The highest-intention animation was created in a manner similar to the Heider and Simmel pattern ( upper panel, which corresponds to i = 1 motion path). The lowest intention (i.e., random) animation was made by simple drifting ( lower panel, which corresponds to r = 1 motion path). We also made a series of different intermediate animation pairs for a total of 25 pairs ranging from (i = 1, r = 1) to (i = 25, r = 25), where i and r indicates intention and random, respectively. Thus, we matched animation to have a similar motion path length and time for all triangles within a pair. Based upon this design, it was expected that participants would judge the triangles in a pair (for example, i = 19, r = 19 shown) to be somehow different from each other in terms of intentionality, while triangles in another pair (for example, i = 1(highest), r = 1(lowest) shown in) would be much different from each other Thus, we created a total of 25 graded steps of stimulus pairs. Of the 25 animations, the mean intentionality score was 5.77 in the “high” group and 1.79 in the “low” group. A two-way repeated-measures ANOVA (intention×animation number) revealed a significant main effect for intentionality \[F(1,14) = 768.9, *p*\<.001\] and stimulus number \[F(24,336) = 4.82, *p*\<.001\]. We also found significant interaction between intentionality and stimulus number \[F(24,336) = 6.35, *p*\<.001\]. Multiple comparisons using Turkey's HSD revealed significant differences between high- and low-rated scores. Thus, we confirmed that the higher-rated group was significantly more sensitive to intention than the lower- rated group. T-tests comparisons between the number of objects (2 to 3 objects) found no differences in terms of intentionality. Based on these preliminary findings, we adopted all the stimulus objects tested for later experiments. ## fMRI session No participants who participated in the preliminary study participated in the fMRI study. In an fMRI session, an animation was presented one second after a beep tone and an evaluation screen appeared which asked the participant to rate the level of intentionality from one (high) to four (low). Participants made ratings by choosing from two sets of four buttons (one set for each hand). One trial took 17 s, resulting in a total of length of 14 min 30 s for each session. For the first session, fifty moving patterns were presented in random order to participants in a counter-balanced manner. Twenty-five patterns were presented to the high group and to the low group, respectively. In the second session, up- down reversed patterns from the first session were presented. The presentation of normal and up-down reversed patterns was counter-balanced for each participant. In the preliminary study, we confirmed that participants could easily decide a response after reading the agent's intention 3 s after presentation. Therefore, the fMRI scan began 3 s after the animation presentation. Animations were back-projected onto a screen viewed through an angled mirror. The size of each animation was 11.5°×11.5°. In one session, participants observed 50 animations presented in random order. The length of each trial was 17 seconds. ## fMRI data acquisition Whole brain images were acquired on a 1.5-T whole-body magnetic resonance imaging scanner (Shimadzu-Marconi Magnex Eclipse, Kyoto, Japan). Head motion was minimized with a forehead strap. Functional MRI was performed with a gradient echo-planer imaging (TR = 3000 ms, TE = 49 ms, flip angle = 90°, 5 mm slice thickness, FOV = 192 mm×192 mm, and pixel matrix 64×64). After the collection of functional images, T1-weighted images (154 slices with no gap) using a conventional spin echo pulse sequence (TR = 12 ms, TE = 5 ms, flip angle = 8°, FOV = 220 mm×220 mm, and pixel matrix 256×256) were collected for anatomical co- registration with the functional images. After image reconstruction, functional images were analyzed using SPM2 (Wellcome Department of Imaging Neuroscience, London, UK). Six initial images were discarded from the analysis to eliminate the non-equilibrium effects of magnetization. All functional images were corrected for between-slice timing differences in image acquisition and realigned to correct for head movement, which was less than 1 mm within runs. The functional images were normalized and spatially smoothed with an isotropic Gaussian filter (6 mm full-width at half- maximum). Low-frequency noise was removed by high-pass filtering (time constant = 128 s). We conducted the analysis using an event-related design. An onset of an event according to the data analysis occurred three seconds after an animation started based on the results of the preliminary study. Data were modeled by convolving the vector of expected neural activity with the canonical hemodynamic response function (HRF) included in SPM2 and modulated by ratings of intentionality (4-point scales: high for 4 and low for 1). Single- participant *t*-contrast images were then entered into second-level analysis using a random effects model for all participants. The levels of statistical significance for these analyses were set to *p*\<0.001 (uncorrected). # Results Two contrasts were specified per single-participant analysis: 1) Low versus High and 2) High versus Low. Low-intention involves activations under participant's button press 1 (highest) and 2 (higher) and high-intention involves that of button press 3 (lowest) and 4 (lower). As shown in and, fMRI revealed activation of three main areas when participants observed 25 low-intention-involved animations (low\>high): the left middle occipital areas including the calcarine sulcus/cuneus (BA17,18), the right lingual gyrus (BA18), and the right prefrontal gyrus in the middle prefrontal cortex (BA9). However, when participants observed 25 high-intention-involved animations and intentional bias was increased (high\>low), the activated areas extended to include the bilateral posterior STS sulcus (BA22/37/39), the right temporal pole (BA38), the bilateral inferior frontal gyrus (BA47:IFG), the premotor (BA6), the inferior temporal gyrus (ITG), the left supramarginal gyrus, and the left superior parietal lobule (SPL). We did not find any activation in the MPFC. # Discussion In this study, we sought to investigate the differential contributions of the areas involved in visual and intentional cognitive processes. Participants conducted tasks that required them to make social interpretations by looking at moving objects that were presented as low- or high-intentionally biased animations. By varying the stimuli, we varied the extent to which intentional cognitive processing was required, which facilitated the analysis of intentional and perceptual influences on various brain regions. Based upon event-related fMRI data, our results revealed activation of several visual areas including the calcarine sulcus/cuneus and the lingual gyrus (BA17, 18), which is near the fusiform gyru when the visual brain operated in a mechanical low-intention-involved context. The middle frontal gyrus is thought to maintain visual attention to groups of moving objects. In contrast, the fusiform gyrus is believed to play a general role in the representation of visual stimuli that signify intent, independent of the visual form. Our finding of activation in the lingual gyrus, which is near the fusiform gyrus corroborates with a previous study. As shown in, when the brain processes high-intention-involved interactive animations, activation in the posterior STS involving part of the supramarginal area increased. It has been demonstrated that the STS becomes activated while viewing animated geometrical figures portraying social interactions, and when evaluating the intentions of others. Using fMRI, Gobbini et al. reported that social animations activated an extensive portion of the STS including areas in the posterior STS as well as the inferior parietal lobule. In an earlier PET study, Castelli et al. presented animations that featured two characters (a large red triangle and a small blue triangle) moving on a framed white background similar to Heider and Simmel's pattern. The investigators presented each participant with three types of animation: 1) ToM (two triangles bluffing one another); 2) goal-directed (two triangles dancing together); and 3) random (two triangles merely drifting). These animations were displayed for approximately 40 s over the course of 12 scans and divided into two consecutive counterbalanced blocks consisting of cued and uncued animations. These animations were designed to evoke mentalizing and elicited activity in the STS relative to a random motion condition. The design of the current study improved that done by Castelli et al. in two ways. First, intentional biases were manipulated continuously from highest to lowest by 25 matched pairs selected from 50 animations using ratings from the participants in a preliminary study. Second, an event-related design was introduced to avoid prior knowledge by using a shorter presentation duration (5 s). Based on our results, it is likely that intentional bias may be controlled more by the STS than by the MPFC, particularly when brain responses to high-intention-involved animations are compared with responses to low-intention-involved animations. The STS has been hypothesized to be closely connected to the perception of biological motion. Studies using transcranial magnetic stimulation and magnetoencephalography have shown that the simulation of human walking induced by moving dots selectively activates a brain area on the ventral bank in the occipital extent of the STS and the right temporo-parietal junction. Furthermore, such animations may be similar to the Heider and Simmel paradigm. We show here that tasks tapping mentalization and agency attribution activated the same brain regions in the STS and temporo-parietal cortices including the supramarginal gyrus, inferior temporal gyrus, the temporal pole, and the SPL. One explanation for why we did not find activation in the MPFC is that we used an event-related design to avoid expectancy with a much shorter presentation time than the 30 s previously reported. Expectancy cueing and longer presentation time could also yield possible contingent activations in the MPFC in addition to controlling intentional bias in the STS. It is highly possible, therefore, that higher-intention-involved animations, such as the fight between the blue and green triangle used in the current study, was perceived by the observer as though he/she was participating in the action against an antagonis. Indeed, humans may possibly detect intentions in shapes, even when those shapes change their motion to face another object. Overall, we assumed that activation in the premotor cortex invoked a mirror system when a human acts and when the person observes the same action performed by anothe. This system may be important for understanding the actions of other people, and that of the geometrical shapes in our animations. Some researchers also speculate that mirror systems may simulate observed actions, thus contributing to ToM skills,. In the premotor area, a functional mirror system estimating others' intentions may contribute to activation of the IFG. In the current study, significant increases of activation in the IFG were observed only when the animations were actively viewed with intention. Therefore, it is possible that the IFG monitors intentional thoughts in the STS. In contrast, activity in visual areas, including the lingual gyrus, which is near the fusiform gyrus, was only found in conditions requiring less intentional involvement and passive viewing. With close interconnections to the STS, the IFG and the temporal pole provide internally-represented self and other's mental states. Rather than the MPFC per se, it is the ventral side of the IFG, close to the orbitofrontal PFC and temporal pole, along with temporo-parietal-junction areas including the posterior STS and supramarginal gyrus) that are possible critical components for the representation of another's mental state. Saxe et al. examined whether activation of the posterior STS, similar to the perception of intentionality, depends particularly on the contingency between an agent's motion and the environment by introducing short and long occlusions of a walking person's animation strip. They showed that right posterior STS activation occurred following the long occlusion (i.e., when a person remained hidden for a few seconds before re-emerging). In the current study, we found activation in the same region; namely, the bilateral posterior STS, using simple geometric animations depicting high-intention-involved action. The present study suggests that the posterior STS is involved in constructing an abstract visual description of another agent's intentional actions, without engagement of the MPFC. Based on the present results, it is possible that incoming animated information is decoded perceptually and integrated with contextual interpretation; the constituent product of these two processes can be understood either in terms of perceptual- or intention-involved behaviors. In their examination of the neural correlates of mentalization, Vogeley et al. (2001) used fMRI to investigate common and differential neural mechanisms underlying ToM and the self during the presentation of a verbal story, finding that a ToM task led to increased neural activity in the temporal pole, whereas the self-task led to increased neural activity in the right temporo-parietal junction involving the STS. Interestingly, our data corroborate theirs regarding the neural correlates of ToM despite the large differences in the methods employed. The ability to model another intentional mind using an animated patter could be an evolutionary innovation in the human social brain that developed from the perceptual brain. Further investigations are necessary in order to clarify this issue. # Conclusion To summarize, we investigated how the visual brain transitions to the social brain using event-related fMRI in the present study. Animations consisted of moving patterns evoking various mental states of attribution based on intentions. Among 25 pairs of animations, each participant rated the higher- and lower-intention animation according to their attribution of agency (i.e., internal or external). Results showed that activations of the posterior STS, ITG, IFG, premotor, temporal pole, supramarginal gyrus, and SPL occurred under high-intention–involved animations, whereas occipital, lingual, and middle frontal gyri were activated under lower-intention-involved animations. Findings of the present study suggest that as intentional stance increased, the portion of the social brain involving the representation of an agent's intentional actions became more activated. Thus, developing the capacity to model another's mind could be an evolutionary innovation in the human social brain that developed from the perceptual brain. Previous studies have implicated regions activated by higher intention in self-monitoring in the perception of biological motion and in the attribution of mental states, and regions activated by lower-intention in simple perceptual processing. In the present study, we report how the visual brain shifts to the social brain in an agency attribution experiment. We suggest that as agent attribution increases, the visual brain changes to the intention-assuming social brain and therefore possesses a flexible network for processing information about social interactions based on agency attribution. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: NO TI MO. Performed the experiments: NO TI MO. Analyzed the data: NO TI. Contributed reagents/materials/analysis tools: NO TI MO. Wrote the paper: NO TI MO.

# Introduction Mitigating the effects of climate change is a major global challenge. In November 2015, the 21^st Conference of the Parties (COP 21) of the United Nations Framework Convention on Climate Change (UNFCCC) in Paris, agreed to limit global temperature rise to less than 2°C compared to preindustrial levels (, Article 2:1(a)); reduce emissions from deforestation and forest degradation, and conserve forests, manage forests sustainably and enhance forest carbon stocks (REDD+). Measurement, Reporting and Verification (MRV) is a set of methods to monitor REDD+ activities and outcomes. MRV activities include measuring carbon stocks and the emissions of carbon dioxide and other greenhouse gases, reporting these measurements to higher (regional, national and global) levels, and verifying that estimations and reporting have been conducted correctly. Apart from major carbon stores, forests are also home to many people whose lives are affected by deforestation. The interests of local communities appear to be poorly reflected in the internationally-driven REDD+ processes (,), and it remains unclear how local people can contribute to and benefit from MRV. REDD+ is increasingly not only about emissions reductions, it also includes social safeguards, livelihood benefits, biological conservation, and sustainable landscape development. As an essential component of REDD+, MRV should monitor all these aspects. The UNFCCC encourages local participation as part of REDD+ implementation and monitoring so that community members and other local stakeholders become empowered participants in the process rather than mere bystanders. Countries are encouraged to include social and environmental safeguards, clarify benefit sharing among the different stakeholders involved in REDD+, and propose alternative livelihoods to prevent further deforestation and forest degradation. One of the ethical underpinnings of REDD+ is ‘to avoid doing harm’ to local communities. These principles also apply to MRV and PMRV. Given the potential value of community participation in MRV for REDD+, how can it operate in practice? Past case studies have focused on tree measurement and biomass estimations (see this collection, henceforth indicated by \*,, ), leaving the other aspects of PMRV largely unexplored (,,). How feasible is PMRV for the whole range of REDD+ concerns (e.g. carbon sequestration, drivers of forest cover change, social and environmental safeguards)? In this collection we address this vital question through two review papers, and nine empirical studies based on research in Indonesia, Ethiopia, Mexico and China. This introductory article synthesizes the general lessons from the collection at the time of publication. We highlight five crucial considerations for ensuring PMRV is feasible and then examine some broader lessons and priorities for future research. The five considerations include the need to: (i) clarify who the **stakeholders** are, (2) understand stakeholders’ **motivation** to engage in MRV, (3) **integrate knowledge** from multiple disciplines and actors into MRV activities, (4) convey MRV information across **multiple levels of governance**, and (5) understand the links between PMRV and **REDD+ safeguards.** # Five crucial considerations for PMRV ## 1. Clarify the stakeholders PMRV depends on the collaboration of local communities, NGOs, local governments, the scientific and the international community. ### Meaningful involvement of local people The role of local people is currently limited to trainees or resource persons for collecting data on land use changes. Their participation needs to be integrated at higher levels, including the design of PMRV activities, and they need feedback from the national level on how decisions about forest management will impact them. Risks for local livelihoods (e.g. important economic activities being forbidden because of their environmental impact, see Bong et al\*) also need recognition. Information based on community perceptions of forest cover changes, and indicators of social and environmental safeguards can ensure that MRV systems monitor beyond emission reductions. However, communities are heterogeneous, with people having different views and aspirations. There are practical limits to generalizing data and describing the population as a whole. Such limits are often unclear in reports that often describe inadequately how the data was collected and aggregated (see\*). PMRV activities that plan to make use of community members’ perceptions need to be mindful of the methods by which the perceptions were elicited from individuals, and used on behalf of a larger community. What are the benefits and caveats of engaging local communities in MRV? Analyzing the literature on PMRV, Hawthorne et al. found several claims to how PMRV can benefit REDD+ programs, especially with regard to producing accurate, low-cost carbon measurements. In contrast, there is little evidence of negative effects of PMRV with regard to equity in benefit sharing and inaccurate reporting by local people if financial rewards are linked to performance. PMRV claims have seldom been based on empirical findings. This may be due to the limited number of empirical PMRV studies, each with its own limitations, and few opportunities to test how PMRV affects existing REDD+ components. ### Involving other stakeholders PMRV requires more than local people. Setting up the program, training, supervision, trouble shooting, quality control and data interpretation will all require a certain expertise. Local NGOs working with local people can facilitate and also relay reported data collected by/with local people to the national level (\*). The government should aggregate data from subnational sites, integrate the data into the national database, validate the data collected and report the outcomes to the international level. ## 2. Understanding stakeholders motivations to participate Understanding the motivations and incentives of local people to participate in an affordable and effective PMRV system is important to create a local sense of ownership and ensure sustainability of PMRV, with local people available and interested in taking part in long-term activities. Ekowati et al\* explored the case of the Indonesian child health monitoring program, centered around the *Posyandu* (village healthcare posts). This PMRV system has existed for decades, reaching the most remote, rural parts of Indonesia and offering only minor financial incentives to the (mostly women) collaborators. The authors found that the main motivations for participating were (1) personal interest in the work, (2) belief that their participation benefits the community, and (3) engagement by respected persons. While collaborators receive small payments, they are incentivized by other, non-financial, benefits, such as recognition by their community and others within the health system, and by the opportunities they gain to participate in training programs. Also Felker et al.\* found that non-financial incentives were important for carbon monitoring (e.g. for tree measurements), especially incentives in the form of increased legal recognition of land rights. However, such incentives are related to local land tenure arrangements, which differ across their study sites. For example, different parts of a village forest are generally under various authorities, based on either statutory or customary claims. These various authorities could guide PMRV actors and activities. Recognition of customary lands and associated rights may encourage communities to participate in REDD+ activities. Often tenure and boundaries of these lands are unclear and contested. Beaudoin et al. \* found that participatory mapping can provide a forum for discussion between villagers and government authorities to clarify village and individual land boundaries. These discussions can help identify and solve possible conflict over land use and land tenure between neighboring villages and/or with the government. Perceived land tenure security/insecurity is probably more important in this regard than legal tenure security. For example, people in the Java study sites do not feel secure with land certificates, which are arguably the most legally secure land documents available. Some have agreements with Perhutani (i.e., the Government owned Forestry Enterprise), which grants them limited access and user rights to the tree plantations. They feel insecure because they do not know how long the agreement(s) will last and what will happen next. In contrast, people in Papua who do not have any land certificates, have a strong sense of ownership and trust in customary land tenure that–according to the villagers interviewed–surpasses statutory tenure. PMRV activities in a village forest would need first to secure authorization from the village head, representing the statutory authority. However, a village forest (or parts of) can traditionally belong to groups or clans under the authority of a traditional leader (e.g. in Papua), and so securing authorization from these traditional leaders would be an essential second step. The traditional leaders could also provide guidance in selecting villagers to help with the activities. When discussing the potential benefits from PMRV activities, the heterogeneous nature of communities must be considered. For example, some villagers might prefer payments to benefit the entire community, while others may prefer individuals to receive the benefits, i.e. those directly involved in the PMRV activities or having customary rights to the land where activities are conducted. Bong et al. suggest benefit sharing is judged on how much local livelihood activities are impacted by REDD+ with those suffering greater costs receiving greater compensation. Benefit sharing arrangements should take into account the potential conflicts that may arise when different user groups within a village lay claim to land, resources and forest-based livelihoods. ## 3. Integrating knowledge and information from multiple disciplines and sources Participatory MRV requires the collection of data at the local level and the use and integration of different sources of information. In some cases, using a combination of modern technology and local knowledge can benefit MRV. The use of GPS and smartphones, as well as biomass assessments, data analysis and reports writing skills can also contribute to MRV. Although local people know the land use history of their territory and the drivers of deforestation and forest degradation, this knowledge will still need to be standardized for it to be integrated into the national MRV system and databases. Data collection can be standardized as part of an integrative and interactive system that builds communication and sharing among the different stakeholders (e.g. villagers, government MRV agencies, civil society). Pratihast et al.\* describe the design and implementation of such a (web-based) system to monitor the UNESCO Kafa Biosphere Reserve in Southwestern Ethiopia. Information collected through this web-based system was consistent over time and aggregated in a standardized way. Their system supports PMRV because it is interactive, dependent on local monitoring data, and provides feedback to local people, something Ekowati et al. highlighted as vital for PMRV. DeVries et al.\* give another example of knowledge integration from Ethiopia. They combined a stream of data from smart phones used by local people with Landsat time series to characterize and map deforestation and forest degradation. The locally recorded forest-changes offer deeper understanding of forest change processes recorded by satellites, especially the changes below the canopy. Bong et al. emphasize the importance of local knowledge of deforestation and forest degradation dynamics at the local level. Integrating this local knowledge is vital for MRV success because it allows for site-specific assessment and monitoring of drivers of land use change. Vega Praputra et al. suggest using simple data formats, based on existing forest reporting systems, which are agreed by all stakeholders. This should include a clear description of community reporting responsibilities and benefits. Beaudoin et al. stress that MRV costs could be minimized by limiting the sampling area for ground-truthing to areas where remote sensing and participatory mapping data are inconsistent. This can save time and may trigger local interest and participation in ground checks (i.e. verification) and carbon measurements. Participatory mapping can provide data on social contexts (e.g., tenure status, historical events) and land use dynamics that are either impossible or difficult to accurately obtain from remote sensing. Pratihast et al. also consider that data collected by local communities can help validate data from remote sensing, and serve as an additional source of verification. Data on tree species and forest composition aids accuracy in estimating carbon sequestration. Mingxu Zhao et al.\* found that local people can identify tree species nearly as well as trained botanists, but at a much lower cost. Such local skills offer considerable promise for lower cost monitoring, though they may not be available in every community. ## 4. Conveying knowledge and information across governance levels The main feature of PMRV is the inclusion of local people in contributing to the data flow used in a national MRV system, and–conversely–in benefiting from feedback from the national system on the state of the forests they manage. Information needs to be conveyed across multiple levels of governance. Vega Praputra et al. explored data flow in the Indonesian forestry sector, and found that locally collected forestry data (e.g. yields and diseases of smallholder forest crops, forest disturbances, logging activities) could be merged with the national database in a consistent manner. For this to happen, they note the need for facilitators to be present at the local level (e.g., civil society organizations, government agencies, trained villagers). These facilitators should understand the national MRV system, possess the skills to aggregate data at the village level (e.g. using appropriate software), and ensure data accuracy, completeness, consistency and comparability. McCall et al.\* found that, in Mexico, community members with appropriate training and a balance of technical skills and experience can competently collect data needed for MRV. In the Indonesian healthcare system, villagers have for decades been taking measurements and reporting to the sub-district government health center, eventually reaching the national level. Data collected by villagers are considered accurate and are used by the national government to guide planning. As explained earlier, Pratihast et al. proposed a simple web-based system to integrate locally collected data into the national database in near real-time. ## 5. Clarifying and enabling the links to REDD+ safeguards REDD+ has adopted a set of social and environmental safeguards to ensure that its activities do not result in negative social and environmental impacts. The following principles were agreed during the 2010 Conference of the Parties (COP) in Cancun: transparency, participation of stakeholders, protection of biodiversity and ecosystem services, and respect for the rights of indigenous and local communities. Several studies have found that local people’s participation in MRV can contribute in checking if REDD+ safeguards have been implemented. PMRV can be used to collect information on synergies and tradeoffs between REDD+, social and economic wellbeing, and environmental integrity that would otherwise be difficult to attain. Bong et al. suggest that communities are best placed to determine promising options to address trade-offs between livelihoods and deforestation and forest degradation. According to McCall et al., local communities have experience and competence in observing and checking (i.e. monitoring) the status of their natural resources, such as water and forest quality, valued wildlife, and territorial infringements. Their interest in this monitoring is often linked to the perceived opportunities for ecotourism. Mingxu Zhao et al., support these observations and suggest that community-led collection of tree diversity data can contribute to monitoring biodiversity safeguards in REDD+ programs. # Is PMRV feasible? The studies in this collection suggest that PMRV is feasible and that adopting PMRV can provide opportunities for carbon and non-carbon monitoring to support REDD+ implementation in other ways too, such as: 1. Monitoring REDD+ safeguards 2. Monitoring co-benefits and benefit sharing of REDD+ interventions on the ground 3. Obtaining more accurate information on drivers of deforestation 4. Using multiple approaches for more robust information/verification systems However, the studies in this collection also indicate several caveats: 1. A feasibility study should be conducted prior to implementing PMRV to check the capacity and possible level of community involvement. Methods of engaging with local communities in MRV will need to be as diverse as the communities involved. 2. Incentives for maintaining local participation need to be developed and should incorporate important non-financial incentives and equity. For example, (i) plan the measurement activities with local people to ensure relevance to the local context (e.g. tenure arrangements, level of ecological knowledge, livelihood activities); and (ii) acknowledge the community’s role and rights as forest managers and beneficiaries, and provide new job and livelihood opportunities for participants. 3. To be ‘PMRV’ ready, national governments need to invest in a reporting and verification system and a national database designed to integrate locally collected data. This should include: (i) building the capacity of facilitators, (ii) simplifying and clarifying reporting procedures across multiple levels of governance; (iii) providing mechanisms to clarify monitoring and reporting responsibilities, and benefits from PMRV, which are often related to local land rights and livelihoods; (iv) using a tiered system that accommodates different levels of reporting capacity across communities and countries involved in REDD+ to progressively achieve the gold standards; and (v) developing PMRV standards and step-by-step guidelines to be included in GOFC-GOLD methods and procedures. In circumstances where local community tenure is unclear and funding and local motivation cannot be secured, a PMRV approach is unlikely to be viable–at least in the short-term. # Next steps for PMRV research Currently, PMRV is not widely practiced, however, some governments, including Indonesia, are considering including it in their national MRV systems. To assist in the building of such a system, we need to: 1. **Assess past experiences systematically**: PMRV pilot activities provide lessons on how to sustain local participation and highlight specific challenges. We need to identify how best to evaluate outcomes and lessons. 2. **Assess costs and benefits at multiple levels**: so far, the costs and benefits of PMRV have only been studied at the local level, e.g. comparing the cost of measuring trees with local people versus experts. We need to explore and determine the costs and benefits of scaling up PMRV at regional and national levels, and include all investments required to implement PMRV. 3. **Operationalize reporting**: practical tools are required to facilitate reporting from local to national levels. While we possess a good understanding of how local communities could be involved in data collection, and national databases are already operational in many countries implementing REDD+, we lack agreed formats for data collection and reporting, procedures for aggregating locally collected date and ensuring its inclusion in the national database. 4. **Operationalize verification**: practical approaches are required to facilitate verification. PMRV can be one among multiple approaches, together offering more reliable information than any one approach alone. Key to improved reliability is the recognition of discrepancies between data sets from different sources and effective actions to address them. Both remain unclear so far. While we are certain that PMRV can benefit from local communities' cross checking and monitoring, we need to develop the procedures to ensure that these benefits are forthcoming. 5. **Move from a project scale to a national scale**: developing a larger scale system for PMRV in REDD+ poses challenges. However, we can utilize long term experiences such as from the Indonesian healthcare monitoring system to learn how to build a viable, national, data collection and reporting system. 6. **Test and refine PMRV**: any PMRV implementation needs to build in learning and adaptation procedures. The pilot projects with REDD+ implementation are small scale and PMRV is not fully operational. A large- scale test of PMRV that includes collecting, reporting, and validating data, then integrating the data into a national database, and finally feedback would provide further lessons to guide these important developments. We would like to thank all the authors of the articles for their commitment and contribution to this collection. We also thank Glen Mulcahy, Miriam van Heist, the anonymous reviewers, and the editors of PLOS ONE for their valuable comments and editing. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** MB MH SA DS. **Formal analysis:** MB MH SA DS. **Methodology:** MB MH SA DS. **Supervision:** MB MH. **Writing – original draft:** MB. **Writing – review & editing:** MB MH SA DS.

# Introduction Evolution has provided us with a unique and valuable tool to respond adaptively to environmental changes through emotions. An increasing number of researchers conceive emotions as discrete, universal, evolutionary, functional, and organized system composed of behavioral, psychophysiological, experiential, and cognitive responses to perceived challenges and opportunities. Particularly, emotions organize the mind’s responses to specific adaptive situational demands. Consequently, recurrent features of the situation have progressively acquired a place in the architecture of the emotional process. For instance, sadness is usually triggered in response to a significant loss; thus, it is also characterized by physiological withdrawal reactions to recover from the loss. The recurrent need to manage a specific situation triggering the most fitness- promoting responses—relevant for adaptation—constantly shapes and refines emotion-related features. Therefore, these regulatory processes vary according to a given emotion \[e.g., 8\]. In the first decade of the twentieth century, several researchers have started a new line of research focusing on the link between specific *discrete emotions* and their management, that is, peculiar *emotion regulation* (ER) strategies \[e.g., 9, 10\]. They showed differential emotion regulation paths, especially for each negative emotion, such as sadness or shame. However, although the interest in regulatory processes and discrete negative emotions-link has increased recently, regulation strategies occurring within the unfolding of positive emotions are still an open issue, maybe, since benefits associated with positive emotions are slower to come, while detrimental effects of negative emotions are immediately evident and urgent. Specifically, positive emotions have been mainly conceived as desirable phenomena that people attempt to maintain or enhance. Only recently, an increasing number of researchers \[e.g., 5, 6, 13\] have devoted their attention to the relationship between other general positive (vs. negative) affect and specific emotion regulation strategies, such as *cognitive reappraisal* and *expressive suppression*. These two strategies can be deployed at the two poles of emotion regulation process, thus, virtually covering the full process of an emotional response. *Cognitive reappraisal* comes early and affects the onset of an emotional response, and it involves re-thinking a situation to change its meaning and emotional effect. On the other hand, *Suppression* is a response-focused strategy that moderates outward signs of an emotional response, already deployed and appraised, by decreasing concurrent emotion-expressive behavior. Cognitive reappraisal displayed a healthier pattern of emotional, cognitive, and social effects compared to expressive suppression. Trait cognitive reappraisal predicted reactivity to affective stimuli by increasing the positive effect and decreasing the negative one. On the other hand, trait suppression appraisal led to more negative affective states. Crucially, Barrett et al. found that individuals who were better able to distinguish discrete emotional states were also more competent in regulating negative and positive emotions compared to those who were less able to differentiate specific discrete emotions. In conclusion, fairly little is known about how these two strategies interact with specific positive emotions, which are crucial for our flourishing, and nothing has been investigated regarding cross-cultural variations of this link. Take the case of compassion. It should be conceived as a positive emotion serving specific interpersonal functions. Compassion can be defined as a “feeling that arises in witnessing another’s suffering and that motivates a subsequent desire to help”. It entails a commitment to and thus facilitation of a relationship. Therefore, this emotion should be downregulated to save resources and direct them only to specific people. This would be far more crucial if dealing with a stable tendency to experience compassion related to stable or recurrent circumstances, as suggested by Goetz, Keltner. With this regard, an emerging trend in emotion theories suggests that positive emotions should be considered not only as discrete states evolutionarily shaped to respond to specific demands but also as stable traits or dispositions to specific eliciting situations. Daily life is imbued with systematic attempts to modify even specific positive emotions that do not fit the social norms of a situation. For instance, we try to avoid an amused laugh every time our colleague we dislike shows up. We try not to exhibit an excessive proudful smile every time we beat all other competitors at work. We attempt to empathize when our tenderhearted friend talks about her new boyfriend at dinner. These eliciting situations can repeatedly occur, thus bringing forth stable emotional tendencies. However, while the relationship between discrete dispositional positive emotions and other durable tendencies, such as personality traits \[e.g., 30\] or specific strengths and virtues, has been deeply investigated, the way people usually regulate them has remained unexplored. Specifically, research has showed that in the domain of stable affect, not all emotion regulation strategies can be considered as equal. Despite this evidence, nothing is known about the unique pathway linking specific emotion regulation strategies to specific discrete positive emotions, whose peculiar functional value it is crucial to support people’ wellbeing and health. ## Measuring the disposition to experience positive emotions In line with this evolutionary-functionalist approach on dispositional positive emotions, Shiota et al., developed a self-reported instrument to assess individuals’ proneness to experience seven discrete positive emotional states, namely, Joy, Contentment, Pride, Love, Compassion, Amusement, and Awe. *Joy* is usually associated with a sense of strength and self-confidence in safe and secure situations that promote the exploitation of new resources and social connections. *Contentment* is a low-arousal emotion related to the evaluation of actual material resources as exceeding individuals’ needs. This emotion invites people to savor their time while becoming aware of the available resources that can be integrated into schemas on self and others. Therefore, *Contentment* can change both worldview and self-perception. *Pride* is a self-conscious emotion arising from success during a socially relevant activity. It increases our status, provides information about the level of social acceptance, and expands the possibility to exploit more group resources. Shiota et al. defined *Love* based on Bowlby’s attachment behavioral system. People perceive a positive emotion of closeness towards trustworthy and dependable persons who take care of them. *Compassion* consists of feelings of concern and nurturance towards other people, even non-kin, especially if they are children, helpless, and vulnerable \[e.g., –\]. Love and compassion facilitate social connectedness. *Amusement* arises when there is a shift between one’s expectancies and one’s experience, and it usually occurs during humoristic situations. Finally, *Awe* is a complex emotion arising when faced with vast and cognitively challenging stimuli. Awe is a prosocial, epistemological, and self-transcendent emotion. Shiota et al. showed that each positive emotion disposition of this scale correlates significantly with at least one of the BFI dimensions (i.e., Extraversion, Agreeableness, Conscientiousness, Neuroticism, and Openness to Experience), in line with the broadening nature of positive emotions. The only exception was Compassion, which showed a small and not significant correlation with Neuroticism. Albeit preliminary, these findings are in line with the assumption that personality traits could predict specific emotional tendencies. Moreover, studies on dispositional discrete positive emotions have been conducted mainly in English-speaking countries. Thus, although DPES was developed to capture potentially universal, functionally distinct, and discrete emotion constructs, as well as their relationship with other universal phenomena, the definitions of each emotion reflect English-language constructs. Therefore, it is relevant to test the DPES factor structure also in Italian, among other languages. In backing up this claim, a six-factor solution (instead of the original seven- factors structure) emerged from the exploratory analysis in the German validation of the scale, which was carried out at the item level and in a more recent validation study carried out by Dixson, Anderson, & Keltner. Specifically, Contentment and Joy loaded on the same factor, while other emotions loaded on the residual five factors. This study showed that emotions could be split into two subtypes. Pride, Joy, and Contentment appeared to reflect *self-oriented* emotions while Amusement, Love, Compassion, and Awe emerged as *other-focused* emotions, depending upon their main elicitors. In this regard, the evolutionary functionalist account suggests that emotions emerge quickly from specific eliciting situations and serve to manage them–usually–in the most appropriate way. Since different emotions arise from different elicitors and trigger different action tendencies, it has been suggested that specific strategies could regulate each emotional state. These strategies, which help people redirect the instinctive emergence of emotions, are considered the components of the “emotion regulation” process. However, to our best knowledge, no study has investigated the link between discrete positive emotions and specific emotion regulation strategies. Therefore, it becomes far more relevant to investigate whether the frequent use of a given emotion regulation strategy is associated with specific positive emotions and whether it can, eventually, predict them. In this study, we sought to investigate the link between specific discrete positive emotions and two main categories of emotion regulation strategies of emotion-regulation, that is, cognitive reappraisal and expressive suppression. Thus, we aimed to determine whether there can be differential paths linking the two emotion regulation categories and each positive emotion. We also aimed to test whether specific BFI factors could predict each discrete emotion disposition. In the following paragraph, we detail the direction of the causal link between emotion regulation strategies and dispositions to experience positive emotions. ## The current study To contribute to the current literature on emotion regulation and discrete positive emotions, we carried out an exploratory study aimed at investigating the relationship between the seven discrete positive emotions of DPES and two major emotion regulation strategies, namely, Reappraisal and Suppression. With this regard, while the relationship between discrete dispositional positive emotions and other durable tendencies, such as personality traits \[e.g., 30\] or specific strengths and virtues, has been deeply investigated, the way people usually regulate them has remained unexplored. Despite it has been demonstrated the unique predictive value of specific emotion regulation strategies on stable affect, nothing is known about the unique pathways linking specific emotion regulation strategies to specific discrete positive emotions considered as stable traits, whose peculiar functional value it is crucial to support peoples’ wellbeing and health. In this study, the DPES by Shiota et al. and the Italian version of the Emotion Regulation Questionnaire (ERQ) were administered to a sample of native Italian participants. ERQ assesses how frequently individuals use cognitive reappraisal or suppression strategies. Given the high degree of differentiation among discrete positive emotions, we hypothesized the existence of different correlation patterns between each emotion and each emotion regulation strategy, which can be explained in light of a functional model of emotions. Specifically, starting from evidence on the impact of peculiar ER strategies on stable affect and emotions, we assumed that the frequent use of a given emotion regulation strategy (cognitive reappraisal or suppression) could predict also discrete stable positive emotions. Moreover, since the definitions of each emotion reflected English-language constructs, as mentioned earlier, we also tested the factor structure of the translated Italian version of the DPES, as well as its concurrent and divergent validity. We also administered the Italian version of Positive and Negative Affect Schedule as a measure of concurrent validity concerning the DPES (see. PANAS is a well-validated measure of positive vs. negative affect in Italian. Finally, we also assessed the Big Five personality traits using the Big Five Inventory (BFI) Italian version. Therefore, the concurrent validity of the Italian DPES was studied by examining both the extent to which the different scales were associated with the two global dimensions of affect, as assessed by PANAS, and with Big Five factors, as assessed by the BFI Italian version. In line with previous validation studies, we expected positive correlations of all DPES factors with PANAS scales, as well as with Extraversion and Openness to Experience personality traits, and negative correlations with Neuroticism trait. Moreover, given preliminary but promising evidence on the link between BFI and DPES factors, we tested also the contribution of BFI factors to each discrete emotion. # Method ## Participants The Italian DPES version was administered to 532 participants from Italy (mean age = 25.2; SD = 5.698, range = 18–49) of whom 439 were women (mean age = 25.05; SD = 5.448) and 93 were men (mean age = 26.42; SD = 6.671). Sample size was established on the base of previous works that validated or used scale in US or European populations. Their religious orientation was as follows: 63.62% was Catholic, 0.38% was Buddhist, 0.56% were Muslim or Protestant, 34.77% indicated practicing a non-conventional religion or being Atheists. Mean years of education of the sample was 16.20 (S.D. = 2.812), ranging from a minimum of 5 to a maximum of 26 years. Mean years of education was 15.58 (S.D. =.310) for men, ranging from 8 to 22 years, and 16.32 (S.D. =.133) for women, ranging from 5 to 26 years. The participants were recruited through social networks and flyers. They completed the online version of the DPES voluntarily and received no credit for their participation in the study. The same participants also completed the Italian Emotion Regulation Questionnaire, Positive and Negative Affect Schedule Italian version, and the BFI Italian version on the same occasion. All the instruments are reported and described in the next section “Materials”. The study was approved by the Ethical Committee of the Università Cattolica del Sacro Cuore of Milan prior to data collection and it was conducted in accordance with the Helsinki Declaration. ## Materials In this study, the Italian version of the DPES was translated from the original English questionnaire, and it uses the same item numbering. The DPES is a self- administered questionnaire, composed of 38 items that measure the proneness to live seven discrete positive emotions. It was structured by seven different scales, each representing a specific positive emotion: Joy (6 items), Contentment (5 items), Pride (5 items), Love (6 items), Compassion (5 items), Amusement (5 items), and Awe (6 items). The 38 items were presented to responders in a seven-step Likert scale ranging from Strongly Disagree (1) to Strongly Agree (7). First, two bilingual translators (one expert in the field of emotions and the other naïve) translated Dispositional Positive Emotions to Italian, as suggested by Borsa, Damásio and Baindeira. Further, the accuracy of the translation was tested by a back-translation from Italian to English done by two different bilingual translators, both fluent in Italian and English. The back-translation was checked by one of the authors of the original version. Afterward, three expert judges compared the original and back to finalize the Italian version. ## Italian version of the ERQ Italian ERQ is a 10-item 7-points likert (1 = strongly disagree; 7 = strongly agree) scale consisting into two scales each measuring an emotion regulation strategy: suppression (4 items) and reappraisal (6 items). Participants were instructed to think about how they control (that is, regulate and manage) their emotions. Cronbach alpha for both original scales was moderate to high (reappraisal =.84; suppression =.72). ## Italian version of the PANAS The Italian PANAS version captures the two main clusters of the affective experience. Specifically, this questionnaire is composed of 20 adjectives measuring the two main dimensions of affective experience: the positive (10 adjectives) \[Positive Affect scale (PA)\] and the negative one (10 adjectives) \[Negative Affect (NA)\]. In this study, we adopted the trait version by instructing participants to report the extent to which they usually feel as described by each of the 20 adjectives. The Italian PANAS version showed an adequate internal-consistency reliability for PA (.83) and for NA (.87) using the trait time instruction. ## Italian version of the BFI The Italian Big Five Inventory (BFI) is a is a self-report questionnaire relying on the work of, which measures the five personality trait dimensions defined by the Five-Factor Theory of Personality: Conscientiousness (9 items), Agreeableness (9 items), Extraversion (8 items), Neuroticism (8 items), and Openness (10 items). It consists of 44 items on a 5-point Likert scale, from 1 “Disagree Strongly” to 5 “Strongly Agree”. The internal consistency range in this sample was α = 0.70−0.84. ## Procedure Participants completed the survey in Qualtrics survey software and were provided with an online informed consent. No minors were included. Data have been anonymized. Participants were fully informed regarding the procedure and objectives of this study. They were recruited through flyers, social networks and by word of mouth. They were first informed that they would be completing some questionnaires about positive and negative emotions. Then, first, they filled the first part of the questionnaire created to gather socio-demographic information. Secondly, they completed the Italian translation of the DPES. Finally, they completed the Italian version of the PANAS and of the BFI. # Results This study was designed to test the relationship between two emotion regulation strategies and seven emotion dispositions. This was the first time that this scale was administered to an Italian sample. Therefore, we first assessed the structure of the DPES to determine the number of components needed to adequately describe the psychological constructs of Positive Emotions Dispositions in the Italian sample. We then assessed concurrent and divergent validity with PANAS and BFI subscales. Finally, we examined the relationship between the two emotion regulation strategies and the resulting emotion dispositions. ## Primary analyses To examine the structure of the Italian DPES, an Explorative Factor Analysis (EFA) analysis was carried out on the original set of items. Data were analyzed by IBM SPSS Statistics software (Version 21, release 21.0.0.0 64-bit edition). No missing values were found. Skewness and kurtosis of DPES items indicated a normal distribution. The means of factors ranged from 5.704 (compassion) to 4.420 (love). To get a clear idea of the relationships between the seven emotional dispositions and their underlying latent constructs, we carried out an EFA. The aim was to uncover the underlying structure of this large set of variables (38 items) in an Italian sample. ## Structure of the DPES scale: Exploratory factor analysis Before carrying out the Exploratory Factor Analysis, a parallel Monte Carlo simulation analysis was run on the 38 items to determine the number of factors to retain in EFA. This analysis revealed a maximum of seven-factor structure. The seven-factor solution was confirmed by the eigenvalues, indicating that 7 factors had eigenvalues greater than 1.0, which accounted for 36.0% of the total variance. Inspection of the correlation matrix showed several coefficients of 0.30 and above. Therefore, we chose to test the 7-factor solution in line with the original structure of the scale, as suggested by Parallel Analysis. Moreover, given the nature of the scale, which assesses positive emotions, we expected that our factors were correlated. Therefore, first, we carried out a Principal Axis Factoring (PAF) analysis with oblimin rotation (delta = 0). Pearson’s r correlations among factors were closer to.3. The 7-factor solution explained 60.301% variance. This solution was also supported by Kaiser-Meyer-Olkin Measure of Sampling Adequacy (.90) and Bartlett’s test of Sphericity \[χ2 (703) = 9818.071; p \<.01\]. However, item 36 had a very low level of communality (.106) and items did not load clearly in the seven factors; therefore, we chose to trim item 36, and we ran the PAF again, with oblimin rotation forcing a 7-factor solution. Global explained variance increased (61.481%). However, most items still double-loaded on several factors, and the resulting factorial structure was not easily interpretable. Therefore, we chose to run PAF again, with oblimin rotation forcing a 6 factors solution, in line with the findings of Gusewell and Ruch (see for Factors correlation matrix using principal axis factoring with oblimin rotation 6-factor solution). This solution explained less variance (58.398%), but all items loaded on each factor more clearly, without double loading items or just one item per factor. Specifically, the final solution had all Joy (i.e., 1, 2, 3, 4, 5, 6) and Contentment (i.e., 7, 8, 9, 10, 11) items loading on the first factor, while the Pride, Love, Compassion, Humor, and Awe items each loaded on separate factors. Only items from the original Joy scale merged with those of the original Contentment scale, while Pride emerged as a separate factor. This is in line with the validation study of the German version of this scale by Güsewell and Ruch. Consistent with Güsewell and Ruch, we selected the 6-factors solution as the best model to explain our variables (see. for the final factor solution). Items loading on **Factor 1**, which we called “Happiness” (ten items), represent a positive attitude towards life in general (e.g., “I am generally a contented person”). This factor encompasses all items from the original “Contentment” scale (except for item 11 that loaded on Factor 5) as well as items from the original “Joy” scale (e.g.., “Good things happen to me all the time”; “My life is always improving”). Items loading on **Factor 2**, called “Compassion” (five items), represent a desire to take care of, or a concern for, another’s well-being, especially for those in need of help and vulnerable. Items loading on Compassion factors are the same as those loading on the original scale. Items loading on **Factor 3**, “Amusement” (five items), refer to positive feelings in humorous situations perceived as inconsistent with one’s expectations. Items loading on **Factor 4**, called “Love” (six items), are the same as the original factors called “Love.” This factor concerns trust towards other people who are conceived as straightforward and authentic. Items loading on **Factor 5**, called “Pride” (six items), the same items as those in the original scale along with the last item from the original “Contentment” scale (item: “My life is very fulfilling), capture a social recognition of self- identity (e.g., “Many people respect me”). Finally, items loading on **Factor 6** “Awe” (five items) all belong to the original scale. This Factor refers to a deep sense of wonder and astonishment when facing something grand to an extent to require an update of individuals’ current mental frames. Only original item 36 was omitted due to the low response variance (i.e., “I often look for patterns in the objects around me”). ## Gender differences for DPES factors in the Italian sample We also tested gender differences in each DPES factor, since Güsewell and Ruch found significant differences between men and women in joy, contentment, compassion, and awe. A one-way ANOVA with gender as factor showed that women (mean = 5.75; S.D. =.791) scored significantly higher compared to men on compassion (mean = 5.46; S.D. =.982) \[F(1) = 9.468; p =.002; partial eta square =.018\], while men reported significantly higher levels of amusement (mean = 5.24; S.D. =.904) compared to women (mean = 4.928; S.D. =.996) \[F(1) = 7.861; p =.005; partial eta square =.018\]. ## Scale reliabilities and concurrent validities A further aim of this study was to assess the internal consistency and reliability of the DPES scales. We calculated the Cronbach’s alpha coefficients for all scales. Internal consistency of each scale was high: “Happiness” scale (.906), Pride scale (.798), and the Awe scale (.794), with.80 for love.80 for compassion, and.75 for amusement. The Awe scale and the overall reliability increased compared to the original scale. This study also investigated the criterion validity of the scale, correlating it first with the two subscales of PANAS, which are self-report measures conceptually similar, assessing the same construct of experiencing positive emotions. Second, the correlations between DPES, BFI and ERQ scores were computed. The internal consistency and criterion validities were calculated using the data from the 532 participants. All correlations between DPES, BFI, and ERQ, as well as PANAS, were carried out considering the Italian structure of DPES (i.e., 6-factors structure). ### PANAS Reliability was calculated using the data from the 532 participants. The scale showed high reliability (Cronbach’s Alpha =.911). To calculate the scale criterion validity, we computed a Pearson correlation between the DPES and the PANAS positive and negative subscales scores (i.e., PA- Positive Affect; NA- Negative Affect). In this sample, the PA subscale of the PANAS had good reliability (Cronbach’s Alpha =.839), as did the NA subscale (Cronbach’s Alpha =.881). Factors from the DPES correlated positively with the Positive Affect Scales and negatively, or not at all, with the Negative Affect Scale. Most correlations were significant (See). ## BFI Reliability was calculated using the data from the 532 participants. The BFI scale showed a good level of reliability (Cronbach’s Alpha =.783). To calculate the scale criterion validity, we computed Pearson correlations between the DPES and the BFI subscales scores (i.e., Extraversion, Agreeableness, Conscientiousness, Neuroticism, Openness). Correlations are reported in. Finally, all emotions displayed significant positive correlations with Openness to experience personality factors. Given the correlations between discrete positive emotions of DPES and BFI factors, we chose to carry out six multiple linear regression models with Extraversion, Agreeableness, Conscientiousness, Neuroticism, and Openness as predictors and each discrete emotion as the response variable. The results are reported in. Given the correlations between discrete positive emotions of DPES and ERQ factors (.), we chose to carry out six multiple linear regressions to test whether trait cognitive reappraisal and trait suppression significantly predicted participants’ ratings of each positive emotion disposition. See for all multiple regression analyses with DPES factors as measures and each emotion regulation strategy (Cognitive Reappraisal and Suppression) as predictors. Cognitive Reappraisal was a significant positive predictor of all emotions of DPES. Suppression emerged as a significant negative predictor of all emotions of DPES, except for Amusement and Awe, which were not significantly predicted by Suppression. # Discussion This is the first study in which the relationship between specific discrete positive emotional tendencies and two of the major emotion regulation strategies of reappraisal and suppression was tested. To this end, we administered one stable measure of emotion regulation (ERQ)–already validated in Italian–and of discrete positive emotion dispositions (DPES), which was translated into Italian. Although DPES was intended to capture potentially universal, functionally distinct emotion constructs, the definitions of each emotion reflect English-language constructs. Therefore, we also tested the DPES factor structure, as well as its concurrent and divergent validity, in Italian. The Italian version of the DPES was validated through an analytic strategy that involved exploratory factor analysis and concurrent validity analysis. These steps yielded a six-factor scale, comprising 37 items out of the original 38 items, with acceptable to excellent alpha coefficients. In this regard, some adjustments had to be made to reflect Italian culture. Specifically, the dimension of the original scales of Joy (items: 1, 2, 3, 4, 5, 6) and Contentment (items: 7, 8, 9, 10, 11) underlined the same dimension in this sample, which we labeled “Happiness.” Moreover, one item of the original Contentment scale (i.e., My life is very fulfilling) loaded on the Pride factor. These findings could be due to the “other-oriented” and interdependent nature of Italian culture as opposed to English-speaking societies. While the first items of joy referred explicitly to joy, the other was related to satisfaction towards life and self-domain in general. Indeed, the concept of Joy in Italy should be based on interpersonal-relational elicitors more than on one’s own life. However, no DPES items explicitly referred to as *interpersonal* situations. Therefore, it could be possible that participants did not recognize peculiar elicitors or prototypical situations of joy while reading the joy-related items. Consequently, a macro factor encompassing all items of life fulfillment and life satisfaction, in general, emerged from a blend of Joy and Contentment. This scale was labeled “Happiness” since it entailed all aspects related to satisfaction with one’s own. Following the Contentment factor, Compassion and Love also showed a high frequency of occurrence, in line with the findings of previous studies on emotional lexicon usage in Italian samples \[e.g., \]. Finally, in this study, one Awe item (i.e., the mental shift: “I often look for patterns in the objects around me”) was removed due to its initial low communality with other items (.141), improving the fit the scale to our data. Finally, Pride had a high mean occurrence frequency. This result could be due to the renowned Italian disposition towards others, which could facilitate social comparison at the base of Pride. Finally, one item from the Pride original scale loaded on the general Happiness factor. It could be due to the semantic domain covered by this item “My life is very fulfilling,” which, in Italian, is close to the domain of self-accomplishment. The concurrent validity of the Italian- DPES scale was evaluated with PA and NA and the BFI traits. First, the results showed that Italian-DPES had moderate to strong positive correlations with positive affect and moderate correlations with negative affect. Specifically, all discrete emotions showed significant correlations with PA; only Amusement and Compassion showed positive (but not significant) correlation with Negative Affect. Furthermore, all dispositions showed significant correlations with BFI personality traits. All factors were significantly positively correlated with Extraversion, in line with Shiota et al.. On the other hand, while in Shiota et al.. All DPES subscales correlated significantly also with Agreeableness, except for Amusement. All DPES scales correlated negatively with Neuroticism, except for Compassion, which is also in line with Shiota et al.. Finally, all emotions displayed significant positive correlations with Openness to experience personality factors. Agreeableness predicted positively Happiness, Love, Compassion, and Awe, consistent with the claim of McCrae and Costa, who suggested that this trait facilitates positive affect. Conversely, highly conscientious people were less inclined to experience life situations as amusing and to feel the love in their daily activities. Conscientiousness is defined as the propensity to respect social conventions to control impulses and to achieve personal goals, and some evidence demonstrated that only certain specific amusing situations were positively associated with this personality trait (i.e., self-enhancing humor as a coping strategy). It would be relevant to investigate amusement-related elicitors that the participants considered while completing the scale. At the same time, Conscientiousness predicted Love negatively. This result is surprising since previous research demonstrated the positive role of Consciousness in determining different facets of love (i.e., intimacy, passion, commitment, relationship satisfaction) \[e.g., \]. Additionally, Shiota et al. showed that Conscientiousness and Love are positively, even though not significantly, related. However, all these studies were conducted in English- speaking countries. A possible explanation of this finding is that the Italian culture defines love differently from other Anglo-Saxon cultures. On the other hand, Conscientiousness predicted Compassion, Pride, and Happiness positively. Neuroticism predicted all emotions negatively except from Compassion, which is also in line with Shiota et al.. Almost all emotions displayed significant positive correlations (and were predicted by) with Openness to experience personality factors, consistent with the literature on positive emotions. However, Openness to experience did not predict Happiness, and it predicted Love negatively. This was surprising since DPES measure a Love directed towards other people. Italian people are perhaps more inclined to address this Love only to their intimate partners and not to others. Finally, and importantly, our results revealed significant correlational patterns between each emotion, suppression, and reappraisal emotion regulation strategies, in line with previous studies. Reappraisal promotes better management of stressful consequences; thus, it can increase the likelihood of experiencing positive states. Cognitive reappraisal was positively associated with all positive emotion tendencies, thus supporting its main adaptive evolutionary function. Crucially, cognitive reappraisal strongly predicted Happiness, Awe, and Pride. This result is in line with the existing literature on Joy and positive emotions-emotion regulation strategies in general and suggests that Italian people may prefer reappraising situations in terms of happiness, pride, and awe, which usually trigger a cognitive restructuring people’ actual worldview. Indeed, reappraisal is an effective strategy also used to decrease positive affect. Therefore, participants could have activated this process also to down-regulate positive emotions, that is, to diminish their intensity. For instance, the high uncertainty related to awe-inspiring situations could have prompted participants to diminish its intensity and turn it into something more manageable and acceptable. On the other hand, suppression also intervenes in modulating the outcome of emotional response at the expressive level; thus, we assumed, again, a causal link from this strategy and each emotion. In line with the literature on positive affect, we also found that suppression predicted Happiness, Pride, Love, and Compassion negatively and significantly. However, amusement was positively, but not significantly, associated with suppression. Maybe because Amusement is associated with both adaptive and maladaptive humorous responses to stressful situations; thus, sometimes, Italian people may need to regulate it to fit the delicate demands of a social situation. # Conclusions In this study, for the first time, the relationship between specific emotion dispositions and the two emotion regulation strategies of reappraisal and suppression was tested. Each positive emotion tendency correlated positively with reappraisal and negatively with suppression, except for Amusement, which showed a positive and even non-significant correlation. This supported the adaptive value of cognitive reappraisal in the emotion regulation process. We also conducted a cross-cultural validation of the Dispositional Positive Emotions Scale in an Italian sample. The results showed an excellent internal consistency of this scale as well as a convergent validity with the PANAS factors. All BFI factors were significant predictors of almost all discrete emotions. Although these findings are promising, future research should consider also other cultural and contextual-specific factors to deepen these aspects. For instance, it could be useful to focus on different religious Italian samples, since some emotions, such as awe or compassion, can be related to this aspect. Specifically, it would be interesting to elucidate the peculiar situational elicitors of cognitive reappraisal of positive emotions, to understand when people tend to reappraise a situation as happy and thus capture cultural variations of this positive emotions. Specifically, DPES cannot connect peculiar positive emotion elicitors with systematic ER strategies. This can be a useful aspect to be deepened by future studies, for instance, by integrating also the Ecological Momentary Assessment or daily diaries. Moreover, the final Italian DPES structure did not match the original English factor structure. Future studies should corroborate our finding and it could be useful to add other items to DPES to differentiate between Joy and Contentment in the Italian population. Further, our sample composed of women for the most part. Therefore, future validation studies should investigate the role of gender in emotion disposition and regulation. Finally, adopting a finer granularity of emotion regulation strategies (e.g., different reappraisals strategies) to manage positive emotions could provide useful information, especially about the understudied Italian population. In this regard, the main discrepancies with previous studies could be explained in terms of cultural variations. This observation may support the assumption that emotions can be conceived as containing prototypical elements (*core themes*), even though their peripheral aspects can vary according to the context. Therefore, the present cross-cultural validation study may allow capturing both core emotional themes, and their variations also related to other phenomena, in this case, personality traits. This result maybe relevant also in the clinical field, since it could help clinicians design tailored interventions for people suffering from personality or affective disorders, in relation to the emotion regulation strategies they usually adopt to manage positive emotions \[e.g., \]. In brief, the Italian DPES is easy to administer and can provide quick information on frequently experienced emotions and their link with other emotional or cognitive processes across a wide array of domains. Overall, the findings demonstrated specific and strong links between cognitive and emotional depositions and evidenced that Italian DPES has good validity and reliability and is suitable to be used with an Italian sample. [^1]: The authors have declared that no competing interests exist.

# Introduction The metabolically versatile gram-negative bacterium *Pseudomonas aeruginosa* has an extraordinary ability to adapt and colonize several ecological niches. This opportunistic pathogen has dispersed globally, causes a variety of infections, and is frequently acquired in hospital environment, such as intensive care units (ICU). Because it represents a major health burden in hospitalized patients, and due to its scientific and medical interests, several studies have investigated the dissemination of virulent or drug-resistant clones. The analysis of bacterial pathogens by various typing methods provides important information for establishing the genetic relatedness among isolates for the purposes of long-term epidemiological studies and/or outbreak investigations. Hence, strain typing has two major aims: (i) to index genetic microvariation for use in outbreak investigations and (ii) to index genetic macrovariation for use in phylogenetic and population-based analyses. Generally, phylogenetic methods were developed to elucidate evolutionary history of strains and depicting their patterns of relatedness. An accurate phylogeny attempts to unravel the crossed wires of evolutionary relationships among samples of populations and elucidates how they are related and diverged from each others. In addition, it provides the ability to predict the phenotypic and genotypic traits, permitting a better understanding of the ecology and the dynamic of the bacterial population biology. Phylogenetic analyses are often conducted to determine whether one particular outbreak may be related to another during times of an epidemic, and whether certain strains are more often associated with outbreaks than others. For over 20 years, genome fingerprinting was a tool of pivotal importance to track the transmission routes of microbial pathogens. It is a highly useful strategy for surveillance and outbreak investigation of human infections and evolutionary analysis. Several molecular techniques have been successfully applied to differentiate bacterial strains and clonal groups. The major success factor of a typing system lies in the discriminatory power, i.e. the ability to distinguish between epidemiologically related and unrelated strains. Each technique has its characteristics and its applicability and the limitation of such techniques can be discerned. In some cases the inference of relationship among isolates by one typing method does not correlate with that obtained by another. Typing methods have evolved from phenotypic methods to genotypic methods. Several schemes have been adapted to type *P. aeruginosa* associated to cystic fibrosis patients and isolates featuring antibiotic resistance determinants. The traditional phenotypic markers are known to be unstable and do not offer satisfactory resolution power for discrimination of strains. A wide variety of molecular genetics methods were generally introduced to fulfill those shortcomings. Some of them were PCR-based methods such as Random Amplified of Polymorphic DNA (RAPD) and rep-PCR. Other techniques were based on Restriction Endonuclease Analysis (REA) of the total genome such as Pulsed-Field Gel Electrophoresis (PFGE), or partial fragments of the genome such as ribotyping. PFGE has been regarded the ‘gold standard’ molecular typing method for a variety of bacterial species for outbreak investigation purposes. Another technique, Amplified Fragments-Length Polymorphism (AFLP) has been shown in some studies to offer a discriminatory power comparable to that of PFGE. Technically it can provide results more rapidly than PFGE, but it has several limitations in terms of being challenging to standardize, labor intensive, and also relatively subjective. Given the need to validate new portable and standardizable approaches, Multi- Locus Sequence Typing (MLST) schemes, were developed, which are capable of measuring the genetic variation in house-keeping genes. MLST is fully standardized for numerous bacterial species, and is able to detect phylogenetically informative genetic variations retrieved from the strictly conserved studied genes. Therefore, this scheme is capable of differentiation between strains, and of accurately tracking the global clonal history of different species. For several pathogens MLST is a preferable tool, especially when epidemiological, geographical and/or evolutionary studies need to be carried out. Another relatively new technique, Multilocus Variable Number Tandem Repeat Analysis (MLVA), has been successfully used for epidemiologically studies for several species. Onteniente et al. developed the first MLVA scheme for *P. aeruginosa* which consisted of seven VNTR markers. Later, the MLVA scheme (MLVA-15) was updated by improving the protocol and adding new discriminative markers, making the assay more robust and discriminatory. The MLVA-15 scheme is thus assumed to be highly informative and useful for epidemiological surveillance of *P. aeruginosa* infections. Moreover, MLVA has become a useful tool for outbreak detection and source tracing in European countries. Interestingly, it has been reported that an accurate phylogeny could be established from an MLVA scheme containing several VNTR loci displaying a wide range of diversity. More recently, a new rep-PCR-based technique was introduced: the DiversiLab system (DL, bioMérieux, Marcy l'Etoile, France). This system gives more reproducible results than standard rep-PCR approaches, as the analysis is fully automated. The new system has been evaluated for its usefulness in determining hospital outbreaks with commonly occurring pathogens. Few reports have so far evaluated the performance of this system for *P. aeruginosa* typing, and there is no clear consensus regarding the performance and usefulness of this scheme. In parallel, it has been suggested that DL can be used for phylogenetic purposes in *Klebsiella pneumoniae*. In a previous study we used PFGE and MLST to investigate the population structure of *P. aeruginosa* isolated from five Mediterranean countries. From this study we selected a representative collection consisting of 133 isolates, which was subjected to typing with MLVA and DiversiLab. The primary aim of this study was to analyze the genetic heterogeneity of *P. aeruginosa* by each technique and compare their discriminative power and congruence, especially with MLST sequence types or clonal complexes, as these are regarded most predictive of long-term epidemiological relationships. # Materials and Methods ## Bacterial strains A representative collection of strains from five Mediterranean countries (Tunisia, Libya, France, Spain and Italy), previously typed by MLST and PFGE were used in the present study. Furthermore, four metallo-β-lactamase (MBL)-producing strains having distinct STs: AK5493 (ST229; Sweden), VR143 (ST227; Italy), 134MG (ST228 Italy) and PA66 (ST230; Sweden), were used. Overall, the bacterial collection consisted of isolates from various geographical origins and derived from several backgrounds of infections. The collection included several important clones such as the CC235 clone and two isolates (environmental and clinical) belonging to the most successful clone C. The latter has shown a remarkable spread in European countries in both the environment and cystic fibrosis patients. Three reference strains (ATCC 27853, PAO1 and PA14) were also included. These strains served for the calibration of the experiment. ATCC 27853 was used as a control for both PFGE and DiversiLab, whereas PAO1 and PA14 were used as control strains for MLVA. ## PFGE and MLST PFGE typing was performed as previously described with minor modifications. In short, all isolates were digested with the *Spe*I restriction enzyme (New England Biolabs, Hirts, United Kingdom). DNA fragments were separated in 1.2% agarose (SeaKem® Gold Agarose) in a CHEF-Mapper (Bio-Rad, Hercules, USA) in 0.5X Tris-Borate EDTA (TBE) running buffer at 12°C and 6 V/cm for 30 hours with pulse time ranging from 1 to 50 s. *P. aeruginosa* ATCC 27853 was used as a reference and included in every 6 lanes to allow normalization of gels. Gels were stained with ethidium bromide, photographed and saved as a TIFF files in Geldoc EQ (BioRad Laboratories, Hercules, CA). The resulting photographic images were analyzed with the GelCompar II software (Applied Maths, NV St-Martens-Latem, Belgium). The band patterns were compared using the Dice-coefficient by using the Unweighted Pair Group Method with Arithmetic Mean (UPGMA) to determine band similarity accordingly to the criteria established by Tenover et al.. Isolates producing PFGE fingerprints with ≤6 bands differences and with ≥80% similarity were categorized as the same PFGE type (PT). MLST was performed according to the protocol of Curran et al. with minor adjustments using the newly designed primers. Briefly, for DNA extraction, overnight cultured *P. aeruginosa* isolates were heated to 100°C for 10 min. The seven housekeeping genes (*acsA*, *aroE*, *guaA*, *mutL*, *nuoD*, *ppsA* and *trpE*) were amplified for all isolates by real-time PCR using Rotorgene 6000 (Corbett Robotics Inc; San Francisco, CA, USA). The PCR reaction was carried out using the QuantiTect SYBR Green PCR mix (Qiagen, Valencia, CA, USA). The amplified products were sequenced on both strands with the published primers and the newly designed primers using the BigDye Terminator Ready Reaction Mix v3.1. Nucleotide sequences were determined for both strands by ABI Prism 3100 Genetic Analyzer (Applied Biosystems, Foster City, CA). The results were further analyzed with the BioNumerics7 software (Applied Maths, St-Martens-Latem, Belgium) in order to assign the allelic numbers and sequence types (STs). The obtained results were compared with the available alleles in the MLST database (<http://pubmlst.org/paeruginosa>). In addition, the BioNumerics software was used to manage the data and perform the clustering analysis. The minimum spanning tree (MST) is an alternative phylogenetic network generated by an algorithm implemented in BioNumerics and created for both MLST and PFGE data. MLST allelic profiles were used as categorical data. Isolates with five or more identical alleles were considered as part of the same clonal complexes (CCs). Concerning PFGE, an MST with permutation resampling was generated based on the band matching class table which served as input data. The available collapsing tool was used to define the closely related isolates with ≤6 band class differences. ## MLVA The annotated genome sequences of the reference strains PAO1 and PA14 were scanned and inspected for the presence of potential VNTR loci by using the strain comparison tool developed by Denoeud and Vergnaud, available at the Microbial Tandem Repeats Database (<http://minisatellites.u-psud.fr/>). Bacterial thermolysates were used as source of DNA. For MLVA typing, the PCR reaction was performed using the published primers adapted from the protocol by Vu-Thien et al.. These authors designed a new MLVA scheme, based on a 15 candidate VNTR loci grouped into two panels: panel 1 with 13 minisatellite loci (ms77, ms127, ms142, ms172, ms211, ms212, ms213, ms214, ms215, ms216, ms217, ms222 and ms223) and panel 2 with 2 microsatellite loci (207 and ms209). The amplification reactions were carried out using PTC 200 thermocycler (MJ Research Inc., Ramsey, MN, USA), as follows: initial denaturation step at 94°C for 5 min, 35 cycles (denaturation for 30 s at 94°C, annealing for 30 s at 60°C, and extension for 45 s at 72°C), followed by a final extension step at 72°C for 10 min. For the minisatellites, PCR products were separated in 2% in Nusieve® 3:1 Agarose using GNA-200 machine characterized by a 20-cm-wide gels made in 0.5X TBE buffer and run at 8 V/cm. The amplified PCR products of the reference strains (PAO1 and PA14) and 100 bp marker (Fermentas GmBH, St. Leon-Rot, Germany) were run on each gel. The same procedure was used for microsatellites, with some modifications. PCR products were separated in a 4% agarose gel made with 2% of Nusieve® 3:1 Agarose plus 2% MetaPhor® Agarose. A 20 bp ladder size marker (Fermentas GmBH, St. Leon-Rot, Germany) was used to analyze the 6 bp repeat units of ms207 ms and ms209 markers. Uncertain band sizes and unexpected PCR products were subjected to DNA sequencing to improve the analysis and facilitate the interpretation. The size of each amplicon was deduced by visual inspection and measured using BioNumerics and the number of repeats was deduced using the MLVA alleles assignment table on the genotyping site (<http://minisatellites.u-psud.fr/MLVAnet/>). The number of repeats in the alleles was estimated by subtracting the invariable flanking region from the amplicon size divided by the repeat unit length, as determined for reference strains PAO1 and PA14 and according to the formula: number of repeats (bp)  =  fragment size (bp) - flanking regions (bp)/repeat size (bp). Each locus was given an allele number and each isolate generated an allelic profile consisting of a string of 15 numbers reflecting the number of repeats in the 15 VNTR loci. The combination of all VNTR repeats generated an MLVA-type (MT). A distinct MT was assigned on the basis of at least one differing number. All profiles among which new MTs were discovered in this study were deposited in the database. The polymorphism index for individual VNTR loci was expressed as the Hunter-Gaston diversity index (HGDI), an application of Simpson's diversity index (SID). The BioNumerics software package was used to perform clustering using the categorical coefficient to generate dendrograms based on the UPGMA method or to draw an MST. The latter method is a graphical tool illustrating the total members of the population and allowing visualization of all genotypes into one single compact image. The same setting adopted for both MLST and MLVA data is that an MST was created based on the categorical coefficient and a priority rule consisting of the highest number of single-locus variants (SLVs). For MLVA, groups or clonal complexes were created if 13 out of the 15 MLST loci were identical. ## DiversiLab (DL) system (rep-PCR) *P. aeruginosa* DNA was extracted using the UltraClean™ Microbial DNA Isolation Kit (Mo Bio Laboratories Inc., Solana Beach, CA) according to the manufacturer's instructions. The DNA quantification was measured using the NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA). Then, rep-PCR was performed using PTC 200 thermocycler and Pseudomonas fingerprinting kit (Bacterial Barcodes, bioMérieux, Athens, GA, USA) in a total reaction volume of 25 µl. The reaction mixture consisted of 18 µl of rep-PCR MM1, 2.5 µl of Gene Amp 10X, 2 µl of primer Mix, 0.5 µl of AmpliTaq DNA polymerase (Applied Biosystems, Foster City, CA, USA) and 2 µl of genomic DNA (25-50 ng/µl). Thermal condition was as follows: initial denaturation step at 94°C for 2 min, 35 cycles (denaturation at 94°C for 30 s, annealing at 50°C for 30 s, extension at 70°C for 90 s), followed by a final extension step at 70°C for 3 min. The PCR products were analyzed using the Agilent 2100 BioAnalyzer (Agilent Technologies, Santa Clara, CA, USA). Then, the amplified fragments (sizes from 100–1000 bp) were electrophoretically separated with the microfiluidic labchip. In order to monitor reproducibility, the reference strain of *P. aeruginosa* ATCC 27853 was used as a control in each PCR reaction and each chip. Electropherograms were downloaded and automatically analyzed by the DiversiLab® (version 3.4). All fingerprint patterns were normalized, then, the Pearson correlation coefficient was used in order to calculate the distance matrices among all samples. Based on the UPGMA and the multidimensional scaling, the DiversiLab® software created a customized report presenting a dendrogram, electropherograms, virtual gel images and scatter plots. The relatedness among isolates was deduced as previously described : linked isolates (similarity above 95%) and different (similarity less than 95%). An alternative cut-off (similarity index: 93%) was tested to evaluate whether it had a better correspondence to MLST findings. The DiversiLab data were visualized in BioNumerics as an MST with permutation resampling, showing all genotypes into one graph. Closely related isolates differing by a maximum of two band classes were collapsed into one node. ## Index of diversity and concordance of techniques The index of diversity and the degree of congruence among different typing schemes were determined via an online tool (<http://www.comparingpartitions.info/>; accessed on 23/01/2013). The discriminatory ability of the above described techniques was evaluated using the Simpson's index of Diversity (SID: with 95% confidence intervals) as described by Hunter and Gaston. An index greater than 0.90 is considered desirable if the typing results are to be interpreted with confidence. Inter-method concordance was calculated using the adjusted Rand and Wallace coefficients; the adjusted Rand index (R) shows the proportion of agreement corrected for the presence of chance agreement whereas the Wallace coefficient (W) indicates the probability that two strains classified as the same type by one method will also be classified as the same one when using the other method. # Results ## PFGE Using a similarity cut-off of 80%, the 133 isolates produced 90 PFGE-types, designated PFGE 1 to 90. In addition, the results revealed twenty clonal groups consisting of two or more isolates, representing closely related patterns. Group 17 was the largest with 8 isolates. The remaining 70 unique PFGE-types were categorized as singletons. ## MLST MLST was performed for all strains. Sixty-seven STs were assigned and segregated into 11 clonal complexes. The main clonal complex was MLST-CC235, consisting of five STs (235, 989, 979, 230 and 227, with ST235 as the primary founder) followed by the clonal complex CC244 with five STs (244, 990, 986, 993 and 654 with ST244 as the primary founder). The remaining groups were doublets or had a simple association among couples of STs. Thirty-nine unrelated STs were considered as singletons. ## MLVA With some exceptions (n = 17) that did not generate any band for a few markers, all isolates were successfully typed. For 8 isolates, unexpected bands above 1500 bp were observed. This is probably due to the presence of an insertion sequence (IS) in the VNTR especially for ms214. MLVA-15 was able to subdivide the studied population into 85 MTs. Most of them were new MTs (n = 70). As shown in the MST, the resulting MTs were segregated into 11 MLVA clonal complexes. MLVA-CC62 was the largest lineage, consisting of 12 MTs, and was shared by 43 isolates belonging to MLST-CC235. The remaining 10 MLVA-CC were as follows: MLVA-CC79 (70, 71), MLVA-CC82 (76, 77, 78 and 79), MLVA-CC12 (17 and 99), MLVA-CC75 (103 and 104), MLVACC-34 (90 and 91), MLVA-CC24 (101 and 102), MLVA-CC26 (39, 108 and 109), MLVA-CC11 (15, 112 and 113), MLVA-CC77 (96 and 97) and MLVA-CC89 (86 and 87). ## DiversiLab All isolates were typed by the automated rep-PCR-based DiversiLab system. By using a similarity cut-off of 95%, DL typing differentiated the 133 isolates into 61 DL-types, whereof 36 were singletons or unique patterns. The remaining DL groups represented the clonally related isolates comprising two or more isolates. DL 1, 10 and 27 (arbitrary numbers) were the largest groups, comprising 14, 13 and 16 isolates, respectively. Interestingly, strains belonging to DL 1 and 27 belonged to MLST-CC235 and MLVA-CC62, respectively. The alternative cut-off of 93% decreased the total cluster by subdividing the whole population into 46 clusters. ## Simpson index of diversity (SID) and concordance between typing schemes expressed by the Wallace coefficient (W) The SID of all the studied techniques was above 0.9 indicating a relatively high level of discrimination of all techniques. As suspected, PFGE had the greatest index (0.989; cut-off ≥80%), followed by MLVA (0.980), DL (0.961/0.937; cut-off 95/93%) and MLST (0.906). The clustering analysis of the studied collection by all typing schemes revealed a lot of related and distinct genotypes but not totally overlapping from one technique to another. The adjusted Wallace coefficient indicated that MLST-type at the ST-level was well predicted by MLVA (0.930) and PFGE (0.919), but less well predicted with DL (0.642/0.562; cut-off 95/93%). Conversely, MLST, as expected, was not able to predict types retrieved using any of the other techniques. MLVA-type was not predicted by the other techniques except to some extent with PFGE (0.635). DL-type was predicted better by PFGE (0.617/0.786; cut-off 95/93%) than other techniques. PFGE-type could not be predicted by any of the other techniques. We found high bidirectional agreement between MLST-CC and MLVA-CC. However, neither MLST-CC nor MLVA-CC could predict PFGE and DL. PFGE predicted the MLST- CC-type at 100% level, whereas the DL predicts the MLST-CC-type at a 74/65%; cut-off 95/93%) level. When analyzing separately the 44 MLST-CC235 isolates, they were split into 19 PFGE-types (SID = 0.926), 13 MTs (SID = 0.840), 9 DL- types (SID = 0.754) and 5 STs (SID = 0.175). All these isolates belong to the MLVA-CC62 except the isolate FP10, which had a divergent MLVA profile. # Discussion The population biology of *P. aeruginosa* has been extensively investigated. There is a consensus that the population structure can be defined as a non- clonal, including several heterogeneous clones but punctuated by some major epidemic clones or clonal complexes,. This study was carried out using a set of isolates collected from several Mediterranean countries. The strains were characterized and typed by four molecular typing schemes (MLST, PFGE, MLVA and DL). ## Assessment and nature of genetic variations retrieved by the four genotyping schemes The selected isolates were all typeable by the selected genotyping schemes. PFGE showed a high index of diversity (SID = 0.989), as discussed in other studies. The amount of PFGE variation is due to genetic events that alter the size of the restriction fragments, such as modification of restriction sites by point mutations or insertions/deletions of longer DNA sequences. The MLST variation is restricted to changes within housekeeping genes, while MLVA variation is restricted to the tandem repeats on the selected VNTR loci. The DL variation is mediated by non-coding repetitive sequences dispersed along the genomes. The high resolution power of PFGE makes the method well suited for outbreak investigations, but less well suited for tracking long-term epidemiological relations. MLVA-15 subdivided the population into 85 genotypes or MTs indicating a high level of discrimination (SID = 0.98). The available MLVA profiles in the database are skewed towards clinical isolates from cystic fibrosis patients, or MDR isolates from patients of other categories, ,. Even though these studies use different MLVA schemes, they have pointed out the usefulness and discrimination of the MLVA protocol for *P. aeruginosa* genotyping. More recently, a new automated protocol of MLVA, named MLVA-16 which is assayed in two multiplexed PCR, was developed. The MLVA-16 contains all the MLVA-15 VNTR loci, and an additional ms61, which is a highly polymorphic loci, previously used in the MLVA-9. Importantly, this additional microsatellite increases the discrimination of the method by adding further insight into the strain evolution during long- term colonization. Remarkably, one interesting feature of MLVA is that the index of diversity can be increased to a great extent by including new VNTR markers. The findings with the MLVA-15 used in our study are in line with earlier observations by displaying diversity values ranging from 50 to 88%. The VNTR loci displaying lower or moderate diversity are useful for establishing deeper phylogenetic relationships and for the definition of clonal complexes, whereas markers with higher level of diversity offer a greater discriminatory power among closely related isolates. DiversiLab was found to have a relatively high resolution power, separating isolates into 61 DL-types with an index of diversity of 0.961, at the 95% similarity cut-off level. When using the relaxed cut-off at 93% similarity, the correlation with MLST decreased. Few reports have evaluated the utility of DiversiLab system, however, some have found an equal capability to that of PFGE for the confirmation of the circulation of multidrug-resistant *P. aeruginosa* clones in different hospital wards. Other studies have typed a collection of *P. aeruginosa* obtained from different epidemiological situations and provided results similar to those obtained with PFGE.. According to our findings, DL could identify closely related isolates identified by PFGE or MLST in some cases. On the other hand, it overestimated the relatedness since some isolates belonging to the same DL groups were differentiated by PFGE into several types. Other isolates belonging to the same MLST-CC or PFGE group were grouped differently based on DL. The DiversiLab system has been applied for many bacterial species, it was evaluated for its performance and feasibility for the identification of hospital outbreak, and it has shown to be useful for some species and less for others like *P. aeruginosa*. MLST subdivided the population into 67 STs with an index of diversity of 0.906. Although MLST has limitations in terms of resolution power, the method is now clearly considered the gold standard for phylogenetic studies. The method identifies SNPs as well as genomic rearrangements located in the seven conserved genes, allowing phylogenetic analysis by tracing the evolutionary history between strains and identifying a subset of isolates genetically appearing related through a common ancestor. The main drawback of MLST is that the method is time-consuming and costly. In order to overcome these limitations and given the importance of the scheme, Boers et al. developed an MLST protocol for *P. aeruginosa* and other species using next-generation sequencing (NGS). The high throughput MLST (HiMLST) protocol consisted of two standardised steps increasing the typing potential of MLST and promising a high quality and cost effective typing of bacterial strains. In parallel, in order to speed up the current MLST protocols, new microfluidic platform technologies are currently developed. ## Congruence of the three schemes with MLST at the ST and clonal complex levels The adjusted Wallace coefficient indicates the probability that pairs of isolates which are assigned to the same type by one typing method are also typed as identical by the other. MLST-types were best predicted by MLVA, followed by PFGE and DL. By contrast MLST is not a good method to predict MLVA-, PFGE- and DL-types. The discrepancies were mainly observed in some isolates having the same STs, but dispersed into different PFGE groups, MLVA-types or DL groups. PFGE has a good unidirectional prediction to MLST, MLVA and DL, but due to the higher resolution of PFGE other methods perform poorly in terms of predicting PFGE-types. Regarding MLVA, all MLST-CC235 isolates except one belonged to the same MLVA- CC62. Our findings confirm the high level of agreement between MLST and MLVA at the clonal complex level. Likewise the bidirectional congruence between MLST-CC and MLVA-CC was well supported by the Wallace coefficient in and well visualized in MSTs in. On the other hand, the congruence was univocally increased among PFGE to MLST-CC (W = 100%), DL to MLST (W = 74%). As seen in, the unidirectional congruence of techniques to others and absence or limited congruence of the counterpart is difficult to refer to one factor but could occur for several reasons. A large amount of polymorphisms was detected and the phylogenetic incongruence is mainly due to the significant amount of homoplasy, homologous recombination and lateral gene transfer. However, it is difficult to find optimal markers establishing a real phylogenetic history. Ideally, SNPs which are relatively rare and scattered through the genome are more evolutionary informative than any other markers. These SNPs display low mutation rates, and can provide a relatively high level of phylogenetic resolution,. Interestingly, the genetic data provided by the fingerprint-based methods such as PFGE and DL could somehow predict the clonal level, but provides no specific DNA sequence information. The rep-PCR relies on short repeated stretches of DNA abundant in eubacterial genomes. The PFGE strategy makes uses of rare cutting sites dispersed in the whole genome. The DNA information provided by both approaches is therefore clearly affected by recombinational events. Hence, they are useful for identifying closely related isolates in short time scale but cannot be used for understanding the underlying genetic diversity and evolutionary history of population isolates. Generally, for populations such us *P. aeruginosa* undergoing significant recombination, phylogenies inferred by the PFGE or DL data are inaccurate and therefore misleading. MLVA can enhance the resolution within epidemiologically related isolates clustered by MLST. Currently the genotyping by MLVA is in progress, and more investigation is needed to fully appraise the potential of this approach as a typing tool. MLVA allows the differentiation of isolates in the form of numerical codes reliably stored in databases, henceforth, we appreciate that MLVA would be used as a reasonable proxy for MLST. A good correspondence between MLST and MLVA was highlighted in previous studies. Therefore the MLVA scheme was accurate in identifying the population structure of *P. aeruginosa*. ## Phylogenetic relatedness within closely related isolates exemplified with CC235 The prominent clonal lineage detected in our collection was MLST-CC235. To gain additional resolution of the molecular evolution and dissemination of such epidemic clones, we analyzed this clonal complex separately. The 44 isolates belonging to this lineage were shown to be heterogeneous by PFGE (19 types), followed by MLVA (13 MTs), DL (9 types), and finally MLST (5 STs). This finding highlighted the “violation” of ST assignment of CC235 isolates similar to the observation of Deplano et al. when analyzing the epidemiological concordance of ST235 isolates. This study revealed that ST235 isolates were split into diverse DL types. These results clearly shed light on the features of each typing technique and their relative capacities to discriminate and differentiate among isolates belonging to the same lineage. Importantly, this lineage is characterized by a great complexity reflected by micro-evolutionary events visualized by multiple patterns resolved by each technique. The international spread of the CC235 isolates has been reported on several occasions,. In fact this is an important successful clone disseminated over a long period and eminently associated with serotype O11, presence of the *exoU* virulence gene, and characterized by frequently being multidrug-resistant. Likewise, Harris and coworkers have analyzed the population structure of a prominent lineage of *Staphylococcus aureus*, MRSA ST238, which has shown an intercontinental spread. By conducting a high throughput genomic approach based incisively on the SNPs that reside in the core genome, it was possible to trace an ideal phylogeny permitting to interrogate the geographical origin and intra- hospital spread of this pathogenic clone. However, MLST seems less suitable when trying to distinguish various clonally related isolates i.e. the *P. aeruginosa* CC235 isolates, *Escherichia coli* ST131, methicillin-resistant *S. aureus* and particularly monomorphic pathogens. The limitation of MLST here is referred to the crude estimates of SNPs restricted within the housekeeping genes. The rapid evolution of *P. aeruginosa* clones such as the CC235 clone might decrease their genetic relatedness and obscure their epidemiological linkage, and eventually conflict the phylogenies extracted simultaneously by several techniques. # Conclusions In this study, we have compared the properties of four genotyping schemes for the phylogenetic analysis of *P. aeruginosa*. Although MLST is highly informative, it has a limited resolution, especially when applied to closely related isolates as shown for the prominent lineage CC235. MLVA was able to predict the MLST-type with high accuracy, and even higher at the clonal complex level, making it the best surrogate method for MLST in phylogenetic studies. Also, MLVA, as a relaxed scheme, provides sufficient resolution for strain differentiation and clustering analysis, and could also be used for outbreak investigation. The low throughput PFGE and the automated DL are useful for outbreak investigation, but to a lesser degree for long-term epidemiological studies. They are efficient to dissect subtle genomic differences especially among very closely related isolates. # Supporting Information We thank Inga Karlsson and Marie Andersson for excellent technical assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MM CGG CP. Performed the experiments: MM MAH ATR. Analyzed the data: MM CGG ATR AI CP. Contributed reagents/materials/analysis tools: CGG AB OS. Wrote the paper: MM CG.

# Introduction Phosphorus is an important factor in numerous biological processes and exists in the form of inorganic phosphates in the body. The intake of dietary phosphate has been gradually increasing with changes in life style over the past several decades. In healthy subjects, present-day levels of dietary phosphate are not likely to cause imminent hyperphosphatemia; however, excessive intake could present serious problems, particularly for chronic renal patients. Therefore, our understanding of the mechanisms of phosphate homeostasis is extremely important. The major organs involved in phosphate homeostasis are the small intestine, kidney, parathyroid gland, and bone. Serum phosphate levels are tightly- regulated through the action of humoral factors such as parathyroid hormone (PTH), fibroblast growth factor 23, and 1α,25-dihydroxyvitamine D (also known as calcitriol). The expression or synthesis of these factors is coordinately regulated in response to changes in dietary and serum phosphate levels. However, the mechanism of regulation of phosphate homeostasis, including effector molecules such as phosphate transporters, remains to be elucidated. The kidney plays a pivotal role in phosphate excretion and has been the focus of many studies, particularly those studying the effects of a high phosphorous (HP) diet in experimental animals. Administration of a HP diet causes renal calcification and infiltration of inflammatory cells. Furthermore, previous reports have suggested that there are major alterations in the mRNA expression profile in response to a HP diet, including downregulation of sodium-dependent phosphate transporter (NaPi)-IIa and NaPi-IIc, and upregulation of osteopontin in the rat kidney. However, relatively few studies have examined the influence of a HP diet on renal gene expression in a comprehensive manner. Dietary phosphate-responsive genes have been reported in rainbow trout kidney, but the global effects of a HP diet on mammalian gene expression in the kidney have yet to be reported. To investigate the mechanism(s) by which phosphate levels are maintained in the body, gene expression in the kidney of HP diet-fed rats was assessed by DNA microarray analysis. The effect of administration of a HP diet on overall gene expression in the kidney as well as the induction of specific genes such as NaPi-IIb in response to a HP diet was demonstrated. # Results ## Food intake and body weight Food and calcium intake, body weight at baseline and study end, and average weight gain were not significantly different between control and HP-fed rats. As expected, phosphorus intake, calculated from measured food intake and the known phosphorus content of the diet, was significantly higher in the HP group than in control animals. ## Serum phosphorus, calcium and urea nitrogen levels HP rats had significantly lower serum phosphorus concentrations, while serum calcium concentrations were similar between the two groups. To determine whether the intake of a HP diet affected renal function, blood urea nitrogen (BUN) was measured, which is an index of renal function. BUN levels did not differ significantly between the two groups. On the other hand, fractional excretion of phosphate was significantly increased in the HP group, while fractional excretion of calcium was similar. ## Effects of a HP diet on phosphorus and calcium balance Net intestinal phosphorus absorption was significantly increased in the HP group, whereas phosphorous retention between the two groups was similar. Thus, an excessive amount of the ingested phosphorus was passed through the body in HP diet-fed animals. There were no significant differences in parameters of calcium balance between the two groups. ## Kidney weight and mineral content Kidney weight, both wet and dry weight, was significantly increased in the HP group. Kidney calcium and phosphorus content was also significantly increased in the HP group compared to controls. These results suggested that there was some degree of calcification and hypertrophy of the kidney in HP diet-fed rats. ## Histological analysis To evaluate the effect of a HP diet on kidney morphology, histology was performed on kidney tissue sections using hematoxylin and eosin (H&E) staining. In rats fed a HP diet, cyst-like swelling was apparent in the renal cortex and medulla of the kidney. Fibril formation was also noted, especially in the renal medulla. These results suggested that intake of a HP diet induces morphological abnormalities in the kidney. ## DNA microarray analysis of gene expression The distribution free weighted method (DFW) quantified microarray data were subjected to hierarchical clustering analysis. Each experimental group formed a large and separate cluster, which indicated that a HP diet can induce changes in gene expression profile in the rat kidney. To identify differentially expressed genes (DEGs) in response to a HP diet, the rank products (RP) method was applied to DFW-quantified data. The RP method combined with a DFW preprocessing algorithm has been shown to be one of the best ways to accurately detect DEGs. Applying a significance value for the false discovery rate (FDR) of \<0.05, we identified 1056 upregulated probe sets (838 genes) and 712 downregulated probe sets (536 genes) in the HP group compared to the control group. The full list of DEGs is shown in. ## Gene Ontology (GO) analysis DEGs were classified into functional categories according to GO. GO terms that were significantly enriched within the two sets of DEGs (upregulated and downregulated genes) are summarized in and, respectively. The overrepresented GO terms in the upregulated genes were further classified into three major GO term clusters: “ossification”, “collagen fibril organization” and diverse clusters of “inflammatory and immune response”. Among the DEGs related to ossification we found secreted phosphoprotein 1 (SPP1; also known as osteopontin). This result was in good agreement with a previous study showing a marked induction of SPP1 in response to a HP diet. Also identified were genes for non-collagenous bone matrix proteins including glycoprotein (transmembrane) nmb (Gpnmb/osteoactivin), secreted protein acidic and rich in cysteine (SPARC/osteonectin), and fibronectin 1 (Fn1). The GO term ossification also included genes for several collagens such as collagen type I alpha 1 (Col1a1), type V alpha 2 (Col5a2), and type XI alpha 1 (Col11a1), as well as humoral factors such as bone morphogenetic protein (Bmp6) and stanniocalcin 1 (Stc1). Within the GO term collagen fibril organization were a number of genes encoding fibrous collagen proteins, including collagen type I alpha 2 (Col1a2), type III alpha 1 (Col3a1), type V alpha 1 (Col5a1) and alpha 2 (Col5a2), all of which are components of the extracellular matrix. Transforming growth factor beta 2 (TGFb2) was also included within this term. The third GO cluster in the upregulated genes included many genes related to inflammatory or immune responses. Quick GO analysis revealed that these genes were strongly associated with two GO terms, “positive regulation of Type IIa hypersensitivity” and “positive regulation of T cell mediated cytotoxicity”, and both groups included genes encoding complement component 3 (C3), Fc receptors and major histocompatibility complex (MHC) class I molecules. Downregulated genes in HP rats included three overrepresented GO terms, “oxidation reduction”, “carboxylic acid biosynthetic process” and “response to nutrient”. The genes included within these GO terms are listed in. ## Changes in gene expression of transporter or channel for phosphate, calcium and water Gene-annotation enrichment analysis (as above) is able to detect only those genes with the same GO annotation that are statistically enriched in a given population of DEGs. To complement the GO analysis, we selected DEGs related to transporters or channels for phosphate, water and calcium for further analysis. Among known phosphate transporters, solute carrier family 34 (Slc34a) members 1, 2 and 3 (Slc34a1/2/3), also known as type II sodium-dependent phosphate cotransporters (NaPi), were differentially expressed (NaPi-IIa and -IIc were downregulated, while NaPi-IIb was upregulated) in response to a HP diet. Aquaporin11 (Aqp11), a water channel, was downregulated in the kidney in HP rats. Furthermore, differentially expressed genes for calcium channels and transporters were as follows: ATPase Ca²⁺ transporting plasma membrane 1 (Atp2b1), type 2C member 2 (Atp2c2), Calcium channel voltage- dependent T type alpha 1I subunit (Cacna1i), and alpha 2/delta subunit 1 (Cacna2d1). However, these four genes were only small part of many calcium channels and transporters. In fact, there were no significant differences in fractional excretion of calcium between the HP and control groups, it seems likely that these four genes are not critical for calcium reabsorption and secretion in kidney. ## *In situ* hybridization analysis of NaPi-IIb in the rat kidney *In situ* hybridization was carried out to confirm the differential expression of Slc34 family mRNAs in the rat kidney in response to a HP diet. Weak NaPi-IIb- positive signals were observed in the cortex but not the medulla of the kidney in the control group. In contrast, the corresponding signals clearly localized to the kidney cortex and medulla in the HP group. No signals were detectable in the negative control samples (data not shown). The downregulation of NaPi-IIa mRNA expression was also confirmed by *in situ* hybridization (data not shown). ## Immunohistochemical analysis of NaPi-IIb in the rat kidney To determine whether NaPi-IIb localization exhibited any polarity, kidney tissue sections were analyzed by immunohistochemistry using an anti-NaPi-IIb antibody. NaPi-IIb localized predominantly to the cytosol in the cortex and medulla of the kidney in the control group. In the HP group, NaPi-IIb also localized to the cytosol in the cortex, whereas in the medulla, NaPi-IIb was located at the basolateral membrane of the epithelial cells surrounding the urinary duct. # Discussion DNA microarray analysis was used to investigate the effects of a HP diet on gene expression in the rat kidney. HP rats were administered a diet containing 1.2% phosphorus, which was 4-times the amount in the AIN-93G control diet, for 24 days. Under these conditions, there were no significant differences between control and HP rats in terms of final body weight and food intake. Apparent phosphorus absorption as well as urinary and fecal phosphorus excretion was significantly increased in the HP group, whereas there was no significant difference in phosphorus retention between the two groups. These results suggest that a large amount of the phosphorus that was ingested in the HP group passed through the body. Serum phosphorous concentrations were significantly lower in the HP group compared to the control group. This discrepancy may be due to the timing of blood sampling, and the fact that the fractional excretion of phosphorus was increased in the HP group. Rats were sacrificed after a 4 h fast, during which there was no further ingestion of phosphorus. Thus, during this period, blood phosphate was being actively excreted, which could lead to lower serum concentrations in the HP group than in the control animals. In a previous study, there was a negative phosphorus balance in rats given a 1.2% or 1.5% phosphorus diet, supporting the idea that enhanced excretion of phosphorus occurs as a result of ingestion of a HP diet. There was a significant increase in kidney weight in the HP group, indicating renal hypertrophy, and significant increases in the phosphorus and calcium content of the kidney, indicating nephrocalcinosis. These results suggest that an excess amount of phosphorus passing through the body might be sufficient to induce a high-phosphorus phenotype, even though phosphorous retention and serum concentrations were not elevated in the HP group. Hierarchical clustering analysis of DNA microarray data generated using mRNA isolated from control and HP rats revealed that there were marked differences in the gene expression profile of the kidney between the two groups. GO terms that were significantly enriched in the set of upregulated genes fell into three categories: ossification, collagen fibril organization, and inflammatory (or immune) response. Nephrocalcinosis, fibrosis and inflammation are major symptoms induced in the kidney by a HP diet. In fact, phosphorus and calcium content was increased in the kidney in the HP group, and histological analysis by H&E staining revealed fibrosis-like regions in kidney sections from HP rats. Thus, the microarray data and gene expression changes correlated with morphological and biochemical phenotypes in rats fed a HP diet. Among the DEGs that mapped to the GO term ossification, we found a member of TGFβ superfamily gene, Bmp6. Previously, it was shown that exogenous Bmp6 induces the upregulation of a set of osteoblast-related genes, including SPP1, in human mesenchymal stem cells. In light of these previous results, the data from the current study suggest a mechanism in which the upregulation of Bmp6 induces the differentiation of osteoblast-like cells in the kidney, resulting in the progression of nephrocalcinosis. The GO term collagen fibril organization included Col1a2, Col3a1, Col5a1, Col5a2, and TGFb2. Previous reports have shown a transient upregulation of TGFb2 in acute-phase experimental nephritis, and this change in expression correlated with fibrosis progression. In a separate study, the addition of exogenous TGFb2 resulted in strong upregulation of Col1 and Col3 in keloid and burn hypertrophic scars. Thus, upregulation of TGFb2 in response to a HP diet may induce the expression of extracellular matrix proteins that contribute to the fibril formation observed in the kidney medulla. Many genes related to inflammatory or immune responses were differentially expressed in response to a HP diet. Quick GO analysis revealed that these genes mapped overwhelmingly to two GO terms, positive regulation of Type IIa hypersensitivity and positive regulation of T cell mediated cytotoxicity. These two categories included genes encoding C3, Fc receptors and MHC class I molecules. These results indicate that in HP rats, the complement pathway might be activated by IgG or IgM binding to sites in the kidney, leading to subsequent cell lysis or tissue damage by phagocytic cells such as macrophages or neutrophils (i.e. type II allergy-like immunoreactions). The upregulation of genes encoding MHC class I molecules also suggests the activation of macrophage or neutrophil chemotaxis in response to a HP diet. Col4 is a major component of the renal basement membrane and contains epitopes that trigger autoantibody formation. Some chronic renal diseases are known to be caused by Col4 protein- autoantibody immune complexes. The current results showing that Col4 was upregulated in response to a HP diet are consistent with these observations. From the full set of DEGs, genes related to phosphate, calcium and water transport were extracted. Aquaporins (Aqps) are known to play major roles in water transport. There are 13 mammalian Aqps identified to date. Of these, only one, Aqp11, was altered (downregulated) by a HP diet; the expression levels of the others did not change significantly. Aqp11 is associated with polycystic renal disease and Aqp11 knockout animals develop cysts in the kidney. We observed cyst-like swellings in kidney sections from the HP group, indicating that significant downregulation of Aqp11 may play a role in cyst formation in rats fed a HP diet. Of the many calcium transporters or channels that have been identified, only 4 genes appeared to be differentially expressed in response to a HP diet: Atp2b1, Atp2c2, Cacna1i and Cacna2d1. However, since fractional excretion of calcium was not significantly affected by a HP diet, it is likely that these 4 genes are not involved in calcium homeostasis, particularly calcium excretion. The three types of NaPi have been identified, NaPi-I, II and III corresponding to genes encoding Slc17, Slc34, and Slc20 solute carrier families, respectively. NaPi-IIa and -IIc have been shown to be important for phosphate homeostasis. They are expressed on the apical side of proximal tubule epithelial cells and play a pivotal role in phosphate reabsorption in the kidney. NaPi-IIa and -IIc are downregulated in response to increased levels of dietary phosphate and serum PTH. Consistent with these observations, we found that NaPi-IIa and -IIc were downregulated in response to high dietary intake of phosphorous in rats. On the other hand, NaPi-IIb (Slc34a2) was significantly upregulated in HP rats. This upregulation of NaPi-IIb was confirmed by *in situ* hybridization, with marked expression observed in medullary epithelial cells. Immunohistochemical analysis showed that NaPi-IIb localized to the basolateral membrane of cells lining the collecting duct and/or the ascending limb of the loop of Henle. This cellular location of NaPi-IIb implies that it has a role in phosphate homeostasis, particularly excretion. Fractional excretion of phosphate was significantly increased in the HP group, with values exceeding 100%. This increase in phosphate excretion could not be explained solely by the downregulation of NaPi-IIa and -IIc. Thus, there appears to be other active excretion mechanism at play under conditions of a HP diet. The current data suggest that NaPi-IIb is upregulated in response to high dietary intake of phosphorous and results in enhanced phosphate excretion. NaPi-IIb is expressed in the intestine, liver, epididymis and various other tissues, including the kidney. A previous study using a conditional knockout approach demonstrated that NaPi-IIb plays a major role in phosphate absorption in the intestine. Moreover, in mice, deletion of NaPi-IIb results in embryonic lethality, suggesting that NaPi-IIb is essential for survival. The function of NaPi-IIb in the kidney, however, remains to be fully elucidated. Apical expression of NaPi-IIb has been demonstrated in the intestine, whereas in the kidney, the current data indicate that NaPi-IIb is expressed at the basolateral side of renal epithelial cells. From a broad localization standpoint, these results are not necessarily contradictory; in both cases, phosphate was transported from the outside of the cell to the interior. Thus, in HP rats, upregulation of NaPi-IIb can be thought of as enhancing basal phosphate excretion levels in the kidney. The other cotransporters NaPi-IIa and -IIc were shown to express on apical side of kidney epithelial cells both *in vivo* and *in vitro*. These reports support that the role of NaPi-IIa and -IIc are exclusively phosphate reabsorption. On the other hand, under normal phosphate concentration condition, NaPi-IIb was shown to express on both apical and basolateral side of kidney epithelial cells *in vitro*. While additional experiments using cultured kidney cells (MDCK or opossum kidney cells) will be needed especially under high phosphate conditions to define the events involved in phosphate transport, as well as the mechanisms underlying the polarity of NaPi-IIb expression, we are convinced that the current findings uncover a new facet of the mechanism of phosphate homeostasis. # Materials and Methods ## Animals and diets Male Wistar rats (4 weeks old) were purchased from Japan SLC Co. (Hamamatsu, Japan) and individually housed in metabolic cages under controlled conditions of 22±1°C and a 12-hour light/dark cycle (lights on from 08:00 to 20:00 daily). Two different diets containing 0.3% phosphorous (control diet) and 1.2% phosphorous (HP diet) based on the AIN-93G diet were purchased from Research Diets Inc. (New Brunswick, NJ, USA). All rats were fed the control diet for a 7-day acclimatization period. After acclimatization, rats were divided into two groups of similar mean body weight (n = 5 each) and then fed either the control or the HP diet for 24 days. The animals were allowed to eat *ad libitum* and had free access to water (MilliQ water). Urine and feces were collected from day 20 to 23 for balance studies. At the end of the experimental period, all rats were sacrificed under anesthesia and blood and kidney samples were taken for analysis. Serum and urine samples were stored at −20°C until use. The right kidney was used for mineral analysis and the left kidney was used for DNA microarray analysis. For fractional excretion and histochemical analyses, an additional animal experiment was performed (n = 6 or 7) as described above except that the urine was collected on the last day and the left kidney was removed and sectioned. Tissue samples were frozen in liquid nitrogen immediately after excision and stored at −80°C until use. To measure calcium and phosphorus content, feces and kidney tissue were dried, ashed and then demineralized with a solution of HNO₃ (0.1 mol/l). Calcium concentration was analyzed by SPS 1200VR Inductively Coupled Plasma (ICP)-Atomic Emission Spectrometry (Seiko Instruments Inc., Chiba, Japan). Phosphorus, BUN and creatinine were assayed using Phosphor C, Urea N B, and LabAssay™ Creatinine kits (Wako Pure Chemical Industries, Osaka, Japan), respectively. The protocol for the animal experiments was approved by the Animal Use Committee of the Faculty of Agriculture at The University of Tokyo (approval number: P09-283). ## DNA microarray experiments Total RNA was isolated from the kidney using the TRIzol reagent (Invitrogen Life Technologies, Carlsbad, CA) and then purified using an RNeasy mini kit (Qiagen K.K., Tokyo, Japan). The quality and quantity of purified total RNA were verified by agarose gel electrophoresis and spectrophotometry, respectively. We selected four average rats from each group, according to determined by urine volume (vs. control group) and kidney phosphorus content (vs. HP group). DNA microarray analysis was carried out as described previously with slight modifications. In brief, biotinylated aRNA was obtained from 100 ng of purified total RNA using a GeneChip 3′ IVT Express Kit (Affymetrix, Santa Clara, CA, USA). The aRNA was purified, fragmented and then hybridized to an Affymetrix Rat Genome 230 2.0 Array containing probe sets for over 30,000 rat genes. Following hybridization at 45°C for 16 h, the arrays were washed and labeled with phycoerythrin. Fluorescence signals were scanned using the Affymetrix GeneChip System. Affymetrix GeneChip Command Console software was used to reduce the array images to the intensity of each probe (CEL files). All the microarray data are MIAME compliant and have been deposited in a MIAME compliant database, the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (<http://www.ncbi.nlm.nih.gov/geo/>, GEO Series accession number GSE31973), as detailed on the MGED Society website (<http://www.mged.org/Workgroups/MIAME/miame.html>). ## DNA microarray data analysis The CEL files were quantified with the DFW using statistical language R (<http://www.r-project>) and Bioconductor (<http://www.bioconductor.org/>). Hierarchical clustering was performed using the pvclust function in R. To identify DEGs, the RP method was applied to DFW quantified data, with the number of permutations set at 1,000. Probe sets with an FDR \<0.05 were regarded as having different expression levels between the two groups (i.e. were differently expressed). The annotation file for the Rat Genome 230 2.0 Array was downloaded from the Affymetrix web site (August 10, 2010, Rat230_2.na31.annot.csv). Gene-annotation enrichment analysis of DEGs was performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID; <http://david.abcc.ncifcrf.gov/>) and Quick GO (<http://www.ebi.ac.uk/QuickGO/>). EASE scores, which are modified Fisher's exact test *p*-values, were used to extract statistically overrepresented GO terms from the DEGs; Benjamini and Hochberg FDR corrections were used to correct the results by multiple testing. GO terms with FDR-corrected *p*-values of \<0.05 were regarded as significantly enriched. ## In situ hybridization Kidney tissue from control and HP rats was embedded in Tissue-Tek O. C. T compound (Sakura Finetechnical, Tokyo, Japan) and frozen in liquid nitrogen immediately after excision. Frozen tissue samples were sectioned into 7 µm-thick slices. The sections were fixed with paraformaldehyde (PFA), carboethoxylated and then treated with 10 µg/ml proteinase K for 30 min, followed by 4% PFA in PBS for 10 min. Sections were subjected to hybridization with riboprobes in hybridization buffer. The antisense riboprobes were synthesized from DNA fragments subcloned into pBluescript II SK(−) (Stratagene, LA Jolla, CA, USA) and hydrolyzed to approximately 500 bases before hybridization. As a negative control, *in situ* hybridization was performed using sense riboprobes. ## Immunohistochemistry PFA-prefixed cryosections were prepared as described for *in situ* hybridization. Tissue sections were treated with Target Retrieval Solution (Dako, Carpinteria, CA, USA) for 20 min at 95°C and then cooled for 20 min at room temperature. After the inactivation of endogenous peroxidase with PBS containing 0.3% hydrogen peroxide for 30 min at room temperature, the sections were blocked with normal horse serum in PBS containing 0.1% Triton X-100 for 30 min, and then incubated at 4°C overnight with 3 µg/ml of rabbit anti-mouse NPT2b IgG (Alpha Diagnostic Intl. Inc., San Antonio, TX, USA). The sections were then incubated with ImmPRESS™ Reagent Horse Anti- Rabbit IgG (Vector Laboratories, Burlingame, CA, USA) for 30 min at room temperature. Visualization was performed with DAB Metal Enhanced Substrate kit (Pierce Biotechnology, Rockford, IL, USA). # Supporting Information We thank Drs. Sei Sasaki (Tokyo Medical and Dental University), Ken-ichi Miyamoto and Hiroko Segawa (University of Tokushima) for valuable advice on histochemical analysis. We also thank Dr. Mitsuru Abo (The University of Tokyo) for assistance in measuring the serum calcium concentrations by ICP-AES. [^1]: Conceived and designed the experiments: YN KA. Performed the experiments: TS SO TI KI. Analyzed the data: TS YN. Contributed reagents/materials/analysis tools: SO TI KI. Wrote the paper: TS KA YN. [^2]: The authors have declared that no competing interests exist.

# 1. Introduction Aging persons living with HIV, well-controlled by antiretroviral treatment(ART), present a high prevalence of age-related cardiovascular and metabolic comorbidities, higher than the prevalence observed in non-infected individuals with similar risk factors. Therefore, in these patients, it is mandatory to favor ART with minimal metabolic and cardiovascular toxicity. Some contemporary used protease inhibitors (PI) have been associated with an increased cardiovascular risk, in part related to the boosting concentration of ritonavir, which leads to raised LDL-cholesterol and triglycerides levels. This has been clearly shown for ritonavir-boosted lopinavir, and ritonavir-boosted darunavir. Ritonavir-boosted atazanavir (ATV/r) has been associated with a lower cardiovascular risk than ritonavir-boosted darunavir. This could be related to the ability of ATV/r to increase bilirubin levels, because bilirubin has been related to cardio-protective anti-oxidant effects. Integrase strand transfer inhibitors (INSTI) have been initially considered as lipid- and metabolic-friendly. Because they exert potent anti-viral activities, they are recommended at present for treatment initiation in ART-naïve patients and also for switch strategies in ART-controlled patients with comorbidities. A decreased level of proatherogenic lipids has been consistently reported in patients switched to INSTI. As well, no increased cardiovascular risk has been observed and, even, recently, a lower cardiovascular disease risk has been associated with INSTI versus other regimens after a median follow-up of 18 months in more than 20 000 ART initiators. However, recently, several reports revealed treatment with that some INSTIs resulted in weight gain, both in ART-initiated and ART-experienced patients switched off PIs to INSTI. This may represent an undesirable effect placing patients at higher risk for cardiovascular and metabolic complications on the long term. Discrepant results were reported regarding the impact of INSTI on insulin sensitivity, some studies arguing for an improvement, and others for no change or even a worsening of insulin resistance. Regarding maraviroc (MVC), its ability to modulate atherosclerotic progression and to improve endothelial function was shown in two small studies, this improvement being associated, in one of them, with decreased inflammatory markers. We have previously reported that DTG, raltegravir, MVC, ATV/r and ritonavir- boosted darunavir differentially affected endothelial cells. DTG exerted anti- inflammatory effects and reduced senescence, as did MVC to a lesser extent, while we observed the reverse with the two PIs. The aim of the present paper was to further decipher the cellular pathways leading to inflammation, senescence and insulin sensitivity in HCAECs and which were altered by MVC, DTG and ATV/r. At first, we tested the possibility that the SIRT-1 pathway, which activation results in decreased inflammation and senescence in endothelial cells, could explain the effect of the antiretroviral molecules. We could not confirm this hypothesis. Thus, we used a non-targeted transcriptomic approach and identified USP18 as a probable signaling step in the effect of DTG and ATV/r, but not MVC, on inflammation and senescence. Moreover, DTG and ATV/r, but not MVC, altered insulin sensitivity and their effects were not dependent on USP18. Taken as whole, DTG and ATV/r appeared to oppositely alter the same pathways different from those used by MVC. # 2. Material and methods ## 2.1 Cell culture and cell treatments We used Human primary Coronary Artery Endothelial Cells (HCAEC, PromoCell, Heidelberg, Germany) isolated from the coronary arteries from adult donors. To perform the entire set of experiments we used coronary endothelial cells issued from four different donors: two male and two female subjects, devoid of cardiovascular morbidity, non-obese, non-diabetic, aged 27, 35, 40 and 57 years old, two Caucasian, one Hispanic and one Hispanic/Black. We used, for all cultures, the endothelial cell growth medium from Promocell (C22010) supplemented with 5% fetal calf serum. Cell were grown in 6-well dishes from passages 2–8 and used when confluent. We verified by Western blot (antibody MEM-111, ref ab2213 from Abcam, Cambridge, UK, dilution 1/1000) that the CCR5 receptor was expressed on endothelial cells, as previously shown. We confirmed its expression in each coronary cell line. Moreover, when cells were incubated for 15 days with MVC, DTG and ATV/r, the mean level of CCR5 was not modified (one-way Anova with Dunnett’s multiple comparisons tests NS). ATV, ritonavir and MVC were purchased from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA, USA), DTG from Medchem Express (Princeton, NJ, USA). HCAEC were treated for 15 days with DMSO 0.1% without (control) or with the drugs dissolved in DMSO at Cmax plasma concentrations according to the literature or to the Liverpool HIV-interactions website ([www.hiv- druginteractions.org](http://www.hiv-druginteractions.org/)): MVC 0.9 μg/ml, DTG 3.7 μg/ml, ATV/r 5.2 and 0.94 μg/ml. Three independent experiments were performed for each protocol. The results are all expressed as the mean values of the 3 experiments related to the % of the control established at 100%. ## 2.2 SIRT-1 pathway We analyzed the effect of a SIRT-1 activator SRT1720. Regarding dose-response experiments, at first, we evaluated concentrations between 0 and 10 μM for 48h. However, since the higher concentrations (2.5–10 μM) induced cell toxicity, we further analyzed the effect of SRT1720 at concentrations 0.5, 1 and 2 μM. The absence of toxicity was evaluated by visual inspection and by the verification that the total protein amount in each dish was equivalent to that of the control at the end of the experiment. The efficacy of SRT1720 was evaluated by its ability to inhibit NFκB. We determined that the concentration of 2 μM was optimal to induce SIRT-1 and inhibit NFκB. We also evaluated the effect of a SRTI-1 inhibitor, splitomicin. We tested splitomicin at concentrations ranging from 0 to 500 μM according to the literature. We discarded the 500 μM concentration which was toxic and evaluated the ability of splitomicin to induce NFκB at concentrations of 50, 100, 200 μM. The dose-response evaluation allowed to determine that the concentration of 100 μM was optimal to decrease SIRT-1 and to activate NFκB. Thereafter, these concentrations of SRT-1720 and splitomicin were used in all the experiments performed. The cells were cultured during 15 days with/without the antiretroviral molecules, SRT1720 or splitomycin being added for the last 3 days. ## 2.3 Western blotting Whole cell lysates were subjected to sodium dodecyl sulfate/polyacrylamide gel electrophoresis and Western blotting. We used antibodies against USP18 (sc-374064), AKT (sc-8312), phospho-AKT (sc-7985R), phosphotyrosine (sc-7020) from Santa Cruz at dilution 1/500; against NFκB p65/RelA (8242S), phospho(S536)-NFκB p65/RelA (3033S), insulin receptor beta-subunit (3025), JNK (9252), phospho-JNK (9251S), TAK1 (45055) and phospho-TAK-1(4531) from Cell Signalling Technology, Inc. (Danvers, MA, USA) at dilution 1/1000; against p16 (10883-1-AP) and p21 (10355-1-AP) from Proteintech (Proteintech Europe, Manchester, UK) or from BD Biosciences (San Jose, CA, USA) at dilution 1/1000; against ICAM-1 (ab2213) and VCAM-1 (ab134047) from Abcam (Abcam, Cambridge, UK) at dilution 1/1000; against tubulin, used as an index of the cellular protein content, from Sigma-Aldrich (Saint Louis, MO, USA) at dilution 1/1000. For antibodies against SIRT1 we used at first antibodies from Santa Cruz (sc15404) at dilution 1/500 then from Abcam (ab32441) at dilution 1/1000. The activity of NFκB was measured by the ratio of the level of phosphorylation of the p65/RelA protein to the total level of p65/RelA, as measured by Western blot. ## 2.4 Secretion of pro-inflammatory cytokines and adhesion molecules The secretion of interleukin-6 (IL-6), interleukin-8 (IL-8), soluble intracellular cell adhesion molecule-1 (sICAM-1) and soluble vascular cell adhesion molecule-1 (sVCAM-1) in 24-h culture supernatants was evaluated by ELISA assays (R&D quantikine HS600C and HS800 for human IL-6HS and IL-8HS, DVC00 and DCD540 for human VCAM-1 and ICAM-1) or using the ELLA system (Protein Simple, Simple Plex Cartridge for IL-6 and Simple Plex Cartridge for IL-8, Bio Techne, Minneapolis, MN, USA). Cells were incubated for 14 days with/without the drugs. The culture medium was replaced for the last 24 hours by a medium without foetal calf serum with/without the drugs and collected after 24 hours to measure the level of secreted cytokines and adhesion factors. ## 2.5 RT-PCR Cells were incubated with drugs during 15 days. Total mRNA was extracted followed by reverse transcription (capacity cDNA reverse transcription kit, \#4368814, Life Technologies). Quantitative PCR was performed with LightCycler 480 using LightCycler 480 SYBR Green I Master mix (#04887352001, Roche Diagnostics, Meylan, France). The primers were designed by using the “Assay design center” of Roche. <https://lifescience.roche.com/en_gb/brands/universal-probe-library.html> and ordered to Invitrogen (Thermo Fisher Scientific, Waltham, MA, USA). The primer sequence for USP18 was: `hUSP18 S: TCC CGA CGT GGA ACT CAG h USP18 AS: CAG GCA CGA TGG AAT CTC TC` We used two reference genes: HPRT-1 and PPIA. ## 2.6 Transcriptomics RNA-quality was evaluated according to good practice of Genom’IC facility (Institut Cochin, Paris, France) by Agilent Bioanalyzer 2100 before being spotted on human Gene 2.0 ST arrays. After an RMA (Robust Multi-array Average) with Bioconductor, differentiated gene analysis of data was performed by One-Way Repeated Measure ANOVA (paired). Enrichment analysis of this gene set (p\<0.05 over treatment vs. control) was carried out with IPA software (Ingenuity pathway analysis, [www.ingenuity.com](http://www.ingenuity.com/)). ## 2.7 Gene silencing We used the protocol from Santa-Cruz biotechnologies for siRNA-mediated inhibition of gene expression. The specific USP18 siRNA and the scramble siRNA (SC-37007), which contained a scrambled sequence that will not lead to the specific degradation of any known cellular mRNA, were ordered from Santa Cruz. The transfection was performed in the presence of the transfection medium (SC-36868 and SC-29528) in 80% confluent endothelial cells in 6-well dishes, for 6h at 37°C. ## 2.8 Senescence markers We evaluated the protein level of two cell-cycle inhibitors associated with cellular senescence, p16^INK4 and p21^WAF1 by Western blot as previously described. ## 2.9 Insulin sensitivity The basal level of insulin sensitivity was evaluated by the production of nitric oxide (NO) as previously described. NO production was assessed with the cell-permeant NO indicator 4-amino-5-methylamino-2′,7′-difluorofluorescein diacetate (DAF-FM; D23844; Molecular Probes, Life Technologies, Carlsbad, CA, USA). Cells were cultured in 96-well plates, washed and incubated with DAF-FM (12.5 μmol/l) or Hoechst 33258 (0.01 mg/ml) in DMEM without fetal bovine serum for 30 min at 37°C in the dark. Quantification was performed with a plate fluorescence reader (Infinite M200; Tecan-France, Trappes, France) at 515 nm (DAF-FM) and 460 nm (Hoechst 33258). We evaluated the cell response to insulin at two key steps of the [insulin signaling](https://www.sciencedirect.com/topics/medicine-and-dentistry/insulin- signaling) pathways, [tyrosine phosphorylation](https://www.sciencedirect.com/topics/medicine-and- dentistry/protein-phosphorylation) of the insulin receptor-[β subunit](https://www.sciencedirect.com/topics/medicine-and-dentistry/beta-chain) (level of tyrosine-phosphorylated β-subunit reported to the level of total insulin receptor β-subunit), and activation of AKT (level of phospho-AKT reported to the level of total AKT) as previously published. In brief, cell grown in 6-well dishes were cultured to confluence, then depleted from fetal bovine serum for 16 h, and stimulated for 15 min with insulin at 100 nM. [Western blotting](https://www.sciencedirect.com/topics/medicine-and- dentistry/western-blotting) analysis used antibodies as indicated above. ## 2.10 Statistics Comparison of the different points with the cells incubated with the different drugs dissolved in DMSO to the control value being set at 100% was performed at first by using one-way ANOVA (or two-way ANOVA for experiments with USP18 silencing) with Dunnett’s multiple comparisons tests, by comparing each experimental point to the relevant control with the Prism software. If ANOVA indicated a significant difference, we further evaluated the differences between the respective control and each drug treatment by using the Student T-test with the Welch’s correction with the Prism software. The results regarding significant differences obtained with one-way ANOVA with Dunnett’s multiple comparison tests and with Student T-test with Welch’s correction were globally similar. # 3. Results ## 3.1 Effect of the antiretrovirals and of SIRT-1 activation or inhibition on the SIRT-1 level and on inflammation ### 3.1.1 Effect of antiretrovirals and of SRT1720 or splitomycin on the protein level of SIRT-1 We observed that DTG and DTG+MVC increased SIRT-1 respectively by 34% and 43% while MVC had no effect. The SIRT-1 activator, SRT1720, tended to increase the SIRT-1 level (+54%, p = 0.07) while the inhibitor splitomycin decreased this level by 44%. However, in the presence of splitomicin, DTG alone or associated with MVC kept its ability to increase the level of SIRT-1 (+64% and +61% as compared to splitomicin alone). ### 3.1.2 Effect of antiretrovirals and of SRT1720 or splitomycin on NFκB activation and secretion of cytokines and adhesion molecules As expected, DTG, DTG+MVC and SRT1720 inhibited NFκB while splitomycin activated it. In the presence of splitomicin, DTG and the association DTG+MVC were still able to decrease NFκB activation. MVC had no effect. DTG and DTG+MVC significantly decreased the secretion of IL-6 and IL-8 by endothelial cells and this was also the case for SRT1720. Conversely, splitomycin increased the secretion of the two cytokines. This effect was reversed in part by DTG and DTG+MVC. MVC has a mild inhibitory effect on IL-6 and IL-8 secretion. We obtained similar results for the secretion of sICAM-1 and sVCAM-1. Treatment with DTG and DTG+MVC resulted in decreased secretion (by 46% for ICAM-1 and 43–49% for VCAM-1) of the two adhesion molecules. Splitomycin increased ICAM-1 (by 44%) and VCAM-1 (by 52%) secretion. DTG and DTG+MVC were still able to decrease ICAM-1 (by 20–22%) and VCAM-1 (by 44–45%) secretion in splitomicin-treated cells. MVC had no effect. Thus, DTG, and to a lesser extent MVC, reduced NFκB activation, resulting in decreased secretion of IL-6, IL-8, sVCAM-1 and sICAM-1. Their effects were not mediated by SIRT-1. ## 3.2 Transcriptomic analysis of the effect of antiretrovirals in endothelial cells To further analyze the pathways that were modified by a treatment with DTG, MVC and ATV/r we performed a global untargeted transcriptomic analysis, comparing the mRNA profile obtained in the presence of each drug to the control condition. We identified several genes which expression was differentially affected by the drugs. Among these different genes, we selected USP-18 since it has been involved in interferon signaling but also in insulin sensitivity and inflammation\[–\]. The results of the transcriptomic analysis can be accessed through the link [*https*:*//www*.*ncbi*.*nlm*.*nih*.*gov/geo/query/acc*.*cgi*?* acc=GSE137247*](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE137247) RT-PCR analysis confirmed that that the mRNA level of USP18 was increased by MVC by 57% but not by DTG or ATV/r. ## 3.3 Silencing of USP18 ### 3.3.1 Effect on the protein level of USP18 We analyzed the protein level of USP18 in endothelial cells incubated with the specific USP18 siRNA or with a scramble siRNA taken as a control. MVC tended to increase USP18 level (p = 0.12) in accordance with increased mRNA level. USP18 silencing resulted in a 61–67% reduction in the USP18 protein level as compared to the level observed in cells incubated with the scramble siRNA. ### 3.3.2 Activation of NFκB and secretion of cytokines and adhesion molecules Silencing of USP18 resulted in a 6-fold increased basal level of NFκB activation. MVC decreased by 49% this level of activation while DTG and ATV/r had lost their effect. As well, in USP18-silenced cells the basal level of IL-6 secretion was 69% higher than in cells treated with the scramble construction. In USP-silenced cells, MVC decreased IL-6 secretion by 33% while DTG and ATV/r had no significant effect. Accordingly, IL-8 secretion was increased by 67% when USP18 was silenced, MVC reduced this secretion in USP-silenced cells by 30% while DTG has a minor effect (13% decrease) and atazanavir no effect. We observed similar results for sICAM-1 and sVCAM-1. Moreover, we evaluated the cell-associated level of ICAM-1 and VCAM-1. This level was reduced by MVC and DTG and increased by ATV/r. Silencing of USP18 resulted in an increased level of cell-associated ICAM-1 and VCAM-1. In silenced cells, the effects of MVC were maintained but those of DTG and ATV/r were reduced or lost. Thus, MVC increased the mRNA but not the protein level of USP18. Its inhibitory effect on NFκB activation and cytokine/adhesion molecules secretion was enhanced in USP18-silenced cells. Conversely, the effect of DTG and ATV/r was lost or reduced in USP18-silenced cells indicating that USP18 was required for the anti- and pro-inflammatory effects of DTG and ATV/r on NFκB activation and cytokine/adhesion molecules secretion. ### 3.3.3 Effect of USP-silencing on the level of cell-cycle inhibitors We have previously shown that antiretrovirals were able to modulate the basal level of endothelial cell senescence. To go further, we analyzed whether this parameter was altered in USP18-silenced cells by measuring the protein level of the two cell-cycle inhibitors strongly linked to the senescence phenotype, p16^INK4 and p21^WAF1. We observed that their level was increased in silenced cells and that MVC was still able to decrease their level. Conversely, the ability of DTG and ATV/r to respectively decrease and increase the level of the senescence proteins was lost or minored, suggesting that USP18 was involved in the effects of DTG and ATV/r on senescence but not in the effect of MVC. ### 3.3.4 Insulin sensitivity Then, we addressed the ability of the drugs to modify insulin sensitivity. In the basal state, we evaluated the production, in the culture medium, of nitric oxide, reflecting the state of insulin sensitivity. We observed that DTG increased, while ATV/r reduced, NO production, indicating respectively enhanced and decreased basal insulin sensitivity. MVC did not modify it. To go further, we analyzed cell response to insulin by measuring the level of tyrosine phosphorylation of the insulin receptor β-subunit and the level of phosphorylation of AKT (ratio of phospho-AKT to total AKT), involved in the metabolic insulin signaling pathway. DTG increased and ATV/r decreased (or tended to decrease p = 0.1) insulin receptor and AKT activation while MVC had no effect. The total protein level of the insulin receptor β-subunit reported to that of tubulin did not differ in cells treated or not with the drugs. We then evaluated the level of insulin resistance in USP18-silenced cells by addressing AKT activation. In cells treated with the scramble construction, we confirmed that DTG increased and ATV/r decreased insulin sensitivity. USP18 silencing increased basal insulin resistance (decreased AKT activation). However, the effect of DTG was maintained in silenced cells, indicating that it did not involve USP18. We also analyzed steps involved in insulin resistance upstream of AKT, at the level of the TGFβ-activated kinase (TAK-1) and of the two Jun kinases, JNK1(p46) and JNK2 (p54) by evaluating the ratio between the phosphorylated and the total form of each enzyme. USP18 silencing increased the basal level of activation of TAK-1, JNK1 and JNK2 (Figs and). In silenced cells, in accordance with their effect on insulin sensitivity, DTG decreased TAK-1, JNK1 and JNK2 activation while ATV/r tended to increase this level (increased activation in each experiment but large inter-experiment variations in the % of increase). Therefore, DTG improved and ATV/r worsened insulin sensitivity in USP18-silenced cells indicating that their effects were not dependent on USP18. MVC did not alter insulin signaling. # 4. Discussion We show here that HIV antiretrovirals from three classes differently affected inflammation, senescence and insulin sensitivity in endothelial cells. The PI ATV/r enhanced inflammation, senescence and insulin resistance while the INSTI DTG had opposite effects. The CCR5 inhibitor MVC was mildly anti-inflammatory and reduced senescence with no effect on insulin sensitivity. The ability of PIs to increase the cardiovascular risk has been previously addressed, mainly for ritonavir-boosted lopinavir and ritonavir-boosted darunavir. In in vitro studies, we and others have observed that the different protease inhibitors were able to enhance inflammation and induce endothelial dysfunction to different extents, with lopinavir presenting the worse profile, ATV/r exerting milder and darunavir/r even milder effects. Regarding INSTI, a recent clinical study indicated that their use was associated with decreased occurrence of cardio-vascular outcomes as compared to other ART (PI or non-nucleoside analogue reverse transcriptase inhibitors), even if the individual INSTI were not analyzed in this study. Also, all studies evaluating switch strategies from PI towards INSTI reported improved lipid levels (total and LDL cholesterol, triglycerides) after the switch. We previously reported that, in endothelial cells, raltegravir had no effect on endothelial function while DTG improved it. Few clinical studies, including a low number of patients, have evaluated the effect of MVC intensification on vascular parameters. In 15 aviremic PI-treated HIV-infected patients, MVC intensification resulted in a reduction in intima- media thickness and pulse wave velocity, supporting a reduction in the cardiovascular risk and an improvement in endothelial function. As well, MVC intensification in 22 HIV-infected patients at high cardiovascular risk improved several markers for cardiovascular risk, endothelial dysfunction, and arterial stiffness. The ability of PIs or INSTI to modulate inflammation has been evaluated in several studies switching off PIs to INSTI. However, in that setting, it is difficult to address the respective role of each antiretroviral class. In the SPIRAL study, in which patients were switched off PIs to raltegravir, levels of CRP, monocyte chemoattractant protein 1 (MCP-1), TNF-α and IL-6 were decreased. As well, patients switched off PIs (mainly ATV/r and ritonavir-boosted darunavir) to raltegravir or DTG presented markedly decreased levels of IL-6. In the NEAT022 study, switching off contemporary PIs to DTG resulted in decreased levels of soluble-CD14 with a tendency for CRP and oxidized-LDL. Patients from ETRAL switched off contemporary PIs to raltegravir plus etravirine presented decreased levels of interferon-gamma inducible protein 10 (IP10) and solubleCD14 but increased levels of D-dimer. Accordingly, in the present study, we observed that ATV/r exerted pro-inflammatory and DTG anti-inflammatory effects and we have previously reported that raltegravir did not modify inflammation in endothelial cells. One study on MVC intensification reported that IL-6, sICAM-1 and sVCAM-1 levels were decreased but the inflammatory markers remained unmodified in another one, suggesting that the effect of MVC on inflammation is probably limited. We have previously shown a mild anti-inflammatory effect of the drug on endothelial cells that was also found in the present work. Several PIs have been associated with increased insulin resistance. In healthy non-infected controls, a 10-day treatment with ritonavir-boosted lopinavir was associated with markedly increased insulin resistance and a similar trend was reported for ATV/r to a lesser extent. We report here that ATV/r worsened insulin resistance in endothelial cells. Regarding INSTI, clinical data on insulin sensitivity are discrepant. In the SPIRAL study, switching off PIs to raltegravir improved insulin sensitivity. As well, patients switched off PIs to raltegravir or DTG presented improved insulin sensitivity. Otherwise, when evaluating either ART-naïve or ART-experienced patients, insulin resistance (evaluated by the HOMA-IR index) increased similarly in subjects receiving raltegravir or DTG and in those receiving other antiretroviral regimens. In the ETRAL study, PI-controlled HIV-infected patients switched to a dual raltegravir-etravirine therapy gained fat and increased their insulin level, indicating increased insulin resistance. Greater weight gain has been reported both in cohort studies and clinical trials with INSTIs, as compared to PIs and non-nucleoside analogue reverse transcriptase inhibitors, a worrisome outcome, }. This weight gain was more marked with DTG than raltegravir and in a recent study equivalent between DTG and bictegravir. Weight gain following ART initiation could be attributable in part to a “return to health” phenomenon in patients with severe HIV infection, but this is not the case for switched patients. Weight gain is also associated with personal factors as sex, age and ethnicity. This increased weight has been associated, for DTG and raltegravir, with a decreased circulating level of the adipokine adiponectin, an insulin-sensitizing and anti- inflammatory adipokine, suggesting a deleterious impact of these INSTIs on adipose tissue. This situation could suggest that INSTI differently impact different tissues. They could exert neutral or favorable effects on the vascular wall and adverse effects on adipose tissue. Moreover, the nucleoside reverse transcriptase inhibitor backbone could also be involved in fat gain as shown by the fat gain-favoring effect of tenofovir alafenamide (TAF) plus DTG as compared to tenofovir disoproxil fumarate (TDF) plus DTG. We decided to further decipher the pathways involved in the effect of DTG, MVC and ATV/r in endothelial cells. At first, we evaluated the role of the SIRT-1 pathway, since we have previously observed that the SIRT-1 protein level was modified by the different antiretrovirals, with an effect opposite to their ability to activate NFκB. Importantly, SIRT-1 has been shown to play an important role in endothelial cells, being able to reduce the production of reactive oxygen species, prevent premature senescence and inhibit NFκB. Therefore, we thought that SIRT-1 was a credible candidate to explain modified NFκB activity in response to antiretroviral molecules. SIRT-1 activation decreased the level of inflammation, as expected, and the reverse was observed when SIRT-1 was inhibited, confirming the role of SIRT-1 in the basal inflammatory state in endothelial cells. However, the ability of DTG to decrease basal inflammation in cells exposed to the SIRT-1 inhibitor was preserved, indicating that DTG and SIRT-1 were using different pathways to decrease inflammation. In this setting, MVC had no effect on inflammation. To better approach the mechanisms involved in the drug effect, we performed a non-targeted transcriptomic analysis and evaluated the level of the genes altered by the different drugs. Among different genes of interest, we selected, at first, USP18, a specific ISG15 isopeptidase and an inhibitor of the interferon type 1 signal, which exerts different protease-dependent and independent effects. The ability of USP18 to decrease inflammation has been previously reported in T cell. USP18 was shown to inhibit the NFκB pathway and TAK1. However, the catalytic deubiquitinating activity of USP18 was required for TAK-1 but not for NFκB inhibition, indicating that different signaling pathways were involved. Accordingly, since TAK1 is required for JNK activation, leading to insulin resistance, and since USP18 inhibits TAK-1, an insulin-sensitizing effect of USP18 has been shown in the liver. By using the siRNA strategy, we silenced the gene and reduced by 60–70% the level of the USP18 protein. Interestingly, the level of inflammation and insulin resistance was increased in cells with reduced USP18 level. Therefore, our results, which demonstrate the ability of USP18 to inhibit TAK-1 and JNK activation resulting in enhanced insulin sensitivity, agree with other studies obtained in T lymphocytes and the liver. Interestingly, we observe that the effect of DTG and ATV/r on inflammation required USP18 but that USP18 was dispensable for their effect on TAK1, JNK and AKT activation, indicating that they affected insulin signaling independently of USP18. We have previously shown that some ART were able to affect senescence in endothelial cells, DTG and MVC preventing and ATV/r increasing the senescence level. We show here, in addition, that USP18 was also able to prevent senescence, since its basal level was increased in USP18-silenced cells. Moreover, in these cells, MVC, but not DTG and ATV/r, was still able to reduce the level of senescence markers. It will be interesting, in a future work, to analyze the interactions between inflammation and senescence in this setting. Therefore, DTG and ATV/r exerted opposite effects on the same three pathways, insulin resistance, senescence and inflammation, respectively independent and dependent on USP18. The effect of MVC on inflammation and senescence was independent on USP18. A schematic view of these different pathways is presented in. Our work has limitations. We used an in vitro model of endothelial cells and our results need to be validated in the clinical situation. Nevertheless, in vitro models allow to analyze separately different tissues and drugs, while, in the clinical situation, the evaluation is more global. Therefore, to perform in vitro studies could be relevant in the situation of INSTI, since INSTI probably present discrepant adverse effects, with a possible deleterious impact on adipose tissue but a neutral or favorable one on lipids and vascular wall. Indeed, we previously observed that raltegravir was neutral and we confirm here that DTG exerted favorable effects in endothelial cells. We provide here results suggesting links between different signaling pathway and enlightening the importance of USP18 in several of these pathways. Further studies will be important to better decipher these interactions, in particular the link between SIRT-1, USP18, inflammation and senescence. # 5. Conclusions We have evaluated the effect of individual drugs from three ART classes on endothelial cells as a surrogate of arterial wall. MVC exerted no or mild effects on insulin sensitivity, senescence and inflammation. DTG presented a favorable profile regarding these three parameters conversely to ATV/r. The effects of the drugs signaled through different pathways, and in particular, involved USP18 for the inflammatory and senescence pathways. Therefore, this enzyme could represent a potential target to modulate inflammation and senescence in ART-controlled patients. # Supporting information This work has been presented in part at the 20^th International Workshop on Comorbidities and Adverse Drug reactions in HIV, New-York, October 2018 [^1]: I have read the journal's policy and the authors of this manuscript have the following competing interests: JC received honoraria for academic presentations from ViiV Healthcare, MSD, Janssen and Gilead. This does not alter our adherence to PLOS ONE policies on sharing data and materials. The other authors have declared that no competing interests exist.

# Introduction Analyzing gene expression in multicellular organisms involves a tradeoff between the spatial precision of imaging and the efficiency and comprehensiveness of genomic methods. RNA *in situ* hybridization (ISH) and antibody staining of fixed samples, or fluorescent imaging of live samples, provides high resolution spatial information for small numbers of genes. But even with automated sample preparation, imaging, and analysis, *in situ* based methods are difficult to apply to an entire genome's worth of transcripts or proteins. High throughput genomic methods, such as DNA microarray hybridization or RNA sequencing, are fast and relatively inexpensive, but, at least for the small species worked with in most labs, the amount of input material they require has generally limited their application to homogenized samples, often from multiple individuals. Methods involving the tagging, sorting, and analysis of RNA from cells in specific spatial domains have shown promise, but remain non-trivial to apply systematically, especially across genotypes and species. Recent advances in DNA sequencing suggest an alternative approach. With increasingly sensitive sequencers and improved protocols for sample preparation, it is now possible to analyze small samples without amplification. Several years ago we developed methods to analyze the RNA from individual *Drosophila* embryos. As we often recovered more RNA from each embryo than was required to obtain accurate measures of gene expression, we wondered whether we could obtain good data from pieces of individual embryos, and whether we could obtain reliable spatial expression information from such data. To test this possibility, we chose to focus on anterior-posterior (A–P) patterning in the early *D. melanogaster* embryo, as the system is extremely well-characterized and the geometry of the early embryo also lends itself to biologically meaningful physical dissection by simple sectioning along the elongated A–P axis. # Results To test whether we could consistently recover and sequence RNA from sectioned *D. melanogaster* embryos, we collected embryos from our laboratory stock of the line CantonS (CaS), aged them for approximately 2.5 hours so that the bulk of the embryos were in the cellular blastoderm stage, and fixed them in methanol. We examined the embryos under a light microscope and selected single embryos that were roughly halfway through cellularization (mitotic cell cycle 14; developmental stage 5). We embedded each embryo in a cryoprotecting gel, flash- froze it in liquid nitrogren, and took transverse sections along the anterior- posterior axis. For this initial trial we used 60 µm sections, meaning that we cut each approximately 350 µm embryo into six pieces. We placed each piece into a separate tube, isolated RNA using Trizol, and prepared sequencing libraries using the Illumina Tru-Seq kit. In early trials we had difficulty routinely obtaining good quality RNA-seq libraries from every section. We surmised that we were losing material from some slices during library preparation as a result of the small amount (approximately 15 ng) of total RNA per slice. To overcome this limitation, after the initial RNA extraction we added RNA from a single embryo of a distantly related *Drosophila* species to each tube to serve as a carrier. We used RNA as a carrier, instead of a standard carrier like salmon sperm DNA or linear acrylamide, so that the carrier was present throughout the experiment, and we used RNA from multiple *Drosophila* species in particular so that the sequence reads from the carrier RNA would not be wasted. In this first experiment the carrier RNA was part of an experiment examining gene expression in early embryos of other *Drosophila* species. We only used embryos from species that were fully sequenced and sufficiently diverged from *D. melanogaster* to allow us to readily separate reads derived from the *D. melanogaster* slice and the carrier species computationally after sequencing. With the additional approximately 100 ng of total RNA from the carrier in each sample, library preparation became far more robust. We independently sliced three CaS embryos, prepared libraries from the sliced RNA using the standard TruSeq RNA kit, and sequenced them using an Illumina HiSeq 2000, obtaining approximately 40 million 50 bp paired-end reads for each slice+carrier sample. We aligned these reads to the *D. melanogaster* and carrier genomes using TopHat, and identified between 1.7 and 31.4 percent of reads as having come unambiguously from *D. melanogaster*. We then used Cufflinks to infer expression levels for all annotated mRNAs using the *D. melanogaster* reads alone. The data for each slice within an embryo were generally highly correlated, reflecting the large number of highly expressed genes with spatially uniform expression patterns. The data for equivalent slices of embryos 2 and 3 were also highly correlated, while the slices for embryo 1 were systematically less well matched to their counterparts in embryos 2 and 3, suggesting that it may have been sampled at a slightly different developmental stage. To examine how well our data recapitulated known spatial profiles, we identified a panel of genes with known anterior-posterior patterns of gene expression and compared our data to their published expression patterns. shows RNA in-situ hybridization patterns from the Berkeley Drosophila Genome Project (BDGP) alongside the expression data for that gene from our sliced embryos, demonstrating a close qualitative agreement between the visualized expression patterns and our sliced RNA-seq data. In order to more quantitatively compare our data to existing patterns, we constructed a reference set of spatial expression patterns along the A-P axis using three-dimensional “virtual embryos” from the Berkeley Drosophila Transcription Network Project, which contain expression patterns for 95 genes at single-nucleus resolution. We transformed the relative expression levels from these images into absolute values (FPKM) using genome-wide expression data from intact single embryos. We compared the observed expression for these 95 genes from an average of each of our slices to all possible 60 µm slices of these virtual embryos. High scores for most slices fell into narrow windows, with the best matches for each slice falling sequentially along the embryo with a spacing of about 60 µm, the same thickness as the slices. We next used the program Cuffdiff to identify 85 genes with statistically significant differences in expression between slices (; this is a very conservative estimate). We compared these genes to those examined by the BDGP, the most comprehensive annotation of spatial localization in *D. melanogaster* development that we are aware of. Of our differentially expressed genes, 21 had no imaging data available, and 33 were annotated as present in a subset of the embryo (the annotation term meant to capture patterned genes); the remaining 31 genes showed either clear patterns that were not annotated with the most general keyword, or no clear staining. There were 194 genes tagged by the BDGP as patterned that were not picked up as having statistically significant patterns in our data. However, most of these had primarily dorsal-ventral patterns, faint patterns, later staging in the images used for annotation, or had good qualitative agreement with our data but fell above the cutoff for statistical significance. As a more sensitive approach to finding patterned genes, we applied k-means clustering to our data. We first filtered on expression level (at least one slice in one embryo with FPKM \>10) and agreement between replicates (average Pearson correlation between embryos of \>0.5), then clustered based on normalized expression (k = 20, centroid linkage; 20 was chosen empirically as smaller k's merged genes with different patterns and larger k's provided no additional useful information). We identified several broad classes of expression, including localization to each of the poles, and five different gap gene-like bands along the AP axis. Of the 745 genes, only 349 had images in the BDGP set. Staining for these genes is sometimes undetectable and well-matched stages are often missing from the databases, but where comparisons were possible, the BDGP image data agrees with our RNA-seq patterns. To extend our dataset, we collected individual embryos from seven different time points based on morphology—stage 2, stage 4, and 5 time points within stage 5—and sliced them into 25 µm sections, yielding between 10 and 15 contiguous, usable slices per embryo. For these embryos we used total RNA from the yeasts *Saccharomyces cerevisiae* and *Torulaspora delbruckii* as carrier, which are so far diverged as to have fewer than 0.003% of reads ambiguously mapping. These finer slices are better able to distinguish broad gap-gene domains, with several slices of relatively low expression between the multiple domains of *hb*, *kni*, and *gt*. Excitingly, we can also distinguish the repression between stripes of pair-rule genes like *eve* as well. Given the non-orthogonal orientation of the anterior-most and posterior-most *eve* stripes relative to the AP axis, we do not expect to see all 7 pair-rule stripes, but at least three can be unambiguously observed. Putting the 60 µm and 25 µm slice datasets together, we find a large number of genes with reproducible patterns in the 60 µm slices whose formation over time can be clearly seen in the timed 25 µm slices, including many without previously described early patterns. # Discussion The experiments reported here demonstrate that slicing and sequencing animal embryos is a practical and effective method to systematically characterize spatial patterns of expression. While we are by no means the first to dissect samples and characterize their RNAs—Ding and Lipshitz pioneered this kind of analysis twenty years ago —to our knowledge we are the first to successfully apply such a technique to report genome-wide spatial patterns in a single developing animal embryo. Given the degree to which the *D. melanogaster* embryo has been studied, and the presence of at least two large *in situ* based studies whose goals were to systematically identify and characterize genes with patterned expression in the embryo, we were surprised by the large number of genes we find as clearly patterned that had not been previously described as such. We note in particular a large number of genes with expression restricted to the poles, most with no known role in either anterior patterning or pole cell formation or activity. This emphasizes the potential for sequencing-based methods to replace *in situ* based studies in the systematic analysis of patterned gene expression, as they are not only simpler, cheaper, and easier to apply to different species and genetic backgrounds, but appear to be more sensitive. The data we present here are far from perfect - the relatively small number of reads per slice (due to the presence and sequencing of carrier RNA) means that the slice by slice data are somewhat noisy. However the consistency between replicates and the agreement between the 25 µm and 60 µm data demonstrate that the experiment clearly worked, and additional sequencing depth and better methods for working with small samples should greatly reduce the noise as we move forward. Obviously, to truly replace *in situ* based methods, sequencing based methods will need to achieve greater resolution than presented here. One can envision several basic approaches to achieving the ultimate goal of determining the location of every RNA in a spatially complex tissue. Sequencing RNAs in place in intact tissues would obviously be the ideal method, and we are aware of several groups working towards this goal. In the interim, however, methods to isolate and characterize smaller and smaller subsets of cells are our only alternative. One possibility is to combine spatially restricted reporter gene expression and cell sorting to purify and characterize the RNA composition of differentiated tissue—c.f.. While elegant, this approach cannot be rapidly applied to different genetic backgrounds, requires separate tags for every region/tissue to be analyzed, and will likely not work on single individuals. Sectioning based methods offer several advantages, principally that they can be applied to almost any sample from any genetic background or species, and allow for the biological precision of investigating single individuals. The 60 µm and 25 µm slices we used here represent reasonable tradeoffs between sequencing depth and spatial resolution given the current limits of sample preparation and sequencing methods, but with methods having been described to sequence the RNAs from single cells, and with sequencing costs continuing to plummet, it should be possible to obtain far better resolution in the near future. A rough estimate suggests that a single embryo contains enough RNA to sequence over 700 samples to a depth of 20 million reads. Thus it is theoretically possible to dice an embryo into 20 µm cubes and sequence each one to obtain genome-wide three- dimensional expression data, although this presents several difficult but likely solvable technical challenges, especially handling and tracking hundreds or thousands of tiny samples. # Materials and Methods ## Fly Line, Imaging, and Slicing We raised flies on standard media at 25° in uncrowded conditions, and collected eggs from many 3 to 10-day old females from our *Canton-S* lab stocks. We washed and dechorionated the embryos, then fixed them according to a standard methanol cracking protocol. Briefly, we placed embryos in 20 ml glass vials containing 10 ml of heptane and 10 ml of PEM (100 mM PIPES, 2 mM EGTA, 1 mM MgSO4) and mixed gently. We then removed the aqueous phase, added 10 ml of methanol, shook vigorously for 15–30 seconds, and collected the devitellinized embryos, which we washed several times in methanol to remove residual heptane. We then placed the fixed embryos on a slide in halocarbon oil, and imaged on a Nikon 80i with DS-5M camera. After selecting embryos with the appropriate stage according to depth of membrane invagination and other morphological features, we washed embryos with methanol saturated with bromophenol blue dye (Fisher, Fair Lawn NJ), aligned them in standard cryotomy cups (Polysciences Inc, Warrington, PA), covered them with OCT tissue freezing medium (Triangle Biomedical, Durham, NC), and flash froze them in liquid nitrogen. We sliced frozen embryos on a Microm HM 550 (Thermo Scientific, Kalamozoo, MI) at a thickness of 60 µm or 25 µm. We adjusted the horizontal position of the blade after every slice to eliminate the possibility of carry-over from previous slices, and used a new blade for every embryo. We placed each slice in an individual RNase-free, non-stick tube (Life Technologies, Grand Island, NY). ## RNA Extraction, Library Preparation, and Sequencing We performed RNA extraction in TRIzol (Life Technologies, Grand Island, NY) according to manufacturer instructions, except with a higher concentration of glycogen as carrier (20 ng) and a higher relative volume of TRIzol to the expected material (1 mL, as). For the 60 µm slices, we pooled total RNA from each slice with total RNA from single *D. persimilis*, *D. willistoni*, or *D. mojavensis* embryos, then made libraries according to a modified TruSeq mRNA protocol from Illumina. We prepared all reactions with half-volume sizes to increase relative sample concentration, and after AmpureXP cleanup steps, we took care to pipette off all of the resuspended sample, leaving less than 0.5 µL, rather than the 1–3 µL in the protocol. Furthermore, we only performed 13 cycles of PCR amplification rather than the 15 in the protocol, to minimize PCR duplication bias. Libraries were quantified using the Kapa Library Quantification kit for the Illumina Genome Analyzer platform (Kapa Biosystems) on a Roche LC480 RT-PCR machine according to the manufacturer's instructions, then pooled to equalize index concentration. Pooled libraries were then submitted to the Vincent Coates Genome Sequencing Laboratory for 50bp paired-end sequencing according to standard protocols for the Illumina HiSeq 2000. Bases were called using HiSeq Control Software v1.8 and Real Time Analysis v2.8. ## Mapping and Quantification Reads were mapped using TopHat v2.0.6 to a combination of the FlyBase reference genomes (version FB2012_05) for *D. melanogaster* and the appropriate carrier species genomes with a maximum of 6 read mismatches. Reads were then assigned to either the *D. melanogaster* or carrier genomes if there were at least 4 positions per read to prefer one species over the other. We used only the reads that mapped to *D. melanogaster* to generate transcript abundances in Cufflinks. ## Data and Software We have deposited all reads in the NCBI GEO under the accession number GSE43506. The processed data are available at the journal website and at <http://eisenlab.org/sliceseq> with a search feature for the 25 µm dataset. All custom analysis software is available <https://github.com/eisenlab/SliceSeq>, and is primarily written in Python. Commit b0b115a was used to perform all analyses in this paper. # Supporting Information We thank peer reviewer Boris Adryan for helpful comments, and many readers who contributed feedback on a preprint of the manuscript posted on MBE's blog and the arXiv. We also thank members of the Eisen lab for their assistance, especially Susan Lott and Jackie Villalta. [^1]: MBE is a cofounder and member of the Board of Directors of PLOS. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: PAC MBE. Performed the experiments: PAC. Analyzed the data: PAC MBE. Contributed reagents/materials/analysis tools: PAC MBE. Wrote the paper: PAC MBE.

# Introduction Hypertension is influenced by multiple risk factors, among which high NaCl intake is one of the risk factors that has been studied the most. It is believed that increased NaCl intake elevates blood pressure and thus favoring the development of hypertension. Considerable evidence indicated that it is the combination of Na⁺ and Cl^-, rather than Na⁺ per se, is responsible for the development of salt-sensitive hypertension. For example, in Dahl rats (an animal model of salt-sensitive hypertensive rat), selective loading with only Na⁺ or Cl⁻ but not both failed to induce changes in blood pressure. It has been indicated that decrease in absorption of Cl⁻ in renal tubular might be the reason for why sodium salt in the absence of chloride could not be able to increase blood pressure in the Dahl salt-sensitive rat. Cl⁻ is the richest anion in both extracellular and intracellular environments of our body. Therefore, it is not surprise that Cl⁻ transport along the nephron is important in the regulation of extracellular fluid volume as well as blood pressure. These results suggest that control of not only Na⁺ absorption but also Cl⁻ absorption is critical for regulation of blood pressure and thus contributes to the development of salt-sensitive hypertension. The thick ascending limb (TAL) of the loop of Henle reabsorbs over one-fifth of the filtered NaCl while absorbing no water. By doing so, the TAL segment helps the nephron to establish and maintain the hypertonic medullary solute gradient, generate dilute tubular fluid, and thus plays important roles in regulating fluid volume and blood pressure. Improper regulation of Cl⁻ absorption by this segment has been implicated in salt-sensitive hypertension. Cl⁻ absorption is a two-step process in the TAL. Cl⁻ entries across the apical membrane via Na⁺-K⁺-2Cl⁻ co-transporter. Generally, the proper function of Na⁺-K⁺-2Cl⁻ co-transporter requires the simultaneous presence of all three ions, which means that the transport of Na⁺ and Cl⁻ across the apical membrane is dependent on each other. Once inside the cell, Cl⁻ exits across the basolateral membrane down a favorable electrochemical gradient through the basolateral Cl⁻ channels. Ion-substitution experiment showed that the basolateral Cl⁻ channels provide the major pathway for Cl⁻ exit. These results suggest that Cl⁻ channels in the basolateral membrane play an important role in controlling Cl⁻ absorption in the TAL. Several researches using patch-clamp technique have found that two different Cl⁻ channels exist in the TAL: the smaller one with conductance of \~10 pS and the larger one with conductance of \~30–40 pS. It has been demonstrated that the smaller one is the major type of the basolateral Cl⁻ channels in the TAL. For example, our previous study has showed that the single channel activity of the 10-pS Cl⁻ could be obtained in 314 patches among the total of 1,147 patches investigated, whereas the single channel activity of the 30-pS Cl⁻ channel was observed in only 68 patches. Furthermore, we have demonstrated that nitric oxide (NO) synthase substrate (L-arginine) could inhibit the basolateral 10-pS Cl⁻ channel in the TAL through the cGMP/PKG signaling pathway. As a NO releasing agent, S-nitroso-N-acetyl-penicillamine (SNAP) might be a potential candidate for regulation of NaCl absorption by affecting the basolateral 10-pS Cl⁻ channel activity. In the present study, we studied the detailed molecular mechanism of SNAP on the regulation of this 10-pS Cl⁻ channel. # Materials and methods ## Preparation of the TAL for single channel recordings The pathogen-free C57BL/6 mice (male, 5 wks old) were from Laboratory Animal Center of Xiamen University (Xiamen, China). All animal procedures were in strict accordance with the National Institutes of Health’s Guidelines for the Care and Use of Laboratory Animals. All the experimental protocols were approved by the Animal Care and Use Committee of Xiamen University following the Guide for the Care and Use of Laboratory Animals. The mice had free access to water and were fed with a control diet. Every effort was made to minimize the number of animals used and to minimize suffering. After the mice had been euthanized by pentobarbital administration (150 mg/kg) followed by cervical dislocation, the kidneys were removed immediately and thin coronal sections (1-mm) were cut with a razor blade. The TALs were dissected in a HEPES-buffered NaCl solution containing 1 mg/mL collagenase type 1A (Sigma, St. Louis, MO) at 37°C for 55 minutes. Then, the dissected TAL was transferred onto a cover glass (5 mm × 5 mm) that was coated with poly-lysine (Sigma) overnight to immobilize the tubule. The cover glass was placed in a chamber mounted on an inverted microscope (Olympus) for single channel recordings and the tubules were superfused with HEPES-buffered NaCl solution containing (mM): 5 KCl, 140 NaCl, 1.8 MgCl₂, 1.8 CaCl₂ and 10 HEPES with pH of 7.4. ## Single channel recordings Single-channel recordings were obtained from cell-attached configuration of patch-clamp technique. The pipette solution contains (mM): 1.8 MgCl₂, 140 NaCl and 10 HEPES with pH of 7.4. Single channel activity was defined as *NP*_o, a product of channel number (*N*) and open probability (*P*_*o*). The *NP*_o was calculated from single channel recordings using the following equation: $$NP_{o} = \sum\ (1t_{1} + 2t_{2} + \ldots it_{i})$$ where *t*_i is the fractional open time spent at each of the observed current levels. For the single channel activity, the single channel recordings with duration of 60 seconds, which were obtained at least 2 min after addition of SNAP, were analyzed. ## Chemicals KT-5823, SNAP, carboxy-PTIO and 1H-Oxadiazolo\[4,3-a\]quinoxalin-1-one (ODQ) were from Sigma (St. Louis, MO). LY-83583 and BAY-60-7550 were from Santa Cruz Biotechnology (Santa Cruz, CA). ODQ, LY-83583 and BAY-60-7550 were dissolved in dimethyl sulfoxide (DMSO). To ensure that the channel activity was not affected by DMSO, its final concentration in the bath was less than 0.1%. ## Statistical analysis An IC50 value was obtained by fitting concentration dependence data from single channel recordings to the following equation: $$I\ (\%) = \left\lbrack {SNAP} \right\rbrack^{H}/(IC_{50}{}^{H} + \left\lbrack {SNAP} \right\rbrack^{H})$$ in which I (%) is the percentage of inhibition, H represents the Hill coefficient and \[SNAP\] represents concentration of SNAP. The percentage of inhibition at the test potential is calculated by the following equation: $$I\ (\%) = \lbrack 1‐NPo_{({SNAP})}/NPo_{({control})}\rbrack \times 100$$ where *NPo*_(SNAP) and *NPo*_(control) represent *NPo* of the channels in the presence of SNAP and under control condition, respectively. Data are shown as mean ± SEM. Differences in means were tested with paired sample t-test and were accepted as significant if *P* \< 0.05. # Results In most of the experiments, the *NPo* was different from different patches even under control condition. To avoid the effect of the values of different *NPo* of different patches on the experimental results, we applied different concentrations of SNAP on the same patch and then using paired sample t-test for statistical analysis. is a set of representative single channel recordings from the same cell-attached patch, showing that the *NPo* of the channel in the same patch was gradually decreased with the increase in \[SNAP\] from 0 to 10 μM. The statistical analysis on dose-response effect in showed that the mean *NPo* of the channel was 1.23 ± 0.08 before application of SNAP, and it decreased to 0.93 ± 0.04, 0.74 ± 0.07, 0.56 ± 0.07, and 0.48 ± 0.11 when \[SNAP\] was increased to 2.5, 5, 7.5 and 10 μM, respectively. A nonlinear curve fit of the Hill equation to the data points in yielded an IC₅₀ value of 6.62 ± 0.20 μM and a Hill coefficient of 1.31 ± 0.08 (n = 5), respectively. Because 5 μM SNAP was a suitable concentration to inhibit the channel, we used this concentration in the following experiments. To judge whether the inhibitory effect of SNAP on the single channel activity of the 10-pS Cl^- channels is due to its release of NO, we used a NO scavenger (carboxy-PTIO) to eliminate NO. illustrates a set of single channel recordings from the same patch under control condition, in the presence of carboxy-PTIO as well as in the presence of both carboxy-PTIO and SNAP. Application of 10 μM carboxy-PTIO alone did not affect the single channel activity. However, SNAP failed to inhibit the channel activity if carboxy-PTIO was applied. Statistical results from 6 set of experiments demonstrated that 10 μM carboxy-PTIO did not significantly affect the single channel activity, but the inhibitory effect of SNAP on the Cl^- channels was abolished in the presence of carboxy-PTIO, suggesting that the effect of SNAP to inhibit the channel was due to its release of NO. To investigate whether soluble guanylate cyclase (sGC) was involved in the inhibitory effect of SNAP on the channel activity, ODQ, a sGC inhibitor, was used in the following experiments. is a set of single channel recordings from the same patch under control condition, in the presence of sGC blocker ODG (10 μM) as well as in the presence of both ODG (10 μM) and SNAP (5 μM). Statistical results from 5 set of experiments demonstrate that ODG alone did not affect the single channel activity. However, application of ODG blocked the inhibitory effect of SNAP on the channels. To further confirm this result, we examined whether LY-83583, another specific sGC blocker, could also prevent the effect of SNAP on the channel. is a set of single channel recording from the same patch showing that even though 10 μM LY-83583 alone did not affect the channel activity, it could prevent the inhibitory effect of SNAP on the channels. Statistical analysis from 5 sets of experiments demonstrate that application of SNAP did not significantly inhibit the Cl⁻channel activity in the TAL that was pre-treated with LY-83583. These results suggest that the inhibitory effect of SNAP on the channel is due to its activation of sGC. To examine whether cGMP-dependent protein kinase (PKG) signaling pathway is involved, the effect of SNAP on the 10-pS Cl⁻ channels was investigated in the TAL pretreated with PKG inhibitor KT-5823. A set of single channel recordings from the same patch and statistical summary showed that KT-5823 (5 μM) alone had no significant effect on the single channel activity of the 10-pS Cl⁻ channel. However, inhibition of PKG could eliminate the effect of SNAP on the channel activity after the TAL was pretreated with KT-5823. Finally, to test whether the cGMP-stimulated phosphodiesterase II (PDE2) is involved, BAY 60–7550, a specific PDE2 inhibitor was used in the following experiment. A set of single channel recordings from the same patch and statistical summary showed that BAY-60-7550 alone did not significantly affect the 10-pS Cl⁻ channel activity. However, SNAP has the similar inhibitory effect on the channel even in the presence of BAY-60-7550, suggesting that PDE2 is not involved in mediating the inhibitory effect of SNAP on the single channel activity of the 10-pS Cl^- channel. # Discussion The basolateral Cl^- channels of the TAL belong to the ClC family. Two family members of ClC are found in the TAL: ClC-K1 and ClC-K2, corresponding to CLC-NKA and CLC-NKB in humans. For a functional channel, barttin subunit is necessary to facilitate the insertion of the channel into the plasma membrane of the cell. Although both ClC-K1 and ClC-K2 are expressed in the TAL, several lines of evidence indicate that ClC-K2 is the basolateral 10-pS channel; 1) Only ClC-K2 can be observed by immunohistochemistry technique in the basolateral membrane of the TAL; 2) Judged by electrophysiological properties, especially ion selectivity of the channel, it has been suggested that CLC-K2 is the dominant Cl⁻ channel in the TAL; 3) Single channel recordings indicated that the 10-pS channel is dominant Cl^- channel in the TAL. 4) Electrophysiological study on knockout mice has showed that ClC-K2 is the 10-pS Cl^- channel and it accounts for most of the basolateral Cl^- current in the TAL. Furthermore, as the major pathway for Cl^- exits across the basolateral membrane, this 10-pS ClC-K2 channel has been suggested to be essential for NaCl absorption in the TAL, which in turn plays roles in salt- sensitive increases in extracellular fluid volume and blood pressure regulation. Mutations that lead to loss-of-function of the channel cause Bartter’s syndrome in humans indicates that ClC-Kb is responsible for NaCl absorption in the TAL. Importantly, overexpression of ClC-K2 in Dahl rats increased Cl^- channel activity, which might contribute to the elevation of blood pressure of the rats. It has been found that T481S mutation in ClC-Kb strongly increased the Cl- conductance by a factor of 20. Interestingly, there is a strong association between the ClC-Kb^T481S carriers and higher average blood pressure as well as fraction of participants who had hypertensive blood pressure levels, suggesting that the mutation ClC-Kb^T481S of the channel may predispose to the development of hypertension. Electrophysiological studies performed on knockout mice indicate that ClC-K2 is essential for salt absorption in the TAL. Taken together, these results suggest that increase in salt absorption by activation of the basolateral 10-pS Cl^- channel (ClC-K2) in the TAL is associated with hypertension. Historically, the sodium absorption pathways have been studied much more than that of the chloride pathways. However, as a matter of fact, the absorption of both ions is mutually dependent with each other, which is critical for regulation of extracellular fluid volume as well as blood pressure. In the present study, we found that SNAP inhibited this basolateral 10-pS Cl^- channel in the TAL in a dose-dependent manner with an IC50 value of 6.6 μM. Furthermore, we showed that the inhibitory effect of SNAP on the channel was through activation of the cGMP-dependent PKG pathway but not through the cGMP-stimulated PDE2 pathway. A model of the signaling pathway for the inhibitory effect of SNAP on the basolateral 10-pS Cl^- channel in TAL is presented in. NO is released exogenously by SNAP (NO donor), diffuses across the plasma membrane. In the cytoplasm, NO reacts with GC, and stimulates the production of cGMP. This intracellular messenger in turn activates PKG, and inhibits the basolateral 10-pS Cl channels in TAL. Increase in activity of this channel in the TAL has been shown to be associated with hypertension. It is reasonable to assume that inhibitors of the 10-pS Cl-channel, such as SNAP, may have the potential of being developed as anti-hypertension agents. # Supporting information # Ethics statement Animal care and experiments were performed in accordance with procedures approved by the Animal Care and Use Committee of Xiamen University. [^1]: The authors have declared that no competing interests exist.

# Introduction The second generation antipsychotic (SGA) olanzapine is prescribed to treat schizophrenia and a growing number of other disorders in adults and children, but can cause adverse metabolic side-effects including increased body weight, caloric intake, and abdominal adiposity, and reduced physical activity. Metabolic side-effects are a growing concern due to co-morbidities such as diabetes and obesity, and are a risk factor for medication non-compliance. A number of potential mechanisms for SGA-induced metabolic dysfunction have emerged over the past few years. In particular, the histaminergic, serotonergic and dopaminergic neurotransmitter systems are thought to be highly implicated in SGA-induced body weight gain. However, SGAs have a broad receptor binding profile that allows direct and indirect effects on multiple neural and peripheral signalling pathways, and further research into other candidate systems is required. The hypothalamic arcuate nucleus (Arc) and the dorsal vagal complex (DVC) of the brainstem are well-documented for their role in appetite and energy homeostasis –; responding to the acute nutritional status and long-term regulation of energy stores in the body. Neurons of the Arc and DVC express G_i/o-coupled cannabinoid CB1 receptors (CB1R) , which facilitate the effects of cannabinoids on appetite and energy metabolism. Weight gain during olanzapine and clozapine treatment is associated with a CB1R gene polymorphism in individuals with chronic schizophrenia, and chronic high-dose risperidone treatment increases cannabinoid receptor agonist, \[³H\]CP-55940, binding density in the Arc of male rats. We previously demonstrated a decrease in \[³H\]CP-55940 binding density in the DVC of rats treated with olanzapine, but not aripiprazole or haloperidol. However, whether changes in receptor density were attributed to the CB1R is unclear due to the low specificity of the ligand used and localisation of cannabinoid CB2 receptors in the brain. Moreover, the effects of olanzapine on CB1R density in the Arc remain unknown. The appetite enhancing effects of the major neuronal inhibitor, γ-aminobutyric acid (GABA) in the hypothalamus were reported more than 30 years ago. GABAergic neurons in the Arc are sensitive to leptin, and GABA receptor agonists and antagonists stimulate and suppress feeding behaviour, respectively. Down- regulated expression of glutamic acid decarboxylase (GAD, the GABA synthesising enzyme) has been observed in individuals with schizophrenia, bipolar and mood disorder, whereas antipsychotic drug treatment increases cortical GAD expression in rats and primates. GAD exists as two isoforms, 65 and 67; the latter is found throughout the neuronal cytoplasm, whereas GAD₆₅ is located primarily in the axon terminal – and is the predominant transcript in the hypothalamus of the adult rat brain. However, to our knowledge the effects of olanzapine on GAD₆₅ mRNA expression in the hypothalamic Arc or the DVC have not been investigated. The Arc and DVC both express mRNA for orexigenic neuropeptide Y (NPY) and anorexigenic pro-opiomelanocortin (POMC). The POMC gene encodes for neuropeptides such as adrenocorticotropic hormone, β-endorphin and α-melanocortin stimulating hormone; the latter of which exerts its anorexigenic effects largely through melanocortin-3 and melanocortin-4 receptor (MC4-R) subtypes. Conversely, the central application of NPY induces food intake in a number of species, hypolocomotor activity in rats, and can lead to obesity following chronic over-exposure. Therefore, it is possible that interference in the balance of POMC and NPY by olanzapine may contribute to the drug's obesogenic liability. Several reports demonstrated increased NPY immunoreactivity in the Arc of clozapine-treated rats, whereas chronic risperidone treatment in male rats had no effect on POMC or NPY expression, or body weight, which may be due to the lower sensitivity of male rats to SGA- induced metabolic side-effects compared to females. Other studies have examined antipsychotic effects on NPY mRNA expression in the brain with region-dependent outcomes. The effects of antipsychotics on POMC or NPY in the brainstem have not been examined and studies on hypothalamic appetite-regulating peptides during olanzapine treatment are confounding; one group reported an increase in orexigenic NPY and AgRP and a concurrent reduction in appetite-inhibiting POMC and cocaine- and amphetamine-related transcript (CART), whilst another reported no change in several hypothalamic peptides, including NPY and POMC. A key factor that may contribute to the difference in findings is drug dosage (i.e.: 1 mg/kg olanzapine *vs.* a supratherapeutic dose of 3 mg/kg olanzapine (b.i.d.). Indeed, metabolic outcomes can differ with antipsychotic dosage – and increased dose induces greater metabolic dysfunction in the rat, however, high antipsychotic dosages in the rat may not represent the clinic. In addition, both studies had a large dosage interval, i.e.: 6–7 then 17–18 hourly treatments, b.i.d.. As the half-life of olanzapine is 5.1 hours in the rat brain with high levels remaining after 8-hours, compared to approximately 75.2 hours in the human brain, multiple dosages are required in the rat in order to minimise drug fluctuations below sub-therapeutic D2 receptor occupancy levels. Therefore, it may be possible to model olanzapine-induced metabolic dysfunction in the rat using low olanzapine dosages when administered in accordance with the half-life of the drug, i.e.: 8 hourly (t.i.d.) within in 24-hours. Using an established rat model of olanzapine-induced metabolic dysfunction,, this study aimed to investigate the mechanisms underlying weight gain associated with olanzapine treatment by examining its effects on POMC, NPY and GAD₆₅ mRNA expression, and CB1R binding density (using the CB1R-specific ligand \[³H\]-SR141716A) in the Arc and DVC. Statistical correlations between these parameters in the brain and body weight, food intake and visceral adiposity were investigated. To identify the minimum dosage threshold required to induce metabolic change, rats were treated with different clinically-relevant olanzapine dosages, calculated based on comparable therapeutic *in-vivo* dopamine D2 receptor occupancy levels and differences in body surface area between species. Collectively, the present study demonstrates that olanzapine changes the balance of anorexigenic POMC and orexigenic NPY mRNA expression in the Arc, does not alter POMC or NPY in the DVC, and increases GAD₆₅ mRNA expression but reduces CB1R density in both the hypothalamus and brainstem. These largely dose-sensitive changes may underlie a shift in energy balance that favours weight gain during olanzapine treatment. Metabolic dysfunction can be modelled in the female rat using low olanzapine doses when administered in-line with the half-life of the drug. # Methods ## Ethics Statement All experimental procedures were approved by the Animal Ethics Committee (Approval \#: AE06/32), University of Wollongong, and complied with the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes (2004). All efforts were made to minimise animal distress and prevent suffering. ## Animals and Treatment Seven week-old female Sprague Dawley rats (Animal Resources Centre, Perth, WA, Australia), housed in 12-h light–dark cycle (lights on 07:00, 22°C) were habituated for 1-week, then randomly assigned to 0.25, 0.5, 1.0 or 2.0 mg olanzapine/kg (Zyprexa, Eli Lilly, Indianapolis, IN, USA) or vehicle (control) (n = 12), administered three-times daily in a sweet cookie dough pellet, as described previously. Briefly, olanzapine tablets were de-coated and pulverized then the assigned dosage was added to measured dry ingredients. Water droplets were added immediately prior to administration to achieve a dry-dough consistency. After a 1-week teaching period, rats learnt to voluntarily self- administer a 0.3 g cookie-dough pellet either containing the assigned dosage of olanzapine, or plain cookie-dough without the drug (control group), offered by a metal spoon at 8-hourly intervals (3 pellets/day) for 14-days. Consumption of each pellet was observed to ensure complete dosing. Body weight and food intake measurements were recorded (n = 12). Animals were allowed *ad libitum* access to water and standard laboratory chow diet throughout the study. Animals were fasted for 4–6 hours then euthanized using sodium pentobarbitone 10–12 hours after the last treatment. Brain tissue was immediately frozen in liquid nitrogen and stored at −80°C. Visceral (perirenal and periovary) white fat pads were dissected and weighed (n = 12). Six brains were randomly selected from each treatment group for use in the mRNA expression and receptor binding experiments. Tissue was sectioned (14 µm, −18°C) along the coronal plane then stored at −20°C. ## NPY, POMC and GAD₆₅ mRNA In-Situ Hybridisation POMC mRNA expression was observed using in-situ hybridisation techniques previously described by our laboratory, using the following specific antisense hybridisation probe: 5′-CGTTCTTGATGATGGCGTTCTTGAAGAGCGTCACCAGGGGCGTCT-3′ (J00612, 547–591). NPY mRNA expression was observed using in-situ hybridisation techniques previously described by our laboratory, using the following specific antisense hybridisation probe: 5′-GAGTGTATCTGGCCATGTCCTCTGCTGGCGCGTCCTCGCCCGG-3′ (M15792, 1650–1693). GAD₆₅ mRNA expression was observed using the following specific antisense hybridisation probe: 5′-GGCGTCCACACTGCAAGGCCTTGTCTCCCGTGTCATAGGACAGGTCAT-3′ (NM_012563.1, 1419-1372), as previously described by Ling et al.. Oligonucleotide probes were terminally labelled using \[³⁵S\]dATP (1000 Ci/mmol, Perkin Elmer, Waltham, MA, USA) in 10-fold molar excess and terminal transferase (Promega, Madison, WI, USA), then purified using a MicroSpin G-50 column (GE Healthcare Ltd, Buckinghamshire, UK). Hybridisation was performed by incubating slides in hybridisation buffer (4× SSC, 1× Denhardt's solution, 50% de-ionised formamide, 200 µg/ml sperm DNA, 100 µg/1 ml polyA, 120 µg/ml heparin, 20 mM sodium phosphate and labelled probe, pH 7.0) for 18-hours at 37°C. Slides were then washed in 1× SSC buffer at 55°C (3×30-minutes each) and incubated for 1-hour in SSC buffer at room temperature. Sections were dipped in Milli-Q water followed sequentially by 70% then 95% ethanol, and dried under a gentle stream of air. Autoradiographic images were captured on film (Kodak BioMax MR film, Rochester, NY, USA) exposed for 3-weeks. Films were quantified using a GS-800 Densitometer (Bio-Rad Laboratories, Inc), and analysis software (Quantity One, v4.6.7, Bio- Rad Laboratories, Inc, CA, USA). Values were derived from a standard curve generated from a \[¹⁴C\]-labelled autoradiographic standard (GE Healthcare Ltd, Buckinghamshire, UK) (mean binding nCi/g tissue equivalent vs. density). Slides were dipped in Emulsion solution (GE Healthcare Ltd, Buckinghamshire, UK) and exposed for 6-weeks, then stained with cresyl violet (Nissl stain) (Sigma-Aldrich, NSW, Australia), to allow further examination of positive signals at the cellular level. ## CB1R Binding Density CB1R binding density was detected using methods previously published by our laboratory. Briefly, air-dried slides were pre-incubated for 15 min in incubation buffer containing 50 mM Tris–HCl buffer (pH 7.4) and 0.1% bovine serum albumin, at room temperature. Sections were then incubated with 10 nM \[³H\] SR141716A (52 Ci/mMol, Amersham, UK), a CB1R-specific inverse agonist, in buffer (pH 7.4) at room temperature for 60 minutes to determine total binding. Non-specific binding was determined by incubating subsequent sections in 10 nM \[³H\] SR141716A in the presence of 100 µM CP-55940, in buffer (pH 7.4) for 60 minutes at room temperature. Slides were washed in ice cold buffer (pH 7.4), (2×30 minutes), then dipped in distilled water and dried under a gentle stream of cool air. CB1R autoradiographic images were captured using a Beta Image camera (BioSpace, Paris, France), which counts the amount of β-particles emitted from the tissue (3.5 hours exposure) to determine the level of radioactivity bound to the brain sections. Radioactive levels were obtained in counts per minute per square millimetre of tissue (cpm/mm²), converted to nCi/mg tissue equivalent using standard tissue sections calibrated with commercial standards (Amersham, Buckinghamshire, United Kingdom), then transformed into fmol/mg tissue equivalent by taking into account the specific activity of the radioligand (52 Ci/mMol). Quantification was conducted using β-Image Plus software (version 4, BioSpace, Paris, France). ## Quantification and Statistical Analysis Quantification of autoradiographic images was performed on the hypothalamic Arc and the DVC of the brainstem, which were confirmed using a corresponding set of cresyl violet-stained slides and a standard rat brain atlas. Data were analysed using SPSS (version 17.0, SPSS, Chicago, IL, USA). All data points were within ±2 standard deviations. One-Sample Kolmogorov-Smirnov tests revealed normal data distribution. One-way ANOVAs were employed to determine the effect of treatment on percentage body weight change, food intake, visceral adiposity, as well as NPY, POMC and GAD₆₅ mRNA expression, and CB1R binding density in the hypothalamus and brainstem. ANOVAs were followed by multiple comparisons using post-hoc Dunnett-T tests where relevant (*p*\<0.05). Correlations were identified using Pearson's correlation tests. # Results ## Body Weight, Food Intake and Visceral Adiposity There was a significant effect of treatment on the percentage of body weight change from treatment day 0 (*F*_4,55 = 7.68, *p*\<0.01). Compared to controls, olanzapine significantly increased percentage of body weight change in the 0.5 mg/kg (*p*\<0.05), 1.0 mg/kg and 2.0 mg/kg (*p*\<0.01) treatment groups, but not in the low dosage group of 0.25 mg/kg (*p*\>0.05). Mean cumulative food intake significantly increased in the 2.0 mg/kg olanzapine dosage group compared to the control group (318.6±11.2 g *vs.* 269.4±11.1 g, *p*\<0.05). An increase (7–8%) in food intake was also observed in the 0.5 mg/kg and 1.0 mg/kg olanzapine treatment groups, but did not reach significance compared to the controls (0.5 mg/kg: 289.1±13.9 g *vs.* 269.4±11.1 g, 1.0 mg/kg: 291.6±12.1 g *vs.* 269.4±11.1 g, *p*\>0.05), and the low dosage group (0.25 mg/kg) did not differ to the control group (265.4±11.7 g *vs.* 269.4±11.1 g). Olanzapine treatment had a significant effect on visceral adiposity (*F*_4,55 = 4.60, *p*\<0.01), with a significant increase observed in the 2.0 mg/kg olanzapine treatment group (*p*\<0.05) and a trend for an increase in the 1.0 mg/kg dosage group (*p* = 0.09), but not in the lower dosage groups (*p*\>0.05). ## POMC and NPY mRNA Expression Examples of POMC and NPY mRNA expression in the hypothalamus are shown in. Olanzapine had a significant effect on POMC mRNA expression in the Arc (*F*_4,25 = 8.32, *p*\<0.01), not in the DVC (*F*_4,25 = 1.44, *p* = 0.25). Post-hoc analysis identified a significant decrease in POMC mRNA expression in the Arc following dosages of 0.5 mg/kg, 1.0 mg/kg and 2.0 mg/kg (*p*\<0.01) olanzapine, but not 0.25 mg/kg olanzapine (*p* = 0.90), compared to controls. There was also a significant effect of treatment on NPY mRNA expression in the Arc (*F*_4,25 = 8.55, *p*\<0.01), with a significant increase in the Arc following 1.0 mg/kg and 2.0 mg/kg olanzapine, but not in the lower dosage groups (*p* = 0.34 and *p* = 0.11 for 0.25 mg/kg and 0.5 mg/kg olanzapine, respectively). NPY mRNA expression in the DVC was unaltered by olanzapine (*F*_4,25 = 0.73, *p* = 0.58). ## GAD₆₅ mRNA Expression Examples of GAD₆₅ mRNA expression are shown in. A significant effect of treatment on GAD₆₅ mRNA expression was observed in the Arc (*F*_4,25 = 5.21, *p*\<0.01) and DVC (*F*_4,25 = 7.73, *p*\<0.01), with an increase following olanzapine dosages of 1.0 mg/kg (Arc *p*\<0.05, DVC *p*\<0.01) and 2.0 mg/kg (both regions *p*\<0.01), but not in the 0.5 mg/kg or 0.25 mg/kg groups (*p*\>0.05), compared to controls. ## CB1R Binding Density An example of CB1R binding density in the hypothalamus and DVC is shown in. There was a significant effect of treatment on CB1R binding density in the Arc (*F*_4,25 = 7.48, *p*\<0.01), with a reduction in all olanzapine treatment groups compared to controls (*p*\<0.01). Olanzapine decreased CB1R binding density in the DVC (*F*_4,25 = 3.48, *p*\<0.05) of animals treated with 0.5 mg/kg, 1.0 mg/kg or 2.0 mg/kg olanzapine (*p*\<0.05), but not 0.25 mg/kg olanzapine (*p* = 0.43). ## Correlations POMC mRNA expression in the Arc significantly correlated to percentage body weight change (*r* = −0.43, *p*\<0.05), visceral adiposity (*r* = −0.49, *p*\<0.01), NPY mRNA expression (*r* = −0.45, *p*\<0.05), and GAD₆₅ mRNA expression in the Arc (*r* = −0.54, *p*\<0.01). There was a significant positive correlation between NPY and GAD₆₅ mRNA expression in the Arc (*r* = 0.69, *p*\<0.01), however the two factors did not correlate to percentage body weight change (*p*\>0.05). CB1R binding density in the DVC correlated to percentage body weight change (*r* = −0.38, *p*\<0.05), visceral adiposity (*r* = −0.38, *p*\<0.05) and GAD₆₅ mRNA expression in the DVC (*r* = −0.52, *p*\<0.01), and a weak correlation was observed between CB1R binding density in the Arc and percentage body weight change (*r* = −0.33, *p* = 0.08). # Discussion We found that olanzapine alters signals in the hypothalamus and brainstem that are implicated in appetite and energy homeostasis in a largely dose-sensitive manner. These changes may underlie a shift in energy balance that favours weight gain. The data support a role for POMC in the mechanisms underlying olanzapine- induced obesity. Reduced POMC satiety signalling leads to obesity in the clinic and in animal models of obesity, for example, POMC mRNA expression is attenuated in genetically obese Zucker rats, tubby mice (*tub* gene mutation) and diet- induced obese mice. In addition, genetic POMC deficiency leads to obesity in humans and mice, and MC4-R deficiency leads to morbid obesity associated with enhanced adiposity and chronic hyperphagia. The result of unaltered POMC mRNA expression in the DVC was not entirely surprising as the role of POMC neurons in the DVC is not well-characterised, and functional and chemical distinctions to the Arc have been identified. NPY mRNA expression was upregulated in the Arc following 1.0 mg/kg and 2.0 mg/kg olanzapine treatment, however no significant correlation with weight gain was observed. This is consistent with some NPY transgenic and deficiency models i.e.: mice and rats that over-express NPY do not have a hyperphagic/obese phenotype, and genetic modelling of NPY-deficiency does not result in reduced body weight, adiposity, or food intake. However, it is possible that NPY had an indirect effect on weight gain in olanzapine-treated animals, for example by inhibiting POMC. Indeed, NPY neurons synapse on POMC cell bodies and can inhibit their spontaneous activity, however, unlike POMC, NPY mRNA expression did not change in the 0.5 mg/kg olanzapine treatment group suggesting a role for other systems in POMC regulation. The dosage response of NPY mRNA expression was in- line with the increase in GAD₆₅ mRNA expression in the 1.0 mg/kg and 2.0 mg/kg olanzapine treatment groups, although GAD₆₅ mRNA expression increased in both the Arc and DVC. Upregulated GABAergic signalling during weight gain is consistent with previous reports that whilst NPY and/or AgRP gene deficiency is insufficient to reduce food intake, ablation of NPY/AgRP/GABA neurons results in acute hypophagia and deletion of vesicular GABA transporter in AgRP neurons (which co-express NPY) results in a lean, obesity-resistant phenotype in mice. GABA is co-localised in approximately 30% of POMC and NPY/AgRP neurons in the Arc, and GABA derived from NPY/GABA axons can suppress spontaneous firing of POMC neurons. In addition, a dense population of leptin- responsive, largely non-AgRP GABAergic neurons that increase inhibitory post- synaptic currents in POMC neurons were recently identified in the Arc. Therefore, the increase in Arc GAD mRNA expression observed in the present study may have arisen from a number of GABAergic sources. Olanzapine treatment elicited a robust reduction in CB1R density in the Arc and DVC. Using the CB1R-specific ligand, \[³H\]-SR141716A, we confirm that our original findings of a reduction in \[³H\]CP-55940 binding density in the DVC during olanzapine treatment were attributed to the CB1R sub- type and extend these findings into the hypothalamus. CB1R number and cell signal transduction pathways decrease following over-exposure to agonists and animal models of obesity, for example obese *db/db* and *ob/ob* mice, and fatty Zucker rats, exhibit elevated hypothalamic endocannabinoid levels. Therefore, reduced CB1R binding density following olanzapine treatment may be a result of increased endogenous cannabinoids. Endogenous cannabinoids play an important regulatory role in synaptic transmission by modulating neuronal excitatory and inhibitory input. Interestingly, POMC neurons secrete endocannabinoids under basal conditions that retrogradely activate CB1Rs expressed on GABAergic neurons. G-protein sub-units coupled to the CB1R inhibit the opening of calcium channels, which reduces vesicular release of GABA. CB1R activation can relieve inhibitory input to the post-synaptic POMC neuron –. We suggest that reduced CB1R density during olanzapine treatment may diminish cannabinoid-regulated inhibition of GABA, and therefore enhance GABAergic input to POMC neurons, suppressing POMC and encouraging body weight gain. In addition, anandamide and CP-55940 increase NPY release in the hypothalamus, therefore, increased endocannabinoid levels may contribute to an increase in NPY during olanzapine treatment that further suppresses POMC. CB1Rs can also modulate GABA and glutamate release in the DVC, however the functional implications of changes in CB1R density and GAD₆₅ mRNA expression during olanzapine treatment require further investigation. Additionally, the influence of olanzapine on other neurotransmitter systems may play a role in the mechanisms underlying SGA- induced weight gain. For example, olanzapine is a potent histamine H1 receptor antagonist and antipsychotic affinity for the H1 receptor can predict its weight gain liability, however the underlying mechanisms for the effect of the H1 receptor on antipsychotic-induced body weight may be independent of melanocortinergic neurotransmission. On the other hand, dopamine D1 and D2 receptor antagonism influences hypothalamic NPY mRNA expression – and serotonin 5-HT_2C receptor agonists can activate POMC neurons , therefore, the antagonistic affinity of olanzapine to D2 and 5-HT_2C receptors, may contribute to its weight gain side-effects. These receptors may form broader components of the mechanism proposed in the present study, however further research is necessary. Our finding of a decrease in POMC and increase in NPY mRNA expression during olanzapine treatment coincide with Ferno et al, but contrast to the lack of change reported by Davoodi et al. As discussed earlier, these studies differ in olanzapine dosage and treatment interval. Additionally, in Davoodi's study animals were not fasted and PCR methods were used to detect expressional changes in the whole hypothalamus, whereas rats were fasted prior to euthanasia and *in- situ* hybridisation techniques were utilised to target expression specifically in the Arc in the present study and. Furthermore, patterns of daily changes in hypothalamic NPY and POMC gene expression have been reported, therefore timing of euthanasia may also confound results. A previous study from our laboratory reported a drug withdrawal response of NPY mRNA expression to olanzapine treatment cessation, i.e.: no change in Arc NPY mRNA expression after 2-hour drug washout and a decrease after 48-hour withdrawal after 5-weeks olanzapine treatment. As body weight associated with olanzapine treatment follows a ‘peak- and-plateau’ trend over time, the lack of change in NPY mRNA expression may be related to compensatory mechanisms that coincide with a plateau in body weight. Further investigation into the time-dependent pattern of NPY mRNA expression during chronic olanzapine treatment would be useful. Secher et al. reported increased \[³H\]CP-55940 binding density in the Arc following 28-days risperidone treatment, and observed a significant correlation between plasma drug levels and visceral adiposity. These results are similar to our study as olanzapine influenced CB1R density in the Arc and changes in CB1R density correlated with adiposity. Differences in the direction of CB1R density change may be attributed to several differences in experimental design in Secher et al.'s study, including drug dosage above the upper clinical limit, administration method i.e.: continuous drug application via mini-pump with no drug washout period, and treatment duration as time-dependent changes in CB1R density and transduction pathways have been reported. Neither drugs have an affinity for the CB1R (\>10,000 K_i (nM),), indicating that effects on the CB1R are secondary changes and exactly how these SGAs influence CB1Rs should be investigated in future studies. An olanzapine-induced decrease in CB1R binding density seems contrary to the orexigenic influence of CB1R activation, and appetite suppression of CB1R blockade. However, there is vast potential for the endogenous cannabinoid system to modulate metabolism, including central and peripheral effects on food intake and reward aspects of feeding, glucose and lipid metabolism, and energy expenditure –; aspects of which may contribute to the weight loss efficacy of rimonabant. Olanzapine-induced weight gain is associated with increased GABA and decreased CB1R density, whereas anorexigenic rimonabant, decreases GABA release from NPY/AgRP/GABA neurons, possibly via modulation of cannabinoid-sensitive opioid peptides. This suggests that although olanzapine and rimonabant influence the CB1R, they exert their effects through different modes of action. The doses used in the present study were selected based on the recommended clinical olanzapine dosage range of 5–20 mg/day, excluding the 0.25 mg/kg treatment group, which was included as a minimum response threshold. Olanzapine was administered every 8-hours, based on the half-life of olanzapine in the rat brain, to minimise inappropriate peaks and troughs in drug levels between treatments. The present study demonstrates that olanzapine-induced metabolic dysfunction can be modelled in the female rat using low olanzapine dosages when treatment is administered in accordance with the half-life of the drug. In addition, treatment was voluntarily self-administered orally in a cookie-dough pellet, which aimed to minimise potential handling stress and maintain a consistently high drug dosage in the brain. Oral drug administration in rats requires a teaching period to ensure voluntary pellet consumption, however this method resembles clinical administration and may circumvent limitations reported using other administration techniques, such as mini-pump, injection and gavage,. Consistent with the clinic, olanzapine has a sedative effect in the rat at high doses and we previously reported a general trend of reduced locomotor activity in response to increasing olanzapine dosage. It is possible that sedation plays a role in weight gain during olanzapine treatment, however, as hyperphagia was only apparent in the high dosage group (2 mg/kg olanzapine) in the present study, it is unlikely that sedation influenced the animal's ability to consume food. In conclusion, our data demonstrates that olanzapine, an antipsychotic drug with a high metabolic liability, alters key metabolic signals in the hypothalamus and brainstem in a manner that favours positive energy balance and may contribute to its weight gain/obesity side-effects. Olanzapine decreases anorexigenic POMC, increases orexigenic NPY, and alters CB1R and GABAergic signalling in a largely dose-sensitive manner. Low dosages of 0.5 mg/kg and 1.0 mg/kg olanzapine (t.i.d.) were sufficient to induce metabolic changes. Drug dosage may contribute, in-part, to inconsistencies observed between reports in the literature. Enhanced body weight and visceral adiposity during olanzapine treatment are associated with reduced anorexigenic POMC mRNA expression. We propose that increased NPY and enhanced inhibitory GABAergic input, possibly through reduced CB1R density, may contribute to POMC inhibition. However, the present study has several limitations, firstly, statistical correlations do not provide direct evidence of a causal link, and secondly, changes to mRNA and receptor binding density may not reflect a functional protein change, therefore further studies are required to confirm the mechanism proposed in the present study. Examination of olanzapine's effects on the GAD₆₇ isoform and other hypothalamic neuropeptides, such as AgRP and CART, would be useful, as well as investigation into the time-response of all parameters at different intervals during treatment. Finally, as CB1R density decreased in all olanzapine dosage groups, experiments using lower dosages are required to identify the minimum dosage threshold. Taken together, this study supports a role for the melanocortinergic, GABAergic and cannabinoid systems in the underlying mechanisms contributing to olanzapine-induced weight gain side-effects and provides direction on dosage consideration for future animal modelling studies. We sincerely thank Dr Kai Kang, Dr Mei Han, Ms Kelly Liu and Ms Jiamei Lian for their expert technical assistance. [^1]: Conceived and designed the experiments: KWG XH CD. Performed the experiments: KWG. Analyzed the data: KWG XH CD. Contributed reagents/materials/analysis tools: KWG XH CD. Wrote the paper: KWG XH CD. [^2]: The authors have read the journal's policy and have the following conflicts: CD received an honorarium from Eli Lilly, Australia, for presenting at the Cutting Edge Debate, Melbourne (2010). This does not alter the authors' adherence to all PLoS ONE policies on sharing data and materials.

# Introduction Deep learning (DL) technology has facilitated technological innovations in various fields, including computer vision, natural language processing, and speech analysis. These technologies are also being exploited in the medical field, with devices incorporating DL in endoscopy, radiology, and histopathology being used in actual clinical practice. For example, cytology is a less invasive procedure than histopathology for collecting cells from the lesions of patients directly. Recently, machine learning or DL technologies used for diagnosing smears and liquid-based cytology (LBC) specimens obtained from the cervix have been investigated utilizing digitized glass slide images known as whole-slide images (WSIs). The use of DL technology in cytology has the potential to enable quantitative, objective, and reproducible testing. Training high-accuracy DL models necessitates enough high-quality labeled datasets. However, there are no open- source datasets with sufficient high-quality data in the field of cytology. Therefore, the data for specific purposes must be collected manually, which is time-consuming. In addition, previous studies on cytological evaluations using DL methods were typically meant to classify and detect atypical epithelial cells at a higher magnification, such as 20× or 40×, resulting in the burden of analyzing as many as 2,000 to 5,000 single cells present in a specimen. A systematic review also revealed that most studies on cytological diagnosis using DL methods had been conducted in experimental settings and have not yet been implemented in clinical practice. Cervical cancer is the fourth most frequent cancer in women worldwide, with an estimated 604,000 new cases and 342,000 deaths in 2020, with low- and middle- income countries accounting for approximately 90% of the cases and deaths. For early diagnosis of cervical cancer, its screening using cervical cytology specimens is performed based on the Bethesda System. However, compared to high- income countries, cytology is less extensively employed in low- and middle- income countries due to a lack of healthcare infrastructure and a paucity of cytologists or cytopathologists. In the first step of cytology screening, an entire region on a glass slide is observed under low magnification, and the areas where abnormal cells are suspected are observed in detail under higher magnification in the second step. More than 90% of cervical cancer screening specimens are free of atypical cells and are diagnosed as negative for intraepithelial lesion or malignancy (NILM). In most cases, cytologists or cytopathologists can usually diagnose NILM in the first step. Therefore, establishing a DL model that can evaluate NILM at low magnification, as human diagnosticians usually do, is beneficial for reducing labor and supplementing human resources for screening. However, no studies have been conducted using a DL model for low-magnification observation. This study aimed to develop a DL model for use at a low magnification that classifies cervical LBC images with less labeled data. We trained a DL model based on a convolutional neural network (CNN) by introducing semi-supervised learning and verified its performance in predicting normal and abnormal images at low magnification. We also evaluated the performance of the model as a screening tool for NILM cases. # Materials and methods ## Data selection Cervical specimens were obtained from JA Shizuoka Kohseiren Enshu Hospital patients from October 2020 to October 2021. Only cervical specimens from the patients who did not undergo a hysterectomy or cervical conization were used in the study. Only the first specimen was used for the same patient sampled multiple times during this period. Each specimen was subjected to BD SurePath™ (Becton Dickinson, Inc., Franklin Lakes, NJ, USA) LBC and standard Papanicolaou staining. Two cytologists (with more than 20 years and 10 years of experience in cytology diagnosis, respectively) and two cytopathologists (each with more than 10 years of experience in cytology diagnosis) diagnosed the LBC specimens. Overall, 140 cases were randomly selected from the above. According to the Bethesda System, 100 of them were diagnosed as intraepithelial lesions: 50 with low-grade squamous intraepithelial lesions (LSIL) and 50 with high-grade squamous intraepithelial lesions (HSIL); the remaining 40 were diagnosed with NILM. The Ethics Review Committees of Hamamatsu University School of Medicine and JA Shizuoka Kohseiren Enshu Hospital approved this study (21–131). We obtained written opt-out consent. ## Data processing and assigning pseudo labels The LBC specimens were scanned at 40× magnification using a whole-slide scanner (NanoZoomer 2.0-HT; Hamamatsu Photonics, Hamamatsu, Japan) and converted into WSIs. The WSIs were divided into small patches of 1,024 × 1,024 pixels (0.92 microns/pixel), called tiled images, equivalent to a 10× objective lens of an optical microscope. The number of pixels excluding the background of the tiled images was used to determine cell volume per tile, and images with a cell volume of 30% or more were kept for later evaluation. depicts an overview of the learning pipeline. The CNN model training was divided into three stages. Further, of the 100 WSIs diagnosed as intraepithelial lesions, 10 from LSIL and 10 from HSIL were randomly selected. From these cases, 7,493 tiled images were extracted, and each tile was labeled as normal (when no abnormal cells appeared in the tiled image) or abnormal (when abnormal cells that might be used for cell diagnosis appeared). The tiled images labeled as normal were randomly downsampled to equalize the number of the images labeled as abnormal, and 2,600 images were used as training data for training a teacher model in the first stage. Next, of the remaining 80 WSIs, 20 from LSIL and 20 from HSIL were randomly selected. As an unlabeled dataset, 19,437 tiled images were obtained from these cases. In the second stage, the teacher model obtained in the first stage was used to assign pseudo labels to the unlabeled data. To select images for pseudo labeling, a confidence score for the prediction of each image was used. Images with a confidence score of 0.8 or higher were selected, and the prediction was applied as a pseudo label on the image. The pseudo- labeled data and 7,493-labeled data were combined; tile images labeled as normal were randomly downsampled to equalize the number of images labeled as abnormal, and 13,814 images were used as training data for training a student model in the second stage. In the third stage, the model obtained in the second stage was used as a teaching model to assign pseudo labels to the unlabeled images. Images with a confidence score of 0.9 or higher were selected. The same operations as in the second stage were performed. From the remaining 20 LSIL and 20 HSIL cases, 8,950 tiled images were obtained and were manually labeled as a test set to evaluate the model performance. From 40 NILM cases, 21,116 tiled images were obtained, and all were labeled as normal. Ideally, all tiled images obtained from NILM cases should be classified as normal by the model; subsequently, we evaluated the confidence score of each image and abnormal ratio (AR) for each case to assess the model performance as a screening tool for NILM. AR was calculated by dividing the number of images classified as abnormal by the total number of images. A human cytologist manually reviewed images classified as abnormal by the model, and the regions in the image that influenced the prediction were visualized using the Gradient- weighted Class Activation Mapping (Grad-CAM) technique. ## CNN training Noisy Student Training was used as the learning approach in this study. Compared to other semi-supervised learning methods, the Noisy Student method has been widely used for various tasks, including machine-learning competitions. We adopted it because of its ease of implementation. It does not require large amounts of labeled data, and it uses two models: a teacher and a student model. The teacher model is trained on labeled data, following which the model generates pseudo labels for unlabeled data. Then, by combining the labeled and pseudo-labeled data, a student model is trained with noise added to the data. To make the student model equivalent to or better than the teacher model, these training processes are iterated a few times. In this study, two-stage student learning was performed as described above. EfficientNet was used for a CNN architecture in this study. EfficientNet is a CNN model released in 2019 with a high-performance architecture with fewer parameters than traditional models. The model was pre-trained in ImageNet-1k, which provides eight levels of models (B0−B7) at different scales, and EfficientNet-B3 was used in this experiment. summarizes the number of images and training parameters used to train the model. We did not scale up the model at every stage. For example, we used data balancing without changing the batch size ratio of unlabeled and labeled data because our model was not very large and the dataset was small. Instead of learning from scratch, we used a pre-trained model from ImageNet to make the best use of our relatively restricted computational resources and accelerate the learning process. The dataset was divided using the holdout method and fine- tuning so that no duplicate cases were found in the training and validation data. The training was performed using an RTX A6000 GPU single graphics card (NVIDIA, Santa Clara, CA, USA) with 48 GB memory, with PyTorch serving as the framework. ## Data augmentation During training, basic augmentations were performed using the Albumentations library with the following augmentations: VerticalFlip (50%), Rotate (50%), RandomGridShuffle (50%), RandomBrightnessContrast (30%), and RandomGamma (30%). These augmentations were applied to the training data based on established probabilities, and for each epoch, either RandomBrightnessContrast or RandomGamma was applied. In the first stage, only basic augmentation was used to train the teacher model, while in the second and third stages, Mixup, CutMix, Drop Out, and Stochastic Depth were used to train the student model. Augmentation was not applied to the validation data. ## Model evaluation The area under the curve (AUC), accuracy, and F1 score were calculated at each stage. In addition, the false positive rate (FPR) and false negative rate (FNR) at each stage for the test data were calculated. FPR and FNR were calculated as follows. <img src="info:doi/10.1371/journal.pone.0285996.e001" id="pone.0285996.e001g" /> F P R = F P F P \+ T N <img src="info:doi/10.1371/journal.pone.0285996.e002" id="pone.0285996.e002g" /> F N R = F N F N \+ T P FP, TN, FN, and TP stand for false positive, true negative, false negative, and true positive, respectively. Even images with low prediction probability are classified into one of the two classes in a binary classification task. Therefore, cytologists or cytopathologists should re-confirm images that the DL model classifies with low confidence for screening purposes. We used Youden’s J statistic to determine the confidence score cutoff value. This comprehensive assessment is performed considering sensitivity and specificity, which are important factors in determining diagnostic accuracy. If a calculated confidence score of an image by the model is below the cutoff, the image is classified as abnormal. Youden’s J statistic (*J*) was calculated as follows: $$J = TPR - FPR$$ $$TPR = \frac{TP}{TP + FN}$$ TPR represents the actual positive rate. # Results ## Classification performance During the training phase, the AUC, accuracy, and F1-score increased with each successive stage, with the highest score in the third stage (AUC: 0.910, accuracy: 0.911, F1-score: 0.910). For the test data, the receiver operating characteristic (ROC) curve revealed AUCs of 0.909 and 0.908 for the second and third stages, respectively. The calculated Youden’s J statistic was 0.7, which was used as the confidence score cutoff value. shows the confusion matrix. We increased sensitivity while maintaining high specificity and a high F1 score. In addition, cutoff values were adjusted to reduce the FNR, improve sensitivity, and maintain high specificity and a high F1 score. This DL model achieved the best performance for the screening process at the third stage. The median confidence score was the lowest at the first stage and increased significantly (normal: p\<0.001, abnormal: p = 0.0028) at the second stage for both groups predicted as normal and abnormal. However, no significant difference was observed between the second and third stages. The ranges of confidence score values for the normal group were smaller than those for the abnormal group. The median confidence score for the abnormal classification was slightly lower than that for the normal classification at the second and third stages. ## Evaluation of NILM cases Forty NILM cases were evaluated using the third-stage DL model with a cutoff value of 0.7. presents detailed information on the cases and results. The median AR of these cases was 0.114 (IQR: 0.014–0.309), and 27 of the 40 cases (67.5%) had an AR of \< 0.2. Among 13 cases with AR \> 0.2, 7 (53.8%) presented cellular changes associated with atrophy and were over 50 years of age, suggesting that the observed cellular changes were related to aging or postmenopausal changes. Another four cases (30.8) revealed significant cell overlap. In the images that the DL model classified as abnormal, cellular changes associated with atrophy (Cases 1–7, 9, 12, 15, 21, and 26;), bacterial flora (Cases 14, 22, and 27;), squamous metaplasia (Cases 10 and 17;), endocervical cells (Case 19), cellular overlap (Cases 8, 11, 13, 14, 16, 18–20, 23–25, and 27;), and cell clusters (Cases 10 and 19;) were observed. The areas that Grad- CAM heatmaps highlighted in the images confirmed these findings. # Discussion Cytologists and cytopathologists can quickly recognize the wide background part and shapes of multiple cells in a cervical cytology smear at low magnification. If no atypical cells are found, the specimen is diagnosed as NILM. In other words, NILM may be diagnosed at low magnification by recognizing the texture of the low-magnification image as a single image and matching it with normal images throughout their careers. Therefore, this study focused on developing a DL model that, at low magnification, identifies images without abnormal cells as being normal, which is the first step in the standard cervical cytology screening process. Previous studies focused on diagnosing appeared cells, whereas our DL model aimed at detecting abnormal cells. Noisy Student Training achieved high reliability in classifying cervical cytology images using less labeled data, where only around one-tenth of the total data was used to develop the model. Additionally, because of its high specificity and low AR, our model is suitable as a screening tool for NILM cases. Specimens requiring careful observation include those from older women (\>50 years), which tend to present cellular atrophy related to aging or postmenopausal changes, and those showing cell overlapping. The model may regard these changes as abnormal. Further, given the above, a working example can be suggested: when the AR is \< 0.2, only the tiled images evaluated as abnormal should be checked by humans, while when the AR is equal to or \> 0.2, the physical glass slide should be observed under a microscope. The above operation, which uses the developed DL model, will allow cytologists or cytopathologists to concentrate on cell observation under high magnification and spend more time determining and classifying atypical cells when found. Cytology is a cost-effective screening method for detecting cervical cancer early. Moreover, even in low- and middle-income countries, the DL model developed in this study may be operated on premises using a compact and inexpensive WSI scanner and a laptop equipped with a GPU. In addition, with the rapid advancement of information and communication technology and the widespread use of mobile devices in low- to middle-income countries, online web applications could be one of the strategies used to engage patients in screening programs. By making the newly developed DL model available as a web application in low- and middle-income countries, it will be possible for cytologists or cytopathologists in these countries to obtain support from their counterparts worldwide. In addition, the use of the application will supplement scarce human resources. Most previous studies on cervical cytology using DL technology aimed to classify or detect atypical epithelial cells at a single-cell level, where many single cells in the image were classified or detected one by one under high magnification. However, depending on the WSI scanner model and the imaging range, a WSI will generate approximately 900 tiled images at 10× and 14,000 at 40×. Further, if a DL model evaluates all the 10× and 40× images, it will take approximately 16 times longer to process images at 40× than at 10×. In other words, developing a DL model that evaluates low-magnification tiled images will significantly reduce WSI processing time. To introduce DL models into the clinical practice of cytology, the models should be developed to enable image evaluation without difficulty. For example, our dataset contained an average of 4.7 epithelial cells per tiled image at 40× magnification. Therefore, if each tiled image contains at least one cell, approximately 65,000 (14,000 × 4.7) tiled images (epithelial cells) need to be evaluated. This requires approximately 70 times longer to finish processing a case at 40× magnification than at 10× magnification. This may be one of the reasons why the DL models developed have not been introduced into cytological clinical practice. New strategies for screening and diagnosing cervical cancer or precancerous lesions have been studied, including the use of artificial intelligence and novel biomarkers. These strategies use various data, such as age, number of sexual partners, age at first sexual intercourse, childbearing history, smoking history, and high-risk HPV genotypes. The model developed in this study exclusively uses cell imaging in cytology. However, there is enormous potential to create a multimodal ensemble model using a large-scale model, including the model we developed and other essential data besides images, for various purposes, including predicting the occurrence and recurrence risk, in addition to cervical cancer diagnosis. Furthermore, the development of multimodal models using diverse data has great potential for various applications, such as difficult treatment decision-making, determining follow-up frequency, and making decisions about the use of low-invasive surgery, which requires a wide range of operations. However, the datasets now available are limited, and the approaches are diverse. Further, given disparities in accuracy due to racial or cultural diversities, it may be necessary to construct a large-scale global dataset. Artificial intelligence has also been developed for rapid WSI diagnosis; however, validation is limited, and more testing is necessary using benchmark datasets with large computing resources, datasets, and algorithm development methodologies. Adding RandomGridShuffle to the primary augmentation contributed to improved performance. In general, an augmentation that swaps patch images, such as RandomGridShuffle, is rarely used because it significantly changes the structure of the image and is thus used in a minimal range of applications. To our knowledge, this is the first experiment in the cytology field to use RandomGridShuffle, and its application to low-magnification cytology images was successful. We used the RandAugment automated data augmentation method, which is a robust data augmentation method that applies rotations and transforms to image data while searching for suitable parameters for augmentation. However, it did not improve the performance of our model. RandAugment searches for augmentations that process existing data while preserving the meaning through geometric and color space manipulations. Nonetheless, it excludes augmentations that change the structure of the image. Therefore, because the texture of each image appears similar, low-magnification tiled images may create over-fitting. Therefore, we used RandomGridShuffle, an augmentation that generates artificial data from known data by dividing the image into *n* × *n* patch images and randomly replacing them. Further, while a larger *n* result leads to more information loss and lower accuracy, it was possible to maintain local texture as long as *n* was not too large. We assumed that RandomGridShuffle brought diversity to low-magnification tiled images in our data and suppressed over-fitting. LBC can generate uniformly distributed cells on slides and reduce cellular artifacts, and it is challenging to reduce false positives, which can occur in normal images due to overlapping cells. The use of Z-Stack and generative adversarial networks (GAN) have also been investigated as techniques to minimize the effects of cellular overlap in cytology images. However, Z-Stack has the technical problem of long scan times during WSI creation and substantial WSI data volume, and GAN requires a large amount of data, making it an arduous task. Given the difficulty of the analysis technique, improving LBC specimen preparation techniques and minimizing cellular overlap are required. # Conclusions In this study, we used semi-supervised learning to develop a DL model for screening cervical cytology specimens. By integrating Noisy Student Training, which reduces the amount of labeled data needed for training, we were able to achieve an AUC of 0.910 for the test data. Furthermore, we found the optimal threshold for confidence score and the optimal augmentation for low- magnification tiled images. The DL model we have developed is expected to be utilized in screening work for cervical cytology, as it can be used to evaluate normality and abnormality in low-magnification tiled images accurately. # Supporting information We thank Oi Yoshihiro and Katsuhide Kume for the preparation and diagnosis of cytology specimens. We also thank Masahiko Shibuya and Sanshiro Togo for their maintenance of the GPU servers. In addition, we would like to thank Mitsue Kawashima, Nao Muranaka, and Yuka Homma for their assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Hamamatsu University School of Medicine, Hamamatsu City, Shizuoka, Japan [^3]: Current address: Cancer Institute Hospital, Tokyo, Japan [^4]: Current address: Shizuoka Kohseiren Enshu Hospital, Hamamatsu City, Shizuoka, Japan

# Introduction Programmed cell death (PCD) provides protection against biotrophic microbial pathogens and is a hallmark of plant immune reactions. Execution of pathogen- triggered plant cell death, often referred to as the hypersensitive response (HR), is generally controlled by two distinct immune receptor classes, membrane- resident pattern recognition receptors (PRRs) and intracellular nucleotide- binding domain leucine-rich repeat (NLR) proteins, which are the most abundant type of plant resistance (*R*) proteins. Upon recognition of pathogen structures or pathogen-induced changes in the host cell, these receptors initiate a number of cellular events, like calcium influx, burst of reactive oxygen species (ROS), and accumulation of salicylic acid (SA) that are assumed to serve as signal molecules that eventually trigger HR. It is unclear, however, how the activation of plant immune receptors eventually translates into a cell death reaction. We study the *R* gene *Bs3* from pepper (*Capsicum annuum*) which mediates recognition of the *Xanthomonas* transcription activator like effector (TALE) protein, AvrBs3. TALEs are one class of bacterial effectors that, upon injection into host cells, translocate to the plant nucleus where they bind to an approximately 20 base pair long effector binding element (EBE) and transcriptionally activate the downstream host susceptibility (*S*) gene to promote disease. Some genotypes of otherwise susceptible plant species contain TALE-compatible EBEs upstream of transcriptionally controlled cell death executor genes. TALE-induced transcriptional activation of these EBE-containing executor alleles triggers HR and thereby stops proliferation of the biotrophic pathogen *Xanthomonas*. Accordingly, executor alleles with TALE-compatible EBEs act as plant *R* genes although the intrinsic function of executor genes might lie in other processes, such as for example developmentally-regulated cell death. Executor-type *R* genes have distinct functional modules for effector recognition (*R*-gene promoter) and orchestration of the immune response (executor R protein). This is in striking contrast to constitutively expressed NLR type R proteins, which are known to mediate both, detection of effectors and induction of an immune reaction. At the mechanistic level, there are resemblances between executor type R proteins and activated NLRs since both trigger HR. However, whether or not executors employ canonical immune pathways that are utilized by NLRs needs to be elucidated. Six executor-type *R* genes have been cloned so far. With exception of the rice executor R proteins Xa10 and Xa23 that share about 50% sequence identity, executor R proteins show neither sequence relatedness to each other nor homology to any other NLR-type or PRR-type immune receptor. Within the class of executor R proteins, Bs3 is exceptional as it is the only one that shares homology to a protein class of known function. More specifically, Bs3 shows homology to flavin-containing monooxygenases (FMOs), a class of enzymes that uses molecular oxygen (O₂) for oxygenation of metabolites. In plants, FMOs are a large and diverse group, but only a few members that have been found to function in hormone production or pathogen defense have been characterized so far. Bs3 is most closely related to YUCCA proteins, a plant-specific FMO subgroup that catalyzes the final step in tryptophan-dependent auxin (Indole-3-Acetic Acid; IAA) biosynthesis but despite its similarity to YUCCA, *Bs3* expression does not lead to increased auxin levels. The most prominent difference of Bs3 and YUCCAs is a stretch of \~70 amino acids, that is conserved across all YUCCAs but absent from Bs3 and which harbors a endoplasmic reticulum anchor sequence that can not be exchanged in between YUCCAs and Bs3 without loss of protein function. *Bs3* expression coincides with the accumulation of the immunity related metabolites salicylic acid and pipecolic acid. Yet, in contrast to other plant FMOs that have a function in immune-signaling or chemical defense, Bs3 is unique as it is the only FMO known to trigger cell death. Homology of Bs3 and the well-studied FMOs provides a unique opportunity to establish testable biochemical models of how Bs3 triggers HR. Given that Bs3 structurally resembles FMOs, we tested if Bs3-induced cell death can be explained by the well-studied FMO enzymatic cycle. Evidently, reduction of the bound FAD cofactor in FMOs by NADPH and binding of molecular oxygen results in a C4a-(hydro) peroxyflavin (C4a) intermediate, which is ready to oxygenate suitable substrates. If no metabolic substrate is available, the FMO C4a intermediate breaks down without substrate oxygenation, a process referred to as the uncoupled reaction, where reduction equivalents are released as H₂O₂. While it is unclear if H₂O₂ production by FMOs via the uncoupled reaction serves a biological function, the role of H₂O₂ as a signaling molecule in plant immune reactions is well established. We therefore speculated that unlike YUCCAs, which bind and oxygenate IPA, Bs3 does not oxygenate a metabolic substrate but instead acts exclusively as an NADPH oxidase that produces H₂O₂ to trigger HR. In this study, we show that *Bs3* expression not only triggers HR in plant cells but also inhibits proliferation of yeast cells. *In vitro* studies of recombinant Bs3 protein and *in planta* studies with the redox-sensitive roGFP2 suggest that Bs3 oxidizes NADPH and produces H₂O₂. To analyze if H₂O₂ released via a Bs3 uncoupling reaction is causal for HR, we created a Bs3 derivative that due to a mutation in its NADPH binding site has elevated NADPH oxidase activity. While, this Bs3 derivative showed elevated NADPH oxidase activity *in vitro*, its expression did not affect yeast growth and failed to trigger HR *in planta*. These observations are not consistent with a model where ROS produced by Bs3 is the metabolic trigger of Bs3-dependent HR and suggest that Bs3 converts a so far unknown substrate to trigger cell death. # Results ## Bs3 but not Bs3_S211A triggers HR *in planta* and growth arrest in yeast Previous studies on the rice executor-type R protein Xa10 from rice uncovered that Xa10 triggers cell death not only in plant cells but also in human cells. Inspired by this finding we wondered whether function of the executor R protein Bs3 is restricted to plants or if it would be functional in the budding yeast *Saccharomyces cerevisiae*, a model system of eukaryotic genetics. To study Bs3 function in yeast, we cloned the wild-type *Bs3* gene into a yeast expression vector under control of a galactose inducible promoter (*pGAL1*). Yeast transformants were grown in liquid medium, dropped onto agar plates containing either glucose (repressing) or galactose (inducing) and incubated at 28°C for two days. We found that yeast containing wild-type *Bs3* grew only on glucose- but not on galactose-containing agar plates, suggesting that *Bs3* expression inhibits proliferation of yeast cells. FMO function generally depends on transient binding of the NADPH cofactor, and mutations in codons that translate into conserved glycine residues within the NADPH binding site of FMOs have been shown to cause loss of enzyme activity. Mutational studies also uncovered specific mutation types within the NADPH binding site of FMOs that lead to derivatives producing increased amounts of H₂O₂ relative to the corresponding wild-type protein. For example, a serine to leucine change within the NADPH binding site of human FMO2 (GxGx**S**G → GxGx**L**G) or a serine to alanine change within the NADPH binding site of the *Aspergillus fumigatus* FMO SidA (GxGx**S**G → GxGx**A**G), result in a derivative with increased NADPH oxidase activity. The respective serine residue within the NADPH binding site is present in Bs3 and all Arabidopsis and pepper YUCCA proteins. We aimed to echo such mutations in the context of the pepper Bs3 protein to generate a mutant derivative with increased NADPH oxidase activity that would possibly induce a faster Bs3 HR. To do so, we mutated the triplet encoding the serine at residue 211 to an alanine codon, to create Bs3_S211A, a Bs3 derivative which would conceivably produce increased H₂O₂ levels. To test whether or not Bs3_S211A triggers HR *in planta*, we agroinfiltrated *35S* promoter driven *Bs3-GFP* or *Bs3*_*S211A**-GFP* into *N*. *benthamiana* leaves. These agroinfiltration assays showed that Bs3, but not Bs3_S211A triggered HR in *N*. *benthamiana* leaves. Immunoblot analysis showed that the Bs3-GFP and Bs3_S211A-GFP fusion proteins were equally abundant *in planta* suggesting that the S211A mutation affects Bs3 function but not protein stability. Correspondingly, we expressed *pGal* driven *Bs3*_*S211A* *in S*. *cerevisiae*. No growth defect of the strain carrying *Bs3*_*S211A* could be observed on inducing medium after two days of incubation at 28°C. Therefore, the serine to alanine mutation within the NADPH binding site resulted in a non-functional Bs3 derivative that does neither trigger HR *in planta* nor inhibit growth of yeast cells. ## Execution of Bs3 HR correlates with accumulation of H₂O₂ *in planta* To clarify whether or not occurrence of Bs3 HR would correlate with accumulation of H₂O₂ within infected tissue, we performed histochemical staining of pepper leaves using 3,3-diaminobenzidine tetrahydrochloride (DAB) as described previously. We infected the pepper variety Early Calwonder 123 (ECW123;) that contains the bacterial spot (*Bs*) plant *R* genes *Bs2* and *Bs3* with a *Xanthomonas euvesicatoria* (*Xeu*) strain that lacks the *avrBs2* and *avrBs3* genes (82-8 universally susceptible \[uns\]) or isogenic transconjugants containing either *avrBs2* or *avrBs3*. Brown staining, indicative of accumulation of H₂O₂, was observed in leaf tissue infected with *Xeu* strains delivering either AvrBs2 or AvrBs3 after 30 hours. The DAB staining shows that the execution of an HR by the NLR protein Bs2 or the executor R protein Bs3 in pepper leaves both correlate with a local increase of H₂O₂ levels. Similarly, HR induced by *Agrobacterium tumefaciens* mediated delivery (agroinfiltration) of a *35S* promoter-driven *Bs3* T-DNA in *Nicotiana benthamiana* leaves correlates with DAB staining in *N*. *benthamiana* leaves that becomes visible two days after inoculation. Our studies in pepper and *N*. *benthamiana* are compatible with a model in which Bs3 triggers HR via production of H₂O₂. However, it cannot be determined by histochemical staining whether H₂O₂ is directly produced by Bs3, or if a Bs3-dependent immune pathway eventually results in expression or activation of proteins that produce H₂O₂. ## Recombinant Bs3 and Bs3_S211A proteins bind FAD Plant immune reactions generally correlate with the release of H₂O₂. Accordingly, our observation that Bs3-triggered HR correlates with release of H₂O₂ does not clarify if the detected ROS is produced by Bs3 itself, or by a potential downstream immune signaling component that Bs3 recruits to trigger HR. To clarify if Bs3 indeed has NADPH oxidase activity, we studied recombinantly-expressed *Bs3* by *in vitro* studies. *Bs3* and *Bs3*_*S211A* were expressed in *E*. *coli* and soluble Bs3 and Bs3_S211A protein were affinity purified with yields of two milligram per liter of culture. Given that Bs3 has homology to FMOs, we expected that a bound FAD cofactor was crucial for enzymatic activity. Indeed, supplementation of the lysis buffer with FAD cofactor was necessary to obtain active Bs3 protein and to increase protein yield. In contrast to the purification protocol established for the *Arabidopsis thaliana* YUC6 protein, which uses extraction buffers containing 0.5 M sodium chloride, Bs3 purification required low sodium chloride conditions (\< 100 mM). The purified Bs3 and Bs3_S211A proteins were bright yellow and the UV-vis spectra show characteristic peaks similar to FAD. Notably, we observed a slight shift of the local maxima of Bs3_S211A (at 368 nm and 453 nm) compared to the wild type Bs3 protein (373 nm and at 448 nm;). This observation is consistent with the expected slight structural changes in Bs3_S211A as compared to wild-type Bs3. ## Bs3_S211A has higher NADPH oxidation activity and produces more H₂O₂ than Bs3 *in vitro* To compare NADPH oxidase activities of recombinant Bs3 protein and its mutant derivative Bs3_S211A, we mixed corresponding protein fractions with the cofactor NADPH and monitored NADPH consumption via spectrophotometric measurements at 340 nm. The concentration of active protein was calculated from absorbance at 450 nm and the extinction coefficient of FAD (ε = 11300 M^-1 cm^-1). At 25°C, we measured an NADPH oxidation activity of 63 nmol/mg\*min for Bs3 and 137 nmol/mg\*min for Bs3_S211A. These *in vitro* studies show that the Bs3_S211A mutant has approximately two-fold higher NADPH oxidase activity compared to that of the Bs3 wild type protein. This is in accordance with our expectation that the mutation of the conserved serine to alanine within the NADPH binding site destabilizes the C4a intermediate of Bs3_S211A and favors release of H₂O₂. ## Competitive inhibitors of YUCs inhibit Bs3_S211A to a lesser extent than the Bs3 wildtype protein We tested by *in vitro* assays if the two chemicals yucasin and methimazole (MMI), which are known competitive inhibitors of YUC function, had an influence on NADPH oxidation by recombinant Bs3 or Bs3_S211A proteins. We observed that both Bs3 and Bs3_S211A have reduced NADPH oxidase activity upon inhibitor treatment. Yucasin and MMI reduced the NADPH oxidase of Bs3 activity by 95% and 79%, respectively. These findings are in agreement with inhibitor studies on YUCs where yucasin was found to be a stronger inhibitor than MMI. Competitive inhibitors bind to the active site of the enzyme and prevent the substrate from binding. Thus, our observation that competitive inhibitors of YUCs also inhibit Bs3 suggests that the substrate binding sites of YUCs and Bs3 are structurally similar. It is worth noting that yucasin and MMI reduced the NADPH oxidase activity of the Bs3-derivative Bs3_S211A by only 39% and 16%, respectively. The observation that both competitive inhibitors had less pronounced effects on Bs3_S211A as compared to the Bs3 wild- type protein suggests that the S211A mutation affects the topology of the substrate binding site of Bs3. ## Fusion of redox sensitive roGFP2 to Bs3 has no impact on function The plant intracellular environment is likely distinct from the *in vitro* conditions in which we demonstrated NADPH oxidase activity of Bs3 resulting in H₂O₂ production. In particular, the putative metabolic substrate of Bs3 can be assumed to be present *in vivo*. Unfortunately, direct measurement of H₂O₂ and the differentiation from other ROS is not possible *in planta*. Therefore, we utilized the reduction-oxidation- sensitive GFP-derivative roGFP2 to measure increases in oxidation level, which are indicative of H₂O₂ production. Two cysteine residues in roGFP2 mediate its redox-sensitivity, which correlates with changes in its fluorescent properties ultimately allowing for ratiometric measurements in plant cells. To measure Bs3-dependent redox changes we translationally fused roGFP2 to Bs3, thereby positioning the redox-reporter in spatial proximity of Bs3. We first tested if the redox reporter roGFP2 had an impact on HR induction when translationally fused to Bs3. *Bs3-roGFP* was expressed via agroinfiltration of *35S* promoter-driven T-DNA constructs in *N*. *benthamiana* leaves (Figs and). Three days post inoculation (dpi), leaf discs were harvested for ion leakage assays which allow to monitor HR via an increase of conductivity. At four dpi, full leaves were harvested and cleared with ethanol to visualize the HR. *35S*-promoter-driven *Bs3-roGFP2* but not *roGFP2* induced visible cell death and high conductivity in *N*. *benthamiana* leaves, demonstrating that the translationally-fused roGFP2 reporter does not interfere with the Bs3-dependent HR. Overall, the observed *in planta* reactions induced by Bs3-roGFP2 fusion proteins are consistent with the previously observed phenotypes of corresponding Bs3-GFP fusion proteins. ## Phenotypes induced by Bs3 mutant derivatives are consistent in plants and yeast To analyze how distinct mutations in cofactor binding sites affect activity of Bs3 *in planta*, the redox-reporter was fused to the Bs3-derivative Bs3_S211A and the previously studied Bs3 derivatives Bs3_G209A (mutation in NADPH binding site \[Gx**G**xSG → Gx**A**xSG\]) and Bs3_G41A (mutation in FAD binding site \[Gx**G**xxG → Gx**A**xxG\];). The *Bs3* mutant derivatives were expressed via agroinfiltration of *35S* promoter-driven T-DNA constructs into *N*. *benthamiana* leaves (Figs). At 3 dpi, *Bs3*_*G209A**-roGFP2* caused a similar increase in electrical conductivity as *Bs3-roGFP2*. At four dpi *Bs3*_*G209A**-roGFP2* induced a less distinct HR phenotype compared *Bs3-roGFP2*. The Bs3 mutant derivatives *Bs3*_*G41A**-roGFP2* and *Bs3*_*S211A**-roGFP2* did not induce HR or an increase in conductivity in agroinfiltration assays. To analyze if the graduated HR intensities, induced by Bs3 mutant derivatives *in planta* are reflected in yeast, we studied the effect of *Bs3*, *Bs3*_*G41A*, *Bs3*_*S209A* and *Bs3*_*S211A* expression on yeast grown in liquid medium. To do so, yeast cultures were diluted to a starting OD₆₀₀ of 1 in inducing medium containing galactose and yeast growth was monitored in a time-course experiment over a period of three days. Strains containing the mutant derivatives Bs3_G41A and Bs3_S211A, that both do not trigger HR *in planta*, showed no yeast growth inhibition. Strains containing Bs3 and Bs3_G209A that trigger HR *in planta*, show impaired growth. In accordance with the *in planta* phenotype, expression of *Bs3*_*G209A* causes less severe inhibition of growth compared to *Bs3*. Immunoblot analysis of *Bs3* and its derivatives in yeast revealed that Bs3 and Bs3_G209A proteins were similarly abundant, while the mutant derivatives Bs3_G41A and Bs3_S211A showed somewhat lower levels. Altogether, severity of *in planta* HR and inhibition of growth in yeast correlated consistently across all tested Bs3 derivatives. ## roGFP2 reporter assays indicate *in planta* oxidase activity for Bs3 and Bs3_S211A To analyze changes of the intracellular oxidation state caused by Bs3 and its mutant derivatives, the different *roGFP2* fusion constructs with full, reduced and no capacity to trigger HR (*Bs3-roGFP2/ Bs3*_*G209A**-roGFP2/ Bs3*_*G41A**-roGFP2 and Bs3*_*S211A**-roGFP2*) and *roGFP2* were expressed in *N*. *benthamiana* leaves (Figs). Thirty hours post inoculation (hpi) confocal laser scanning microscopy (CLSM) was used to determine the fluorescence intensity upon excitation at 405 and 488 nm. Oxidation of roGFP2 causes an increase of fluorescence at 405 nm excitation and a corresponding decrease at 488 nm excitation. Therefore, a higher relative fluorescence intensity (ratio 405nm/ 488nm) is indicative for a higher oxidation state. We found that the HR-inducing proteins Bs3-roGFP2 and Bs3_G209A-roGFP2 had about two-fold higher RFI values than the roGFP control. This observation suggests that Bs3 and Bs3_G209A have indeed NADPH oxidase activity *in planta*, which would be consistent with a model where ROS produced by Bs3 triggers HR. Bs3_G41A-roGFP2, which does not trigger HR, had similar RFI values as the roGFP2 control, is still in agreement with our proposed model. However, the NADPH-binding site mutant Bs3_S211A-roGFP2, which does not trigger HR *in planta*, had higher RFI values than the HR inducing Bs3-roGFP2 fusion protein. Indeed, the elevated oxidation state of roGFP2 induced by Bs3_S211A *in planta* is consistent with our *in vitro* studies that showed increased NADPH oxidase activity in Bs3_S211A as compared to the wildtype Bs3 protein (Figs). In summary, we found that the mutant derivative Bs3_S211A does not trigger HR, but induced a similar increase of the intracellular oxidation compared to the Bs3 wild-type protein, which suggests that the putative release of H₂O₂ by Bs3 is not sufficient to trigger the Bs3-dependent HR. # Discussion ## Competitive inhibitors of YUCCA proteins inhibit Bs3 Pepper Bs3 is one out of six currently known executor R proteins. Bs3 is exceptional since it is the only executor that shares sequence homology to proteins of known function. Due to the sequence homology of Bs3 to YUCCA proteins, which catalyze conversion of IPA to IAA, it would seem plausible that the enzymatic capabilities of Bs3 and YUCCAs are similar. The 70 amino acid long stretch that is present across all YUCCAs but absent from Bs3 might contain domains that mediate metabolic feedback regulation of YUCCAs. For example, one could envision that in absence of IPA the production of H₂O₂ by YUCCAs would stop enzymatic activity YUCCAs before H₂O₂ accumulates to cytotoxic levels. Accordingly, the lack of this 70aa stretch in Bs3 will possibly cause accumulation of Bs3-dependent metabolites to cytotoxic levels. Previously, we showed that Bs3, in contrast to the related YUCCA proteins, does not synthesize IAA and ultimately that Bs3-dependent HR does not involve changes in IAA levels. While Bs3 and YUCCAs seem to differ in their metabolic product, it would seem possible that both enzymes use IPA as a metabolic substrate. Unfortunately, IPA is a rather unstable metabolite that spontaneously converts into IAA *in vitro*. In consequence, quantification of IPA consumption by recombinant Bs3 protein *in vitro* proves to be rather challenging. However, the previously identified competitive inhibitors of YUCCA proteins that reduce enzymatic activity by binding to the substrate binding pocket provide an alternative means to compare metabolite binding sites of Bs3 and YUCCA proteins. Yucasin and methimazole, which are both competitive inhibitors of YUCCA proteins, reduced NADPH oxidase activity of both Bs3 and Bs3_S211A thereby suggesting that Bs3 contains a substrate binding site that is structurally related to that of YUCCAs. The less severe inhibition of Bs3_S211A compared to Bs3 by yucasin and methimazole is in line with the notion that the S to A mutation does change the substrate-binding site but still allows NADPH oxidation. Reduced affinity of Bs3_S211A to a metabolic substrate would favour C4a break down without substrate oxygenation and would explain the increased NADPH oxidase activity of this particular Bs3 mutant derivative. ## H₂O₂ production by Bs3 is not sufficient to trigger HR Based on the fact that YUCCAs and other FMOs can release substantial amounts of H₂O₂ by oxidizing NADPH without substrate conversion in a process referred to as “uncoupling”, we hypothesized that Bs3 would not bind a substrate but induce HR by H₂O₂ production. We envisioned two possible ways in which Bs3 could trigger HR by the production of H₂O₂. Firstly, Bs3 might produce excessive amounts of H₂O₂ that cause oxidative damage to the plant cell, or alternatively, Bs3 produces H₂O₂ as a signaling molecule that activates a cascade leading to plant defense and HR. Indeed, our *in vitro* studies revealed that recombinant Bs3 protein produces substantial amounts of H₂O₂, and our analysis of the intracellular oxidation state by roGFP2-based reporter assays suggests that Bs3 releases H₂O₂ not only *in vitro* but also *in vivo*. To reinforce the possible causal link between direct production of H₂O₂ by Bs3 and HR we studied the function of Bs3 mutant derivatives. For example, the Bs3 mutant derivative Bs3_G41A, which based on roGFP2 reporter assays, does not produce H₂O₂ did also not trigger HR *in planta*, being consistent with a model in which H₂O₂ produced by Bs3 triggers HR. To further support this hypothesis, we replicated a mutation in the fungal FMO SidA that was previously shown to cause increased NADPH oxidase activity. Indeed, *in vitro* and *in vivo* studies of the Bs3_S211A mutant showed increased NADPH oxidase activity and H₂O₂ production (Figs). Yet, while Bs3_S211A has higher oxidase activity than the wild-type Bs3 protein, it does not trigger HR *in planta* (Figs). This implies that release of H₂O₂ by Bs3 is not sufficient to trigger HR. We therefore postulate, that Bs3 converts a metabolite which is possibly structurally related to IPA into a yet to be identified product that triggers HR. Whether or not this metabolic product of Bs3 is simply cytotoxic or if the metabolic product acts as a defense signaling molecule remains to be seen. ## Differentiation of direct and indirect ROS production during Bs3 HR We studied H₂O₂ synthesis during Bs3 HR *in planta*. The results of our measurements via the roGFP2 redox reporter and by DAB staining strongly suggest that the roGFP2 reporter and DAB staining detect distinct H₂O₂ pools. While roGFP2 based measurements revealed high intracellular oxidation states for both, Bs3 and Bs3_S211A, only expression of Bs3 but not Bs3_S211A produced brown precipitates in DAB staining being indicative for accumulation of ROS. The direct correlation of HR with DAB staining but not the roGFP2 reporter readout indicates that the H₂O₂ detected via DAB staining was not produced directly by Bs3, but originated from other sources like membrane-bound NADPH oxidases or apoplastic peroxidases that are activated in the course of Bs3 induced HR. This is supported by the finding that the intensity of DAB staining is independent of Bs3 protein amount. The DAB staining intensity increased until the onset of cell death, even though a decrease of Bs3 protein levels can be observed in *N*. *benthamiana* at 48 hpi. In contrast to the DAB staining, ratiometric measurements are independent of expression levels and our roGFP2 based measurements were able to detect changes in the oxidation state as early as at the onset of Bs3-roGFP2 protein accumulation. RoGFP2 studies were carried out at 30 hpi, a timepoint at which no DAB staining was visible. The higher oxidation values measured for Bs3-roGFP2 compared to roGFP2 suggests that low amounts of H₂O₂ are produced directly by Bs3. The higher oxidation values obtained for Bs3_S211A-roGFP2 show that these low amounts of H₂O2 –presumably produced via the uncoupled reaction–are not sufficient to trigger cell death. However, while H₂O₂ produced by a Bs3 uncoupling reaction is not sufficient to trigger cell death it cannot be excluded that local changes in the oxidation state contribute to a signaling cascade that results in execution of HR. ## Bs3 derivatives that trigger HR *in planta* consistently inhibit proliferation in yeast cells We found that expression of *Bs3* limits proliferation of yeast cells, which are particularly amenable to genetic screens and can be used to dissect the Bs3-triggered cell death reaction in the future. The functional analysis of Bs3 and derivatives thereof in the plant- and yeast systems were consistent in all cases. For example, Bs3 and the NADPH-mutant derivative Bs3_G209A triggered HR *in planta* and also inhibited proliferation of yeast cells. Bs3_G209A induced a somewhat weaker HR than the wildtype Bs3 protein *in planta* and the same trend was observed in the yeast growth assay where expression of Bs3_G209A or Bs3 induced moderate and strong inhibition of yeast growth, respectively. Moreover, the Bs3 derivatives Bs3_G41A and Bs3_S211A, did not trigger HR *in planta* and also failed to inhibit proliferation of yeast cells. Comparison of expression levels of Bs3 and derivatives in yeast seem to indicate that cell-death inducing Bs3 derivatives show generally higher expression levels as the Bs3 derivatives that do not trigger growth arrest. However, *in planta* expression levels of the cell inducing Bs3 wt protein were equal or lower than expression levels of Bs3_S211A that does not trigger HR *in planta*. In summary these observations demonstrate that capability of the studied Bs3 derivatives to induce cell death can not be explained by variation in protein expression levels but resembles the biochemical properties of the studied protein variants. The observed consistency of the Bs3-dependent phenotypes in yeast and plant cells possibly suggest that both phenotypes have a common molecular basis. Therefore, it stands to reason that the observed phenotypes in yeast and plants are caused by the same metabolite that Bs3 presumably produces in plant and yeast cells. ## Recombinant Bs3 protein as a tool to uncover the metabolic basis of the Bs3-triggered HR Previously conducted biochemical studies uncovered that recombinant Arabidopsis YUCCA6 protein uses NADPH and oxygen to convert IPA to IAA. Here, we established a protocol for purification of enzymatically active Bs3 protein that now enables us to compare enzymatic activity of Bs3 and the YUCCA proteins by *in vitro* assays. Given their high structural relatedness, it seems likely that Bs3 and YUCCA proteins are also similar with respect to their enzymatic features. Indeed, our studies revealed that Bs3, just like YUCCA6, has NADPH oxidase activity. Moreover, inhibitor studies suggest that YUCCA proteins and Bs3 have the same or at least structurally related metabolic substrates. We envision future studies where incubation of metabolic candidate substrates, with enzymatically active Bs3 protein could provide the possibility to identify metabolic products of Bs3 by mass spectrometry. In summary, the established protocol for purification of recombinant enzymatically active Bs3 protein is the first step towards exploitation of its biochemical properties and holds the key to uncover the basis of Bs3-triggered immune reactions by metabolic studies. # Material and methods ## Plants and growth conditions *N*. *benthamiana* and *Capsicum annuum* (cultivar ECW123) plants were grown at 20–24°C at 35–60% humidity with a light intensity of 12,3 klx and 16 hours light/ 8 hours dark cycle. Four to six weeks old plants were used for experiments. ## Plasmid construction For *in planta* expression, the *Bs3* CDS was assembled with a *35S* promoter and *eGFP* into the *LIIa* expression vector via Golden Gate cloning. The serine to alanine mutation was introduced by PCR (PrimerFW: `P-GCCGGGATCGATATCTCACTTG` PrimerREV: `ATTGCCACAGCCAACCGC`). For bacterial expression, the Bs3 coding sequence was cloned into the pET-53-DEST (Novagen) expression vector via Gateway cloning. Protein solubility could be dramatically improved by deletion of the nucleotides encoding for the recombination site and the codons accounting for the two N-terminal methionines of Bs3. Deletion of these nucleotides was done by PCR mutagenesis (PrimerFW: `P-GTGATGGTGGTGGTGATGTG` and PrimerREV: `AATCAGAATTGCTTTAATTCTTGTT` CAC). For expression in yeast, the Bs3 CDS was cloned into the pYES-DEST-52 vector via Gateway cloning without further modification. ## DAB staining Leaves are vacuum infiltrated with DAB staining solution (10 mM Sodium phosphate, 1 mg/ml Diaminobenzidine-tetrahydrochloride, 0.1% Tween-20, pH = 7.2), incubated at room temperature with gentle shaking for at least 5 hours and subsequently de-stained with 80% ethanol at 60°C. ## *Xanthomonas* and *Agrobacterium* infiltration *Xanthomonas* (82–8 uns\*) carrying the pDSK602 vector with either *AvrBs3* or *AvrBs2* were grown at 28°C in NYG medium (5 g/L peptone, 3 g/L yeast extract, 20 g/L glycerol) containing Rifampicin and Spectinomycin at a final concentration of 100 μg/ml for 1 day. Cultures were pelleted and re-suspended in water to OD₆₀₀ = 0.4. Leaves of *C*. *annuum* were infiltrated with a blunt end syringe from the abaxial side. *Agrobacterium tumefaciens* (GV3101) carrying the respective binary plasmids were grown over night at 28°C in YEB medium (5 g/L beef extract, 1 g/L yeast extract, 5 g/L peptone, 5 g/L sucrose, and 0.5 g/L mM MgSO₄, pH 7.2) containing Rifampicin and Spectinomycin at a final concentration of 100 μg/ml. Cultures were pelleted and resuspended in water to OD₆₀₀ = 0.4. Leaves of *N*. *benthamiana* were infiltrated with a blunt end syringe from the abaxial side. ## Protein expression *E*. *coli* Rosetta (Novagen) were transformed with *pET-53-DEST_Bs3* or derivatives, plated on LB Agar containing Ampicillin (100 μg/ml) and Chloramphenicol (15 μg/ml) and incubated at 37°C over night. Several colonies were pooled to inoculate LB medium (10 g/L NaCl, 5 g/L yeast extract, 10 g/L tryptone) supplemented with Ampicillin (100 μg/ml) and Chloramphenicol (15 μg/ml). After over night incubation at 37°C and 180 rpm, this starter culture was used to inoculate 2 L TB Medium (24 g/L yeast extract, 20 g/L tryptone, 4 ml/L glycerol, 0.072 M K₂HPO₄, 0.017 M KH₂PO₄) supplemented with Ampicillin (100 μg/ml) at a starting OD₆₀₀ of 0.05. The culture was incubated at 37°C with shaking at 120 rpm until it reached an OD₆₀₀ of 1. Cultures were then cooled down for 15 min in ice water and protein expression was induced with a final concentration of 1 mM Isopropyl 1-thio-β-D-galactopyranoside (IPTG). The cells were incubated for another 2,5 h at 18°C with shaking and followed by centrifugation (4500 x g, 30 min, 4°C). Pellets were stored at -20°C until further use. ## Protein purification The bacterial pellet was re-suspended in lysis buffer (50 mM Potassium phosphate, 10% glycerol, 30 mM Imidazole, 1% Tween, protease inhibitor, 1 μM FAD, 1 mM DTT, pH = 8) using 5 ml of buffer per gram of pellet. 30 ml cell suspension were sonicated for 5 min (5s on/10s off, 60% amplitude) with a sonicator (EpiShear, Active Motif) equipped with a ¼” microtip probe. The lysate was centrifuged (16000 x g, 4°C) for 30 min to pellet cell debris. An ÄKTA Pure 25 FPLC system, equipped with a 5 ml HisTrapFF Crude Column (GE Healthcare), was used for affinity purification. After column equilibration with 10 column volumes (CV) of wash buffer (50 mM Potassium phosphate, 10% Glycerol, 30 mM Imidazole, pH = 8), the supernatant was loaded onto the column and washed with 20 CV wash buffer. The protein was eluted with 2 CV elution buffer (50 mM Potassium phosphate, 10% Glycerol, 500 mM Imidazole, pH = 8). Imidazole was removed by dialysis and the protein was frozen in liquid nitrogen and stored at - 80°C until further use. ## NADPH oxidation Spectroscopic assays were carried out in quartz cuvettes with a UV-900 UV-vis spectrometer (Shimadzu) equipped with a temperature controlled cell holder (TCC-100) set to 25°C. Samples were diluted with 50 mM potassium phosphate buffer (pH = 8) containing NADPH. Exact NADPH concentrations were calculated from its absorption and the extinction coefficient at 340 nm (ε₃₄₀ = 6220 M^-1 cm^-1). ## Detection of H₂O₂ with HyPerBlu Protein was mixed with NADPH solution (100 μM) and incubated at RT for 15 min. Per replicate, 5 μl of this solution were transferred to a white 385 well plate, mixed with 5 μl of HyPerBlu solution (Lumigen) and incubated in darkness for 15 min at RT. Subsequently, luminescence was measured using a Berthold Tristar LB 941 plate reader. H₂O₂ concentrations were calculated using a standard curve prepared with known concentrations of H₂O₂. ## Redox reporter microscopy Constructs in which *roGFP2* was fused to *Bs3* and its derivatives were transiently expressed in four to six week old *N*. *benthamiana* plants via *Agrobacterium* mediated transient transformation. 30 hpi, images were acquired using a Leica TCS SP8 confocal microscope by successive excitation at 405 nm and 488 nm and emission at 498 to 548nm. Images of nuclei were taken using a 63x water immersed objective and 10x digital magnification. Argon laser intensity was adjusted in a way so that pixels were close to saturation in samples with highest *roGFP2* expression. UV laser intensity was adjusted allowing imaging of samples with lowest *roGFP2* expression. Fiji was used to crop the surrounding of the nuclei and to calculate mean pixel intensity of the fluorescent area. ## Ion leakage measurements Ion leakage measurements were conducted using the CM100-2 conductivity meter (Reid & Associates). Each well was filled with 1 ml ultrapure water. Leaf discs (Ø 4 mm) were harvested three days post infiltration. One disc was added per well and incubated at RT. Ion leakage was measured after 20 hours of incubation. # Supporting information We thank D. Holmes for helpful comments on earlier versions of the article. [^1]: The authors have declared that no competing interests exist.

# Introduction Nephrolithiasis is a common disease with a high worldwide prevalence that ranges from 7 to 13% in North America, 5–9% in Europe, and 1–5% in Asia. Geography, diet, fluid intake, genetics, climate, age, and occupation are important factors that can affect the incidence of this disease. Kidney stone management is expensive and surgery is often the gold standard treatment. The European Association of Urology Guidelines define External Shock Wave Lithotripsy (ESWL) or Endourology as the first-choice treatments for 1–2 cm kidney stones, where the term ‘Endourology’ encompasses all Percutaneous Nephrolithotomy (PCNL) and Ureterorenoscopy (URS) interventions. For stones bigger than 2 cm the gold standard is PCNL treatment. Nowadays, the panorama of PCNL instrumentation is extremely wide and includes Micro-PCNL with 4,8 Ch, UltraMini-PCNL (UMP) with 11–13 Ch, Mini-PCNL with 14–20 Ch, Regular-PCNL with \> 20 Ch access diameters. The miniaturization of instruments has impacted indications, as smaller tracts theoretically reduce complications such as blood loss and renal parenchyma injury. Retrograde Intra Renal Surgery (RIRS) is included among the first line treatments for kidney stones between 1 and 2 cm, but could be proposed as a viable alternative therapy to PCNL for stones larger than 2 cm in special groups of high-risk patients (e.g. those with bleeding disorders, obesity, renal congenital abnormalities, or solitary kidney). The guidelines consider each of these techniques but don’t provide specific indications for the different sized PCNL techniques and do not univocally define the best treatment option for kidney stones, especially for calculi between 1 and 2 cm. ## Purpose The aim of our study is to highlight surgeons’ preferences among RIRS, Regular, Mini-, UltraMini- and Micro-PCNL in stone treatment in a real-life setting, based on a Survey approved by the European Association of Urology Section of Urolithiasis (EULIS). Moreover, we compared the effectiveness and safety of these techniques in stone treatment, with a particular interest on the grey zone of 1–2 cm stones. # Materials and methods We performed a Survey approved by the EULIS board in Cape Town on December 2014 regarding the employment of RIRS and regular and small sized PCNL among physicians who attended the Copenhagen EULIS meeting in 2013. The project was carried out in three steps using *Survey Monkey* and took place from February 2015 (collection of the first and second-step results) to May 2015 (third step). This article is focused on the results from the third step. An introductory email was sent to all participants, describing the study and providing the link to the online questionnaire. Participation was voluntary. The first two steps aimed to delineate the departments’ profiles in terms of instrument availability and performed techniques, and to collect experts’ personal opinions on RIRS and regular and small size PCNL. In the *third step* responders were asked to report the number of procedures performed during the previous year and to share information regarding the last 5 cases treated in their centres with each of the techniques under study for which they had at least one-month of follow-up. They were asked to report data concerning different pre-, intra- and post-operative items. Twelve of these items were common for all procedures: 5 were related to patient, stone or operation characteristics (i.e. age, sex, comorbidities, stone size and exit strategy), 2 were considered *effectiveness* outcomes (i.e. *one month stone- free status* and *need for retreatment)* and 5 were *safety* outcomes (i.e. *operation time*, *Hb drop*, *need for transfusions*, *hospital stay duration* and *complications* according to the Clavien/Dindo scale). Stone size was defined as the maximum stone diameter. In cases of multiple stones the cumulative maximum stone diameter was used. Stones were categorized in three groups according to their size (i.e. \< 1 cm, 1–2 cm, \> 2 cm). Operation time was dichotomized by surgeries lasting ≤ 60 minutes and those lasting longer. Stone free status was defined as the total absence of residual fragments at the 1-month follow-up evaluation. The imaging modality used to determine stone presence/absence was dependent on each centre’s local protocol. The questionnaires we sent to collect data on RIRS and PCNL procedures are available in the supporting information files ( and Figs, respectively). Procedures were stratified according to stone size and each centre’s surgical volume (\>/\<50 PCNL procedures performed during the previous year). We explored whether or not effectiveness and safety outcomes differed in relation to stone size, using the Chi-squared test for qualitative outcomes and linear regression analysis for quantitative outcomes, previously log transformed if appropriate (i.e. hospital stay and haemoglobin drop). Effectiveness and safety outcomes for the different techniques were compared using multiple regression models (logistic for categorical outcomes and linear for quantitative outcomes) after correction for stone size, age and gender. The same analyses were also performed on a subgroup of 1–2 cm stones in order to analyse a more homogeneous group in terms of number of procedures. Statistical analyses were performed with the Stata 10 package considering statistical significance for P values \< 0.05. This study protocol was reviewed and approved by the ethics committee of the University of Milan and all clinical investigations were conducted according to the principles expressed in the Declaration of Helsinki. Every patient (parents if minor or guardians if incompetent) signed a written informed consent for his/her clinical information to be used and shared for research purposes. # Results We sent the first and second step questionnaires to 360 people and obtained a 24% response rate (88 responders). Seventy-eight physicians agreed to receive the third step questionnaire and 38% of these (30 responders) shared data from their recent procedures using each technique. Not all responders shared 5 cases for each technique. We collected a total of 420 procedures: 140 RIRS from 28 centres, 141 Regular PCNL (\>20 Ch) from 27 centres, 67 Mini-PCNL (14–20 Ch) from 15 centres, 28 UMP (11–13 Ch) from 7 centres and 44 Micro-PCNL (4,8–8 Ch) from 8 centres. The 420 procedures were performed on an equal number of patients. The sample population had median age of 51 years (2–89) with a slight prevalence of males (57.86%). The techniques performed by the single centres were stratified according to their surgical volume as shown in. Responders reported statistically significant differences in the chosen technique in relation to stone size. As shown in, the effectiveness and safety outcomes were also influenced by stone size, independently of the technique used. The results of the multiple regression analysis assessing whether or not effectiveness outcomes are influenced by the chosen technique (on all procedures, N = 420 and on procedures for 1–2 cm stones, N = 172) are reported in. The stone free rate (SFR) was significantly lower in patients treated with Micro-PCNL compared to Regular PCNL, when considering procedures performed for all stone sizes (OR 0.36; P = 0.019). The distribution of safety outcomes in relation to the different techniques and the results of the multiple regression analysis exploring how the outcomes were influenced by the chosen technique (on all procedures N = 420 and on procedures for 1–2 cm stones N = 172) are reported in. All considered techniques (i.e. Mini-PCNL, UMP, Micro-PCNL and RIRS) presented a lower complication rate than Regular-PCNL, with Mini-PCNL being the most protective technique compared to Regular-PCNL. The total number of transfusions reported was 17: fourteen (82.35%) in patients undergoing Regular-PCNL (3 of which for 1–2 cm stones); one (5.88%) in a patient undergoing Mini-PCNL, one (5.88%) in a patient undergoing UMP and one (5.88%) in a patient undergoing RIRS. # Discussion This article is based on the first survey within the EULIS group. The main goal in stone treatment is to achieve the highest effectiveness with the lowest morbidity. Over recent years, many authors have made comparisons among the different available techniques but still no clear and univocal indications have been delineated, and a wide heterogeneity exists in the application of the single procedures. While only a randomized controlled trial (RCT) could completely resolve this issue, our study outlines the real-life clinical practice among EULIS members. ## Techniques application Our data show that the techniques are employed differently according to stone size. In particular, Regular-PCNL is the most used technique for stones \> 2 cm, followed by Mini-PCNL and RIRS. For stones \< 1 cm, RIRS is widely applied. In cases of 1–2 cm stones the indication is less univocal with a slight prevalence of PCNL (evenly distributed among the different tract size options) over RIRS. The distribution of the employed techniques is not uniform among the respondent centres. In particular, while RIRS and Regular-PCNL are performed by almost all of the responders, Mini-PCNL is practiced by only half of them, and UMP and Micro-PCNL, as expected, are the least performed techniques. The choice of technique also seems to be related to the centres’ surgical volume. Specifically, centres performing more than 50 PCNL/year more often apply miniaturized PCNL techniques (including Mini-PCNL, UMP and Micro-PCNL) than centres performing less than 50 PCNL/year. This could be explained not only by the more robust surgical experience but also by the availability of mini invasive instruments and the related costs. ## Effectiveness ### Background S. Mishra et al. demonstrated that Mini- and Regular-PCNL can achieve equally effective results in terms of SFR for 1–2 cm stones (96% and 100% respectively, P value = 0.49). K. Wilhelm et al. reported comparable results for UMP and RIRS in 10–35 mm stones (SFR 92% and 96% respectively, P value = 0.561). Other studies have obtained comparable stone clearance rates with micro-PCNL and RIRS for stones up to 1,5 cm (97.1% and 94.1% respectively, P value = 1.0 by R.B. Sabnis et al.; 83.3% and 86.7% respectively, by Kandemir et al.). A review by De et al. reported that minimally invasive percutaneous procedures, including Mini- and Micro-PCNL, could provide higher stone-free rates than RIRS. Nevertheless, many studies in the literature report SFRs over 90% for RIRS performed for stones larger than 2 cm, and thus consider it an attractive alternative to PCNL. Furthermore, Micro-PCNL has been described as an adequate treatment method for peculiar indications such as solitary renal stones with a volume \<1 000 mm³ and low density (HU). Therefore, great discrepancies exist in the reported outcomes of the available procedures. Moreover, the imaging modality and the timing of stone clearance assessment are variable in the cited studies, with some authors using X-ray KUB at 1 or 3 months, others using endoscopic inspection and immediate postoperative ultrasound or low-dose CT 4–8 weeks after the procedure and still others using non-contrast spiral CT at 3 months. This renders the comparisons between techniques even less reliable. The lack of consensus on the best follow-up imaging modality was probably reflected in the series we analyzed, in which different local protocols were adopted. ### Overall analysis In our series, considering all performed procedures, no significant differences were observed in terms of SFR between Regular-PCNL and Mini-PCNL, UMP and RIRS, even if Mini-PCNL seemed to be more effective than UMP and RIRS according to the higher odds ratio. Moreover, Mini-PCNL was the only technique studied that resulted in a significantly lower need for second procedures. Micro-PCNL, rather, presented a significantly lower SFR than Regular-PCNL; nevertheless, the retreatment rate after this technique didn’t differ in a statistically significant way from the other techniques, suggesting that the clinical outcomes may be satisfactory enough to not require a second operation. ### 1–2 cm stones group In the 1–2 cm stone group, no statistically significant differences were observed among any of the techniques with respect to Regular-PCNL. However, the results were similar to those described for the total group, with the best outcomes being observed for Mini-PCNL and the lowest SFR for Micro-PCNL. ## Safety ### Background As already stated in the literature, reductions in both haemoglobin loss and hospital stay are obtained by reducing the invasiveness of the procedure; S. Mishra et al. demonstrated that Mini-PCNL for 1–2 cm stones is limited by longer operative times than standard-PCNL but is characterized by significantly less bleeding, shorter hospital stays and a similar safety profile. K. Wilhelm et al., in a comparison of UMP and RIRS for 10–35 mm stones, demonstrated that operating times and hospital stays were significantly longer in the UMP group. No patient in their study required a blood transfusion. R.B. Sabnis et al., after randomizing patients to either RIRS or Micro-PCNL for stones \< 1.5 cm, demonstrated that Micro-PCNL was associated with greater haemoglobin loss, increased pain and more analgesic requirements, while RIRS was associated with a higher requirement for JJ stenting. A. Kandemir et al., comparing the same techniques for lower pole stones up to 1.5 cm, found similar complication rates but longer scopy times and hospital stays associated with Micro-PCNL. A.Yamaguchi et al., in their study of 5537 patients who underwent PCNL, observed that bleeding and transfusions tended to increase with an increasing size of the access sheath (18 Ch to 34 Ch). De et al. found that PCNL is related to higher complication rates, blood loss and longer lengths of stay than RIRS, with no differences in surgical time and secondary procedures. Minimally invasive percutaneous procedures, including Mini- and Micro-PCNL, were also found to be related to added morbidity and longer hospital stays than RIRS for stones \<2 cm. ### Overall analysis In the overall group, our results confirmed that each of the small sized PCNL techniques and RIRS produced less haemoglobin loss and shorter hospital stays than Regular-PCNL. With regard to complications, Mini-PCNL was revealed to be the safest among the PCNL techniques. UMP and Micro-PCNL also appeared to be safe procedures, but didn’t result in statistically significant differences, probably due to the smaller number of patients treated with these techniques in our series. RIRS also, and in line with the literature, was related to significantly lower complication rates than Regular-PCNL. Operative times were comparable to those of Regular-PCNL for all techniques except Micro-PCNL, which appeared to be the most time-consuming procedure. ### 1–2 cm stone group Safety results within the 1–2 cm stone group were similar to those of the overall group. Haemoglobin drop and hospital stay progressively decreased with the reduction of invasiveness. For all techniques, high-grade (Clavien ≥2) complication rates were lower than they were for Regular-PCNL. Micro-PCNL appeared to be the most time-consuming technique in this group as well. ## Limitations One major limitation of this study, intrinsic to the nature of surveys, is the low response rate. This was probably due to the lack of ready availability in all centres to the quested data. Another limitation of the present study is represented by its multicentre retrospective design, which may have led to a misestimation of some of the outcomes. In particular, the stone free status was assessed with different imaging modalities in accordance with the individual centres’ local protocols, which were not investigated in our survey. This reduces the comparability of the results. Nevertheless, the retreatment rate could be considered as a suitable surrogate for the effectiveness of the procedures. One further limitation is the small number of miniaturized PCNL procedures reported. This is due, as explained above, to the infrequent application of these techniques in lower surgical volume centres. We decided to keep the questionnaire as concise and short as possible to maximize the possibility of obtaining responses, only recording fundamental variables. This may have caused us to overlook some characteristics such as stone location and number of stones that could have been useful for the comparison among techniques. # Conclusions Stone size seems to be an important factor driving treatment choice. Miniaturized PCNL techniques are widely applied in the treatment of 1–2 cm stones, in particular in higher surgical volume centres. Mini-PCNL and RIRS are gaining in popularity also in the treatment of stones \> 2 cm. The miniaturization of PCNL appears to result in a safer procedure without compromising effectiveness. In particular Mini-PCNL, with accesses ranging from 14 to 20 Ch, seems to be a good compromise, being the most effective and the safest procedure among the PCNL techniques both in the overall group as well as the 1–2 cm stones group. Micro-PCNL appears to be the most time-consuming technique and probably requires highly selected indications. RIRS continues to be a procedure characterized by satisfactory stone-free rates and low complication rates. Nevertheless, considering the limitations of our survey and its retrospective and multicentre design, we cannot draw firm conclusions on the best technique to treat kidney stones. Future RCTs would be the best option for shedding more light on the subject. # Supporting information The authors wish to thank the following doctors and centres for their contribution: K.Petkova—Military Medical Academy–Sofia—Bulgaria G.Zeng—The First Affiliated Hospital of Guangzhou Medical University—Guangzhou—China A.Hoznek—CHU Henri Mondor—Créteil—France O.Traxer—Hopital Tenon—Paris- France A.Papatsoris—Sismanoglio General Hospital—Athens—Greece A.Gabani—Gabani Kidney Hospital—Surat—India I.Aghaways—Sulaimani Surgical Teaching Hospital—Sulaimani- Iraq A.Saita—Vittorio Emanuele Hospital—Catania—Italy A.Trinchieri—Alessandro Manzoni Hospital—Lecco—Italy G.Giusti—Ville Turro, San Raffaele Hospital—Milano—Italy E.Montanari—San Paolo Teaching Hospital—Milano—Italy F.Germinale—Ospedale Giovanni Borea—Sanremo—Italy A.Bosio—Città della Salute e della Scienza—Torino—Italy C.M.Scoffone—Cottolengo Hospital—Torino—Italy E.Cicerello—Ca’ Foncello Hospital—Treviso—Italy G.Zanetti and I.K.Goumas—Ospedale Civile di Vimercate—Vimercate—Italy M.Morozumi—Saitama Medical University—Kawagoe/Kamoda—Japan N.Ishito—Kurashiki Medical Centre—Kurashiki—Japan A.Gaizauskas—Republic Vilnius University Hospital—Vilnius—Lithuania M.Lezrek—Military Hospital Moulay Ismail Meknes/Voluntary work—Meknes—Morocco J.De la Rosette—AMC University Hospital—Amsterdam—The Netherlands P.Luque and M.Costa—Hospital Clínic—Barcelona—Spain G.Ibarluzea—Clinica IMQ Zorrotzaurre—Bilbao—Spain J.A. Galan—Hospital Universitario del Vinalopó –Elche—Spain F.Ramón de Fata—Universitary Hospital Getafe—Madrid—Spain D.Pérez-Fentes—C.H.U. Santiago de Compostela—Santiago de Compostela—Spain M.Cepeda—Hospital Universitario Río Hortega—Valladolid—Spain K.Sarica—Dr.Lutfi KIRDAR Research and Training Hospital—Istanbul—Turkey A.Bourdoumis—Torbay Hospital—South Devon Healthcare NHS Trust—Torquay—UK S.Mackie—Eastbourne District General Hospital—Eastbourne—London—UK [^1]: The authors have declared that no competing interests exist.

# Introduction Oral pre-exposure prophylaxis (PrEP) with daily tenofovir disoproxil fumarate (TDF) and emtricitabine (FTC) by HIV-negative people can prevent contracting HIV and is a key intervention that can support reducing HIV infections by 90% by 2030. Initially, in 2012, the World Health Organization (WHO) recommended PrEP for special populations such as sero-discordant couples (SDCs), men who have sex with men (MSM), and transgender women who have sex with men. However, in 2015, WHO expanded their recommendation to include all population groups at substantial risk of HIV infection as part of a comprehensive package of HIV prevention interventions. The UNAIDS fast track programme identified Zimbabwe as one of the countries that contributes to 89% of new HIV infections globally. These countries are encouraged to take up new innovations such as PrEP in order to meet their 95-95-95 targets by 2030 and reduce the number of new infections. Zimbabwe has a generalized epidemic and would benefit from offering comprehensive HIV prevention and treatment services to the general population. Other evidence exists on the factors related to uptake and continuation of PrEP in sub-Saharan Africa; however, this evidence comes almost exclusively from participants in clinical trials or other population-targeted programs. PrEP has been described by clients as an HIV prevention tool that allows the user to control their life with minimal cooperation from their partner, especially when condom negotiation is difficult or as additional protection. Perceived risk is suggested to be key for uptake and support for adherence to PrEP by partners and has been highlighted as critical to continued use. In Kenya, Tanzania, and South Africa, adherence was also facilitated by establishing a routine for taking PrEP. However, stigma was also expressed as a barrier to taking PrEP. In Zimbabwe, PrEP was included in the 2016 National Guidelines for Antiretroviral Therapy for the Prevention and Treatment of HIV. In Zimbabwe, an estimated 38,000 new adult HIV infections occurred in 2018, up from 32,000 in 2016. Young women aged 15–24 years account for 9,000 new infections compared to 4,200 young men in the same age group. Women account for 60% of people living with HIV (PLHIV). Infection rates are higher among sub-populations, such as adolescent girls and young women (AGYW), SDCs, and MSM. PrEP was first introduced in Zimbabwe in private sector demonstration projects for the aforementioned populations. However, the Ministry of Health and Child Care (MoHCC) plans to add PrEP to the HIV prevention package in the public sector based on a client’s risk profile. The risk profile is built from information on sexual practices and history in the six months preceding their appointment using a risk assessment tool and follows the eligibility guidelines laid out in the MoHCC’s 2016 National Guidelines for Antiretroviral Therapy for the Prevention and Treatment of HIV. Those at risk are offered PrEP, or a client may request PrEP regardless of the risk assessment outcome. In addition to high-risk groups such as MSM or AGYW, other individuals eligible for PrEP may perceive themselves to be at risk of contracting HIV, possibly due to having multiple sexual partners or stable partners of people with multiple sexual partners. Zimbabwe’s growing epidemic of extra-marital affairs popularly known as “small houses” that have been cited as key drivers of Zimbabwe’s HIV epidemic coupled with the country’s socio-economic challenges necessitates a broader HIV prevention and targeting strategy. Accessing such individuals may require different programmatic approaches. PrEP is a potential user-controlled HIV prevention option with important utility, especially when condom negotiation is difficult. Information about motivations for the general at-risk population in a public sector setting choosing PrEP is limited worldwide and particularly for generalized HIV epidemics like Zimbabwe. The objective of this study was to, therefore, explore the experiences of clients that both accept and decline PrEP offered at a public facility. Specifically, we aimed to understand the factors that motivate clients to accept, decline, continue, or discontinue PrEP. Given the resource limitations of the MoHCC, it is important to understand how to effectively deliver PrEP in the public health system, which may not have the level of resources of well-funded early demonstration projects. # Methods A qualitative study was carried out from January to June 2018. This study was approved by the Medical Research Council of Zimbabwe (Protocol MRCZ-A-2234) and the Chesapeake Institutional Review Board (Protocol Pro00023165). The MoHCC offered PrEP in two semi-public Zimbabwe National Family Planning Council (ZNFPC) clinics, a parastatal under the MoHCC mandated to coordinate the provision of family planning services in Zimbabwe. The MoHCC plan for PrEP rollout was also planned to start in these ZNFPC sites before scaling up to other public sector facilities. These facilities were chosen because they were expected to be good environments to reach members of the general population who were seeking services such as cervical cancer screening, VMMC, and family planning services. The two sites were selected because they had the highest HIV testing volumes in their respective settlements. The two PrEP learning sites were selected purposefully to include one urban and one rural site, and PrEP implementation at the learning sites was operated with support from CHAI and the Bill & Melinda Gates Foundation. There were no other sites offering PrEP to the general population near the rural site. In the urban setting, two other partner- funded sites were offering PrEP: one site with outreach specifically targeted towards female sex workers and the other with a more general population focus. All nurses at these sites were trained on antiretroviral therapy (ART) and TDF- FTC-based PrEP by MoHCC. Clients were eligible for PrEP if they were at least 16 years and had an HIV-negative test result. After HIV testing, clients were screened for PrEP eligibility according to a tool based on government guidelines, and all clients determined to be eligible for and offered PrEP at one of the two sites were eligible for interviews. Clients were offered PrEP on the same day as eligibility was determined and, if they accepted, were given a one-month supply of PrEP on that day. They were then scheduled for a follow-up visit at one month, where they received a three-month supply. Thereafter, they were scheduled for visits every three months and provided with a three-month supply at each visit. During recruitment, the intervention was not formally advertised; however, clients were informed about PrEP when the nurses gave general health education to all clients during the appointment intake. Nurses would then proceed to conduct client HIV risk assessment during routine consultation. Once a nurse determined a client to be at risk, they would then speak to the client about HIV prevention in general and the possibility of them taking up PrEP. Thereafter, word of mouth from the clients served as the main means of advertising. In the rural site, the Ministry held a meeting with the local chiefs and village heads in order to gain acceptance of the intervention. The village heads proceeded to sensitise their communities on the intervention, and the nurse counsellor at the ZNFPC site would actively recruit clients based on his knowledge of the community around the site. Interview candidates were categorized into five sub-groups to ensure diversity of perspectives, including people in the following categories (abbreviated with letters for later reference) (1) declined PrEP (D); (2) accepted PrEP but missed their first scheduled appointment (AM); (3) accepted PrEP and attended their first scheduled appointment (AA); (4) accepted PrEP, attended their first follow-up, but did not attend their second follow-up appointment (AAM); (5) accepted PrEP and attended their first and second follow-up appointments (AAA). shows the timing of the interviews with each population and the number of interviews conducted with each group. For clients found to be eligible for and offered PrEP, health care workers (HCWs) described the study and requested their permission to be contacted for an interview. If an individual agreed to be contacted, MMG contacted them by phone. Follow-up phone calls were conducted only for the purpose of recruiting participants and not to encourage retention. Clients that declined PrEP were called approximately one week after they declined, and clients that accepted PrEP were called approximately 7–10 days after either their first or second scheduled appointment regardless of whether they had attended or not. During the initial call, the study was again described. If clients agreed to participate in an interview, an in-person interview was arranged. Written informed consent was obtained from participants at the beginning of the interview. Interviews were conducted and recorded by Zimbabwean interviewers in English or Shona. Aside from the participant and interviewer, no one else was present. Interviewers were trained qualitative data collectors with no link to the communities. Preferences from clients about interviewer gender were respected. Interviews were conducted in-person at the PrEP site or at the client’s preferred location. The interview guide was based on the Extended Health Belief Model that explores health behaviour constructs, including perceived susceptibility, perceived severity, perceived benefits, perceived barriers, cues to action and self-efficacy. and are the interview guides used for clients who declined or accepted PrEP, respectively. Interviews lasted 15–60 minutes with the average being 45 minutes. Participants were provided with an allowance of US\$8 for transport and other costs of participation. Based on a previous VMMC study in Zimbabwe, an \$8 allowance was deemed to cover costs without manipulating clients to participate in the study. Interviews were conducted and analysed on a rolling basis over five months. Recordings were de-identified, transcribed and translated by the interviewer, and the study coordinator (MMG) listened to the audio recordings and checked all the transcriptions to ensure accuracy. Three members of the research team (MMG, MLP, BEC) participated in coding the transcripts, using the constant comparative method to systematically code the transcripts. In the beginning, all members of the coding team independently coded the first three transcripts looking for key themes emerging from study participant responses. Based on this initial reading, the team created a codebook that was then applied to subsequent interviews and adjusted as needed. When all members of the coding team had triple-coded approximately half of the transcripts with any coding discrepancies discussed in the group to reach a consensus, the final codebook was agreed. At this point, the team switched to having two coders code each transcript. At the end of the coding process, the final codebook was applied to all transcripts coded earlier in the process to ensure all interviews followed the same codebook. The coding team provided on-going feedback to interviewers about topics that should be further explored and determined when theoretical saturation was achieved based on review of code reports. Dedoose software was used to organize coded data. # Results Three PrEP decliners declined to be interviewed and many more refused to be documented by the nurses as having declined PrEP. The characteristics of the sample and the sub-population they were a part of is shown in. In this paper, each quotation will be identified using the gender, age, location of services (RUR/URB), primary risk factor, and sub-population (D, AM, AA, AAM, AAA). This paper presents key themes on factors that motivate clients to accept, decline, continue or discontinue PrEP. The emerging factors are summarized in and described below. ## Facilitators and barriers for PrEP uptake and continuation ### High HIV risk perception Participants who considered themselves to be at high risk of contracting HIV were often interested in PrEP and continued taking it. Most individuals acknowledged being a part of an established risk group, such as sero-discordant couples (SDCs) or commercial sex workers (CSWs), or were concerned their partner may expose them to HIV due their confirmed or suspected extramarital affairs. Almost all men who took up PrEP were in SDCs and the other variations in risk profiles were among women. SDCs indicated a willingness to take PrEP indefinitely in order to live a “normal” sexual life with their HIV-positive partner. > “*My husband and I were quarrelling at home because he was not taking > his ART medication*. *Then he would want to be so defensive and harsh > and I told him that we cannot stay together if you are not protected I > feel uncomfortable so you should use protection and he used > protection…*. *My husband defaulted ART the whole of 2017 starting > from 2016/2017 and he stopped taking the medication now so he is > sick*. *I was offered PrEP and I accepted*. *Now we no longer > quarrel”–Female*, *59*, *in SDC (RUR06) (AAA)* Also, among SDCs, conception without the risk of contracting HIV was also cited as motivation for taking and continuing PrEP. > “*The main reason that made me decide to take PrEP was that I wanted a > child*. *\[…\] all along I tried going to the doctors and that was > very expensive*. *I am now nearing a tricky age*. *So when I heard > about PrEP I was very happy*.*”–Female*, *37*, *in SDC (URB01) (AA and > AAA)* Many participants not in SDCs also described concern that their partner could expose them to HIV because they observed risk factors in their partners’ behavior issues such as having recurring STI infections, receiving sexual messages on their phones, and refusing to get tested for HIV. > “*I saw a lot of messages from different girls on my husband’s phone*, > *I spoke to him about it*, *but I was surprised to be diagnosed with > an STI twice*, *I realized I was talking to myself and decided to take > PrEP*.*”–Female*, *32*, *partner HIV status unknown (URB06) (AM)* One participant acknowledged that if she had multiple concurrent partnerships, her partners may also have other sexual partners that could in turn increase her risk for HIV. > “*Sometimes you can get carried away and you go ahead and do it > unprotected… If I have two boyfriends*, *how many girlfriends do they > have*?*”–Female*, *23*, *partner HIV status unknown (URB09) (AM)* Finally, those that engage in sex work reported being highly aware of their HIV risk; therefore, they were motivated to get regular HIV testing and to accept PrEP when offered. > “*When I heard about PrEP*, *I came running because the nature of my > job*. *\[…\] I can meet an individual who is HIV-positive in my job*. > *God blessed me so far that I am HIV-negative but I knew I was not > going to continue to be skillful*, *so I decided to \[take PrEP\] and > protect myself*.*”–Female*, *31*, *CSW (RUR04) (AAM)* ### Confidence in PrEP as an HIV prevention method Most participants felt confident that PrEP was effective in preventing HIV, and as a result, they felt empowered that HIV prevention was in their hands. > “*PrEP changed the way I stay with my husband*. *I always thought of > divorcing him and going to my parents’ home and work for myself > there*, *without any stresses because he would get infected with HIV > where he does his promiscuity*. *But from the time that I started > taking PrEP I am now a mother who thinks of improving her household*, > *without thinking of divorce*.*”–Female*, *28*, *partner HIV status > unknown (RUR12) (AA)* Several female participants mentioned that they are able to use PrEP where condom use presents challenges. > “*I will take PrEP for life because I can no longer be infected by > HIV*. *In addition*, *my husband was cruel as he would tear the > condoms sometimes and he would pretend as if it had burst*. *I was > really happy that I now have backup*.*”–Female*, *32*, *in SDC (RUR33) > (AAA)* PrEP is seen as a way to achieve peace of mind and confidence in being protected, even in cases where a condom breaks or a partner refuses condom use. In fact, using PrEP was even described as being able to allow participants to ask for and enjoy sex more: > “*As a woman*, *\[PrEP\] can boost your confidence in asking for it > \[sex\] because you know you are protected*.*”—Female*, *37*, *in SDC > (URB01) (AAA)* ### Perceptions of living with HIV Perceptions and consequences of living with HIV also motivate some people to take PrEP. Generally, participants strongly believed that antiretrovirals (ARVs) work and that HIV-positive people could have “normal healthy lives” if they adhere to medication. They noted that it would be difficult to cope with the infection but a positive mindset was essential to survival. Despite these beliefs, participants were still strongly motivated to prevent HIV. In considering the potential impact of HIV, many participants thought about the impact on their families and their children. Many participants described a fear of loss of economic productivity if they were infected with HIV. They expressed that HIV positive people are too weak to work in their fields or engage in day- to-day activities, and therefore, they and their families would suffer. Female participants expressed a desire to ensure that they remain healthy to care of their children so they wanted to protect themselves from HIV. They expressed concern that if they were no longer there to take care of their children, no one would care for their children as they would. > “*What will really affect me about HIV is that the life that I will be > living will hurt me a lot*. *In that now that I will be sick*, *my > children pain me a lot because the father will be sick and the mother > will be sick*. *That hurts me a lot*. *Who will take care of my > children*, *because I will die and that father will die*. *Who will > take care of them*, *when we are all gone*? *No one will take care of > them and you cannot really on the aunt that they will take care of > them because you cannot rely on them and so I should really focus a > lot on these pills \[PrEP\] that will make my life a bit longer as my > children grow*, *it is better*.*”–Female*, *28*, *in SDC (RUR01) > (AAA)* > > “*I liked \[PrEP\] because we want to take care of our families*. *As > women*, *we will not know what we will do with our families \[if we > are gone\]*. *You will be scared of what will happen to your family*. > *I did it for my children*. *No one will work for my family if I am > gone*.*”–Female*, *33*, *partner of unknown HIV status (RUR03) (AA)* ### Barriers to PrEP uptake Some participants were deemed to be eligible, yet chose to decline PrEP and did not attend any follow-up appointments for PrEP. In the PrEP pilot, HCWs did not comprehensively document decliners, so our potential to identify decliners for interviews was limited. However, there were cases where participants who ultimately accepted PrEP described initial hesitation or barriers. ### Concern for pill burden A few participants expressed a dislike for pills or concerns about the size of PrEP pills. Several participants indicated that they thought adherence would be difficult. > “*I was just afraid to take pills*. *\[…\] Growing up*, *I did not > take any pills*. *Sometimes if I take even paracetamol the pills will > be smelling in my body*. *Whenever I would take the pills*, *I would > notice something is happening in my body which will be caused by the > pills”–Female*, *42*, *partner HIV status unknown (RUR21) (D)* ### Needing partner’s consent or fearing partner reaction to PrEP Whereas some participants appreciated being able to use PrEP privately without consent or knowledge of their partners, other participants said they declined PrEP due to social circumstances such as needing to get permission to take PrEP from their partners. One participant went on to explain that, while she was trying to protect herself from her partner’s risky behavior, he might interpret her taking PrEP as a sign of her own risky behavior. > “*I declined PrEP because my husband would accuse me of having another > sexual partner while he is away*. *\[…\] So I think it is best for me > to ask for permission to take PrEP and if he agrees then I will come > with him*.*”–Female*, *20*, *partner HIV status unknown (RUR20) (D)* > > …. *even if I am faithful but you cannot trust men they can sleep > around …*..*I talked to my husband \[about taking PrEP\] and he told > me that he is HIV negative he even asked me that if you want to take > the Prep what will be your reasons since we were all tested HIV > negative…*. *he did not deny getting tested \[on his own\]*. *I found > it better not quarrel with my partner if he is suggesting other issues > rather it will be better to wait for the way forward from him*.*- > Female*, *42*, *partner HIV status unknown (RUR21) (D)* Disclosure of use of PrEP was also cited a consideration for adherence. In some cases, clients received more general emotional support from their partner or family. Simply telling others about PrEP can be liberating or empowering. > “*As for me*, *my brothers*, *mother*, *everyone*, *all my siblings > staying here–there is no one who does not know that I am taking PrEP*! > *So you find for me it’s easy to keep my bottle in the open and not > stashing it in my bag where I can forget it for sure*. *\[…\] That > helps a lot*.*”–Female*, *23*, *two partners*, *HIV status unknown > (URB09) (AM)* While the quotation from the previous participant describes sharing the news of taking PrEP widely, others chose to be more selective. Similar to the decliners, some participants faced challenges in telling their partners about PrEP. They feared being accused of unfaithfulness or promiscuity, therefore opting to discontinue PrEP, especially where conflicts arose. > “*I told my husband that I was taking PrEP*, *and he said I am the one > who wants to acquire the HIV that I wanted to prevent*. *And so I > realized that it was better for me to just stop because it would > create problems*.*”–Female*, *20*, *partner HIV status unknown (RUR34) > (AM)* ### Feeling satisfied with current method of HIV prevention Another reason for declining PrEP was the preference for condom use. One male participant reported being happy with condoms or other current methods for prevention. Males’ ability to control the use of condoms puts them in a position to decline other HIV prevention methods in comparison to their female counterparts. However, one participant noted that they might consider PrEP in the future. > “*I was told about PrEP and it was ok*, *but I told them that condoms > were working well for me and I can stick to them for the meantime*. > *But if they were not working well for me*, *I was going to take > PrEP*. *\[The HCW\] said it was ok*., *they were not going to force me > to take PrEP*, *because I am the one who knows best on what I need*. > *I just want to use condoms only*.*”–Male*, *36*, *in SDC (RUR24) (D)* ## Facilitators and barriers to PrEP adherence and continuation Participants described the challenges they faced and the approaches they used to help them continue PrEP. This section is organized in terms of facilitators and barriers within the following areas: risk perception, routine and logistics, partner and family interaction, comprehension and counseling, and side effects. ### High HIV risk perception From some participants, the continued exposure to perceived HIV risk was a strong motivator to continue PrEP. Most participants in SDCs described feeling that every sexual encounter provides HIV risk exposure, and this motivated them to continue to attend their appointments for PrEP refills. > “*My wife and I agreed to keep on taking PrEP forever unless we get > different advice from the health worker*.*”–Male*, *69*, *in SDC > (RUR10) (AAA)* While HIV-negative partners in SDCs described long-term risk perceptions, other participants described how their risk perceptions may change over time, leading to a change in their motivation to continue PrEP. For example, the following participant described how she would stop taking PrEP if her partner agreed to HIV testing: > “*I don’t think I will keep taking PrEP*. *I think that I will be able > to convince him \[my husband\] to go for an HIV test but I am not > really sure* … *So even up to now I am scared until I get convinced we > can then use condoms*.*”–Female*, *33*, *partner of unknown status > (URB12) (AA)* ### Routine and logistics Most female participants were already accustomed to the routine of taking family planning pills on a daily basis, so they decided to take all pills at the same time Participants also talked about storing their PrEP and family planning pills together, or in a place where they would easily see it every day. > “*What helps me not forget is that I would like to take good care of > my health*. *That is the reason that I placed them where I store my > family planning tablets*. *I cannot forget because I want to take care > of my family and so I know that if I put them in the same place I will > not forget*.*”–Female*, *28*, *CSW (RUR07)*, *(AA)* Other participants established a routine to help them remember to take their pills daily during the same activity, such as waking up, going to bed, or cooking dinner. Others take the pills at the same time every day, often by setting an alarm on their phones. > “*I take my PrEP around 6pm that is when we will be done with our work > in our fields*, *when the sun goes down*. *So when I get into the > house before cooking*, *that is when I take my PrEP \[…\]”–Female*, > *28*, *partner HIV status unknown (RUR07) (AA)* However, two participants explained that they had never taken daily pills before and found it difficult to adjust. Similarly, for most, the size of the PrEP pills was a challenge. > “*Just starting to take pills when you have never taken them \[is > difficult\]*. *Plus*, *with how big it was…iiih*! *I would at times > take \[the pill\] and it would take me some minutes feeling that it is > still sitting on my throat*.*”–Female*, *29*, *partner HIV status > unknown (URB10) (AM)* Even when clients remembered to or intended to take PrEP daily, some individuals faced logistical challenges. Appointments would be missed due to having gone out of town, such as for a funeral or to buy goods in another country for business. A few were in town, but the hours of the PrEP clinic conflicted with work hours, or they lacked money to pay for transport to the clinic. Transport and logistical challenges were more common in the urban site where some participants were formally employed and also required transport funds to move around. > “*Sometimes I won’t have money*. *Sometimes I will have it*. *Or maybe > I have just found enough to go there \[facility\]*, *but not to come > back*.*”–Female*, *31*, *in SDC (URB07) (AM)* > > “*It is due to work for the past two weeks \[that I missed my > appointment\]*. *I used to go for day duty and when I am on day duty I > work from 7am to 7pm*, *therefore I did not get the chance to go there > \[to the clinic\]*.*”–Female*, *aged 33*, *partner of unknown HIV > status (URB11)*, *(AM)* While travel was a key barrier to returning on time for appointments, some overcame this challenge by planning around their travel. Some participants would travel with enough medication for the duration of their trip or would come to the clinic early if they realized they would run out of pills while traveling. This was observed among both males and females. > “*I came today though my date was tomorrow*. *Because tomorrow I have > a journey*, *\[I\] thought I should come today to minimize any > challenges*.*”–Female*, *aged 59*, *in SDC (RUR06) (AAA)* ### Partner and family interactions Support or challenges from partners and family members in relation to PrEP was not directly addressed in interview questions, but this issue came up often. Partners were reported to directly support clients by reminding them to take their pills or about appointments. Indirect support was also provided, such as by providing childcare during appointments. Accounts of these types of support were particularly common within SDCs. > “*\[My husband\] helps me by asking if I have taken my pills*. > *Sometimes I forget*, *so he will be reminding me to take my > PrEP*.*”–Female*, *21*, *in SDC (RUR16) (AA)* In the case where a male participant was taking PrEP because he had multiple wives, including one with HIV, taking PrEP was an effort to protect not only himself but also his other wives. Support for clients to communicate effectively with all parties in such situations is critical. In this case, the man reported the clinic supporting his family by talking directly to the HIV-positive wife at the clinic and sharing written educational materials for other wives. > “*When I first started PrEP it was difficult*, *but I never lost > hope*. *\[…\] It’s very scary to have unprotected sex with an > HIV-positive person*. *Even my HIV-positive wife was so afraid asking > how we can have sex without using condoms*. *\[…\] Then I took her to > the clinic so that she could understand better from the health > workers*, *she accepted the idea of me taking PrEP*. *She was afraid > that she will get me infected with HIV because she does not want me to > get infected by HIV since I have many wives and I will spread the > infection to them*.*”–Male*, *43*, *in SDC (RUR18) (AAM)* ### Comprehension and counseling Another challenge with adherence and retention was participants misunderstanding clinical guidance on how often to take PrEP. For instance, an HCW stressed to a client that they should take PrEP for at least seven days (Zimbabwe guidance) before engaging in risky sex to be protected. The participant misinterpreted this as taking PrEP for seven days before expected exposure, with a hiatus until the next expected exposure. Therefore, several months after her original appointment, the participant still had not exhausted her 30-day supply of pills. Another participant describes a similar understanding below of how HIV protection for PrEP works: > “*I am not taking it continuously*. *I take it take a break*, *take > it*, *have a break*. *\[…\] I do not take it so often*. *I took it for > the first time just to find out if I had a good relationship with the > drug and if I did not have negative side effects occurring*. *I found > out that I have a couple of negative effects and then I said*, *I > ultimately decided that if I were to take PrEP*, *it has to be seven > days before I have sexual encounter which is unprotected with somebody > who is not tested*.*”–Male*, *34*, *partner of unknown status > (URB02)*, *(AA)* ### Side effects Side effects were not reported as a major reason for poor adherence or retention. Many participants reported no side effects. Those who experienced side effects reported them as minor, subsiding after a few days. Reported side effects included headache, dizziness, drowsiness, fatigue, nausea, vomiting, diarrhea, extended or heavy menstrual periods, loss of appetite, and increased appetite. Participants developed coping strategies to deal with side effects, such as eating before taking PrEP pills to avoid nausea or consulting HCWs about them. For example, one participant who experienced nausea found that if she ate a meal before taking her PrEP pills, the symptoms of nausea were not as bad. > “*\[Side effects\] started from the day I started to take the PrEP > pills*. *Sometimes I could even feel a headache*. *I could even feel > dizzy and I would also feel sleepy*. *Sometimes I would take > paracetamol to manage the headache*, *but as for now the headaches are > now better*. *As for now if I take PrEP orally I will feel dizzy*, *so > I will take the pills in the evening before I go to bed and I will > feel dizzy and sleep*.*”–Female*, *aged 21*, *in SDC (RUR02) (AAA)* Based on participant responses, having information available about side effects was critical to allowing them to overcome this initial challenge without stopping PrEP. In some cases, participants felt comfortable enough and were able to return to the clinic to report their symptoms to a health worker and get advice: > “*I faced a challenge when I first took my pills*. *I felt weak when I > woke up in the morning*. *And I thought I should stop taking \[PrEP\] > but I decided I should not stop without consulting \[a health worker\] > on what was taking place*. *When I consulted they told me that*, *‘I > should keep on taking them*. *Maybe they will work well on you*.*’ As > for now as I am taking \[PrEP\]*, *and I no longer feel weak and no > more nausea*. *It happened 1 week only*.*”–Female*, *31*, *commercial > sex worker (RUR04) (AAM)* Similarly, another participant went back to the facility to get pamphlets to take home and read for more information about side effects and other issues related to PrEP. For example, the following participant was able to request the information she needed to understand and cope with side effects: > “*I once had a problem of fatigue and dizziness and I decided to come > back to ask further about the pill they had given me and how it works > because once I took it the next day in the morning when the sun rose I > had fatigue and head ache*. *It happened for a few days of which now > it has passed*. *\[…\] I told myself that I will not stop \[PrEP\] but > rather rush to the clinic to ask those who gave me the medication why > it happens that way*. *They gave me those pamphlets but at times > English is difficult to understand*, *and I asked them to give me > Shona ones \[…\] so that I can sit and read comfortably*. *I was given > and I read them and understood*.*”–Female*, *28*, *in SDC (RUR01) > (AAA)* There were no differences between the results from rural and populations regarding perceived risks, barriers and facilitators. The only differences observed in these interviews related to how participants found out about PrEP with active demand creation by the nurse in the rural site as opposed to mass media communication or nurse referral from another HIV prevention study in the urban location. # Discussion This qualitative study on the introduction of PrEP into the public sector found many themes that will help guide the roll-out of PrEP in the public sector by providing insights into reasons for PrEP initiation and ways to improve PrEP adherence. The effectiveness of PrEP is dependent on its acceptability, accessibility, adoption, and sustainability as part of a comprehensive HIV prevention package. If these components are not there, even the most highly efficacious PrEP medication will have little to no impact in reducing HIV infections. Therefore, HCWs should be trained and familiarized with all themes regarding reasons for initiating, declining, and barriers or facilitators for adherence in order to ensure high-uptake of PrEP within the general population. Among the participants in this study, who were largely in FSWs or in SDCs or had a partner of unknown status, the perception of risk in these groups of people facilitated their uptake of PrEP, which is consistent with other studies in the United States where PrEP was integrated into a family planning clinic. For all groups, the perceived severity of living with HIV was a key driver of PrEP uptake and the use of PrEP was associated with comfort during sex with an HIV sero-discordant partner. This was also reported in a study among gay bisexual men in HIV sero-discordant male relationships. With regards to SDCs, this work found that women start PrEP as a means to protect themselves from contracting HIV from their partners. In many parts of Africa, the risk of HIV infection is found in marriages where husbands may have acquired the virus through extramarital sex. In some cases, the added threat of violence affects women's power and ability to negotiate the conditions of sexual intercourse, especially condom use. A woman’s sexuality within the Zimbabwean context is often controlled by her husband or her male partner such that condom negotiation or refusing sex may be difficult. The theme of limited ability to negotiate condom use has emerged in other studies, particularly when considering relationships between AGYW and older men in Africa. The quotations from several participants capture the conservative nature of rural Zimbabwe, where staying in a marriage is more respectable than divorce, and therefore, PrEP becomes a means to keep the peace in a marriage. However, a study conducted in Kenya on PrEP acceptability revealed that divorce rates among SDCs were high, highlighting a need to counsel SDCs on how to avoid HIV transmission within the marriage. With regards to male partners in SDCs, the results from this work also show that the men who took up PrEP in this study were in SDCs; however, there was one male decliner preferring condoms to any biomedical HIV prevention method. The reasons for declining PrEP found in this study included the fear of pill burden, mistrusting of biomedical means of HIV prevention, preferring condoms and discouragement from taking PrEP by family members. These themes are often supported by those found among MSM in the United States. This study showed how critical support, whether from partners or other family and friends, is to PrEP uptake and adherence. Although PrEP may give women the ability to prevent HIV without demanding condom use, many individuals reported difficulty concealing PrEP use from their partners and therefore sought permission before starting PrEP. Accounts from other women in this study described how having support from sexual partners and others improved their adherence. The findings on family and partner support and adherence are supported by evidence from a study among females in Kenya looking at factors affecting PrEP adherence. HCWs should be prepared to counsel clients on talking about PrEP with their partners or family members. This study was conducted to inform program planning for public sector PrEP rollout by looking at barriers and facilitators to initiation, adherence and retention in an effort to better understand how HCWs can discuss PrEP with clients. The findings from this study provide insight into the types of issues that HCWs and other program staff should be prepared for as they interact with clients around PrEP. For example, sexual relationships can be complex, and HCWs must, therefore, be prepared to counsel clients on their risk level based on a range of situations. Once clients begin taking PrEP, HCWs must ensure that they understand how long they must take PrEP prior to and after possible exposure to ensure maximal protection from PrEP, which may take up to seven days prior to exposure. The counseling needs to stress that daily PrEP is recommended for the majority of persons who cannot predict when they will have sex. However, if their risk changes (e.g., they are not sexually active or have one partner known to be HIV negative), they can discontinue PrEP after at least 28 days after their last exposure. Qualitative studies that were conducted in Kenya and the United States also highlight the key roles played by HCWs in counseling clients for PrEP adherence, other prevention methods and community awareness in order to improve PrEP acceptability. Future studies should focus on the perspective of HCWs as they perform counseling and rollout of PrEP to identify ongoing challenges. The concern that taking PrEP will make clients more promiscuous was not confirmed by participants that were on PrEP. Most participants mentioned that they had not changed their sexual practices or added sexual partners due to the lower perceived risk. Instead, participants continued with the same sexual practices but with less anxiety over contracting HIV after taking PrEP. However, some participants reported decreased condom use after PrEP initiation, and in some cases it was reported that HCWs had recommended this for clients in stable marriages. While other research has shown that stereotypes of promiscuity were a deterrent to taking PrEP in the United States, this issue was not raised in the Zimbabwean setting. ## Study Limitations Few decliners were documented by HCWs. We invited all the decliners for interviews and worked with HCWs to encourage improved documentation of decliners. Participants may have been reluctant to talk about sensitive issues. However, we trained interviewers to help participants feel comfortable in interview settings, matched gender where necessary and held interviews in locations selected by participants. In addition, these participants all self- identified as being motivated to seek PrEP by risk factors related to sexual encounters; this study does not explore access and retention these among other means of risk, such as injection drug use. Information was not available on the HCW perspectives on services and training received as part of this study but should be explored in future studies. # Conclusions This study highlights how HIV risk perception, confidence in PrEP, and perceived challenges of living with HIV drove PrEP uptake in this population, while concern about pill burden and side effects and about partner reactions to PrEP, and reliance on other HIV prevention methods led others to choose not to take PrEP. Issues around HIV risk perception, routine or logistics, comprehension and counseling, partner and family support, and side effects were critical in either enabling or prevention clients from continuing PrEP. The introduction of PrEP added a valuable HIV prevention tool to the comprehensive care package in the two study facilities. It has addressed a gap where clients were previously not able to use existing HIV prevention methods for whatever reason. Based on feedback in this study, some of the greatest benefactors of PrEP were women who through this option are empowered to take HIV prevention into their own hands in instances where condom negotiation is difficult or whose sexual partners engage in risky behavior. SDCs were a specific at-risk group who required a backup method for protection or an HIV prevention lifeline, and this study shows that this significant unmet need was addressed through the introduction of PrEP as a prevention option. By understanding the factors that influenced uptake and continuation, future programs can be improved to better serve the target population. Based on client feedback, we offer recommendations about how to enhance HCW training and service delivery models in order to meet the needs of clients and encourage PrEP uptake for those at risk in the general population. For example, clients may need support on disclosure of taking PrEP and planning appropriately for travel to ensure adherence. Also, going forward, the appropriate timing of use of PrEP needs to be clearly explained to HCWs for counseling purposes to clients. There were a few misconceptions among clients about how to take PrEP, especially if clients had intermittent risk exposure. The use of other prevention methods to complement PrEP was also cited as a point that needed clarity from HCWs, and MoHCC can provide some policy guidance on this issue. At the same time, when considering these recommendations, it is important to take into account the human resources constraints on PrEP programs previously highlighted in the literature. These results have been incorporated into policy documents such as the Zimbabwe National PrEP Implementation Plan and Health Care Worker Training Manual, and the HIV Prevention, Treatment, and Care Communication Strategy. These findings may also be relevant in other countries interested in expanding public sector PrEP access. # Supporting information We would like to thank the Zimbabwe Ministry of Health and Child Care and the Zimbabwe National Family Planning Council for support and collaboration in carrying out the evaluation, the study participants for sharing their time and invaluable perspectives. We would also like to thank the interviewers and other CHAI Zimbabwe staff that provided coordination and logistical support to this work, in particular: Mwaonedza Mushango (the late), Nisbert Teshe, Yemurai Katanda, Alison Erlwanger, Tinashe Sabiti, Euclidita Gurupira, and Tatenda Mudehwe. 3DE Demand Driven Evaluations for Decisions AGYW Adolescent girls and young women ARVs Antiretrovirals AVAC AIDS Vaccine Advocacy Coalition CHAI Clinton Health Access Initiative DFID United Kingdom Department for International Development FSW Female Sex Worker FTC Emtricitabine HIV Human Immunodeficiency Virus MoHCC Ministry of Health and Child Care MRCZ Medical Research Council of Zimbabwe MSM Men who have Sex with Men PEP Post-Exposure Prophylaxis PrEP Pre-Exposure Prophylaxis SDC Sero-Discordant Couple STI Sexually Transmitted Infection TDF Tenofovir Disoproxil Fumarate VMMC Voluntary Medical Male Circumcision WHO World Health Organization [^1]: The authors have declared that no competing interests exist.

# Introduction Water is practically essential for all of the functions of our bodies. Its regulation is based on a complex endocrine system that balances its intake (through food, beverages and metabolic water production) and loss (through skin, sweat, lungs, urine and feces). Maintaining an adequate hydration status is crucial because of water’s role in our bodies. The amount lost in urine depends on water consumed, the solute content of diet (high intakes of salt or protein will increase the daily fluid replacement because of the limited capacity of the kidneys to concentrate urine) and on the total losses. If water intake is restricted, the kidneys will conserve water by producing more concentrated urine. Equally, the body cannot store excess water, so the kidneys get rid of any temporary excess by producing a large volume of dilute urine. Adequate hydration in schoolchildren can be related to better performance at concentration tasks and better physical performance, especially in hot environments. Healthy children may also be at risk of dehydration if there is a sudden increase in water loss for any reason, and physically active children will be at particular risk during periods of warm weather. The large surface area:volume ratio of children means they gain more heat through the skin when the environmental temperature exceeds skin temperature. In this situation, an increased rate of evaporative cooling is achieved at the expense of an increased loss of water from the body. Currently, there is no “gold standard” method to determine an individual's hydration status, although it is important to the human body. The latest studies establish urinary osmolality and dietary water intake as two of the methods that are more frequently used. Additionally, American College of Sports Medicine also recommends thirst or urine color scales and plasma osmolality. In our study, thirst scales have not been considered appropriate because of the complexity of filled the questionnaire by parents through the impressions of children. Urine color scales have not been taking into consideration to evaluate the hydration status due to the lack of facilities in our study circumstances to realize an assessment of the urine sample by trained professionals on-site, who realized the anthropometry. Plasma osmolality is also an excelled standard, but it could be considered an expensive and invasive method and urine osmolality has been considered a valid and reliable indicator of hydration status and could be used in field settings. For this reason, urine analysis has been chosen as a practical, fast and less expensive method. The dynamicity of the water balance makes it especially important to study factors that may modify it; physical activity (PA) and sedentary behaviors can be included among these factors. Regarding the role of inadequate hydration in PA, Stachenfeld et al. highlighted the importance of improved thermoregulation through the maintenance of blood volume and the amount of sweat produced. In this sense, the sweat volume produced during PA practice directly affects the hydration status. Despite the massive amount of knowledge about the physiological inter-regulation between these factors, very few studies have evaluated the hydration status in schoolchildren who maintain active lifestyles and perform PA for recreational purposes and in an unplanned manner. Assessing physical activity behaviors in children is a complex issue. The nature of children’s physical activity behaviors (short, intermittent and, in some cases, unstructured) make it difficult to measure children’s PA. The most feasible measures for schoolchildren are indirect reports by parents or teachers or objective methods such as direct observations, heart rate monitors or motion sensor such as accelerometers. Sedentary leisure behaviors are gaining notoriety among the lifestyle factors. Currently, sedentary behaviors are considered an independent factor of PA that could be interacting with dietary and hydration patterns. Sedentary behavior is any waking behavior characterized by an energy expenditure ≤ 1.5 metabolic equivalents (METs), while in a sitting, reclining or lying posture. Very few studies relate certain sedentary behaviors to different intake of food and beverages (in a quantitative or qualitative manner). This interaction could affect the hydration status. Therefore, we consider it especially interesting to study dietary habits, PA and sedentary behaviors in order to quantify the influence of lifestyle factors on hydration. The aim of the present study was to evaluate the hydration status and dietary water intake in a sample of 7- to 12-year-old schoolchildren and their association with different lifestyles according to their PA and sedentary behaviors. # Materials and methods The design, recruitment of subjects, and methodology regarding anthropometry and urine analysis have been described elsewhere already. The study protocol was approved by the Ethics Committee for Clinic Review of the Clinic San Carlos Hospital, which is part of the Complutense University of Madrid (Madrid, Spain) (Ref 15/522-E). ## Subjects This observational study involved eight primary schools from various Spanish provinces (Madrid, Córdoba, Segovia and Ciudad Real), including the capital of the each province and a semi-urban/rural city. All schools were chosen randomly, looking to include children from different locations of the Spanish geography. Of the several schools contacted by telephone, eight accepted the invitation to be included in the study. Permission was requested to meet the parents of children in the age group 7–12 years. Once permission was given, the details of the study were explained to parents and all questions answered. Written informed consent was then sought to include their children in the study. After this step, lifestyle information and a 3-day dietary record were collected, and the necessary material to collect 24-hour urine samples (24-h) was provided. A total of 1,232 children was given the opportunity to participate in the study. Of these, 278 children provided written informed consent; 16 were excluded because they did not collect the urine sample correctly and 20 were excluded because they did not fully complete the lifestyle questionnaire. The final sample, therefore, compromised 242 from 7 to 12 years old; 49.17% of them were females, and 50.83% were males. The fieldwork was performed between February and March 2014, when the average temperatures were 13.5°C and 14.5°C, respectively and the range of temperatures were 3.6–14.7°C in February and 4.6–16.8°C in March. The inclusion criteria were as follows: living in the provinces of the study, being between 7 and 12 years old and attending the second to sixth grades of primary school, being in compliance with the informed consent via parents or legal guardians, and not presenting any clinical problem that could modify the results, such as liver or kidney diseases, diabetes, and/or hyper/hypokalemia. In addition, the participants had received no pharmacological treatments in the 3 months prior to the study with corticosteroids, insulin or diuretics. ## Anthropometric data All measurements took place at the schools in the mornings and in accordance with norms set out by the WHO (World Health Organization). The weights and heights were measured using a digital electronic balance (range 0.1–150 kg; precision, 100 g; Alpha; Seca, Igni, France) and a digital stadiometer (70–205 cm; 1mm; Harpenden Pfifter, Carlstadt, NJ, USA). The body mass index (BMI) was then calculated and compared with the IOTF cut-off points. ## Lifestyle study A lifestyle questionnaire, previously applied in other Spanish populations of children and adults, was filled out by the parents or legal guardians. The questionnaire included questions to quantify the primary activities that could indicate an active lifestyle and sedentary behaviors, both on weekdays and weekends. This questionnaire collected the frequency of active play, physical activity as a subject in school, extracurricular physical activities and the time spent on each activity. Additionally, questions about the hours spent in front of different displays were used as an indicator of sedentary behavior (personal computer/video game console/television), both on workdays and weekends. Parents, legal guardians or caregivers of schoolchildren collected this information through active observation except for physical activity as subject in school and extracurricular physical activities. In the last two cases teachers provided the information. In the case of active plays, information provided by schoolchildren on the activities carried out during the break has also been taken into account. When all of the information was collected, the reported mean of PA hours per week was calculated (activities that increase the energy expenditure above the BMR (Basal Metabolic Rate) according to the Compendium of Physical Activities (2011). These activities included: active play, PA performed in school and extracurricular physical activities) and sedentary behaviors (hours spent in front of the PC (personal computer), video game console and TV (television)). To estimate the time spent on PA, a classification of children into “actives” was created for those participating in greater than the WHO's recommendation of 1 hour or more of performing moderate to vigorous physical activities to maintain health from 5 to 17 years. However, children were classified as “sedentary” when they spent more than 2 hours performing this type of sedentary activities, according to the latest guidelines of the Australian government for children within the same age range. Afterwards, a classification of the study sample was performed by accounting for these two variables to establish the four following different groups: “sedentary active”, “non-sedentary active”, “sedentary inactive” and “non-sedentary inactive”. ## Dietetic study Parents or legal guardians filled out a dietary record about the consumption of food and beverages by the schoolchildren over 3 days (2 weekdays and 1 weekend day). The days chosen to collect the information were Thursday and Friday as weekdays and Sunday and weekend day. The questionnaire collected the amounts of food and beverages consumed and the time of each intake. In the case of lunch on weekdays that was not consumed with parents, the information collected was checked against the menus provided by school catering service. Once collected, the data were processed in DIAL software for nutritional assessment. The results were focused on estimating the dietary water intake (coming from food and beverages, expressed in ml/day), the percentage of water coming from beverages, and the ml/day of different beverages consumed, from water as a beverage, sugary drinks, sweetened drinks, sports drinks, natural juices, commercial juices and nectars, milk and milkshakes. Specifically, water as a beverage is defined as water in the form of tap water, mineral water and sparkling mineral water. The dietary water intake classification was established on the basis of the Adequate Intake (AI) by the European Food Safety Authority (EFSA), which is 1,600 ml/day for 4-8-year-old children, 1,900 ml/day for females 9 to 13 and 2,100 ml/day for males aged 9 to 13. ## Urine testing A 24-hour urine sample was collected on a weekend day, and it coincided with the dietary record. All of the participants received written instructions and individual containers for the correct collection of the sample. All micturitions were stored immediately in preservative-free, 1-L plastic containers at temperatures \<-12°C before transfer to the laboratory. The volume as well as the urinary urea, sodium, potassium and creatinine were determined. To confirm appropriate collection of 24-h urine, the correlation between urinary levels of creatinine and muscular mass of each subject was taken into account. Fat-free mass was calculated bearing in mind the creatinine excreted over 24 h in urine using the following equation: $$Fat‑free\ mass\ \left( {kg} \right) = 0.02908\ X\ creatinine\mspace{2mu}\left( {mg/day} \right) + 7.38$$ The osmolality was calculated according to the following formula: Osmolality = \[{sodium (mEq/l) + potassium (mEq/l)}\*2 + urea (mg/dl)\]/5.6. The children were considered to have adequate hydration (AH) when the osmolality was equal to or lower than 800 mOsm/kg in accordance with the criteria of other authors. ## Statistical analysis The data were analyzed using the statistical software IBM SPSS Inc. (version 21.0). Descriptive analysis of the values shows the mean ± standard deviation (m±sd). The Kolmogorov-Smirnov test was used to test the normality of the variables. Student's t-test was used to study the normal variables, and the Mann-Whitney U test was used for those that were not normal. For the categorical variables, the Z and Chi-Square proportion tests were used. The Bonferroni test was used to adjust the values in multiple comparisons. To evaluate the influence of the lifestyles, a 2-way ANOVA was performed on the urinary osmolality and different biochemical parameters, and the water dietary intake and beverages were explored with a post hoc Bonferroni analysis. The association of an individual’s sex, compliance with the practice of PA and sedentariness (independent variables) and risk of dehydration (dependent variable) were analyzed using a logistic regression analysis to calculate three models of the odds ratio (OR). In addition, the OR adjusted by sex and the OR adjusted by sex and other lifestyle factors were calculated. Statistical differences were assigned for a value of p \<0.05. # Results The sample description is shown in. The males performed significantly more hours of total physical activities and extracurricular activities, in addition to a greater number of hours of PC/video game console/TV use than females. To this finding, we must add that the percentage of males who complied with the PA guidelines is higher than the percentage for females, without significant differences in compliance with the sedentary guidelines. shows a description of the different parameters related to hydration and the dietary water intake studied here. The results showed that male schoolchildren had a greater excretion of creatinine, urea, sodium and potassium, in addition to higher urinary osmolality. Children with an inadequate hydration status (≤ 800 mOsm/kg) (IH) had higher sodium excretion (145,8±53,4 mEq vs. 119,4±46,2 mEq/24h vs. in AH, p\<0.001), higher urea excretion (19,1±5,7 g/24h vs 15,9±4,7 g/24h in AH, p\<0.001), while potassium excretion was similar. IH children had also lower diuresis (760,4±232,7 ml/24h vs. 1012,6±290,3 ml/24h in AH, p\<0.001), while dietary water intake was similar in both groups. We performed a linear regression analysis to analyze the risk of dehydration considering all these variables. A higher diuresis was a protective factor of dehydration (OR = 0.948 (95%CI = 0.925–0.972)), while urea excretion (OR = 4.174 (2.096–8.315)) and sodium excretion (OR = 1.164 (1.078–1.257)) were risk factors of dehydration. On the other hand, water intake was not associated to the risk of dehydration. shows the biochemical and dietary results according to four lifestyle groups. With regard to the biochemical data, “non-sedentary inactive” schoolchildren presented a significantly higher percentage of adequate hydration. In the sedentary groups, creatinine and sodium excretion was higher. Notably, greater osmolality appeared in those who were more active. shows crude and adjusted ORs that relate different aspects of lifestyle to the risk of dehydration. After adjusting the OR, the only variable that maintained its association with a greater risk of dehydration (OR = 1.756 (1.008–3.060), p = 0.047) was PA (when performing 1 hour or more a day). However, the results showed that neither complying with the PA recommendation nor with the sedentary recommendation was related to non-compliance of the AIs (PA: sex-adjusted OR = 1.118 (0.568–2.000), p = 0.784 and sedentariness: sex-adjusted OR = 2.242 (0.930–5.406), p = 0.072). # Discussion The 24-h urine osmolality result was similar to that in other studies, in schoolchildren, as performed using the same methodology in different countries such as France, Egypt, Greece, Portugal and the United States. We found that males had higher osmolality values than females. This result has been supported by several studies such as Ebner & Manz in German schoolchildren (4–14 years old) from the DONALD study, Kavouras et al. (Greece schoolchildren from 9–13 years old), Bonnet et al. (French schoolchildren, 9–11 years old) and in Portuguese schoolchildren by Rodríguez et al. (9–10 years old) and Padrao et al. (7–11 years old). There are some factors that could explain these disparities between males and females. First, males show higher intake of food and energy, while females show higher intake of water-dense food, specifically fruits and vegetables. Females also have a lower protein, sodium, potassium and phosphorus intake and, consequently, lower total consumption of solutes. Additionally, the different PA patterns between the sexes could play a significant role on these disparities; before and after puberty, males also show a higher PA and higher non-renal water losses. Our results show that the percentage of schoolchildren who presented an osmolality ≤ 800 mOsm/kg was similar to the findings of other studies that employed the same cut-off point as Stookey et al. (37% females and 34% males from Los Angeles and New York, from 9 to 11 years old) and Gouda et al. (43% total, 44.8% females and 41.8% males from Egypt, from 9 to 11 years old). Regarding the dietary data, the mean dietary water intake showed similar results to those of the ANIBES study on Spanish schoolchildren from 9 to 12 years old, the Cuenca Study (Spanish schoolchildren from 9 to 11 years old) and other international studies in Mexico (5–11 years old), Canada (4–8 and 9–13 years old) and France (6–11 years old). Finally, our results were lower than those of Kavouras et al. in Greek schoolchildren, when using 2-day beverage records to collect the quantity, the type and the timing of consumption for water and different beverages that could influence the disparity between the results. Sodium intake can also influence hydration status since the excretion of sodium excess requires the excretion of water. The best method to assess sodium intake is to measure the excretion of sodium in urine for 24 hours. The sodium intake in our population is higher than that recommended and it was even higher in IH children. Urea excretion is also higher in IH, suggesting a higher protein intake. And finally, although water intake was similar in both IH and AH children, the former had also lower diuresis, which suggest that IH children could have higher extra-renal losses of water. In evaluating the adequacy for the AI of the EFSA, our results showed lower percentages than those reported by Kavouras et al., which included 48% of schoolchildren above the AI (50% females and 47% males). This difference could be based on the disparity between methodologies used to collect dietary water intake information. ## Physical activity In accounting for the PA performed by the schoolchildren, more than 60% met the recommendations for total PAs. This percentage falls within similar ranges to those obtained in the ANIBES Study in the 9- to 12-year-old cohort (51.6%; of those, 61.1% were males and 37.9% were females). Males had higher adherence to the recommendations, spending more time performing active play and extracurricular physical activities. Other authors also found statistical differences in these types of PAs according to sex, although comparisons between studies in no easy become there may be cultural aspects that affect boys’ and girls’ PA differently. Ishii et al. studied a sample of Japanese schoolchildren; they observed the same differences in the type of PAs performed and the leisure physical activities based on sex, which were lower in females than in males. Yamamoto-Kimura et al. studied Mexican schoolchildren; their results showed that males practiced more extracurricular physical activities than females. Other relevant aspects of the results associate compliance with the PA recommendation and being a boy with a higher risk of IH in the study sample. Males may have a higher risk because of their body composition, which can present differences even at these ages, with more body fat and less body water than females. However, adherence to PA guidelines may be a risk factor for IH if the water intake does not increase proportionally to the losses associated with higher PA (in the case of our study, there are no differences in the water intake depending on the lifestyle). Additionally, other socioeconomic factors such as accessibility to beverages may also be influence on the hydration status, so they have to be considered as other potential risks of IH. As far as we know, the effects of compliance with the PA guidelines on the state of hydration in schoolchildren have not been evaluated. Similar data have been found in the PHYSMED study conducted in a Spanish adult population aged 55 to 88 years old in which, similar to the results obtained in the present study, significant differences were observed in the plasma osmolality between the "sedentary inactive" and "non-sedentary active" groups. These data highlight the importance of raising awareness about the hydration status in schoolchildren and their environment in which the primary source of hydration should be water as a beverage. It is especially important when it comes to implementing a strategy to promote the performance of PA, with nutritional education playing a key role. Cleary et al. showed a clear example of this promotion; they evaluated a sample of female adolescents who played volleyball as an extracurricular sports activity. These girls had IH on average before and after sports practice. This situation reversed over the period of time (4 weeks) during which they were subjected to nutritional education in the field of hydration for sports. During this training period, the results showed a suitable improvement in the individual hydration state after PA practice due to correct fluid replacement. Unfortunately, once the training was completed, the adolescents again had IH levels before and after sports practice. Kavouras et al. designed an intervention program to increase fluid intake in children who exercise with heat. In this study the children had free access to fluids and water accessibility was facilitated in training, dinning and resting areas. During the intervention period hydration status and performance in an endurance run was improved. In both cases, the fluid replacement regime was developed by the health professionals who were conducting the study due to the lack of standardized hydration guidelines adapted to different sports. In this sense, it is vitally important to bear in mind that post-exercise hydration recommendations may vary depending on the sport and intensity. These specificities are not included in the current hydration guidelines. This finding suggests that fundamental support for this group, in addition to training, should be a commitment by professionals involved in sports, health professionals and the family environment. On the other hand, it is interesting to highlight the need to implement other strategists as promoting water consumption in schools in order to encourage AH in this group. ## Sedentary behaviors Our results indicate that sedentary behavior, unlike PA, did not increase the risk of IH, nor did it increase the risk of non-compliance with adequate water intake. The lifestyle of the studied population included a low percentage of schoolchildren who complied with the sedentary guidelines. These results are consistent with those obtained by Rodríguez-Huertas et al. in Spanish schoolchildren aged 6 to 12 years in which only 30% performed less than 2 hours a day of sedentary leisure activities and data from the ALADINO study, which showed that 37.3% of the population dedicates less than 1 hour, 33.3% dedicates approximately 1 hour, and 21.7% dedicates 2 or more hours to these sedentary activities. In addition, our results show that males commit significantly more of their time to PCs/video game consoles/TVs compared to females. De Jong et al. have obtained similar results in their evaluation of the television and computer use habits of two school-aged population groups, from 4 to 8 years and from 9 to 13 years. Their results showed that computer use by this population increased as the age increased and with the presence of computers in the bedroom for males in both age groups. Regarding the relationship between sedentariness and beverage consumption, it is very difficult to compare with other studies since some of them studied different sedentary behaviors and their relations to the consumption of drinks, but they do not show results for guideline compliance in general. The systematic review by Pearson et al. concluded that television viewing was positively associated with the consumption of energy-dense beverages among other dietary patterns, both in schoolchildren and adolescents. In our country, the ALADINO Study showed that more than 2 hours of screen time per day had a significant association with the increased frequency of consumption of smoothies, fresh fruit and sugary drinks. It is of special relevance to note that these results may not be applicable to other subjects or population groups, but they provide an interesting line of research to evaluate the influence of lifestyle factors on hydration. Among the strengths of our study, we highlight the size of the study sample, which is a healthy school population, not athletes, and the collection of 24-h urine to determine the hydration status, which makes it possible to quantify both the diuresis in addition to the osmolality. However, self-reported data on PA and sedentary lifestyles are subject to error, as is the 3-day dietary record, and thus, there may have been problems in terms of the over/underestimation of some of the aspects studied here. In addition, the cross-sectional design of the study does not allow us to establish a cause-effect relationship between variables. # Conclusions Roughly half of the individuals studied here presented an IH status, and this percentage was higher in males than females. It is important to encourage schoolchildren to be active, and it also essential to reinforce the importance of hydration to these groups, taking into special account the awareness of the people in their environment in helping them to achieve this goal and in acquiring hydration habits to help them maintain an adequate hydration status. # Supporting information [^1]: I have read the journal's policy and the authors of this manuscript have the following competing interests. RU works for Coca-Cola Iberia as Health and Nutrition Director. This does not alter our adherence to PLOS ONE policies on sharing data and materials. AP-G, RMO, AMLS have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work.

# 1 Introduction Seeing the need for information in society, television still becomes a high- demand mass media. Though television still survive amid the emergence of new mass media platform recently, but the increasing number of new platforms such as Youtube and Netflix have raised challenges. As a result, television channels as part of business must maintain and improve their ratings to survive and generate profits. One well-known rating system is Nielsen Ratings. It is a system for television channels developed by Arthur C. Nielsen, a market analyst through his organization, Nielsen Media Research. A high Nielsen rating represents more viewers and eventually advertisements. Therefore, this has an impact on the profits generated by the television channel. Television still maintains a high level of viewing and is anticipated to continue for at least the next few decades, despite the claim that it is a conventional media that has been disrupted by online media. Even Wolff acknowledges that some aspects of television media, such as authenticity and reliability, cannot be replicated by digital media in his book "Television is the New Television: The Unexpected Triumph of Old Media in the Digital Age". Furthermore, television that already exists has a chance to compete with digital media, which is thought to just digitalize what already exists and has not been able to develop the characteristics of broadcasting and content development. Additionally, user-generated content on digital platforms has a lower quality of information and tends to be more amateurish than professional news reporting on television. Additionally, television still becomes the primary option for viewers who reside in distant places without access to the internet as an entertainment and information source. The internet, as is well known, needs a network from an operator with a currently constrained service area. In the meantime, 20% of Indonesia’s population is still without internet access. Television channels are ranging from government and private-owned, national and international channels, but practically all of them provide news programming, as news is a daily necessity that people cannot live without. People increasingly monitor both national and international television stations for news as a result, television stations must adapt their programming in light of globalization. The news reporters are undoubtedly impacted by this, who must be alert and perceptive when choosing worldwide phenomena that worth broadcast. Employee creativity may be influenced by perceived organizational support, a proactive personality, and meaning of work. Previously, a link between work engagement and perceived organizational support was also proved. According to a former study, proactive personalities and work engagement go hand in hand. Likewise, as asserted in other research there was a connection between employee creativity and work engagement. Thus, this study attempts to elaborate those studies by taking into account five variables, including perceived organizational support, proactive personality, meaning of work, work engagement, and employee creativity, that have been empirically correlated in other studies. The employee’s view of the organization is considered important in determining performance at work. Perceived organizational support is defined as a general belief that the organization cares about the contributions and welfare of its employees, thus making employees feel safe and comfortable in the organization. Thus, this study will discuss further perceived organizational support and its effect on field news reporters. Then, another assumption arises those proactive individuals have better work performance than those who are not proactive. Furthermore, suggests that a proactive personality refers to someone who is always looking for opportunities, shows initiative, takes action when needed, and maintains an attitude until the desired change is achieved. In his work titled "Man’s Search for Meaning," Viktor Emil Frankl asserts that meaning plays a central role in the lives of an individual. Similarly, in the workplace, the meaning of work will undoubtedly influence work performance. Meaning of work can determine how employees interpret their work and organization, how they do their duties, and what they experience within the organization. Furthermore, research conducted by suggests work engagement as an employee’s attachment to their work, which motivates them to exert more effort since they view the work as an extension of themselves. Due to the significance of creativity in the entertainment and content creation industries, especially television programming, it is crucial to undertake a study examining the elements that drive creativity. Creativity is a crucial quality for a field reporter to possess. This is due to the fact that field reporters are frequently confronted with unforeseen circumstances, such as sudden changes in coverage plans, which demand reporters to utilize their inventiveness. In addition, when conducting research on coverage material, reporters must be able to create news targets that are relevant and engaging in order to raise ratings while preserving aspects of authenticity and truth. Several field reporters, including television reporters, indicated that they were responsible for researching coverage material, conducting field research, arranging coverage treatment, carrying out production, and making rough cuts or rough editing. This demonstrates that a field reporter must be adaptable and creative in order to endure shifting field conditions. Despite the critical role that television news reporters play in the news reporting process, there are unfortunately still very few studies that concentrate on them. Since television news reporters must communicate the news in auditory and visual formats that viewers can directly perceive, which raises the need for creativity on their part, the role of a news reporter comprises challenging obligations and responsibilities. Therefore, it is vital to comprehend the practical factors that influence television news reporters’ creativity. This study aimed to perform additional research on television news reporters in light of the numerous earlier studies that have focused on creativity. As previously mentioned, other researches have sought to identify variables that may significantly influence employee creativity, but this study is the first to use the model proposed here. It is expected that it will be either theoretically useful for studies involving related variables or practically useful for insights on how to stimulate employee creativity. # 2 Literature review ## 2.1 Theoretical basis ### 2.1.1. Perceived organizational support The definition of perceived organizational support is an employee’s general acknowledgment of the extent to which the organization values the contribution and welfare of employees who are perceived to provide meaning and purpose to the employees’ lives. According to perceived organizational support is also known as the global trust that an employee has made regarding their assessment of organizational policies and procedures. Perceived organizational support demonstrates the psychological process that occurs when employees get knowledge about their social environment, based on organizational support theory and social exchange theory, which is regarded as the primary theoretical foundation. According to perceived organizational support is related with organizations’ awareness regarding the extent to which employees’ involvement is valued and how employees’ view their organization having a concern for their welfare. Therefore, employees with high perceived organizational support tend to build a better life purpose and meaning in their career It can then be emphasized that perceived organizational support is employees’ perception of organizational policies, the justice given to employees and how the organization pays attention to socio-emotional needs. Perceived organizational support is also a form of employees’ expectations after what they have given to the organization. ### 2.1.2. Proactive personality Proactive personality show individual behavior to encourage change, recognize opportunities, and control situations to take advantage of those opportunities. Individuals with proactive personalities exhibit a personal disposition to engage in active role orientation, such as initiating change and manipulating their environment, which makes the employee promise change through continued action until significant changes in the achievement of their goals occur. According to that individuals with a proactive personality is a person who is relatively unconstrained by situational forces and capable of influencing environmental change. The proactive personality type can reflect the extent to which individuals tend to find opportunities to make changes in the workplace and act in bringing about these changes. Hence, proactive personality is dubbed as "an anticipatory action that employees use to influence themselves and/or their environment". It should be stressed that individuals with proactive personalities constantly seek change in order to further their aims. Additionally, proactive personality in this study is defined as a personality trait that prefers to take the initiative when confronted with particular circumstances or activities. ### 2.1.3. Meaning of work Meaning of work is defined as a balance or alignment between employee characteristics and employee expectations, which occurs when employees dedicate themselves to value and meaning at work. The study also states that the meaning of work is a construction that depends on the characteristics of individuals and their experiences, as well as their perceptions of how work has a meaning. According to meaning of work is a fundamental factor that determines how members interpret their work and organization, how they perform their duties, and what they experience in the organization. Meaning of work can contribute to an individual’s perception of being motivated, fulfilled, and committed in life in general. Thus, the meaning of work can be expressed as an alignment between the beliefs, values, and behavior of employees, which occurs when the work target is following the values or standards of the employees themselves. A person’s level of work meaning reveals how much they value the purpose and utility of their employment. As a result, this study considers meaning of work as a gauge of how much employees believe their work is significant, worthwhile, and important in their life. ### 2.1.4. Work engagement Work engagement has been considered as an important predictor of employee attitudes, performance, and behavior. Also supported by who stated that employee work engagement is important because it determines employee well-being and is associated with many motivating work-related outcomes. Work engagement is defined as a positive work-related state of mind that individuals perform related to work characterized by vigour, dedication, and absorption. In the workplace, vigor increases energy levels and psychological toughness. Dedication is characterized by drive, enthusiasm, and challenge. Absorption is characterized by a person’s complete focus on one task and his refusal to interrupt it.. Employees with work engagement are enthusiastic, energetic, and fully absorbed in their work and act to advance the goals and prestige of the organization. It may be noted that previous research examined work engagement in depth, starting from the factors that underlie it to something that can be altered by it. This study makes use of Schaufeli’s concept of work engagement, that is a productive state of mind manifested by people who exhibit energy, dedication, and absorption. ### 2.1.5. Employee creativity The notion of creativity refers to creative work that is produced and is considered a new work that is accepted to be a work that can be maintained or useful and satisfying by a group at one time. Employee creativity is defined as the conceptualization, production, and development of new and useful ideas, as well as processes and procedures by employees or a group of people working together. Creativity involves bringing something new to an organization, which may mean something unique, and unusual such as originality, thinking outside the box, new perspectives, and contributing something that didn’t exist before. Employee creativity can find the hidden needs of others and handle problems creatively and effectively, which ultimately creates superior performance. Thus, employee creativity is capable of conceptualizing, producing, and developing new and useful ideas, involving the ability and capacity of employees to create and develop new and useful thoughts about company products, services, and procedures. The definitions include creative solutions to problems or creative business strategies and creative changes in business processes. Employee creativity is, in other words, the degree of creativity that employees possess in relation to the generation of ideas and ideas. Employee creativity, however, goes beyond just that. It also has to do with how well people use their skills and aptitudes to solve issues quickly and effectively and make decisions. Additionally, this study used Tierney’s definition of employee creativity, which takes into account each person’s assessment of their own level of originality. The definition is thought to be the most acceptable because this study cites two components, intrinsic and extrinsic, that are connected to employee creativity. Engagement, the meaning of work, and a pro-active attitude are intrinsic variables, whereas external elements connected to organizational support. ## 2.2 Hypothesis development ### 2.2.1 Perceived organizational support and employee creativity Perceived organizational support can meet various socio-emotional needs, including affiliation, self-esteem, emotional support, and social approval which can be correlated with creativity and innovation, because perceived organizational support can affect employee creativity through intrinsic and synergistic extrinsic motivation. According to the study, employees with perceived organizational support are likely to increase their sense of responsibility to achieve organizational goals which refer to a creativity that is formed from setting difficult goals, choosing challenging tasks, and persisting when obstacles are encountered. It is known that in a competitive business environment, organizations need to encourage the creative performance of employees to survive, and perceived organizational support plays an important role in employee creativity, as it tends to increase the likelihood of creative outcomes. According to perceived organizational support can increase employee creativity behavior by increasing employee interest in their work. By referring to components related to creativity and innovation which are dynamic, have assumed that individuals who receive support from perceived organizational support will change their intrinsic and extrinsic motivation. Research conducted by has also revealed that there is a positive correlation between perceived organizational support and employee creativity through studies conducted. This study needs to examine how perceived organizational support affects employee creativity, especially on news reporters at Indonesian private television stations because in previous studies, the subjects studied were mostly centered on the hospitality, banking, and even multisector sectors. From the descriptions mentioned above, this study hypothesizes that: H1: Perceived Organizational Support has a significant effect on Employee Creativity ### 2.2.2 Proactive personality and employee creativity Employees with a proactive personality can initiate changes within the institution to achieve the desired goals, one way to do this is by manipulating work arrangements to improve performance and stimulate innovation. Research by also argues that a workforce with a proactive personality can seek every opportunity to detect new methods, demonstrate their skills, and explore modern work techniques, therefore, employee creativity is the right result of a proactive personality. Individuals with a proactive personality do not passively adapt to all aspects of the environment in which they live but are motivated to find new and better solutions for various procedures or processes that they consider ineffective in improving their current situation. Hence, individuals with proactive personalities have a real desire to actively shape the surrounding environment to better suit their needs, and proactive personalities are more likely to show creative behavior. According to proactive personality is significantly related to employee creativity. Research by suggests that proactive personality affects employee creativity through informal leadership formed by individuals who have proactive behavior. Individuals with proactive personalities Those with stronger inner thoughts and feelings tend to adopt positive strategies when they engage in the problem-solving process and can generate new ideas, and as a result, generate further resource gains such as employee creativity. From the descriptions mentioned above, this study hypothesizes that: H2: Proactive Personality has a significant effect on Employee Creativity ### 2.2.3 Perceived organizational support and meaning of work Employee empowerment in the form of perceived organizational support can have a positive impact on the meaning of work. This is evidenced in research which found a correlation between perceived organizational support and meaning of work where employees who received perceived organizational support felt that their work had more meaning and meaning. According to that the more individuals feel the meaning of work, the higher the likelihood that their perceived organizational support will increase. This can happen because perceived organizational support shows employee perceptions related to the extent to which the organization recognizes and appreciates its efforts and values, and cares about its welfare in an organization. If employees feel that they are supported by their organization, they will contribute more to organizational outcomes as a way of responding to that organizational support, and they will feel empowered by the support, knowledge, resources, and opportunities such as formal and informal strengths provided by their organization. to be able to feel work becomes more meaningful. From the descriptions mentioned above, this study hypothesizes that: H3: Perceived Organizational Support has a significant effect on the Meaning of Work ### 2.2.4 Perceived organizational support and work engagement Perceived organizational support reflects a growing type of support that can generate a sense of obligation to care for the well-being of the organization and help the organization achieve its goals based on the norm of reciprocity. The research shows that repeated good treatment received from the organization can increase the employee’s sense of obligation to help the organization achieve its goals. Therefore, it makes sense that employees who have perceived organizational support high, leading to work engagement that can contribute more to job performance, compared to those who have perceived organizational support low. According to perceived organizational support has a significant effect on work engagement. This is shown by the relationship when employees believe that their supervisors and organizations care about their well-being, they are more likely to fulfill their obligations by being more involved in the workplace. From the descriptions mentioned above, this study hypothesizes that: H4: Perceived Organizational Support has a significant effect on Work Engagement ### 2.2.5 Proactive personality and meaning of work Individuals with a proactive personality can work hard to control and manipulate the environment to pursue new information and practices to improve performance. The employee will make active efforts not to give up in the face of challenges by taking the initiative. Employees with this proactive personality also always try to take jobs and position themselves in organizations where they can carry out tasks with a meaning of work high. Such people will also take risks, if necessary, to find new jobs that align with their personality traits. Until states that individuals with a proactive personality will give a greater value to the meaning of work. has also proven that a proactive personality has a significant effect on the meaning of work. Because some of these individuals already have openness to new experiences and jobs, they will be attracted to jobs and organizations that provide a high level of. From the descriptions mentioned above, this study hypothesizes that: H5: Proactive Personality has a significant effect on the Meaning of Work ### 2.2.6 Proactive personality and work engagement It is known that proactive personality has a positive correlation with work engagement as research has suggested that proactive personality is related to work engagement in reciprocal relationships. According to proactive personality must be positively related to work engagement because it will show individuals with a proactive personality involved in a work environment also tend to immerse themselves in their work. Furthermore, the actions of individuals with a proactive personality include always wanting to continue to learn about their work by identifying new ways to describe their strengths, and relentlessly reaching out to others such as co-workers, leaders, and customers, and it represents greater work engagement. It is also supported by the statement of who stated that a proactive personality is a rather stable personality characteristic about showing initiative, persistence to bring about meaningful change and identifying opportunities and acting on them as anticipatory actions that employees take to influence themselves and/or their environment leading to work engagement. at workplace. From the descriptions mentioned above, this study hypothesizes that: H6: Proactive Personality has a significant effect on Work Engagement ### 2.2.7 Meaning of work and employee creativity Research conducted by has found that creativity increases along with acceptance and appreciation that their work has meaning. That way, it proves that there is a correlation between the meaning of work and employee creativity. Agreement between individual values and goals as well as organizational values and goals will increase the meaning of work, which makes employees more likely to display desired behaviors such as employee creativity. Besides, when the meaning of work is high, then it can increase the contribution of employees to the organization by increasing other intrinsic motivations. Employees will likely find it difficult to develop creative ideas because their cognitive resources are insufficient to develop creative skills for their work area. According to meaning of work can become the target of an individual because it can support and encourage them to achieve integrated integrity which also leads to employee creativity. However, different subjects of the study will certainly produce different detailed findings. From the descriptions mentioned above, this study hypothesizes that: H7: Proactive Personality has a significant effect on Employee Creativity ### 2.2.8 Work engagement and employee creativity Employees with work engagement more often experience positive emotions that expand their intellectual and psychological resources which encourage them to explore new unconventional ways of doing their jobs. This positive psychological experience can motivate employees to learn, assimilate new information and produce ingenious solutions and ideas and subsequently make them have higher energy needed to pursue creativity than those who are not involved. To support this argument, shows a significant relationship between work engagement and creativity. This is also supported by who has researched the relationship between work engagement and creativity. According to the study, work-engaged employees often experience positive emotions that expand their momentary thought-action repertoire processes and generate personal resources, and those positive emotions facilitate creative behavior by cultivating a thirst to explore and assimilate new information. From the descriptions mentioned above, this study hypothesizes that: H8: Work Engagement has a significant effect on Employee Creativity ### 2.2.9 Meaning of work mediating perceived organizational support and employee creativity According to intrinsic motivation tends to have a direct impact on employee creativity, meanwhile, extrinsic motivation will only increase if employees perceive their work to have meaning. This is following the statement that " meaningfulness is regarded as an important component of intrinsic motivation " which means that meaningfulness is an important component of intrinsic motivation. Therefore, when employees perceive that their work has meaning, their intrinsic motivation and creativity also increase continuously. When employees feel the meaning of work, they will consider it an important component of intrinsic motivation which also has the aim of utilizing the abilities, talents, and creativity of employees. As long as the organization delegates authority that affects employees to take advantage of these opportunities, they will have perceived organizational support that makes them feel that they must always try harder to cooperate with management. Supported again by the meaning of work, able to increase the desire of employees to be creatively involved. So that employees will be encouraged to have a desire to try harder to be creatively involved. Employees who feel limited and do not get organizational support tend to find it difficult to find the value of their work as a result there may be decreasing in the potential for creativity and innovation. Furthermore, in his research which also observed the relationship between perceived organizational support and employee creativity through the meaning of work as a mediator found that organizational support contributed to an increase in work meaningfulness which then had an impact on increasing employee creativity. Therefore, in this study, further observations will be made of how perceived organizational support affects employee creativity with the meaning of work as a mediator. From the descriptions mentioned above, this study hypothesizes that: H9: Meaning of Work mediates the relationship between Perceived Organizational Support and Employee Creativity ### 2.2.10 Work engagement mediating perceived organizational support and employee creativity The sense of belonging to the organization causes employees to evaluate their work by considering the interests of the organization which makes employees prefer to link their interests with the organization as a common interest to develop trust in increasing their creativity for the organization. According to employees who experience perceived organizational support will interpret it twice as much as an organization that shows value and investment in its employees, and the organization’s concern for their welfare. Furthermore, employees will know their position as well as job content, and they will know their shortcomings well enough to create many good ideas for improvement that arise if their creativity is motivated. Higher work engagement is observed to be positive when employees strongly agree with the organization, based on social exchange theory they will generate motivation to pay for their organization to be better as one of them refers to creativity. has shown that work engagement acts as a strategic driver to increase employee creativity, which is also able to be a driver in motivating employees to be able to focus their intelligence and increase their creativity. So from the findings of the experts above, it can be seen that organizational support has a significant impact on work engagement and the results then increase employee creativity. However, it is known that the mediating role of work engagement has not been supported by strong empirical evidence due to the lack of research. From the descriptions mentioned above, this study hypothesizes that: H10: Work Engagement mediates the relationship of Perceived Organizational Support to Employee Creativity ### 2.2.11. Meaning of work mediating proactive personality and employee creativity Organizations mostly demand enthusiastic and proactive employees. Employees with a proactive personality are known to be more enthusiastic about improving their work performance through their efforts, and they are more determined to identify opportunities to achieve their goals without feeling constrained by internal or external constraints. Employees who find their work meaningful and important will believe that they can feel creative, especially individuals who already have a determination to have a proactive personality. Meaning of work is known to depend on the suitability of the diversity of the job to the personality of the employee. Therefore, it is possible that employees who have a proactive personality also have high creativity. This happens because individuals with a proactive personality are better at managing their work roles with expectations, feeling the compatibility between their expertise and work values with the tasks they do in their daily work. In addition, research by found a correlation between proactive personality and employee creativity moderated by the meaning of work. In this study, it was stated that a proactive personality has an indirect impact on employee creativity through the meaning of work. From the descriptions mentioned above, this study hypothesizes that: H11: Meaning of Work mediate the relationship between Proactive Personality and Employee Creativity ### 2.2.12 Work engagement mediating proactive personality and employee creativity Highly engaged employees often feel enthusiasm, excitement, and interest in their work, which then magnifies their thinking and actions. Moreover, enrich their resources by expanding their suggestions and actions. According to being proactive means foreseeing things, stopping problems, and seizing opportunities by initiating efforts to stimulate and bring about change in work and having a different vision for the future such as boosting creativity. The positive emotions and cognitive states of work engagement are also beneficial for proactive employees because they can focus intelligence on their opportunities to increase creativity. A proactive personality has an impact on increasing the tendency of individuals to have high concern and enthusiasm for work. Meanwhile, creativity is related to innovative work behavior and individual capacity to create ideas. Then, work engagement is understood as the relationship between employees and the work in their organization which can later lead to work behavior that is full of enthusiasm and high effort. In short, a proactive personality increases employee engagement, and then the impact affects employee creativity. has shown that work engagement can mediate organizational needs in influencing employee creativity. From the descriptions mentioned above, this study hypothesizes that: H12: Work Engagement mediates the relationship between Proactive Personality and Employee Creativity The hypotheses were then conceptualized in the following framework. # 3 Research methods ## 3.1 Research approach This research was a quantitative explanatory study since variables were examined using hypotheses and path analysis models were employed. An explanatory research model is used to test a theory by collecting evidence that either supports or refutes it. This study’s population consisted of field reporters in the DKI Jakarta region of Indonesia. The sample of this study was 119 news reporters who were selected through a purposive sampling technique based on the following criteria, (a) members of the Alliance of Independent Journalists (AJI) and the Indonesian Television Journalists Association (IJTI); (b) reporters from 14 national private television stations; (c) a permanent news reporter; and (d) more than two years tenure. The journalists who were members of AJI and IJTI were selected because AJI and IJTI are professional organizations that oversee the television journalist profession hence the members were considered credible. Furthermore, the data collection process was carried out by distributing questionnaires using Google Form media. The reason for choosing news reporters who worked for private television stations was because private television stations are organizations that rely on advertising as their main income. In contrast, state television stations receive a budget from the state and have a tendency to broadcast what is mandated by the state. Thus, observing creativity on state-owned stations is not appropriate. As this is a non-interventional study, Universitas Airlangga’s Development and Innovation Institute of Publishing Journal and Intellectual Property Rights (LIPJIPHKI) determined that no ethical approval was needed. This institute is accountable for conducting research and guiding the development of new research products for the benefit of the community. This Institute has the power to provide ethical approval for Universitas Airlangga academics’ research. Following the organization’s policy, the chief executive, who represented the organization as a whole, also provided written informed consent. The consent was also validated by the Development and Innovation Institute for Publishing Journal and Intellectual Property Rights at Universitas Airlangga (LIPJIPHKI). However, the participants were also informed that their information would be kept strictly confidential and used for research purposes only. ## 3.2 Measurement The independent variables in this study are Perceived Organizational Support (POS), and Proactive Personality (PP), then the dependent variable in this study is Employee Creativity (EC), and the mediator variables used in this study are Work Engagement (WE) and Meaning of Work (MOW). Perceived organizational support was measured with 8-item, one sample item was “The organization is very concerned with the welfare of employees”. Furthermore, Proactive personality was measured by 10 items, one of them was “I saw opportunity earlier than anyone else”. Meaning of work was measured with 15 items, such as "Work is seen as giving meaning to the individual" which represents importance of work, "Individual understands the job well" which represents understanding of work, "Individual understands the direction and purpose of work" which represents direction of work, and “Individuals often do not understand the function of work ®” which represents the purpose of work. Work engagement was measured with 17 items, such as "Individuals feel empowered and have power towards their work" representing vigor, "Individuals have pride in their work" representing dedication, and "Individuals have a high focus at work" representing absorption. Then, Employee Creativity was measured with 9 items, such as “Individuals have a desire to try new things”. All of the questionnaire items mentioned above had been tested for reliability in previous studies and had a Cronbach alpha value above 0.60. ## 3.3 Data analysis technique This study used Partial Least Squares-Structural Equation Modelling (PLS-SEM) as an analytical technique to determine the overall interaction between variables and emphasizes hypothesis testing in this study. PLS was used because it is a sophisticated analytical method that can be applied to various sizes of data, meaning the required sample size does not have to be large. This is in accordance with the number of samples in this study which amounted to 119 which was a relatively small number. In addition, no previous research had evaluated this research model in the setting of news reporters, hence the theory was still weak/had not been extensively tested. In PLS-SEM, there are two stages that must be completed: assessing the measurement model or external model by evaluating the validity and reliability of the structure of each indicator, and testing the hypothesis with the bootstrap resampling method and test statistics. ## 3.4 Data analysis The respondents in this study were 119 field news reporters from a total of 14 private television stations in Indonesia. The characteristics used included age, gender, agency, last education, and marital status. Based on, the majority of respondents were male (58%) and aged between 30–40 years old (50.4%). The last degree that they possessed were bachelor’s degree (84%), and most of them were married (65.5%). In this study, to use PLS-SEM for data analysis, it is necessary to go through an outer model evaluation and an inner model evaluation. ## Outer model evaluation The outer model evaluation goes through an evaluation of convergent validity, and composite reliability. In looking at the evaluation at the convergent validity stage, outer loading has a function to describe how big the relationship between indicators is with each construct. Based on , all indicators on research variables already had an outer loading value \> 0.50. Furthermore, the results of the convergent validity and internal consistency tests are shown in. Therefore, all indicators were valid in measuring perceived organizational support, proactive personality, the meaning of work, work engagement, and employee creativity variables and meet convergent validity to be used for further analysis. After that, the discriminant validity test was carried out by comparing the value of the cross-loading item measuring the variable itself with other variables. The item is said to be valid if the cross-loading value is smaller in other variables. The results of the cross-loading calculation are presented in below. It was known that all indicator items met discriminant validity because they had the largest cross loading value for the variables they form (bold values) and smaller in other variables. Thus, all indicators on perceived organizational support, proactive personality, meaning of work, work engagement, and employee creativity variables met discriminant validity. The next evaluation in the outer model analysis was internal consistency. It tests the indicators consistency in measuring a construct. Internal consistency in PLS can use two measures, namely Cronbach’s alpha and composite reliability. shows that the internal consistency value of each research variable had a Cronbach’s Alpha value of more than 0.60 and a Composite Reliability value of more than 0.70. Thus, it could be concluded that each variable perceived organizational support, proactive personality, the meaning of work, work engagement, and employee creativity met good reliability. ## Inner model evaluation Inner model evaluation goes through an evaluation of the Coefficients of Determination (R2), Q-Square, and Path Coefficient assessment stages. The adjusted R2 value for the meaning of work variable is 0.483, as shown in. This indicates that the influence of perceived organizational support and proactive personality on the meaning of work is 48.3%, which falls within the moderate category. Adjusted R-value 2 for the variable work engagement is 0.413, indicating that the influence of perceived organizational support and proactive personality on work engagement is 41.3% and belongs to the moderate range. Adjusted R-value2 for the employee creativity variable is 0.949%, which places perceived organizational support proactive personality, meaning of work, and work engagement in the large category (substantial). In addition, the measurements of Q2 were evaluated using a blindfold test, and a model is considered predictively relevant if the coefficient of Q2 is greater than 0. (56). demonstrates that the Q2 values met the criterion of a value greater than zero, which qualified them as having a medium to large predictive relevance. The predictive relevance of the concept of employee creativity is high, indicating that the variables perceived organizational support, proactive personality, the meaning of work, and work engagement were strong predictors of a field news reporter’s employee creativity. The hypothesis testing stage with path coefficient is the stage to determine whether or not there is a direct relationship between exogenous variables and endogenous variables used in research. Based on, it is known that in the 2-tailed test, the research hypothesis can be accepted if the t-count (T-statistic) 1.96 or p-value is smaller than the error rate (α) 5%. Thus, there were eleven accepted hypotheses and one rejected hypothesis. Furthermore, all exogenous variables, both perceived organizational support and proactive personality, were significantly mediated by the meaning of work and work engagement. In this case, the meaning of work and work engagement fully mediated the influence of Proactive Personality on Employee Creativity. Meanwhile, the two mediating variables partially mediated the effect of Perceived Organizational Support on Employee Creativity. The results of the comparative analysis of the mean and total effect variables concluded that to increase the creativity of field news reporters, the priorities from the highest to the lowest were meaning of work, perceived organizational support, work engagement, and lastly, proactive personality. # 4 Discussion Based on the results of statistical tests, it showed that perceived organizational support had a positive and significant effect on employee creativity with original sample = 0.125 (positive), T-statistics = 3.812 (≥1.96), p-value 0.000 (≤α = 5%). This was in line with research conducted by which revealed that there was a positive correlation between perceived organizational support and employee creativity. It indicated that the greater the organizational support felt by the reporter, the more reporters’ creativity increased. Perceived organizational support could increase employee creativity behavior by increasing the reporters’ work interest, as well as in changing their intrinsic and extrinsic motivation. Reporters with perceived organizational support were encouraged to increase their sense of responsibility to achieve creativity by facing difficult goals, challenging tasks, and persisting when obstacles are encountered. The findings were in line with former research which proved that perceived organizational support had a significant impact on employee creativity working on various sectors. Therefore, this study bolstered the empirical evidence of earlier studies indicating that perceived organizational support is relevant to employees’ creativity regardless of company sector.Subsequent research results show proactive personality no significant effect on employee creativity with original sample = 0.037 (positive), T-statistics = 1.337 (\<1.96), p-value 0.182 (\>α = 5%). This is not in line with the research of who found that proactive personality was significantly related to employee creativity. The results of this study indicate that the higher the proactive personality of reporters, it does not have a real impact on increasing their creativity. One of the reasons is the existence of a code of ethics that regulates the work of news reporters. Reporters who work at private television stations in Indonesia are always required to be creative because their field of work relies heavily on creativity. Therefore, reporters will always create their creativity even if it is not in harmony with their proactive personality. Then the results of the next statistical test showed that perceived organizational support had a positive and significant effect on the meaning of work with original sample = 0.442 (positive), T-statistics = 6.718 (≥1.96), p-value 0.000 (≤α = 5%). This is in line with the research by which found a correlation between perceived organizational support and the meaning of work. The results of this study indicate that the greater the organizational support felt by the reporter, the higher the meaning of a job for the reporter. Perceived organizational support shows the perception of reporters working at private television stations in Indonesia related to the extent to which their place of work has recognized and appreciated their efforts and values, and has even cared about their well-being in an organization. Thus, it will trigger reporters to feel work becomes more meaningful. The results of the next study showed that perceived organizational support had a positive and significant effect on work engagement with the original sample = 0.426 (positive), T-statistics = 5.998 (≥1.96), p-value 0.000 (≤α = 5%). In line with research perceived organizational support has a significant effect on work engagement. The results of this study indicate that the greater the organizational support felt by the reporter, the higher the reporter’s job involvement. The perceived positive value of the influence of perceived organizational support triggers reporters to increase the sense of obligation of employees to help the organization achieve its goals and to be able to contribute more to work performance through the resulting work engagement. Therefore, the findings in this study also support previous research which found a positive effect of perceived organizational support on work engagement in courier logistics companies. Thus, the implications of perceived organizational support on work engagement prove relevant to be applied to a wider subject. In this case, previous studies have proven on the subject of logistics couriers and the results have been shown to be consistent with this study with the research subject of news reporters. Furthermore, the results of the next statistical test showed that proactive personality had a positive and significant effect on the meaning of work with original sample = 0.341 (positive), T-statistics = 5.769 (≥1.96), p-value 0.000 (≤α = 5%). In line with the research of which states that individuals with a proactive personality will give a greater value to the meaning of work. The results of this study indicate that the better the reporter’s proactive personality, the higher the meaning of a job for the reporter. Reporters with a proactive personality will always try to take a job and position themselves in the organization well which makes them feel a high meaning of work. Then the results of further research showed that proactive personality had a positive and significant effect on work engagement with original sample = 0.299 (positive), T-statistics = 4.860 (≥1.96), p-value 0.000 (≤α = 5%). In line with research which states proactive personality significantly affects work engagement. The results of this study indicate that the better the reporter’s proactive personality, the higher the reporter’s work involvement. Proactive personality refers to showing initiative, and persistence to bring about meaningful change and identifying opportunities, and acting on them as anticipatory actions taken by reporters to influence themselves and/or their environment which can lead to work engagement in the workplace. The results of the next statistical test showed that the meaning of work had a positive and significant effect on employee creativity with original sample = 0.553 (positive), T-statistics = 5.752 (≥1.96), p-value 0.000 (≤α = 5%). This is in line with the research of which states that the meaning of work can support and encourage integrated integrity that leads to employee creativity. The results of this study indicate that the more reporters feel the meaning in their work, the higher the reporter creativity. When the reporters feel the meaning of work is high, they can increase other intrinsic motivation that can make it easier for them, or influence reporters to develop their creative ideas because they feel their work is meaningful. Then the results of further research showed that work engagement had a positive and significant effect on employee creativity with original sample = 0.321 (positive), T-statistics = 3.431 (≥1.96), p-value 0.000 (≤α = 5%). In line with the research of which shows a significant relationship between work engagement and creativity. The results of this study indicate that the higher the reporter’s work involvement, the higher the creativity. The work engagement felt by the reporters made them experience positive emotions more often that could motivate them to learn, assimilate new information, and produce ingenious solutions and ideas to pursue creativity. Furthermore, the results of this study on the relationship of indirect influence indicate that perceived organizational support is partially mediated by the meaning of work and work engagement. This shows that increasing the creativity of reporters can only increase perceived organizational support, but if the meaning of work and work engagement is also increased, the reporter’s creativity will increase even higher. Because perceived organizational support can have a direct impact on increasing reporter creativity, or indirectly through mediator variables. The results of this study indicate that organizational support contributes to improving the meaning of work and work engagement of reporters well. so that reporters are encouraged to have an impact on increasing employee creativity. Furthermore, a proactive personality is fully mediation by the meaning of work and work engagement. This finding is different from some previous studies stated that proactive personality significantly affected employee creativity. Proactive personality cannot affect employee creativity in news reporters, it can be caused by several factors. First is the existence of a journalistic code of ethics that limits and regulates the work of a news reporter. Even though a news reporter has a strong proactive personality, it does not mean he can change or exceed the provisions set out in the journalistic code of ethics. Then secondly, there are policies and business orientations of television stations that will determine what kind of broadcast will be broadcast to their viewers. This shows that increasing reporter creativity cannot only rely on a proactive personality but must also be supported by the meaning of work and work engagement so that reporter creativity can increase. Because efforts to increase creativity without paying attention to the meaning of work and work engagement the effect is not optimal in creating creativity. This can happen because the meaning of work can make reporters believe that they can feel creative, and when they feel work engagement, the reporters will focus their intelligence on their opportunities to increase creativity. # 5 Conclusions and implications ## 5.1 Conclusion Based on the results and discussions that have been carried out in this study, the conclusions that can be drawn from a direct influence are that perceived organizational support has a positive and significant effect on employee creativity, meaning of work, and work engagement. These findings reflect that news reporters who get support from the television station where they work show a higher level of creativity. This is because news reporters feel justice and also feel needed, so that their performance is better and ultimately has an impact on creativity. Not only that, support from television stations on news reporters will create emotional and cognitive bonds of reporters on their work. Therefore, the level of work engagement on news reporters is high. Then, the meaning of work is related to the level of news reporters in interpreting their work as valuable and valuable. Furthermore, the support from the television station where the news reporter works makes news reporters interpret their work as meaningful. This is because the support of television stations, especially psychologically, provides encouragement for news reporters to feel meaningful in their work Then proactive personality has a positive and significant effect on the meaning of work and work engagement but not on employee creativity. This is possible because even though news reporters have a proactive personality and try to change their environment, there is a code of ethics and binding work rules. For example, news reporters must write news according to the vision and mission of the television station concerned. In other words, proactive personality and intention to bring about change in the environment cannot have a direct impact on the creativity of reporters because they are limited by the code of ethics and rules of television stations. Furthermore, the meaning of work and work engagement has a positive and significant effect on employee creativity. Meanwhile, the indirect effect shows that all of the hypotheses are accepted, indicating that perceived organizational support and proactive personality are significantly mediated by the meaning of work and work engagement on employee creativity. This shows that if news reporters who get organizational support have attachments and consider their work as valuable, then this perception leads news reporters to be more creative in their work. Not only that, news reporters who have a proactive personality also allow them to be more engaged with work and feel the meaning of their work. In the end, this has an impact in the form of increasing the creativity of news reporters. Reporters work in the television industry which has a very important role in creativity and content creation. Therefore, we need a study that looks at the factors that influence the creativity of reporters in achieving organizational goals well. All the results of this study indicate that reporters have been able to create creativity well in doing their work and in achieving organizational goals with perceived organizational support, proactive personality, the meaning of work, and work engagement. All variables used contribute well and are in line to preserve and enhance creativity. ## 5.2 Managerial implications The results of this study can be taken into consideration for employees and television companies in creating and increasing employee creativity in the workplace. It is hoped that organizations need to increase their concern with employee welfare because it is their assessment of organizational policies and procedures, then employees are also expected to be able to show their ability to see opportunities earlier than others as a form of being proactive in the workplace. In this way, employees also need to view their work as being able to give meaning to each individual from what has been provided by the organization. When employees feel that their work is important, they should feel good involvement with their work until they feel the desire to work starting when they wake up. The main thing is that the organization needs to create revolutionary ideas in a field, and employees should pay attention to all the factors that can be positive for themselves to be able to help the organization make this happen. ## 5.3 Theoretical implication Based on the discussion that has been described, this research can also be used as a recommendation for further research to analyse the effect of perceived organizational support and proactive personality on employee creativity with the meaning of work and work engagement as intervening variables in other organizational contexts. In addition, further research can also explore other determinants that are thought to be able to strengthen employee creativity. This is useful in developing new ideas that are useful for creativity and the innovation process that is needed by the company in creating value and working conditions that are full of creativity. [^1]: The authors have declared that no competing interests exist.

# Introduction Adaptation to local environments is commonplace in nature and drives the evolution of novel ecotypes. Whether adaptive phenotypic divergence plays an important role in the origin of new species has been the subject of much discussion. In this case, adaptation to different ecological conditions should entail a marked fitness reduction for ecotypes outside the habitat to which they are adapted and it should contribute to reproductive isolation. Of particular interest are case studies in which the selective agents and the traits under selection have been identified, because then the mechanisms that link adaptation to the emergence of reproductive barriers can be studied and their importance assessed. For example, some populations of *Rhagoletis pomonella* fruit flies have recently switched from hawthorn to apple hosts and have adapted their phenology and fruit odour preference accordingly. In stickleback fish (*Gasterosteus aculeatus*), the repeated evolution of distinct trophic ecotypes within lakes is mirrored in the intricate, adaptive divergence of cranial morphology. The intertidal snail *Littorina saxatilis* produces shells with different tickness and aperture in wave-exposed versus crab-rich sections of the shoreline, which provide better protection against dislodgement and predation, respectively. These adaptations contribute to reproductive isolation, because each ecotype is less fit when transplanted into the respective other habitat, because F1 hybrids are at a disadvantage and/or because the adaptations cause assortative mating. A genomic dissection of such divergent traits can not only pinpoint the genomic regions under selection but also investigate genetic linkage between ecologically selected and mating traits or, for example, explore the role of inversions in novel adaptations. Ecological divergence has long been assumed to generate barriers to gene flow in the hybrid zone of European fire- bellied toads, for which genomic resources have recently been developed. Here we seek experimental evidence for adaptive divergence in tadpole traits as a first step in establishing a link between ecology and reproductive isolation. In contrast to the very young taxa referenced above, the red-bellied toad (*Bombina bombina)* and the yellow-bellied toad (*B*. *variegata)* are more anciently derived \[MRCA \~ 3.2 mya, \]. Nevertheless, they interbreed wherever they meet and produce a wide range of recombinants in typically narrow (2–7 km wide) hybrid zones. Classic cline theory suggests that the mean fitness of intermediate hybrid populations is reduced by 42% compared to that of pure populations on either side. There is experimental evidence for endogenous selection against hybrid embryos and tadpoles. Profound ecological divergence is also strongly implicated. *B*. *bombina* reproduces in semi-permanent lowland ponds, such as flooded meadows and oxbows that fall dry in winter \[e.g. \]. *B*. *variegata* lays eggs in more ephemeral habitat (‘puddles’) and sometimes also in small steams, typically at higher elevations. Consequently, hybrid zones tend to be located at ecotones. There is a long list of taxon differences that presumably adapt both taxa to their current habitat \[reviewed in 22\]. For example, adult *B*. *variegata* have a relatively stronger skeleton, longer legs and thicker skin, traits which presumably aid their frequent movements over land, to track the ever-shifting mosaic of suitable breeding sites. Ponds are not only more predictable year-on-year but also allow for larger breeding aggregations. With their well-developed vocal sacs, male *B*. *bombina* produce much louder mating calls and form choruses that can be heard from afar. A *B*. *variegata* male would likely be at a disadvantage if it joined such a chorus. Here we focus on tadpoles, because they experience the strongest habitat contrast. Faced with the ever present risk of desiccation, *B*. *variegata* tadpoles hatch from relatively larger eggs, grow faster and metamorphose earlier. They also spend relatively more time swimming and feeding. The less active behaviour of *B*. *bombina* tadpoles should protect them against visually hunting predators in ponds. These taxon differences reflect a general pattern across anurans. More permanent aquatic habitats have a greater abundance and diversity of predators. Anuran species that reproduce in them have typically quiescent tadpoles that take longer to reach metamorphosis than those that breed in ephemeral sites. More active tadpoles species are typically more susceptible to predation. This robust pattern has been interpreted as a trade-off: tadpoles can either be highly active and maximise food intake for fast growth and development in temporary habitats or avoid predators through quiescent behaviour in more permanent sites, but not both. Tadpole activity and growth rate covary as expected in evolutionary contrasts but evidence for a causal relationship is mixed at the intra-specific level. However, the trade-off between rapid growth and predator avoidance is almost universal irrespective of whether activity is the mechanism that links the two. We therefore expect that *B*. *variegata* tadpoles are at greater risk of predation than *B*. *bombina*. Nevertheless, *B*. *variegata* tadpoles do sporadically encounter predators such as newts, dragonfly and dytiscid larvae, water scorpions and salamander larvae. Thus, tadpoles would maximise their chance of reaching metamorphosis by adjusting their phenotype to the local balance of risk. Indeed, a wide range of anurans display phenotypic plasticity in response to predators. Almost invariably, tadpoles reduce their activity level when they perceive chemical cues that indicate predator presence. This response can occur within minutes. Longer term exposure to such cues can cause an increase in refuge use, alter larval morphology, prompt the development of colour spots on the tail, and slow growth and development. These effects are seen across venues from the laboratory to the field. The commonest and most closely analysed morphological response to predators is the development of a relatively deeper tail fin, usually in combination with a shorter tail and a shallower body. A deeper tail may facilitate rapid escape (faster burst swimming speed) or lure attacks away from the more vulnerable body. While there is evidence that natural variation in tail shape affects tadpole swimming performance, studies to date have given stronger support to the lure hypothesis. Irrespective of the particular mechanism, there is evidence that predator-induced phenotypes are eaten less than predator-naïve ones and that survival is correlated with measured traits that show a plastic response. A recent study of *Rana temporaria* makes a particularly strong case for adaptive phenotypic plasticity in both behaviour and morphology, including an assessment of the variability in predation risk within and between ponds and over time that could bring about such a flexible strategy. Both *B*. *bombina* and *B*. *variegata* respond to the presence of predators with reduced activity and by developing a relatively deeper tail fin \[, see also \]. Exposure to chemical predation cues slows their growth and development under laboratory conditions. While there is at present no evidence that the induced phenotypes are better protected from predators, they should be used for a fair comparison of relative predation risk, because tadoles would develop with predators in nature. A previous study using predator-naïve tadpoles, while showing the expected result, could therefore not fully settle the issue. In a second laboratory experiment with predator-induced tadpoles, a strong population effect overwhelmed any taxon differences. In both cases, significant effects were observed only in the earlier of two sets of predation trials, which suggests that relative susceptibility may change over time. Here, we carried out laboratory experiments that improve on the design of the previous two studies by using predator-induced phenotypes throughout, by increasing the level of replication within each taxon and by estimating relative predation risk over a four-week period, from hatching to the time when the first *B*. *variegata* tadpoles start to metamorphose. Dragonfly larvae were used as predators, because they are ubiquitous in ponds and also found in puddles. We also assessed the contributions of activity, body size and tail shape to the observed differences in predation risk. Our experiments confirmed our expectation that pure *B*. *variegata* suffer a higher predation rate than pure *B*. *bombina* due to a higher activity level. In the discussion, we argue that elevated activity contributes to the gene flow barrier across the hybrid zone which keeps the two gene pools from merging into one. # Materials and methods ## Animal collections, pairing scheme and toad care Adult toads were collected from Moravia (Czech Republic) where the Carpathian lineage of *B*. *variegata* meets the Southern lineage of *B*. *bombina*. The experiment was carried out in 2017. Starting in early May, aquatic habitats were visited regularly (14 known *B*. *bombina* sites and 19 known *B*. *variegata* sites). Due to an unusually cold spring, toads did not appear in appreciable numbers until early June and were then promptly collected to ensure that females had not yet laid eggs. Collection sites were located at least 6 km from known areas of hybridisation and deemed 'pure' based on the taxon-specific spot pattern on the toads' bellies. For each taxon, minimum and maximum distances between sites were 1.5 and 43.6 km (*B*. *bombina*) and 0.7 and 14.9 km (*B*. *variegata*). *B*. *bombina* sites were ponds with a minimum width of 5 m and a maximum depth greater than 1 m. They all contained predators (incl. *Aeshna cyanea*, *Natrix natrix*, fish) and, with one exception (site E), they had abundant aquatic vegetation. *B*. *variegata* sites were either water-filled wheel ruts (max. depth \< 30 cm) or in one case (site 2) the shallow end of an abandoned swimming pool. They were devoid of aquatic vegetation and three of nine sites contained no predators at all (see for details). Random numbers were drawn to pair males and females within sites whenever possible. However due to unequal sex ratios per site, four *B*. *variegata* pairs involved toads from separate, but nearby sites (, maximum distance between sites per pair: 0.96 km). Each pair was housed in a 52 x 35 cm plastic box with 7L of well water, a polystyrene island and three pieces of plastic aquarium plants as support for egg batches. As a precaution against the potential spread of *Batrachochytrium* infection, all waste water from the entire experiment was heated to 100° C and treated with UV (Eheim UV reflex UV800) before disposal to kill any fungal spores. On 20 June, a belly swab was taken from each toad to test for *Batrachochytrium dendrobatidis (Bden)* and *B*. *salamandrivorens (Bsal)* infection. The former pathogen is widely distributed in the Czech Republic, but so far no population declines have been observed. The qPCR followed the protocol of Blooi et al. using a Roche Light Cycler480. No evidence of either infection was found. We note that samples in the same qPCR run but from an unrelated study scored positive for *Bden* infection. The adults were subsequently returned to their exact collection sites. ## Predators We used *Aeshna cyanea* larvae as predators. This dragonfly is widely distributed across Europe and was present in most of the habitats we surveyed (see above). Two hundred larvae were collected from one site near Jihlava (Vysočina County) and were kept in groups of 15 in 50L plastic boxes. They were fed every other day with Tubifex worms before the predation trials started and with tadpoles thereafter. ## Oviposition To induce reproduction, 250 I.U. of human chorionic gonadotropin, dissolved in 0.2 ml of amphibian Ringer solution, were injected into the lymph sac of an outer thigh of both males and females on 6 June. Most pairs produced egg clutches overnight. Over the following two days, 30–40 eggs per family were transferred into the rearing set-up (see below). The remaining offspring were raised in the oviposition boxes, with weekly water changes. They were used as food for dragonflies in order to condition the water in the rearing set-up. The adults were moved to new boxes of the same size. Tadpoles hatched on 12–13 June. In the following, we refer to 13 June as day 0 of the experiment. ## Tadpole rearing We constructed two gravity flow circuits. For each of them, five plastic tubs (52 x 37 x 16 cm) were interconnected with silicone tubing such that water could be pumped from a 20 L bucket at floor level into the top tub and then descend stepwise via the other four tubs before returning to the bucket. Inflows and outflows in each tub were positioned diagonally opposite to maximise mixing. Each tub held 27L of water at a depth of 14 cm. In order to maintain uniform water quality, the pumps (Eheim StreamON+ 2000) were operated 24 hours per day at a low flow rate and per circuit 7 L of water were replaced daily. Throughout the experiment, water temperature was 22–24°C and the air temperature was 24–26°C. All water changes for adults and tadpoles were carried out with well water heated to 22°C and sterilised with UV (Eheim UV reflex UV800). The upper four tubs in each circuit were subdivided with fibreglass insect mesh into eight rearing compartments (size: 18 x 11.5 x 16.5 cm, 64 compartments in total). A plastic aquarium plant was added to each compartment for structural complexity. At the start of the experiment, each compartment held 10 hatchlings from a single family. Families (nine per taxon) were distributed over either three or four compartments in total and either one or two compartments in each circuit. Exactly half of the compartments in each circuit were assigned to each taxon. Placement of families within circuits was decided by random draws. Tadpoles were fed boiled nettle (*Urtica dioica*) leaves *ad libitum* and uneaten food was removed every day. In order to simulate predator presence, we let chemical cues of tadpole predation circulate from day 0 onwards. For this, we used the fifth tub in each circuit to feed 1–2 non-experimental tadpoles (depending on size) to one dragonfly larva every other day. Tadpoles of both taxa were used for this in equal proportions. ## Predation trials Predation trials started on day 2 when the tadpoles were active. They were carried out in two plastic boxes subdivided with fibreglass insect mesh into four sections (26 x 26 x 18 cm), each with two plastic aquarium plants (akin to *Myriophyllum*), placed into diagonally opposite corners. They contained a 1:1 mixture of well water and water from the rearing circuits. For each trial, four tadpoles (2 *B*. *bombina*, 2 *B*. *variegata*) were placed into one section and allowed to acclimatize for 30 min. A dragonfly that had not been fed on the previous day was then added and the trial was stopped when the first tadpole had been eaten (maximum duration: 45 min). The identity of the missing tadpole was determined from digital photographs of the animals taken before the trial (see below). The survivors were returned to their rearing compartments. The trials were recorded with a Microsoft LifeCam VX-2000 video camera suspended over each plastic box. Predation trials were carried out on 18 days as follows: two consecutive experiment days were followed by one ‘off-day’ (1 exception); this pattern was repeated nine times until day 29 post hatching, when the first *B*. *variegata* tadpoles entered metamorphosis. There were 20 trials per day between 10:00 and 15:00 hours. A custom Perl script was used to haphazardly draw the composition of each trial per day such that (a) four different families were represented, (b) one tadpole per taxon came from circuit 1 and the other from circuit 2 and (c) each rearing compartment contributed at least one tadpole on a given day. However towards the end of the experiment, a few rearing compartments were empty and the contribution of others had to be increased accordingly. No individual tadpole or dragonfly was used more than once per experiment day. ## Morphometrics Digital images were taken by placing a tadpole into a glass cuvette and photographing it from the side with a Canon EOS 400D camera (lens: Canon EF 28-90mm). tpsDIG2 (F. James Rohlf) software was used to take the following measurements from these images: total length, tail length, body height and maximum tail height. ## Behaviour From the video footage of the predation trials, we recorded the behaviour of a focal tadpole during one 30 s interval per trial for half of the trials on a given day. A custom Perl script provided the pseudo-random start point of the observation interval and allowed behaviour (swimming, feeding, resting) to be recorded with separate key strokes. From one trial to the next, we alternated between a *B*. *bombina* and *B*. *variegata* tadpole. Taxon identification was based on size in the early trials. Morphological differences were used as soon as they became apparent. Footage from the first two days was not used, because the taxa could not be reliably distinguished. We discarded all observation intervals that overlapped with a predator attack. We could not determine whether the focal tadpole survived the trial, because that would have increased the data collection time roughly 10 to 20-fold. Moreover, it was difficult to track an individual over several minutes, especially when it temporarily disappeared together with others under a plastic plant. We also recorded for 68 predator attacks (four on each of 17 experiment days) whether they were initiated by and directed towards an actively swimming tadpole. ## Permits Collection permits were granted by the relevant regional offices (South Moravia and Zlín) of the Czech Ministry for the Environment. The experimental protocol was approved by the ethics committee of the Institute of Vertebrate Biology, CAS, permit no. 15/2016. ## Statistical analyses ### Estimates of predation risk We estimated predation risk per rearing compartment (= the unit of observation, containing offspring of one family) as the number of tadpoles eaten divided by the number of valid trials in which that compartment was represented. This figure subsumes the ontogeny of tadpoles in terms of changes in size, body shape, behaviour and, possibly, experience in predators avoidance. Recall that surviving tadpoles were returned to their rearing compartment and were likely picked again in later predation trials. Just as in nature and in the classic mesocosm studies, in which tadpole cohorts interacted with predators for a substantial part of larval development, older tadpoles were likely to have survived previous predator encounters. Specific to our design is that we staged these encounters and collected additional data about each one of them. Especially with a small number of trials, our estimate of predation risk may be biased if tadpoles from a certain compartment happened to be assigned to trials with, say, particularly well defended congeners. There is a range of statistical methods to estimate an individual's skill from a limited number of contests with opponents of varying strengths (as e.g. in chess), starting with the model of Bradley and Terry. However, none of these match our trial design of a single winner (or rather: looser) and a 'tie' for the other three. We therefore checked for biases due to a limited number of trials as follows. Each trial *k* involves tadpoles from four (of 64) compartments. The risk context *rc*_ik of the tadpole from compartment *i* (1 ≤ *i* ≤ 4) is the mean of the predation risks *r*_j of the other three compartments that are represented: $${rc}_{ik} = {\sum\limits_{j = 1,j \neq i}^{4}\frac{r_{j}}{3}}$$ For a given compartment *m* (1 ≤ *m* ≤ 64), the average over its associated *rc*_ik estimates in valid trials is the mean risk context, *rc*_m. If our estimates of predation risk, *r*_m, are unbiased, then the estimates of risk context from the observed dataset should be very similar to those from a much larger set of trials. To test this, we haphazardly drew compartments for another 2550 trials with the same script as before and used these ‘mock trials’ to re-compute risk contexts using the *r*_m estimates from the actual experiments. We used linear mixed models to investigate the effect of taxon on predation risk and included ‘circuit’ and ‘family’ terms to properly represent the non- independence of compartments. Circuit was treated as a fixed effect, because it had only two levels. We did not include a population term in the model, because populations were unevenly represented and some families had parents from different sites. Any differences between populations should appear as family effects. For these, we fitted not only random intercepts but also allowed the effect of ‘circuit’ to vary among families (random slopes). Model were fitted with the R module *lme4*. We used the ‘REML = FALSE’ option in order to compare models via a likelihood ratio test. ### Morphological correlates of survival Logistic models were used to investigate the effects of taxon, tadpole size and tail shape on survival per trial. The input was generated by sorting all valid trials in chronological order and picking from them alternately (a) a surviving tadpole at random or (b) the non-surviving tadpole. For the entire dataset, the four morphological measurements were log-transformed to improve normality and entered into a principal component analysis (PCA). We treat the first principal component as a measure of tadpole size. From a regression of log(tail height) on PC1 we extracted the residuals as a measure of relative tail height (≈ tail shape). To ensure shared allometry between the taxa, we inspected pairwise plots of the four measured variables. For the focal tadpole per trial, PC1 and residual log(tail height) were expressed as a deviation from the trial mean. All analyses were carried out in R v.3.4.4. # Results ## Taxon differences in predation risk We assessed relative predation risk over a four-week interval. Our experiments started two days after hatching when tadpoles were uniformly at Gosner stage 23. On the last experiment day, *B*. *bombina* tadpoles were mostly at stages 35–36 whereas the commenest *B*. *variegata* stages were 39–40 with some individuals entering metamorphosis (stages 42+). In total, 360 trials were conducted. Of these, no tadpole was eaten in 97 trials, more than one was eaten in five trials and in three trials the two taxa were not evenly represented. Discarding these, 255 valid predation trials remain. The proportion of valid trials in which a *B*. *variegata* tadpole was eaten was 0.65. The taxon difference was particularly strong for 5–12 day old tadpoles, when around 77% of the tadpoles eaten were *B*. *variegata*. There was also a slightly higher risk for tadpoles from circuit 1 as opposed to circuit 2 (proportion = 0.56). This effect was largely restricted to *B*. *bombina*. The number of trials in which no tadpole was eaten was smaller in the first half of the experiment (tadpole age: 2–14 days, *n* = 31) than in the second half (tadpole age: 15–29 days, *n* = 66). The mean durations of valid trials were 9.9 min (fist half) and 13.8 min (second half). Predation risk was computed per rearing compartment as the number of tadpoles eaten during valid predation trials over the number valid trails in which this compartment was represented. The observed risk contexts (as defined in the Materials and Methods) closely matched those re-computed for a much larger number of hypothetical trials ($\overline{n}{per}$ compartment = 159.9): for 56 of 64 compartments the observation deviated by less than 10% from the expectation (max. deviation 14%) and in all cases it was well within one standard deviation of the prediction. We therefore use predation risk per rearing compartment as the response variable in the following analyses. The full linear mixed model (model 1) included taxon, circuit and family (random effect, nested within taxon). We allowed ‘circuit’ to affect each family differently (random slopes). As expected, there was an increased risk for *B*. *variegata* (*β* = 0.2069 ± 0.0357) and a slightly decreased in risk in circuit 2 (*β* = -0.0814 ± 0.0376). Family effects (intercepts and slopes) accounted for 13.9% of the variance. The reduced model 2 without the ‘taxon’ term gave a much worse fit to the data, resulting in a highly significant likelihood ratio test (χ² = 19.12, 1 df, *p*-value = 1.228 x 10⁻⁰⁵). There is thus clear evidence that *B*. *variegata* tadpoles were at relatively greater risk of predation by *Aeshna cyanea* larvae. Nevertheless, a large amount of unexplained variation remains. ## Tadpole size and shape Throughout the experiment, *B*. *bombina* was on average smaller with a shallower tail. But after correction for body size, *B*. *bombina's* tail fin was relatively higher than that of *B*. *variegata*. Note that this is a difference between two types of predator-induced phenotypes. It does not tell us about the strength of the plastic response in either taxon, because predator- naïve tadpoles were not part of this study. Our shearing approach to correct for body size is validated by the shared allometry in both taxa. If size and/or tail shape are functionally important, we would expect to see within-taxon correlations with survival as well. The logistic regression analysis (, see for details) confirmed the significantly lower survival of *B*. *variegata* and also showed a strong interaction of tadpole size (PC1) and taxon. There were no significant effects associated with relative tail height. illustrates that larger *B*. *bombina* tadpoles survived better than smaller ones (t = 3.5993, df = 85.589, p-value = 0.0005). But this effect is not consistent across the whole dataset. *B*. *variegata* tadpoles are overall larger and more at risk of predation, whereas smaller size is associated with greater risk in *B*. *bombina*. There is therefore no evidence that a general preference of *Aeshna* larvae for larger prey causes higher mortality in *B*. *variegata* tadpoles. The analogous plot for relative tail height hints at an advantage for higher tail fins, but the present dataset provides no statistical support for this hypothesis. ## Tadpole activity During the 30 s observation intervals, both taxa spent most of their time either resting or, seemingly, feeding. No food was provided in these trials. So, the tadpoles' persistent scraping of surfaces with their mouthparts, especially during the first half of the experiment, probably provided little, if any nutrition. Swimming typically occurred in short bursts of a few seconds separated by inactive intervals. Ninety-two percent of recorded predator attacks were directed at actively swimming tadpoles. The distributions of the time spent swimming were strongly skewed to the right in both taxa. As expected, *B*. *variegata* spent more time swimming than *B*. *bombina* (means: 4.27 and 1.78 s, respectively; unpaired Wilcoxon rank sum test, W = 2234.5, p-value = 0.0096). Focal tadpoles were chosen based on their clear visibility during the 30 s observation interval. Their survival status is unknown. We can therefore not investigate within-taxon correlations between activity and survival. # Discussion In our experimental setting, tadpoles of the puddle breeder *B*. *variegata* are at greater risk of predation from dragonfly larvae than those of *B*. *bombina* that develop in semi-permanent ponds. This result confirms our expectation based on similar species comparisons. It also agrees with the findings of Kruuk and Gilchrist who conducted predation trials with larger groups of predator- naīve *B*. *bombina* and *B*. *variegata* tadpoles that could only adjust their behaviour. Here, we allowed tadpoles to develop a more complete predator-induced phenotype. Nonetheless, the survival difference between the taxa persisted and it was particularly strong during days 5–12. In previous mesocosm experiments with pure and hybrid *Bombina* and free-ranging dragonflies, the overall predation rate declined sharply from day 15 onwards even though tadpoles continued to be attacked \[, see similar results for *Rana*, \]. Thus, the days with the highest total predation risk overall may coincide with a particularly strong disadvantage for *B*. *variegata*. The increase in no-predation trials in the second half of the experiment is consistent with the mesocosm observations. The overall proportion of such trials (0.27) was high in relation to comparable studies. In our previous *Bombina* study using the same design (4 tadpoles, 1 dragonfly), all trials resulted in predation. But the trial compartments were only half the size and so presumably led to more predator-prey encounters. McCollum and Van Buskirk exposed *Hyla versicolor* tadpoles (two per trial, 1.5 times larger boxes) to predators for up to 500 min. Both tadpoles survived in 7% of cases. Our trials may have led to a roughly similar outcome, had they been extended from 45 to 500 min. The fact that all tadpoles managed to avoid predation in a quarter of cases suggests that we did not create a completely ‘hopeless’ and thus unrealistic situation but instead one that even the more susceptible taxon could negotiate successfully. Many predator attacks were unsuccessful, because tadpoles could escape via sprints or agile movements through the vegetation. Many tadpoles were used in more than one trial. Later trials may therefore have involved individuals that benefited from experience gained and/or were inherently better at avoiding predators. We believe that this resembles the situation in nature more closely than the alternative of shielding tadpoles from predator encounters until the day of their one experimental test. As pointed out by Melvin and Houlahan, typical tadpole mortality in nature (approx. 95% according to their review) may be non-random with regard to traits under study. The generally much more benign conditions in the laboratory may thus produce experimental cohorts that are not representative of the population of origin. Moreover, we specifically set out to integrate predation risk over a substantial part of larval ontogeny with all its dramatic changes. We consider any gain in experience to be part of this process. Nonetheless, this close-up view of predator-prey interactions is a piece in a mosaic that should be complemented with experiments in semi-natural settings and using additional predator species. If *B*. *bombina* and *B*. *variegata* were reproductively isolated, these results would address the question why *B*. *variegata* reproduces only rarely in semi-permanent, predator-rich ponds. Whether or not *B*. *bombina* tadpoles are present in these habitats may not matter, because there is little evidence for competition among tadpoles in ponds. Predators tend to keep tadpole numbers sufficiently low so that competition is negligible. *B*. *bombina* then simply provides a benchmark for a successful 'pond strategy'. *B*. *variegata*'s greater susceptibility to predation shows that its tadpoles fall short of this benchmark even when they display the predator induced phenotype. Other traits such as relatively fewer eggs per clutch may also limit *B*. *variegata's* reproductive success in ponds. Hybridisation shifts the focus away from a comparison between distinct types and towards selection on traits in recombinants. These genotypes predominate in *Bombina* hybrid zones, while F1 individuals are very rare, if present at all. A hybrid genome consists of blocks of varying sizes that are inherited from one or the other species. Within the hybrid zone and with each round of recombination, these blocks get smaller, while the influx of pure genomes from the periphery introduces new large blocks. The chance of, say, a *B*. *variegata-*derived block to cross into the *B*. *bombina* gene pool depends on how it affects the fitness of its carriers along the way. A genetic variant that reduces fitness in ponds is likely to be eliminated from this habitat together with the entire *B*. *variegata*-block in which it resides. It is in this sense that such variants act as barriers to gene flow. This reasoning extends to adaptively diverged traits, no matter whether the taxon difference is based on one locus or on several unlinked, additive loci: any maladapted *B*. *variegata* variant would increase the probability that the block in which it resides will be eliminated from ponds. Smooth clines of six quantitative traits mapped across the hybrid zone near Pešćenica, Croatia, indeed suggested an largely additive genetic basis. For highly integrated compound traits, the gene flow barrier may be further strengthened if intermediate hybrids have reduced fitness due to mismatched phenotypes. This has been demonstrated in F2 crosses of sticklebacks (limnetic x benthic ecotypes), where numerous quantitative trait loci underlying the trophic adaptations are known: F2 hybrids with certain intermediate genotypes expressed mismatched phenotypes and grew more slowly. Overall, the contribution of ecological divergence to the maintenance of distinct taxa should increase as the list of traits under selection gets longer. Prey activity alerts visually hunting predators and correlates strongly with tadpole predation rates in inter-specific comparisons and also within species. Our findings confirm this pattern. Because the taxon difference in activity is seen in animals raised under uniform laboratory conditions, it should have a genetic basis. We do not know how many loci are involved but we expect that each *B*. *variegata*-derived genetic variant that crosses the hybrid zone and increases activity in a *B*. *bombina* genetic background is selected against. It thus reduces gene flow at genetically linked loci. For the sake of argument we consider two alternative scenarios. There may be a hybrid genotype with an activity level that increases fitness over that of either pure taxon. This type of hybrid might then contribute to an increase rather than a reduction in gene flow. However, we do not find it likely that activity has an intermediate optimum that has not yet been reached in either taxon. It could also be that a high-activity variant is closely linked to a universally favourable mutation that arose in *B*. *variegata*. The mean fitness of blocks including both variants might then be higher than that of their *B*. *bombina* counterparts, generating a conduit of gene flow in this genomic region. But the effect of the activity variant would still be to slow the introgression of such a block into the *B*. *bombina* gene pool and the ‘pull’ of the favourable variant would cease as soon as recombination separates the two. On balance, these considerations do not alter our prediction that higher activity most likely contributes to the reduction in gene flow across the hybrid zone. There was no evidence that relatively higher tail fins caused better survival in either taxon. The large variances that rendered differences between group means (taxon, survival) non-significant (Figs) may reflect heritable variation among families in the amplitude and type of morphological plasticity. This trait warrants further study using more detailed morphometrics. Finally, our analyses disproved the hypothesis that an overall preference of *Aeshna cyanea* larvae towards larger prey contributed to *B*. *variegata's* greater mortality. Is higher predation risk in *B*. *variegata* a direct consequence of adaptation to ephemeral habitat? Population studies have shown that successful metamorphosis can be restricted in a given year to less than 50% of sites in which eggs have been laid (depending on annual rainfall) and that desiccation is the main cause of reproductive failure. This is reflected in the growth rate of *B*. *variegata* tadpoles which was the fastest in a set of 15 European and North American anurans. In contrast to the situation in ponds, intra-specific competition should be commonplace for *B*. *variegata* as tadpoles grow and puddles shrink. The typically high activity of tadpoles growing in this type of habitat supports this. In the laboratory, large *B*. *variegata* tadpoles displace smaller conspecifics from food patches with their powerful tail movements. Not only is the abundance and diversity of predators that *B*. *variegata* encounters lower than in ponds \[, see also –\], sometimes these early-successional sites are predator-free. This suggests a hierarchy of risk: rapid development takes priority whether or not there are predators. They add to the already considerable unpredictability of reproductive success. Bet-hedging may be a way to sidestep an unresolvable trade-off. Adult *B*. *variegata* can live up to at least 15 years in nature \[, but see: \] and females are known to lay eggs repeatedly over their lifespan. There may be aspects of *B*. *variegata*’s anti-predator strategy that remain to be discovered. *B*. *variegata* frequently develops dark tail spots that may divert attacks away from the body, whereas *B*. *bombina*'s tail fin is always uncoloured. Moreover, the animals may not have been able to express some behavioural strategies in our experimental setting. Nonetheless, our results show that high activity in *Bombina* tadpoles is associated with higher predation risk that causes relatively higher mortality in the more active taxon, *B*. *variegata*, even when the tadpoles express a predator-induced phenotype. Population genetic theory predicts that *B*. *variegata* alleles underpinning high activity will be selected against in typical *B*. *bombina* breeding habitat and should therefore act as barriers to gene flow across the hybrid zone. In the context of ecological speciation (cf. Introduction), the remaining question is whether these taxon differences reflect selection towards different phenotypic optima. Given that quiescent behaviour benefits *B*. *bombina*, what are the benefits of high activity for *B*. *variegata*? If indeed rapid growth and development in a competitive environment require active tadpoles, then the classic trade-off applies: activity mediates two mutually exclusive strategies. In *Bombina*, this could be tested not just by comparing the pure taxa, but also by investigating phenotypic and genetic covariances in natural hybrids and by mapping the component traits in the *Bombina* genome. # Supporting information We thank Oliver Gast and Zuzana Hiadlovská for help in the field and with the experiments, respectively. Jarrod Hadfield, Lumír Gvoždík and Crispin Jordan offered advice on the experimental design and statistics. We thank Günther Gollmann, Jacek M. Szymura and Josh Van Buskirk for comments on an earlier version of the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Department of Biology, Faculty of Education, Masaryk University, Brno, Czech Republic

# Introduction The World Health Organization (WHO) projects that diabetes will be the seventh leading cause of death in 2030. The most common type of this rising disease is type 2 diabetes mellitus (T2DM). Diabetes reduces not only life-expectancy but also life–quality. However, it has become largely manageable due to advances in medication and diet. In addition to these traditional methods, a third viable strategy has received increasing attention, namely physical activity (PA). In this regard, a systematic review suggested that different types of PA interventions can contribute to (potentially clinically relevant) reductions in blood sugar level (HbA1c) of about 0.5 percentage points. The duration of the interventions included in this review was rather long (at least twelve weeks) and some of these interventions comprised a dietary co-intervention. Nonetheless, studies in the area of diabetes management have rarely provided a(n) (PA) intervention with conclusive, translatable results. Based on these studies, the current pilot study applied a randomized controlled field trial with a six week PA intervention to address four common pitfalls. First, in the PA intervention domain, in relation to diabetes management and to health promotion in general, a trend is emerging to elicit sustained behavior change through an amalgam of techniques, whether based on psychological theories or not. This approach makes it difficult to obtain an overarching communication structure, let alone theoretical coherence. Therefore, in the current intervention, the Self-Determination Theory (SDT) was used as a theoretical framework. This framework offers a clear vocabulary with principled touchstones as well as possible means to meet these principles. The theory stems from an organismic view of man, thereby focusing on humans’ potential for growth under circumstances that satisfy three innate, essential and universal basic psychological needs: autonomy–to act volitional (with choice, self-authored), relatedness–to belong to a caring relationship-structure, and competence–to master the environment. These three needs were supported in the current intervention. Several studies have shown sustainable effects of SDT-based interventions on PA behavior, also in clinical populations. Additionally, it has been shown that changes in HbA1c over time can be predicted by SDT-derived variables. Second, many recruitment and screening procedures have prevented participation of people who could benefit the most from (PA) interventions, for example those from minority groups or those with comorbid chronic illnesses. Such people might be reached through their health insurance fund, at least in Belgium, where health insurance is obligatory. Therefore, in the current intervention, staff members from the Christian Health Insurance Fund of Leuven (CM Leuven) took care of the recruitments. They used a novel method, which included personal writings based on a database of patients with T2DM, information sessions and e-mails. Third, participants’ general practitioner (GP) was involved in the current intervention. This approach has been applied in previous interventions only to a limited extend but seems promising, provided that the extra workload for GPs remains easy to implement. In a similar vein, a study showed small effects of a GP-intervention (prescription and written materials) on self-reported PA among inactive routine care patients at short term (6–10 weeks). Fourth, most studies in patients with T2DM that measured regular PA in an objective way, relied on either pedometry or accelerometry. Therefore, in the current pilot study, a multisensory activity monitor (Sensewear) was used to measure regular PA. This device takes physiological measures (e.g., heat flux) into account as well as mechanical ones (e.g., accelerations). In addition, a self-report questionnaire was used for triangulation. We hypothesized that participants in an intervention condition (IC) would improve significantly more than participants in a waiting-list control condition (CC) from baseline to six weeks (intervention duration) and to six months follow-up, on blood sugar level (larger decrease in IC), physical fitness (larger increase in IC) (i.e., health-related outcomes), and on the behavioral outcome of regular PA (larger increase in IC). These outcomes were selected in line with the underlying study rationale of patients with T2DM improving their health through PA. The results from the current pilot study can inform research and health insurance funds regarding large scale rollout. # Methods ## Design The current pilot study consisted of a randomized controlled field trial. We used a 2 x 3 repeated measures design with condition (intervention vs. control) as between-persons variable and time (pre–post and pre–follow-up) as a within- persons variable. The two between-persons conditions were: an intervention condition (IC), in which participants were measured and received a 6-week intervention immediately after the baseline measurement; and a control (waiting) condition (CC), in which participants were measured during the study, as the IC, but would receive the intervention only after the follow-up measurement. Participants in the CC were told that during the waiting period their health measurements were analyzed. All participants had to pay a fee because CM aimed to initiate an economically feasible project that could be evaluated on its public health impact. The measurement points were scheduled at three moments: at baseline, that is in spring 2014 (pre), six weeks after the baseline measurement, when the intervention had ended (post), and six months after the baseline measurement (follow-up). ## Participants The inclusion criteria for participants were: (1) ≥ 18 years of age, (2) member of CM Leuven, (3) intake of oral diabetes medication (≥ three months since 2012), (4) no intake of Byetta/Victoza/insulin medication since 2012 (to exclude patients with type 1 diabetes mellitus), (5) in possession of a Global Medical File, which can be accessed by GPs and contains a medical history (e.g., examinations, medication, specialized care), and (6) resident within a region of about 40 km around Leuven. Based on these criteria, a CM-database yielded 4273 eligible patients. After randomization (per GP, to avoid contamination between conditions), CM sent an invitation letter for an information session to 2310 patients in the IC and to 1963 in the CC. These information sessions were organized at four locations within the region and were identical in both conditions, except for the project schedule. Of all invitees, 146 patients in the IC (6%) and 125 patients in the CC (6%) were present at one of the information sessions. Of these patients, thirty-six eventually subscribed for participation (26 IC, 10 CC). Two patients who were not present at the information session subscribed on their own initiative, based on the invitation letter (1 IC, 1 CC). Because of the low participation rate in the CC after the information sessions, a sample of patients that had not subscribed for the information session were contacted through e-mail and/or telephone. This resulted in ten additional participants. In sum, twenty-seven participants subscribed for the IC, twenty- one for the CC. All forty-eight participants (27 IC, 21 CC) signed an informed consent form and received medical approval by their GP. Thirty-six participants (20 IC, 16 CC) paid the full fee of €124.9, nine participants (5 IC, 4 CC) paid a reduced rate of €74.9 because of their low income, and three participants (2 IC, 1 CC) ended their participation prematurely before payment. The reasons for this premature cessation were practical difficulties (in two occasions) and/or different expectations (in two occasions). Of the forty-five remaining participants, forty-two participants completed the study, twenty-three in the IC and nineteen in the CC. The reasons for drop-out were practical difficulties (*n* = 1) or different expectations (*n* = 1) in the IC, or a waiting period that was too long (*n* = 1) in the CC. summarizes the recruitment process flow. This study (ML9981) was approved by the Medical Ethics Committee of the University Hospitals Leuven (ClinicalTrialsID: NCT02064335). ## General practitioners All eligible GPs were informed about the project by CM through a standardized information letter. They were asked: to screen patients that were interested to participate, to fill out a brief questionnaire with questions about the project quality and feasibility (if they had a participating patient), to list information about the medication of participating patients, to collect blood sugar and cholesterol levels of participating patients and to discuss these levels with them. These levels were based on blood samples incorporated into the three-monthly routine investigations. ## Intervention The current SDT-based PA intervention focused on the satisfaction of participants’ basic psychological needs (i.e., autonomy, relatedness and competence) and consisted of three key elements: (1) an intake and an outtake session with a professional PA coach, who held the degree of Master in Physical Education and Movement Sciences and who was familiarized with SDT and Motivational Interviewing, (2) a personalized PA program and (3) five weekly PA group sessions. The intake and outtake session included measurements, discussions about personal scores for the measured variables of interest (e.g., blood sugar) and the composition of a personalized PA program. The intake session had been preceded by an acquaintance session, one week earlier. In this acquaintance session, the multisensory armband was handed because this device has to be worn for a week to obtain a valid (pre) measurement. The personalized PA program was drafted through dialogue between participant and PA coach, in line with psychological principles from SDT and related practices from Motivational Interviewing. Physiological training principles were taken into account too: duration, frequency and intensity ranges were adapted to the individual fitness level. The type of PA could be easily integrated into participants’ daily life. Walking and cycling were the most selected personal activities. The participants also received self-monitor tools such as a pedometer and a booklet that included a PA diary as well as practical and health-related information. The weekly group sessions were guided by the PA coach and consisted of either walking (*n* = 10), Nordic Walking (i.e., walking with poles) (*n* = 13) or aerobic and strength training (*n* = 2). This type of activity was free to choose for participants. Each session lasted for about one hour and took place at one of the four locations from the information sessions. Based on participants’ location and PA preference, groups were composed, in which the number of participants varied from one to five. All but two participants completed all five PA group sessions. shows how the three key intervention elements are related to the three dimensions of need support, in accordance with SDT. ## Measurements ### Demographics Demographic sample characteristics were determined by means of a questionnaire and are given in for the baseline measurement. The IC and CC differed significantly at baseline with respect to age (Mann-Whitney U: *p* = 0.012) and work status (2-sided Fisher’s Exact: *p* = 0.021). These results mean that the IC contained a higher proportion of retired (older) people. ### Health-related outcomes HbA1c was used as a measure of blood sugar level and determined by means of the blood sample, taken by participants’ GP. Values expressed in mole/mole (IFCC- HbA1c) were converted to a percentage (NGSP- HbA1c), according to the formula: *NGSP-HbA1c = 0*.*915 x (IFCC-HbA1c) x 100% + 2*.*15%*. Distance walked in six minutes was used as a measure of physical fitness and determined by means of the six-minute walk test, guided by the PA coach. In this test, participants had to walk as many 20m-lenghts as possible within six minutes. Physical fitness is considered relevant to health in its own way and has been shown to be amendable through PA interventions. ### Behavioral outcome Regular PA was assessed in an objective way by means of a validated multisensory armband (SenseWear). The armband had to be worn at the upper arm continuously for one week, except during water activities. A minimum of three valid weekdays and one valid weekend day was considered a valid measurement week. A minimum of twelve awake wearing hours was considered a valid day. The output consisted of the weighted mean daily amount (*5/7 x weekday mean + 2/7 x weekend-day mean*) of time spent on PA at intensity above 1.8 MET, in bouts of at least ten minutes. Because of very low scores on vigorous PA in our sample (at baseline: overall *M* \< 1 min/day), we did not differentiate between PA intensity levels. Regular PA was assessed in a subjective way too by means of an adapted version of a self-report questionnaire (Godin Leisure-Time Exercise Questionnaire; \[23), to be completed at home. Participants had to report how many times during the previous week they had been engaging in light (L), moderate (M) and vigorous (V) PA in bouts of at least fifteen minutes. A total (T) PA-score was calculated according to the formula: *TPA = 3 x LPA + 5 x MPA + 9 x VPA*. Within the current study, this total PA-score seemed the most appropriate subjective measure of variation in PA behavior because it could be compared with the objective PA-score and because walking (often classified as light PA) was a popular activity among participants. ### Covariates Age, sex, BMI, climatic circumstances and dietary changes were assessed as covariates. To determine BMI (i.e., body mass/height²), body mass was measured by means of a digital balance (OMRON) and height by means of a stadiometer. On an interpersonal level, BMI can be used as a feasible proximal measure to detect people with overweight, which might hamper PA behavior. Three categories of the climatic circumstances during the week of PA-assessment (objectively) were determined: mean day length (in Brussels), based on data from the Royal Observatory of Belgium (ROM), mean day temperature and total day precipitation (both in Leuven), the last two based on data from the Royal Meteorological Institute of Belgium (RMI). It has been shown that climatic circumstances play a role in people’s PA behavior. Two categories of dietary changes compared with the baseline measurement were determined: sugar intake and total energy intake, both measured by means of a question to participants with a trichotomous answer possibility (less/equal/more). Because of the ordinal nature of the data and because the intervention did not target diet, these variables were not measured as an outcome in themselves. Based on self-report, there were six smokers at baseline (*n* = 45). Only four of them changed their amount of smoking during the study, and only to a minimal extent (\< 2 cigarettes/day). Therefore, smoking information was not included as a covariate. ## Analyses All statistical analyses were performed by use of SPSS 23. The significance criterion was set two-sided, at *p* \< 0.05. No correction for multiple testing was used. In order to compare progress over time (assumed linear), we composed linear mixed models with condition (and the covariates) and time (measurement point) as (repeated, in case of time) fixed factors. We opted for unstructured covariance matrices and restricted maximum likelihood estimations. # Results ## Sample All twenty-three remaining participants in the IC completed the intake and outtake session with the PA coach and the five group sessions. Shapiro-Wilk tests indicated in general that data on HbA1c and self-reported PA were skewed. Therefore, these data were transformed into their logarithm (after addition of 1), which yielded satisfactory improvements in symmetry. ## Short-term effects From pre to post, the hypothesized interaction effects (condition x time) were not significant, which denotes a lack of difference in progress on the outcome variables between both conditions. However, significant (six-week) time effects across conditions were found for blood sugar levels and for self-reported PA. More specifically, the HbA1c levels decreased significantly (on average: -0.19%), while the self-reported PA-scores increased significantly (on average: +6.46 points). These findings were identical for the model without and with covariates. ## Long-term effects From pre to follow-up, the hypothesized interaction effects (condition x time) were not significant. However, significant (six-month) time effects across conditions were found for physical fitness and for self-reported PA. More specifically, the distances walked in six minutes increased significantly (on average: +10.74m), as did the self-reported PA-scores (on average: +11.70 points). These findings were identical for the model without and with covariates. # Discussion and conclusion ## Discussion The current pilot study evaluated the effects of a six-week need-supportive PA intervention on health-related and behavioral outcomes among patients with T2DM at short term (six weeks) and at long term (six months). The intervention took place in a real-life setting and was based on the principles of SDT. Positive and sustained effects were expected on blood sugar levels, fitness levels and regular PA. In contrast to the hypotheses, no significant condition by time effects emerged at short or long term. This suggests that the intervention as a whole did not produce the expected impact. This finding is in line with the results of a recently conducted randomized controlled trial that aimed at the reduction of sedentary behavior among young adults at risk of T2DM. That study did not show any significant intervention effects on HbA1c levels or PA behavior, nor at three, nor at twelve months. However, three notable time effects across the two conditions were established in the current study. In a review of PA interventions, the authors concluded that eight of 29 (28%) studies reviewed reported meaningful improvements in PA behavior of participants in a control condition. They suggested that repeated measurement and participant characteristics, such as a lower BMI and at risk for chronic disease (as opposed to healthy as well as to with chronic disease), were likely explanatory factors. In the current study, a first time effect was indicated by a significant decrease in blood sugar level across the conditions after six weeks. This short- term effect is unlikely due to test familiarization, as HbA1c levels are based on blood samples, taken by a GP. Moreover, this effect is probably not caused by biased missing data either, as data on both HbA1c levels were available for forty participants. A possible explanation for the time effect lies in participants’ temporary focus on healthy living (related to blood sugar level), created by study enrollment. Participants might have become more aware of a variety of healthy behaviors, including PA but also sedentary time, medication adherence, diet etc., all of which might have influenced their blood sugar level. It is important to notice that this focus was not restricted to the IC; even though information and explicit need supportive counseling in the CC were minimized, it was not possible to ignore questions from patients in the CC altogether, nor was it desirable to thwart their basic psychological needs. Moreover, twenty-six patients in the IC and ten patients in the CC were recruited through an information session, which included PA- and health-related advice. Furthermore, the mere testing can provide informational feedback for self-monitoring. This feedback, in combination with the general advice and the contact with the PA coach (e.g., for measurement purposes), might have satisfied these patients’ basic psychological needs to a sufficient extent to increase their functional goal-directed health behavior. A second time effect of the current study was indicated by significant increases in self-reported PA across conditions, both at short term as well as at long term. These effects are also unlikely caused by biased missing data, considering the sample size preservation rates of thirty-eight (83%) and thirty-nine (85%) participants respectively. Given that climatic circumstances were included as covariates, seasonal effects probably do not explain these effects either. Moreover, the baseline measurement took place in spring and the total period between the first and the last measurement of all participants covered a wide range. Furthermore, about half of the participants were retired, which diminishes the probability of work-related variations. In line with the time effect on blood sugar level, the focus on PA elicited by participation in the study might explain the time effect on self-reported PA. As stated above, a raised awareness could have elicited more goal-directed behavior towards PA both in the intervention condition as well as in the control condition. In addition, this raised awareness could have elicited increased recall of physical activities, independent of behavioral change. This increased recall might account for the discrepancy between self-reported and objectively measured PA. A cross-sectional study concluded that fitter participants showed larger discrepancies between self-reported and objectively measured PA than lesser fit ones because they tended to over-report more. Another possible explanation for this discrepancy lies in the absence of differentiation in intensity levels in objectively measured PA, whereas the total PA-score based on self-report did account for differences in intensities. We opted not to differentiate between intensities in objectively measured PA in order to create a comprehensive and relevant measure of regular PA, with sufficient symmetry and variation in data. In addition, because of the ten- minute bouts, PA minutes from different intensity categories were not simply additive, which would complicate comparison. Moreover, objective data showed that ten-minute bouts of vigorous physical activities were quite rare (in contrast with the self-reported data), according to the cut-off value of 6 MET. When objectively measured light and moderate PA were considered separately in post hoc analyses, Wilcoxon Signed-Rank tests did not reveal any significant differences across conditions between pre versus post measurements (short term), nor between pre versus follow-up measurements (long term). Similar findings were obtained when light and moderate PA were analyzed without ten-minute bouts. Mann-Whitney U tests revealed only one notable significant difference, namely with respect to the long-term difference scores between the IC and the CC for objectively measured moderate PA without ten-minute bout restrictions. More specifically, the median difference score was positive in the IC whereas this difference score was negative in the CC. This finding suggests a positive long- term intervention effect on short periods of objectively measured moderate PA. As a third and final time effect in the current study, the level of physical fitness increased significantly across the conditions at long term. If patients across both conditions indeed increased their regular PA, a likely consequence would be that there fitness levels also improved. This explanation seems especially appealing given the fact that walking was one of the most popular activities (at least in the IC) and that physical fitness was measured by a walking test. The lack of a time effect on physical fitness at short term might suggest a physiological adaptation period to the activity stimuli. It should be noted however that a positive trend was already noticeable at short term (*p* \< 0.1). The current pilot trial included several strengths. First, the study was ecologically valid because it took place in a real-life setting, in collaboration with patients’ health insurance fund. Second, GPs were involved but not burdened with heavy additional work-load. Third, the intervention was founded on one theoretical framework (SDT), which had been proven to be useful in different contexts. Fourth, several possibly important but less frequently handled covariates (e.g., climatic circumstances) were included. Fifth, PA was measured both objectively by means of a multisensory armband as well as subjectively by means of self-report. The current pilot trial included a number of limitations as well. First, the total sample size was rather small. However, a similar study showed by means of power analysis (power = 0.80; *α* = 0.05) that a sample size of twenty patients in each condition should suffice to detect significant effects on regular PA (as measured by the number of steps per day). Moreover, to detect a clinically relevant change in HbA1C of 0.5% at the within-subject level (i.e., across conditions), a sample size of thirteen participants seemed sufficient (power = 0.80; two-sided *α* = 0.05; paired *t*-test; *SD* = 1.0%; *r*_within = 0.80). Second, the multisensory armband (Sensewear) had a few downsides. The device caused practical difficulties and loss of data and might have influenced PA behaviors. Moreover, the validity of this tool during walking has been questioned. Third, no information about medication was taken into account. It should be noted though that during this study GPs only rarely reported changes in medication of participating patients. Nevertheless, future research should investigate changes in medication both as a covariate and as a valuable outcome in itself (i.e., independent of changes in blood sugar level). However, the willingness to experiment with medication taps into a cultural issue, which needs to be addressed but falls beyond the scope of this paper. Fourth, the results cannot simply be generalized to the entire population of patients with T2DM. The fee in particular might have deterred potential participants. As a result, our sample is likely to be more motivated and/or economically advantaged compared with the non-participants. ## Conclusion The PA intervention as a whole did not produce the expected effects in patients with T2DM. However, time effects suggest that certain aspects of the current project are effective. It would be of particular interest if these aspects could be proven to be not only effective but also relatively inexpensive. Hence, this pilot study suggests a potential for brief but regular expert visit and measurement, which could be incorporated into communal information sessions and short meetings with PA coach and GP. # Supporting information [^1]: We declare that one of the authors (SDC) was commercially affiliated with the Christian Health Insurance Fund. This does not alter our adherence to PLOS ONE policies on sharing data and materials. [^2]: **Conceptualization:** JV JS AB SDC FB. **Data curation:** JV AB. **Formal analysis:** JV AB KD. **Funding acquisition:** JS AB FB. **Investigation:** KD SDC. **Methodology:** JV JS AB FB. **Project administration:** JV JS AB KD SDC FB. **Supervision:** JS AB SDC FB. **Validation:** JV AB KD. **Visualization:** JV JS KD. **Writing – original draft:** JV. **Writing – review & editing:** JV JS AB KD SDC FB.

# Introduction *Verticillium dahliae* Kleb is a soil-borne fungus that penetrates the vascular system of the plant and causes vascular wilt disease. This pathogenic fungus survives in the soil for long periods as microsclerotia, tiny structures produced in the plant tissue. It infects hundreds of economically important crops and trees, but unfortunately, no currently available fungicides are able to effectively control the disease. Verticillium wilt in cotton has been reported in most cotton-growing areas, and it has become the most important disease of cotton in the world. For example, 5–6.6 million acres, approximately half of the cotton cultivating area in China, were subjected to this disease in 2009 and 2010 (National Cotton Council of America– Disease Database). So far, the most effective strategy to combat this stubborn disease is to develop tolerant cotton cultivars by incorporating genes from resistant germplasm. However, higher level of Verticillium wilt resistance exists in island cotton but not upland cotton. Isolates of *V. dahliae* have been characterized based on virulence capacity and the phenotypic symptoms occurring in disease development. *V. dahliae* isolates from cotton are considered either defoliating isolates that cause severe leaf wilting and shedding in infected plants or non-defoliating isolates that induce leaf-wilt symptoms, but cause only limited shedding of the leaves during disease progression. In general, defoliating isolates may possess a stronger pathogenic capacity. BP2 and V991 are two widely spread *V. dahliae* isolates found in China and represent medium aggressive non-defoliating and highly aggressive defoliating isolates, respectively. Efforts made to utilize genetic engineering to obtain transgenic cotton resistant to *V. dahliae* have had varying results. Genes used for these studies have included *GbVe*, *Arabidopsis NPR1*, anti-apoptotic gene *p35*, HR-induced *Hpa1Xoo* and some antifungal genes, including chitinase, D4E1, lipid transfer proteins and gastrodianin. The resistance levels of transgenic seedlings range from the complete inhibition of *Verticillium* growth in engineered plant extracts, to enhanced disease resistance in field trials, or increased resistance to only weak pathogen varieties. For example, the number of germinating *V. dahliae* conidia was significantly reduced in plant extracts from transgenic cotton overexpressing a synthetic antimicrobial peptide, D4E1. *AtNPR1*-transgenic cotton exhibited significant resistance against non- defoliating isolates but remained susceptible to the defoliating isolates of *V. dahliae*. Effective control of Verticillium wilt has been reported in specific crops exhibiting race-specific resistance. The *Ve* locus of tomato is the only cloned locus that is responsible for resistance against race 1 strains of *V. dahliae* and *V. albo-atrum*. This locus includes two proteins named Ve1 and Ve2, leucine-rich repeat class of receptor-like proteins (eLRR-RLPs). Both genes confer resistance when expressed in the close relative potato, while only Ve1 provides resistance in both tomato and in transgenic *Arabidopsis*. The eLRR- RLPs class proteins differ from receptor-like kinases in that they lack a cytoplasmic kinase domain and carry only a short cytoplasmic tail that lacks obvious signaling motifs. This class of race-specific R proteins also includes Cf proteins conveying *Cladosporium fulvum* resistance in tomato and apple HcrVf proteins that convey resistance against *Venturia inaequalis*. *Ve1*-mediated resistance signaling is activated by the effector Ave1. Ave1 is encoded by race 1 strains of *V. dahliae* and is homologous to a widespread family of plant natriuretic peptides. Genetic analysis has shown that this signaling pathway in tomato requires *EDS1* (Enhanced Disease Susceptibility 1), *NDR1* (Non-race- specific Disease Resistance 1), *BAK1* (BRI1-Associated Kinase 1), *MEK2* (*MKK2*, MAP kinase kinase 2), and only partially overlaps with signaling mediated by Cf proteins. *Ve*-mediated resistance involves several short- and long-term defensive mechanisms, including the onset of hydrogen peroxide (H₂O₂) production, activities of peroxidase and PAL (Phenylalanine ammonia lyase), and the synthesis of lignins. Four species of the *Gossypium* genus are cultivated in agriculture, including two allotetraploids (*G. hirsutum* and *G. barbadense*) and two diploids (*G. herbaceum* and *G. arboreum*). Island cotton possess higher resistance toward Verticillium wilt than upland cotton, which produces more than 95% of the annual cotton crop worldwide. However, cross-breeding between island and upland cotton is a challenging work. Silencing *NDR1*, *MKK2*, or *Ve1*-like genes in *Gossypium hirsutum* compromises resistance to Verticillium wilt, indicating that *Ve1*-mediated resistance signaling is conserved between tomato and cotton, and required for resistance to *V. dahliae* infection. These findings suggest that resistant seedlings of upland cotton may be obtained by expressing a *Ve*-like gene from *Gossypium barbadense*, which has been confirmed by a recent report in which a *GbVe* was cloned and enhanced resistance to *Verticillium* wilt in transgenic *Arabidopsis* plantlets. Here, a *Ve* homologous gene, *Gbve1*, which conferred resistance to both defoliating and non-defoliating isolates of *V. dahliae* in cotton, was verified. qRT-PCR revealed that the *Gbve1* gene was induced by *V. dahliae*. GUS stain analysis indicated that the *Gbve1* gene driven by its promoter was specifically expressed in the vascular bundles of roots and stems. Silencing of *Gbve1* in Verticillium wilt-resistant cotton H7124 compromised cotton resistance to *V. dahliae*. Furthermore, transgenic *Arabidopsis* and cotton exhibiting overexpression of the *Gbve1* gene displayed strong resistance to *V. dahliae*. # Materials and Methods ## Plant Materials, *Verticillium dahliae* and Culture Conditions Verticillium wilt resistant island cotton cv. H7124 (*G. barbadense*) and susceptible upland cotton cv. Yumian 1 (*G. hirsutum.*) were cultured at the conditions of emperatures ranging from 20°C to 25°C, under a 16/8 h photoperiod and at 80% relative humidity. *A. thaliana* ecotype Columbia-0 were germinated on the half strength MS medium and transferred into pots containing vermiculite soil in the plant incubator with temperature at 25°C day and 20°C night, 60–70% relative humidity, and light intensity of 200 umol/m⁻²/sec⁻¹ on a 16 h light/8 h dark cycle. Highly aggressive defoliating isolate V991 and non-defoliating isolates BP2 of *Verticillium dahlia* were freshly isolated from infected cotton plants and maintained on potato dextrose agar (PDA) at 25°C for 7 d, followed by inoculation into Czapek’s medium in 1 liter of distilled water. These two *V. dahlia* isolates were used to the inoculation of *A. thaliana* and cotton. Before *V. dahliae* inoculation, the conidia were counted and the conidia suspension was adjusted to a needed density with distilled water. ## Gene Cloning, Bioinformatics and Expression Analysis of *Gbve1* Genomic DNA was extracted from plant tissues by the CTAB method. While total RNA was extracted from plant tissues with an RNAiso Kit (TaKaRa) according to the manufacturer’s protocol, then subjected to RNase-free DNase I (TaKaRa) digestion and purification. An aliquot of 2 µg RNA were used to synthesize the first- strand cDNA using a Primscript RT-PCR kit (TaKaRa) according to the manufacturer’s instructions. The primers EST- F/EST- R specific to a cotton EST (TC121084) in gene index (<http://compbio.dfci.harvard.edu/tgi/plant.html>), which was highly homologous to the tomato *Ve1* gene, were used to amplify the genomic DNA of H7124. The amplified sequence was subsequently used as probe to screen a *G. hirsutum* cv. Maxxa BAC library. A clone containing a full open reading frame homologous to tomato *ve1* (named *Ghve1*) was further analyzed in this study. The specific primers Vdr2-cF/Vdr2-cR were designed according the *Ghve1* gene and used to amplify its homology in *G. barbadense* with genomic DNA and cDNA of H7124 and Yumian1, respectively. The upstream sequence of *Gbve1* gene was obtained by genome walking according to the GenomeWalker™ (Clontech) instruction with two pairs of primers AP1/GSP1 and AP2/GSP2. All the PCR products were cloned into pGEM-T Eeasy vector (Promega), transformed into *E. coli* DH5α and sequenced. Nucleotide sequences and deduced amino acid sequences of *Gbve1* gene were compared using DNAman software (Lynnon Biosoft). The putative motif and domains were identified with SMART, TMHMM and SignalP 3.0 tools and software packages. The flanking sequence of *Gbve1* gene was analyzed with the PLACE database at the Advanced Biosciences Computing Center. Multiple sequence alignments were made using Muscle 3.8 under the default settings. The phylogenetic tree of *Ve* like genes was constructed using MEGA 4.1 with Neighbor-Joining method. The bootstrap analyses were performed with 1000 replications. At 2-leaf-stage, cotton plants were challenged by *V. dahliae* by soil drenching, pouring 10 ml conidial suspension (1×10⁷ conidia/ml) in the soil beside the plants in 750 ml- pots. For hormone treatment, seedlings at 4-leaf-stage were sprayed with 10 mM salicylic acid (SA), ethylene released from 5 mM ethephon(ETH), 100 µM abscisic acid (ABA) and 100 µM methyl jasmonate (MeJA), followed by covering with plastic bags to keep 100% humidity. Stem tissues from *V. dahliae* infected plants and leaf tissues from hormone treated plants were harvested at an appropriate time for RNA extraction. Quantitative reverse transcription-polymerase chain reaction (qRT-PCR) for *Gbve1* gene was conducted using a SYBR Premix ExTaqTM II Kit (TaKaRa) with cotton poly-ubiquitin 14 (UBQ14) gene as the internal standard. The PCR program contained an initial denaturation step of 1 min at 95°C, followed by denaturation for 15 s at 95°C, annealing for 20 s at 60°C, and extension for 20 s at 72°C for 40 cycles. The real-time PCR thermal cycler qTOWER 2.0/2.2 (Analytik jena, Germany) was used to obtain relative expression levels of each sample. All qRT-PCR expression assays were independently performed and analyzed three times under identical conditions. All the primers used in this paper were list in. ## VIGS in Cotton The vectors for virus-induced gene silencing (VIGS) in cotton were obtained from Dr. Xueping Zhou in Zhejiang University. Briefly, the constructs containing *Gbve1* gene were generated from the *geminivirus* cotton leaf crumple virus (CLCrV) vectors and modified for agrobacterium mediated infection method. *Agrobacterium* cultures containing CLCrV-A-*Gbve1* and CLCrV-B were mixed at a 1∶1 ratio and infiltrated into two fully expanded cotyledons using a needle-less syringe. After agroinfection, the cotton plantlets were kept in incubator with temperature 25°C and relative humidity 80%. The *ChlI* gene, which encodes one unit of the chloroplast enzyme magnesium chelatase required for chlorophyll biosynthesis, was used as control, since a visible phenotype of yellow-colored leaves was observed when the *ChlI* gene of cotton was targeted by VIGS. 15–20 d post agroinfection, all the VIGS treated plantlets were checked the replication of CLCrV by PCR with specific primers Clcrv-F and Clcrv-R using genomic DNA as template, and positive plantlets were then subjected to *V. dahliae* inoculation by soil drenching with 30 ml conidial suspension (1×10⁶ conidia/ml) for each pot (250 ml). Foliar damage was evaluated by rating the symptom on the cotyledon and leaf of inoculated plant according to the following disease grades: 0 =  health plant, 1 =  yellowing or necrosis of 1–2 cotyledons, 2 =  yellowing or necrosis of 1true leaf, 3 =  more than 2 wilted or necrotic leaves, 4 =  no leaf left or dead plant. The disease index was calculated according to the following formula: disease index = \[(∑disease grades × number of infected)/(total checked plants×4)\]×100. VIGS experiments were repeated three times with more than 15 plants each time. ## Generation and Analysis of Transgenic *Arabidopsis* The full open reading frame of *Gbve1* was amplified with primer pairs Gbve1-*Sma*IF/Gbve1-*Sac*IR, and cloned into the pCAMBIA2301 vector. The promoter fragment of *Gbve1* gene was amplified with primers Pro-F/Pro-R, and introduced into the pBI101.1 binary vector to drive the *GUS* gene. Both constructs were confirmed by sequencing before transformation into *Agrobacterium tumefaciens* strain LBA4404. The *Arabidopsis* transformation was carried out with floral dip, and transgenic plants were selected on MS medium containing 50 mg/l kanamycin and 500 mg/l cefotaxine. To assess the *V. dahliae* resistance of transgenic *Arabidopsis* with *Gbve1* gene, *Arabidopsis* plants in pots (200 ml) were subjected to soil drench inoculation with 10 ml conidial suspension (1×10⁶ conidia/ml). Plants in the control group received same amount of sterile water. The degree of *V. dahliae* infection was divided into five disease grades with disease scores ranging between 0 and 4(0 = healthy plants, no fungal infection, 1 = 25% of the leaves showing yellowing or abnormal yellow spots, 2 = 25 to 50% of the leaves showing yellow spots, 3 = 50 to 75% of the leaves showing brown spots and curled leaf edges with some leaves dropping, 4 = ≥75% of the leaves produce yellow or yellow irregular spots between the main vein of leaves. The disease index was calculated as the method described formula above. At least 16 T3 individual plants per line were subjected to resistant analysis and each experiment was set 3 times. To investigate the resistance mechanism of *Gbve1* gene in transgenic *Arabidopsis*, *Arabidopsis* leaves were inoculated with the 10 µl spore drop of *V. dahliae* (1×10⁷ conidia/ml) and then incubated at 25°C in a moist incubator for 2 days, the infected leaves were then stained with Lactophenol- Trypan Blue and examined by the Olympus 1X71 invert microscope. Moreover, the expression of four internal defense related genes *PR1*, *PR5*, *EDS1* and *GST1* genes were compared by qRT-PCR between transgenic *Gbve1* gene *Arabidopsis* and the wild type, with *TUB* gene as the internal standard. Histochemical localization of GUS activity in transgenic *Arabidopsis* with *Gbve1* promoter construct was performed according to Jefferson. ## Generation and Analysis of Transgenic Cotton The *Agrobacterium* transformation of embryogenci calli of upland cotton WC line was conducted according to the protocol described by Wu. The regenerated T0 and T1 plants were subjected to southern blot and qRT-PCR test, respectively. For *V. dahliae* inoculation, T1 seedlings were planted in vermiculite soil in paper pots (250 ml). At the 2-leaf-stage, cotton plants were inoculated with *V. dahliae* by soil drenching with 30 ml conidial suspension (1×10⁶ conidia/ml) for each pot (250 ml), followed by keeping in the incubator with 25°C temperature, a 16/8 h photoperiod and 80% relative humidity. Foliar damage and disease index in cotton was evaluated by the method described above. # Results ## Isolation of a Receptor-like Protein Gene from Cotton To clone the tomato *Ve* homologous gene in cotton, we screened the cotton Bac libraries using a cotton EST that shares 48% similarity in protein with tomato Ve1 protein. A gene was obtained from a total of 40 independent positive clones. Sequence analysis showed that this gene was 76% homologous to the original EST. Then, we amplified the gene allele from the island cotton cultivar H7124, which is resistant to Verticillium wilt, an upland cotton cultivar Yumian1, which is sensitive, and an upland cotton cultivar Changkangmian, which is moderately resistant. These three genes were designated as *Gbve1*, *Ghve1* and *Ghve1-2*, respectively. Comparing the sequence amplified by DNA and cDNA indicated that *Gbve1* has no introns. The *Gbve1* gene was predicted to encode a protein with 1138 amino acids and it shared many similarities with tomato Ve1 and Ve2 proteins, and other Ve-like proteins in cotton, *Solanum licopersicoides*, and *Solanum torvum*. Each of these proteins contained the predicted signal peptide (domain A), multiple copy LRR domain B, neutral/acidic amino acid-rich domain C, membrane-associated hydrophobic domain D, and a positively charged cytoplasmic domain E. The predicted C-terminus of Gbve1 protein also contained the PEST sequence (domain F) and endocytosis signal (domain G) found in the Ve2 protein. The similarities between both protein sequences and the deduced primary structures of GbVe1 and Ve1/Ve2 proteins indicate that *GbVe1* may play a similar role. ## *Gbve1* Induction by *V. dahlia*, SA, JA, and ET, but not ABA qRT-PCR was performed to evaluate the relative expression of the *Gbve1* gene in stems of resistant (H7124) and susceptible cotton genotypes (Yumian1) after inoculation with the defoliating and non-defoliating isolates of *V. dahliae*, respectively. Although gene expression in both cultivars was highly induced upon infection with pathogens, both response time and induction level were dramatically different between the incompatible and compatible interactions. shows that the expressional level of the *Gbve1* gene in H7124 reached a peak at 4 dpi (days post inoculation) and then dropped to a similar level to that of untreated tissue at 8 dpi. In contrast, induction patterns in Yumian1 were much slower and weaker and with a peak appearing at 10 dpi. Also, induction levels of the defoliating isolate (V991) were more intense than those of the non- defoliating isolates (BP2) in the resistant cultivar H7124. The promoter region (1.5 kb upstream) of the *Gbve1* gene was cloned and sequenced. Bioinformatic analysis revealed many regulated elements related to pathogen infection and defense response, including SA-responsive elements, ETH activation sites, elicitor or wounding responsive transcription elements, MYB transcriptional factor recognition, or ABA-responsive sites. These results indicate that expression of the *Gbve1* gene may be induced by defense signaling molecules in addition to pathogen infection. Therefore, we examined the *Gbve1* expression profiles upon treatment with SA, ETH, and MeJA. The *Gbve1* transcripts were induced by all the three tested hormones. The highest level of transcript was shown at 4, 8, and 12 h after treatment with MeJA, ETH, and SA, respectively. The induced transcript levels of *Gbve1* were 11.3–16.2-fold higher than those of untreated tissue. In a control of induction by ABA, the maximum amount of *Gbve1* transcripts appeared after 12 h, and transcript levels were only 3-fold higher than the untreated control. Furthermore, we investigated the expressional patterns of the *Gbve1* gene in *Arabidopsis* using transgenic plant lines that were modified to have the promoter region of the *Gbve1* gene fused to the *GUS* gene. The leaves and petioles were faintly blue, while roots and stems always showed high GUS activity. Microscopic analysis revealed that GUS activity was found exclusively in the vascular bundles of roots and stems. Thus, this gene may be induced by defense signaling molecules and infection by *V. dahliae* in the vascular regions of roots and stems. The induction was more rapid and dramatic in the resistant cultivar than in the susceptible cultivar. ## *Gbve1*’s Necessity in Verticillium Wilt Resistance in Island Cotton Since virus-induced gene silencing (VIGS) has been successfully used in the cotton previously, we used this approach to further investigate the roles of *Gbve1* in resistance to *Verticillium* wilt. The cotton *chlI* gene was used as a control to monitor the efficiency of VIGS in cotton. Two to three weeks after agroinfiltration with the *chlI* gene, leaves displayed the yellow phenotype. We evaluated the levels of *V. dahliae* resistance in the *Gbve1*- deficient cotton lines and used the infiltration of CLCrV-A empty vector as a control. In general, more than 80% of the *Gbve1* VIGS plants were severely infected by the two tested isolates of *V. dahliae*, causing all the leaves to wilt in some cases. The average disease indices of *Gbve1* VIGS plants were 75 and 70 by V991 and BP2, respectively. Average disease indices of the control plants were 33 and 30. The silencing efficiencies of the *Gbve1* gene were further confirmed by qRT-PCR analysis using RNA isolated from leaves of VIGS and control plants. Considering that the *GbVe* gene shared many similarities with the *Gbve1* gene, we simultaneity measured the expressional levels of the *GbVe* genes. showed that the expression level of both genes dropped approximately 81–89% in VIGS plants compared to the control. As expected, *Gbve1*- silencing plants and plants infected with the empty vector exhibited the similar growth phenotypes as the wilt type. Therefore, VIGS assays confirmed that *Gbve1* and/or *GbVe* genes are important components in island cotton resistance to *V. dahliae* infection. ## *Gbve1* Conferred Verticillium wilt Resistance in Transgenic *Arabidopsis* To further explore the function of *Gbve1* in *V. dahliae* resistance, the coding sequence of this gene was inserted into the plant expression vector pCAMBIA2301 and transformed into a *V. dahliae*-susceptible *Arabidopsis* genotype Columbia. Seventeen independent transgenic lines were obtained by kanamycin-resistance selection and confirmed by PCR verification. RT-PCR analysis further confirmed that *Gbve1* was successfully expressed in 12 transgenic lines. Four-week-old seedlings of a T₃ generation were used to analyze Verticillium wilt resistance by soil drenching methods. At 15 dpi, the disease index of 12 transgenic lines ranged from 0 to 12 due to V991 infection and from 0 to 8.75 because of BP2 infection, with no disease symptoms observed from V991 infection in three lines and no disease symptoms observed from BP2 infection in seven lines. In contrast, a disease index of 54 was scored for V991 infection, and 25 for BP2 infection. At 30 dpi, the disease index in WT plants increased to 95 and 80 from V991 and BP2 infection, respectively, while the index in transgenic lines varied from 20 to 65 due to V991 infection and 16.7 to 55 because of BP2 infection. Transgenic lines R5 and R2 at 30 dpi showed excellent Verticillium wilt resistance with a disease index of less than 30 under either V991 or BP2 inoculation. The resistance levels of transgenic *Arabidopsis* to *V. dahliae* were highly correlated with the expression levels of the *Gbve1* gene in these transgenic lines. We observed the cell death development under microscope and found that the *V. dahliae* hypha in wild type leaves were easy to extend to adjacent cells and none cell death was shown at 2 days. In contrast, the *Gbve1* transgenic lines exhibited strong hypersensitive response (HR) cell death, and it restricted the extension of *V. dahliae* hypha. We also analyzed some genes reported to be involved in pathogen resistance in transgenic *Arabidopsis* and WT plants. Among the four tested genes, PR1 was confirmed to increase about 20-fold in the *Gbve1* transformed plants compared to WT plants. The expression level of *EDS1* and *GST1* in the *Gbve1* overexpressed plants was nearly double that of the WT, whereas no difference in *PR5* was observed between *Gbve1* overexpressed plants and WT plants. ## *Gbve1* Conferred Verticillium Wilt Resistance in Transgenic Upland Cotton The same vector carrying *Gbve1* was also transformed into the genome of *G. hirsutum* var. WC by *Agrobacterium*-mediated transformation of embryogenic calli. Twelve independent T₀ transformed plants were generated under kanamycin selection and transplanted into pots for greenhouse maturation. All of these putative transgenic cotton plants had similar phenotypes to the non- transformed negative controls with respect to growth, leaf shape, and flowering. The integration of the *Gbve1* construction into the cotton genome was confirmed by PCR analysis and Southern blotting with the DIG-labeled *NPTII* gene as the probe. According to the Southern blot result, T₁ plants of four transgenic lines (1, 4, 5 and 6) were subjected to further analysis. The expression of the *Gbve1* gene was measured by qRT-PCR. The transgenic line 1 yielded the highest expression of *Gbve1* at 822- fold than the wild type, while the transgenic line 6, 4 and 5 showed 233-, 20-, and 2-fold expression, respectively, compared to the WT control. Two single-copy transgenic lines (1 and 6) with high expression of *GbVe1* gene, were further applied to analyze the resistance to *V. dahliae* isolate V991 and BP2. Disease evaluation showed that these two transgenic lines displayed disease indices significantly lower than those of the WT throughout the developmental stages. The resistance of the line 1 and 6 to *V. dahliae* was both comparable with that in *G. barbadense* H7124. The disease index of transgenic lines to defoliating isolate V991 was slightly higher than that of H7124. However, the line1 and 6 demonstrated better resistances to non-defoliating isolate BP2 than that of H7124. # Discussion Many island cotton cultivars are resistant or near immune to Verticillium wilt, while most upland cotton cultivars are sensitive to this notorious disease. A single dominant *Ve* locus that encodes *Ve1* and *Ve2* genes confers resistance to race 1 *Verticillium* strains and has been widely used in tomato breeding. However, no race-specific resistant gene has been identified in cotton. Here, we cloned a *Ve1*-like gene, *GbVe1*, from island cotton by screening the BAC library and presented several lines of evidence to support that the *GbVe1* gene might be an important component in protecting island cotton from Verticillium wilt. Importantly, we found that the levels of resistance to *V. dahliae* isolates of two testing transgenic lines were equivalent to that of island cotton. Several *Ve1*/*Ve2* like genes have been cloned from *S. licopersicoides*, *S. torvum*, *G. hirsutum*, and *G. barbadense*. Those proteins share high sequence similarity with each other and contain almost identical domains. Gene silencing of the *GhVe1* gene by VIGS in cotton seedlings increased its sensitivity to *V. dahlia*, which is consistent with the results of the *GbVe1* gene used in this study. The cotton genome might encode several *Ve1*/*Ve2* homologues that share high sequence similarity with each other. Thus, functional analysis by VIGS suggests that this gene family is necessary for resistance against Verticillium wilt. Since *Arabidopsis* is also a host for *V. dahliae*, it serves a good model plant to analyze *Ve*-relevant genes. Transgenic *Arabidopsis* lines with *Ve1* gene and *GbVe* gene had improved resistance levels to non-defoliating isolates of *V. dahliae*. Here, we showed that *GbVe1*-mediated strong resistance to both defoliating and non-defoliating isolates of *V. dahliae* in *Arabidopsis* and upland cotton. We also cloned *GbVe1* alleles from both susceptible and tolerant upland cotton. Transcriptional analysis showed that the response times, and induction levels, were dramatically different between incompatible and compatible interactions. From these results, we inferred that *GbVe1* conferred high resistance to the defoliating and non-defoliating isolates of *V. dahliae*. *Ve1*-mediated resistance in tomato is activated by an avirulence protein (Ave1) encoded by race 1 *V. dahliae* strains. A homolog of the *Ave1* gene was also found in the bacterial plant pathogen *Xanthomonas axonopodis* and in several other plant pathogenic fungi, but not in the tested *V. dahliae* strains that are responsible for cotton Verticillium wilt. *Gbve1*-mediated resistance in *Arabidopsis* and cotton might be triggered by other unknown pathogen-associated molecular patterns (PAMPS) or effectors that are encoded by the tested *V. dahliae* strains. Our results also indicate that the expression of the *Gbve1* gene is activated by SA, ETH, and JA treatment. Those signaling molecules play important roles in basal plant defense responses as well as in gene-for-gene- mediated defense, and are also crucial in Verticillium wilt resistance. We suggest that Verticillium wilt resistance may be part of a complex, multi- hormone signal network, and the expression of the *Gbve1* gene in cotton might be partially through signaling molecules after infection by *V. dahliae*. Regardless of what are the resistance mechanisms and downstream signaling components for the GbVe1 protein, we have demonstrated here that the GbVe1 from Island cotton confers the resistance to *V. dahliae* in cotton and *Arabidopsis*. Our results support the recent findings that a *GbVe1* allele, *GbVe*, from *G. barbadense* also contributes resistance to *V. dahliae* when it is over-expressed in *Arabidopsis*. Currently, the best way to prevent *Verticillium* diseases is the use of resistant cultivars. As noted above, highly resistant cultivars are usually absent in upland cottons, and different strategies to develop transgenic cotton with increased levels of *Verticillium* resistance have been tested. However, the resistance level of the transgenic plant is usually limited and not sufficient to generate durable resistance. For example, *AtNPR1*-transgenic cotton seedlings increase resistance only to the weak pathogens, and the resistance levels of transgenic seedlings were evaluated only in a naturally infested Verticillium wilt nursery, with information on pathogens not being included. Here, we measured the transgenic cotton lines in a greenhouse and found that the *Gbve1* gene dramatically increased the resistance level to both defoliating and non-defoliating isolates of *V. dahliae*. We suggest that both *Gbve1* and *Gbve* genes might be useful in the breeding of cotton varieties resistant to Verticillium wilt. # Supporting Information We gratefully acknowledge Prof. Xueping Zhou (Zhejiang University, China) for sharing the unpublished protocols and constructs for VIGS in cotton. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BZ YC. Performed the experiments: YY TC TL. Analyzed the data: BZ DD YC. Contributed reagents/materials/analysis tools: HL XF YR DS LL. Wrote the paper: BZ YY TZ DD.

# Introduction Antiretroviral therapy (ART) decreases HIV-related morbidity and mortality as well as infectiousness, resulting in a significant reduction in the risk of HIV transmission from a treated infected person to an uninfected sexual partner. However, while HIV care and treatment programmes have provided access to ART for an estimated 26 million people as of 2020 worldwide, ART coverage is still far from optimal, especially in most resource-limited countries, where the HIV incidence remains relatively high. In particular, Sub-Saharan Africa is disproportionately affected by the HIV epidemic and accounts for almost 70% of HIV infections worldwide. Although new infections have been reduced by 52% since the peak in 1997, 1.7 million people had been newly infected with HIV by the end of 2019 in this region. Key populations and their sexual partners are at particularly high risk of HIV; in 2020, they accounted for 65% of new HIV infections worldwide. Men who have sex with men (MSM) and female sex workers (FSW) are, 25 and 26 times more likely to be infected with HIV respectively, than the general population. Although the high prevalence suggests a high incidence, incidence surveys remain necessary to better understand the dynamics of the epidemic among key populations. While some data are available for MSM, few recent incidence surveys have been conducted among FSW. Moreover, HIV incidence is likely to be heterogeneous among subgroups of FSW, depending on local context, type of sex work and whether or not they attend community clinics offering various services, including HIV testing, condom distribution and prevention programmes targeting risky behaviour. With the implementation of large-scale combination prevention strategies, accurate tools to monitor where and among whom new HIV infections are occurring are essential to assess the impact of these strategies and improve the effectiveness of targeted prevention programs. In particular, new prevention tools such as pre- exposure prophylaxis are recommended by the World Health Organization (WHO) for only populations with a substantial risk of infection. Therefore, it is crucial to assess the incidence of HIV infections among FSW. There are several approaches to measure the occurrence of new HIV infections in a population. The gold standard is a prospective cohorts study that analyses HIV seroconversions in uninfected individuals. However, such cohorts studies are very costly and complex to conduct. This is why the most common approach in developing countries has been inference, considering trends in HIV prevalence and assumptions about mortality and the impact of ART coverage on survival. Recently, several laboratory approaches have been developed to distinguish newly acquired HIV infections from long-term HIV infections in cross-sectional surveys. These incident HIV detection approaches are based on the principle that the immunological response to HIV infection evolves over several months after infection, allowing the identification of immunological biomarkers of early HIV disease that can serve as indicators of recent infection. The assay-based approach involves the use of one or more serological laboratory tests that is able to classify HIV infection according to whether the infection was acquired in the recent past. Classification using one or more assays of this kind represents an HIV Recent Infection Testing Algorithm (RITA). If accurate, incidence testing can be a rapid and cost-effective approach to obtaining reliable and up-to date information about the dynamics of HIV transmission for more effective planning. An HIV RITA has been developed and adapted to the Ivorian context. The present paper used this HIV RITA to estimate the HIV incidence and its associated factors among FSW in two regions of Côte d’Ivoire: Abidjan and San Pedro. # Methods ## Study setting The ANRS 12361 PrEP-CI cross-sectional study was designed and implemented by two Ivorian community-based organisations between September 2016 and March 2017. Aprosam works within the city of San Pedro and its surrounding areas, particularly in villages associated with farming businesses (coffee and cocoa production). Espace Confiance operates in several districts of Abidjan, the economic capital of Côte d’Ivoire (Koumassi, Marcory, Treichville, Zone 4 and Port-Bouët including its beaches). These nongovernmental organisations (NGOs) provide HIV prevention and testing services directly at prostitution sites (outreach activities) and provide HIV and sexual health care services for FSW through community clinics. Recruitment of participants for this study was made possible by the Aprosam and Espace Confiance organisations’ networks of peer educators and their access to the target population. ## Study population The study’s purpose was not to represent all FSW in Côte d’Ivoire but rather to represent FSW who could be reached by the two partner organisations and who would potentially benefit from PrEP in a future programme. FSW are identified by the peer educator and they work at sites that are usually visited by them for HIV prevention/screening. Almost all of the FSW work sites were visited by the peer educator. The FSWs were recruited both in prostitution sites (brothel, hotel, bar/maquis, street, beach) and in the fixed clinics dedicated to sex workers of both NGOs. The survey was conducted among FSW aged 18 years and older, who had never been tested for HIV or who had previously tested negative for HIV, and who worked at a sex work site at the time of the survey, in the areas targeted by the two community-based NGOs. ## Sociodemographic and behavioural questionnaire After obtaining informed written consent, a standardised questionnaire was administered by peer educators. The questionnaire collected sociodemographic data (date of birth, nationality, place of recruitment, level of education), as well as sexual practices and behaviours (such as the duration of sex work, the age at which sex work began, the place of meeting and activity with clients, whether sex work was carried out regularly, the use of condoms during sex work, the price of the pass (the price of a single sexual encounter with a client), the number of clients, the number of condoms used on the last day of activity and the number of sexual intercourse encounters for which a condom was not used during the last seven days). ## HIV screening and laboratory analysis HIV screening was carried out by two rapid tests (Determine^®, Alere and Vikia^®, bioMérieux), for all surveyed FSWs, at the sex work sites. In the event of a positive result, HIV infection was confirmed by a rapid test (stat-pack^®). Then, a dried blood spot (DBS) sample was taken and transported to the laboratory of the University Hospital of Tours, France, to determine the window of infection and false positive rate using a recent infection test adapted to the Ivorian context and performed directly on plasma. This recent infection test made it possible to classify HIV infections into two groups: HIV infections contracted less than 6 months prior and those contracted more than 6 months prior. This recent infection test developed by Barin et al. is the Less-sensitive enzyme Immunodominant assay recent infection (EIA- RI/IDE-V3). This assay uses the enzyme immunoassay technique in a 96-well plate, based on the measurement of absorbances (OD) in one well sensitised with an equimolar peptide mixture TM (cons+D), corresponding to the immunodominant epitope of gp41 (consensus sequence envi—1 group M and consensus sequence env—1 subtype D), and in another well, sensitised with a V3 peptide solution (AE), corresponding to an equimolar mixture of the consensus sequences of the V3 region of gp120 of the HIV subtypes A, B, C, D and CRF01_AE. This test is an in- house test, applicable in the HIV-NRC virology laboratory, serology sector. The test can be performed on serum or plasma, as well as serum, plasma or whole blood on blotting paper or DBS. The test uses a mathematical formula that combines the quantitative responses to gp41 antigens in each region to distinguish between recent and established infection. This in-house test has been the subject of preliminary studies using sequential serum samples from HIV-infected Ivorian patients with known dates of infection (the PRECO-CI ANRS 12277 and PRIMO-CI ANRS 1220 projects) and samples from patients at different stages of the disease (the Temprano ANRS 12136 and Trivacan ANRS 1269 trials); which allowed to distinguish a recent infection (≤180 days) from an established infection (\>180 days) with a window of infection (0.3 years) and false positive rate (13‰) for the Ivorian population studied. During the study, FSW diagnosed with HIV infection were referred to community clinics by peer educators for ART. ## Description of the surveyed population Sociodemographic characteristics and sexual practices and behaviours of the surveyed participants were described according to the study setting (San Pedro or Abidjan). Since the sample was not a random sample but rather a convenience sample of women reached by the two organisations, statistical tests such as Pearson’s χ2 test or Fisher’s exact test could not be formally used to compare differences between the populations from the two study settings. We excluded any missing data from the percentage calculations. All analyses were performed with R version R-4.2.1 software. ## Assessment of HIV incidence For a given population and HIV subtype, a RITA has a mean RITA duration *ω*, defined as the mean duration for which newly infected individuals in the population have had a recently acquired infection. A RITA also has a false recent rate (FRR), noted as *ε*, which is the proportion of non recent HIV infections in the population that are misclassified by the RITA as recent. A RITA is used to estimate HIV infection by first classifying cases of HIV infection in the population as recently acquired or not, and then applying a mathematical formula to the resulting counts of recently acquired infections. The annual incidence rate *I*_*r*, is estimated using the following formula: $$I_{r} = \frac{R - \mspace{360mu}\varepsilon P}{(1 - \varepsilon)\omega N}$$ where *N* is the number of HIV-negative persons in the survey, *P* the number of HIV-positive persons, *R* is the number of persons classified as positive by the RITA, *ω* is the mean RITA duration in years and *ε* is the FRR of the RITA. Sweeting et al., in 2010 describe this mathematical approach in more detail. Regarding the RITA used in this study, the mean RITA duration (ω) was 0.3 years and the FRR of the RITA (ε) was 0.013. ## Factors associated with HIV incidence HIV incidence rate and their CIs were computed for the different subgroups. Unfortunately, no comparison test or multivariate analysis is currently available for RITA data. ## Ethical aspects All FSW were informed of the risks and benefits of participating in this study before inclusion. All FSW provided written informed consent. The National Ethics Committee for Life Sciences and Health of Côte d’Ivoire approved the research protocol (N/Ref: 057/MSHP/CNER-kp of 28 June 2016). For ethical reasons, the full survey dataset is available only upon reasonable request at <https://zenodo.org/record/5948841>. An analytical dataset containing only the variables required to replicate the analysis, as well as the corresponding R script, are available in. ## Inclusivity in global research Additional information regarding the ethical, cultural, and scientific considerations specific to inclusivity in global research is included in the Supporting Information. # Results ## Sociodemographic and behavioural characteristics A total of 1000 FSW, including 400 in San Pedro and 600 in Abidjan, were surveyed. The main characteristics of the surveyed population are presented in. The median age was 25 years (interquartile range: 21–29 years). The FSW surveyed in San Pedro had a lower education level than the FSW surveyed in Abidjan. The FSW in Abidjan engaged in more sex work activity than those in San Pedro. There are differences in socio-demographic characteristics between the two cities. Indeed, Abidjan is the economic capital of the country, in full expansion compared to San Pedro where the level of poverty, literacy or education is lower. San Pedro was in the recent past the largest slum in West Africa and most of the FSWs sites are within the perimeters of this slum. ## Incidence of HIV infection Among the surveyed FSW (those never tested or with a previous negative test result), 39 (3.9%) tested positive for HIV (6.3% in San Pedro and 2.3% in Abidjan) during the survey. Of these, seven FSW were classified as being recently infected according to the RITA (average duration of infection 113 days or 0.3 years). The incidence of HIV was estimated to be 2.3 per 100 person-years overall, with 3.3% in San Pedro and 1.6% in Abidjan. ## Associated factors Some trends emerged from the results presented in. There were variations according to the sociodemographic characteristics of the participants, with higher incidence rates among the youngest age group (3.0% in FSW 24 years old or less vs. 1.9% in FSW 25 years old or more), non-Ivorian FSW (3.2% in non-Ivorian FSW vs. 1.9% in Ivoirian FSW) and the group with the lowest education level (4.6% in FSW who never went to school vs. 2.6% in those with a primary education level and 0.8% in those with a secondary or university education level). The incidence of HIV also seemed to be associated with the sex work practice conditions. Indeed, FSW who charged a lower price for sexual intercourse had a higher HIV exposure rate (4.3% in FSW whose usual price was less than 3.5\$ vs. 1.0% in FSW whose usual price was more than 3.5\$). In addition, FSW working in brothels (4.0%), in the streets (5.4%), and in hotels (4.2%) were more likely to be recently infected than those working in bars or “maquis” (0.8%). The incidence was higher among FSW who had a larger number of clients on the last day of work (6.1% in those with 7 clients or more vs. 1.8% in those with 2–6 clients) and who worked in more than one city (3.7% vs. 1.8%). The incidence did not seem to differ according to tenure in the sex industry (2.7% in FSW who performed sex work for 3 years or more vs. 2.0% in FSW who performed sex work for 3 years or less). A higher incidence was observed among FSW who reported not always using condoms when engaging with their clients (8.5% vs. 1.5%); who reported agreeing to sex without a condom in exchange for a large sum of money (10.1% vs. 1.2); who reported contracting a sexually transmitted infection (STI) in the last 12 months (2.6% vs. 1.9%); and who had consulted a health professional more than one year previously (4.1% vs. 1.4%). # Discussion Our results confirm that FSW remain at high risk of exposure to HIV population in Côte d’Ivoire, with an estimated overall incidence of 2.3% (1.6% in Abidjan and 3.2% in the San Pedro region). In Côte d’Ivoire, the HIV incidence among women in the general population aged 15–64 years was estimated to be 0.03%. Specifically, it was 0.04% among women aged 15–24 years and 0.05% among those aged 25–34 years, according to the Côte d’Ivoire Population-Based HIV Impact Assessments (CIPHIA) 2017–2018 survey. Our results in FSW in Côte d’Ivoire are higher than those estimated in China by Wang et al. in 2012 and in Cotonou, Benin by Diabaté et al. in 2018, who reported rates of 1.4% and 1.4%, respectively. On the other hand, our result was lower than 3.5% estimated by Braunstein et al. in 2011 among FSW in Rwanda. Our results are consistent with well-documented risk factors associated with HIV in previous studies. The HIV incidence was higher among those who reported having contracted an STI in the past 12 months, those who reported not always using a condom when engaging in sex work, and those who admitted to agreeing to sex without a condom in exchange for a large sum of money than among their counterparts. This high incidence of HIV infection further evidences this among FSWs in San Pedro compared to those practising in Abidjan. In fact, they charged a lower pass price with a higher number of customers. They also had a lower rate of condom use and were more likely to accept sex without condoms in exchange for a large sum of money, and they reported more STIs than those in Abidjan. All these factors favour exposure to HIV infection. Our results also highlight that the working conditions of FSW effect the risk of HIV exposure and acquisition. Women working in brothels, hotels or on the street had higher exposure rates than those working in bars and “maquis”. It should be noted that sex work associated with bars/“maquis” are usually occasional; therefore, these FSW generally have fewer clients. A large number of clients on the last day of work was also associated with HIV acquisition. Thus, a larger number of clients increases the risk of HIV infection. Generally, a link between precariousness and HIV acquisition was demonstrated in our results. We observed higher incidences among less-educated FSW, younger FSM, FSW who charged a lower price for sex, FSW of foreign nationality (most often with greater social and administrative insecurity), and FSW who practised sex work in multiple cities than among their counterparts. Similarly, those who are located farther from health services have a higher risk, as the incidence was higher among those who reported not having consulted a health professional in the twelve months preceding the survey. These results corroborate the work of Szwarcwald et al. in 2018 and of Muldoon et al. in 2015. We hypothesised before the survey that those who had recently entered the sex work market would be at higher risk because they have less knowledge about prevention and less capacity to negotiate condom use, but we did not observe any differences in exposure data according to the length of time they had been practising sex work. On the one hand, foreign FSW who are new to the country are usually supervised and educated by site managers about the need for systematic condom use. On the other hand, analysis of the questionnaires and qualitative interviews, which were conducted as part of the same survey and previously published, highlighted that despite a high rate of condom use and strong negotiation skills, FSW remain have a high HIV and other STIs exposure rate, as some sexual intercourses events do not involve the use of condoms. FSW’ responses to the question assessing condom use might refer to ‘typical use’ as opposed to specific circumstances. Also the difference between reported STI cases and the proportion of condom use could be the effect of temporality. Indeed, STIs were reported in the last 12 months while responses on condom use during sex were related to current use, in the week or month before the survey. Also, FSW have difficulty negotiating condom use with their boyfriends or husbands, even when they do not know their HIV status and engage in sex with multiple sexual partners. They also willingly accept condomless sex with some regular clients whom they feel they can trust or when they are in high need of money. Our study is one of the first to estimate the incidence of HIV in FSW in Côte d’Ivoire using tests to detect recent infection. Incidence data from at-risk populations are key in designing better programmes and interventions to limit new infections. Recent infection surveillance is a powerful tool with which Cote d’Ivoire’s national HIV/AIDS program may identify geographic areas and demographic groups within which HIV transmission is ongoing. More broadly, similar explorations would lend important insight into transmission dynamics in a high-stigma environment. Another strength of this study is the recruitment, through peer educators, of FSW with diverse profiles from different locations. Yet, we have to acknowledge some limitations. First, as RITA is a biological assay, classification of infections as recent or not does not rely on self- reported information. Some people with long-standing HIV infection and on treatement may be misclassified as newly infected. However, these false recent cases are taken into account, as a false recent rate is applied when estimating incidence. Secondly, our data suffer from a lack of power due to a relative sample size with a few number of FSW recently HIV-infected. In addition, the comparison between San Pedro and Abidjan could not been done with statistical tests because the sample was not a random sample but rather a convenience sample. Also, our data came from a convenience sample. It was not appropriate to present the confidence intervals. We were not able to perform a multivariate analysis. Some of the observed associations may result from interactions between several variables. For example, more foreign FSW than Ivorian FSW work in brothels, resulting in foreign FSW having larger numbers of clients. # Conclusion Although community-based prevention programmes for sex workers have led to the empowerment of FSW and a high rate of male condom use in general, they are not sufficient on their own to completely eliminate the risks of HIV acquisition. This study confirms that FSW, even those who have engaged in sex work for several years, remain highly exposed to HIV infection. Exposure to HIV is also clearly associated with certain sex-work factors and the material conditions of sex work. Efforts in the fight against HIV infection must be intensified to reduce new infections among FSW. There is a need for appropriate people-centred prevention programmes that include new prevention tools, such as pre-exposure prophylaxis, and take into account the living and working conditions of FSW. # Supporting information We would like to thank all participants as well as Aprosam’s and Espace Confiance’s peer educators and the ANRS 12361 PrEP-CI study group: Aboubakar Sangaré (Aprosam, San Pedro, Côte d’Ivoire), Anglaret Xavier (PAC-CI, Abidjan, Côte d’Ivoire / Inserm, Bordeaux, France), Anoma Camille (Espace Confiance, Abidjan, Côte d’Ivoire), Barin Francis (Université François Rabelais, Tours, France), Bazin Brigitte (ANRS, Paris, France), Becquet Valentine (Ceped/IRD, Paris, France), Dabis François (ISPED/Inserm, Bordeaux, France), Danel Christine (PAC-CI, Abidjan, Côte d’Ivoire / Inserm, Bordeaux, France), Eholie Serge (PAC- CI, Abidjan, Côte d’Ivoire), Ekouevi Didier (PAC-CI, Abidjan, Côte d’Ivoire), Fonsart Julien (Hôpital SaintLouis, Paris, France), Gbosi Kate (Aprosam, San Pedro, Côte d’Ivoire), Kwamé Abo (Programme National de Lutte contre le Sida, Côte d’Ivoire), Larmarange Joseph (Ceped/IRD, Paris, France), Masumbuko Jean- Marie (PAC-CI, Abidjan, Côte d’Ivoire), Méda Nicolas (Centre Muraz, Bobo- Dioulasso, Burkina Faso), Moh Raoul (PAC-CI, Abidjan, Côte d’Ivoire), Molina Jean-Michel (Hôpital Saint-Louis, Paris, France), N’dri-Yoman Thérèse (PAC-CI, Abidjan, Côte d’Ivoire), Nouaman Marcellin (PAC-CI, Abidjan, Côte d’Ivoire), Plazy Mélanie (ISPED / Inserm, Bordeaux, France), Soh Kouamé (Aprosam, San Pedro, Côte d’Ivoire), Tanoe Solange (Espace Confiance, Abidjan, Côte d’Ivoire), Yeo Roselyne (Espace Confiance, Abidjan, Côte d’Ivoire). 10.1371/journal.pone.0271988.r001 Decision Letter 0 Sharifi Hamid Academic Editor 2022 Hamid Sharifi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 8 Mar 2022 PONE-D-22-02895Incidence of HIV infection and associated factors among female sex workers in Côte d’Ivoire, results of the ANRS 12361 PrEP-CI study using recent infection assaysPLOS ONE Dear Dr. Nouaman, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Apr 22 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Hamid Sharifi Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf>. 2\. Please include a complete copy of PLOS’ questionnaire on inclusivity in global research in your revised manuscript. Our policy for research in this area aims to improve transparency in the reporting of research performed outside of researchers’ own country or community. The policy applies to researchers who have travelled to a different country to conduct research, research with Indigenous populations or their lands, and research on cultural artefacts. The questionnaire can also be requested at the journal’s discretion for any other submissions, even if these conditions are not met.  Please find more information on the policy and a link to download a blank copy of the questionnaire here: <https://journals.plos.org/plosone/s/best-practices-in-research-reporting>. Please upload a completed version of your questionnaire as Supporting Information when you resubmit your manuscript. 3\. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see [http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data- access-restrictions. ](http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions. ) In your revised cover letter, please address the following prompts: a\) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent. b\) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see <http://www.bmj.com/content/340/bmj.c181.long> for guidelines on how to de- identify and prepare clinical data for publication. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data- availability#loc-recommended-repositories>. We will update your Data Availability statement on your behalf to reflect the information you provide. 4\. Please amend the manuscript submission data (via Edit Submission) to include author Hervé Dao. 5\. One of the noted authors is a group or consortium \[the ANRS 12361 PrEP-CI team\]. In addition to naming the author group, please list the individual authors and affiliations within this group in the acknowledgments section of your manuscript. Please also indicate clearly a lead author for this group along with a contact email address. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: I Don't Know \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: No Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This is a well written paper on the factors that influence HIV positivity in Female Sex Workers from two different cities in Côte d’Ivoire. The factors that favor HIV infection are age of less than 24 years old, being non -Ivorian, less education, if the clients have paid less, larger number of clients on a single day, inconsistent use of condoms by the client, or condomless sex, especially if extra money has been paid for the same, and presence of STDs. They have also calculated the incidence of HIV infection per 100 person years and used RITA to determine recent HIV infections of less than six months. The paper is well written and has nothing new to offer in terms of determination of factors favoring HIV infection in FSW. However RITA is the new factor mentioned in this paper. Some clarifications: 1\. Did all FSW have both rapid tests or only one as a screening test. This is not clear in line 174. 2\. In Table 1, under duration of sex activity should it not be less than 3 years or more than 3 years. It says less than 3 years and more than 4 years. What about those between 3and 4 years. Lines 267 to 269 state only below and above 4 years. Kindly rectify. 3.Abidjan saw more sexual activity than San Pedro. Yet, San Pedro shows shows higher infection rates of Hiv. A sentence or two should be added to explain this in discussion. 4\. The fact that this was a convenience sample rather than a random sample and therefore certain statistical tests could not be done should be shifted to limitations of the study. Also the fact that multivariate analysis is not available for RITA. Reviewer \#2: This is a very interesting paper on the use of HIV Recent Infection Testing Algorithm (RITA) to measure recent HIV infection and HIV incidence in a convenience sample of FSW in two towns in Côte d'Ivoire. This study will be a great addition to the literature, especially as it informs the epidemiology of HIV among key populations. I have a few questions that I hope the authors can address to strengthen the manuscript (below): Introduction • Can the authors explain why RITA needs to be adapted to geographic context if it is an assay that measures immunological response to HIV infection (shouldn’t this biological response not be dependent on geographic context?)? Methods • I appreciate the authors specifying who is the target population for this study (i.e., not all FSW). However, the authors specify that part of the target population are FSW who could potentially benefit from PrEP. Which FSW do the authors believe would not potentially benefit from PrEP in the future, given the high prevalence of HIV among FSW? • Specifically, how was recruitment done for this study? Were FSW incentivized to participate? • Because, as the authors admit, this is a convenience sample, I do not think it makes sense to include CIs for the HIV incidence rate, which assumes the data come from a probability distribution. Could the authors please comment? • How is the HIV incidence calculation impacted by this being a convenience sample? Put another way, how is the HIV incidence calculation robust to this being a convenience and not a random (or otherwise probabilistic) sample? Results • Although results are taken from convenience samples and therefore we should be cautious about directly comparing the samples, differences in the socio- demographic composition of the two samples are striking. Are there differences in the socio-demographic characteristics in the towns themselves that could explain some of the differences in the FSW populations? Reviewer \#3: The manuscript is very well-written, and speaks to an extremely important topic for improving HIV programming. A few issues require addressing in order to strengthen conclusions and improve clarity: 1\) Line 170: “Price of the pass” means what? 2\) Lines 176-181: over the past several years, countries have been utilizing various assays for recent infection surveillance, with varying MDRI and FRR values. The specifics of the assay and algorithm used in this study need to be described in this section. 3\) Lines 311-317: given that 64.7% (per Table 1) of participants reported having had an STI in the past 12 months, while 86.9% reported “always” using condoms, perhaps the authors should comment on the reliability of reported condom use 4\) Lines 318-321: it warrants mentioning that recent infection surveillance is a powerful tool with which Cote d’Ivoire’s national HIV/AIDS program may identify geographic areas and demographic groups within which HIV transmission is ongoing. This study is an excellent example of that, and similar such explorations more broadly would lend important insight into transmission dynamics in a high-stigma environment. 5\) Lines 322-327: if viral load data is not integrated into the RITA, this constitutes a major limitation of the study. Without VL data, reported recent infections may include individuals who did not disclose previously known HIV+ status, who retested having been on ART, and who are therefore misclassified. The possibility of misclassification has not been discussed in this paper and needs to be addressed. Moreover, quantifying re-testers (those who test positive on recency assays, but are actually virally suppressed because of ART) would be programmatically relevant in a highly stigmatized population and environment such as this one. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0271988.r002 Author response to Decision Letter 0 25 Apr 2022 To the editors of the Journal of the PLOS ONE Abidjan, April 25th 2022 Response-to-reviewers letter and revised manuscript entitled “Incidence of HIV infection and associated factors among female sex workers in Côte d’Ivoire, results of the ANRS 12361 PrEP-CI study using recent infection assays” Dear Editors, This document provides point-by-point responses to the reviewers’ comments. Their comments, reproduced in italics, are followed by our responses and, finally by the additional text and changes (in bold red) to the manuscript first submitted. We thank the reviewers for their comments which have helped improve our paper. Reviewer : 1 This is a well written paper on the factors that influence HIV positivity in Female Sex Workers from two different cities in Côte d’Ivoire. The factors that favor HIV infection are age of less than 24 years old, being non -Ivorian, less education, if the clients have paid less, larger number of clients on a single day, inconsistent use of condoms by the client, or condomless sex, especially if extra money has been paid for the same, and presence of STDs. They have also calculated the incidence of HIV infection per 100 person years and used RITA to determine recent HIV infections of less than six months. The paper is well written and has nothing new to offer in terms of determination of factors favoring HIV infection in FSW. However RITA is the new factor mentioned in this paper. Some clarifications: 1\. Did all FSW have both rapid tests or only one as à screening test. This is not clear in line 174. 2\. In Table 1, under duration of sex activity should it not be less than 3 years or more than 3 years. It says less than 3 years and more than 4 years. What about those between 3and 4 years. Lines 267 to 269 state only below and above 4 years. Kindly rectify. 3\. Abidjan saw more sexual activity than San Pedro. Yet, San Pedro shows shows higher infection rates of Hiv. A sentence or two should be added to explain this in discussion. 4\. The fact that this was a convenience sample rather than a random sample and therefore certain statistical tests could not be done should be shifted to limitations of the study. Also the fact that multivariate analysis is not available for RITA. Authors’ response: 1\. All study participants were tested for HIV by two rapid tests. A clarification was made in the text as follows: « HIV screening was carried out by two rapid tests (Determine®, Alere and Vikia®, bioMérieux), for all surveyed FSWs, at the sex work sites » (line 179) 2\. The remark is correct. We have harmonised in the tables and in the text as follows: « \>3 years or more » Vs. « 3 years or less » (table 1, 2 and lines 280-281) 3\. Yes, indeed the reviewer is right. Female sex workers (FSWs) in San Pedro were more likely to be exposed to HIV than those in Abidjan. Indeed, they charged a low price for the pass, which results in a high number of customers. They also had a lower rate of condom use and were more likely to accept sex without condoms in exchange for a large sum of money. They also reported more STIs than those in Abidjan. All these factors favour exposure to HIV infection. The discussion section has added some elements to this effect: « The high incidence of HIV infection further evidences this among FSWs in San Pedro compared to those practising in Abidjan. In fact, they charged a lower pass price with a higher number of customers. They also had a lower rate of condom use and were more likely to accept sex without condoms in exchange for a large sum of money, and they reported more STIs than those in Abidjan. All these factors favour exposure to HIV infection ». (lines 302-307). 4\. The remark is relevant. And we have completed the limitations paragraph of the study as suggested : « Yet, we have to acknowledge some limitations. Fist, as RITA is a biological assay, classification of infections as recent or not does not rely on self-reported information. Some people with long-standing HIV infection and on treatment may be misclassified as newly infected. However, these false recent cases are taken into account, as a false recent rate is applied when estimating incidence. Secondly our data suffer from a lack of power due to a relative sample size with a few number of FSW recently HIV-infected. In addition, the comparison between San Pedro and Abidjan could not been done with statistical tests because the sample was not a random sample but rather a convenience sample». Also, we have already stated in the limitations that multivariate analyses were not available for RITA. (lines 348-356) -------------------------------------------------------------------------------- -------------------------------------- Reviewer: 2 This is a very interesting paper on the use of HIV Recent Infection Testing Algorithm (RITA) to measure recent HIV infection and HIV incidence in a convenience sample of FSW in two towns in Côte d'Ivoire. This study will be a great addition to the literature, especially as it informs the epidemiology of HIV among key populations. I have a few questions that I hope the authors can address to strengthen the manuscript (below): Introduction Can the authors explain why RITA needs to be adapted to geographic context if it is an assay that measures immunological response to HIV infection (shouldn’t this biological response not be dependent on geographic context?)? Authors’ response: The estimation of HIV incidence by RITA depends on two key parameters : the average duration of infection (ω) and the false recent rate (ε) and their associated uncertainties, which also quantify the share of the recent among the infected and the precision of this quantification. These parameters are directly related to population levels of HIV prevalence, which determines the proportion of infected and uninfected ; and the incidence, which determines the share of the recently infected. HIV prevalence is different in different geographical contexts, that is why we say that the RITA test must be adapted to the geographical context, i.e. here to the Ivorian context. (lines 182, 185-186) Methods I appreciate the authors specifying who is the target population for this study (i.e., not all FSW). However, the authors specify that part of the target population are FSW who could potentially benefit from PrEP. Which FSW do the authors believe would not potentially benefit from PrEP in the future, given the high prevalence of HIV among FSW? Specifically, how was recruitment done for this study? Were FSW incentivized to participate? Because, as the authors admit, this is a convenience sample, I do not think it makes sense to include CIs for the HIV incidence rate, which assumes the data come from a probability distribution. Could the authors please comment? How is the HIV incidence calculation impacted by this being a convenience sample ? Put another way, how is the HIV incidence calculation robust to this being a convenience and not a random (or otherwise probabilistic) sample? Authors’ response: We thank the reviewer for this comment. Indeed, our sample is not representative of the whole of FSWs. Nevertheless, in the same study, we assessed the acceptability of offering PrEP to this population. And almost all the FSWs interviewed would accept taking PrEP if it were offered to them. The recruitment of participants for this study was made possible by the peer educator networks of two community organizations. FSW are identified by the peer educator and they work at sites that are usually visited by them for HIV prevention/screening. Almost all of the FSW work sites were visited by the peer educator. The FSWs were recruited both in prostitution sites (brothel, beaches, hotel, bar/maquis) and in the fixed clinics dedicated to sex workers of both NGOs. The recruitment involved FSWs who had never been tested for HIV or who had previously tested negative for HIV, and who were working at a sex work site at the time of the survey, in the areas targeted by the two NGOs. There was no incentive to participate in the study. All that was required was written consent to participate. (lines 160-164). Due to the small sample size, we believe that it is important to provide a sense of the uncertainty of our estimates, in particular to see if differences between groups are relevant or not. Considering that multivariate analysisis not possible with RITA, we believe that 95% CI constitutes a relevant indicator of the uncertainty. We note that the assessment of HIV incidence in this key population is rare in our context, particularly in Côte d'Ivoire where the study took place. And the use of recent infection tests is a first and could provide a basis for future studies of incidence estimates in larger samples and in other populations in Côte d'Ivoire. Results Although results are taken from convenience samples and therefore we should be cautious about directly comparing the samples, differences in the socio- demographic composition of the two samples are striking. Are there differences in the socio-demographic characteristics in the towns themselves that could explain some of the differences in the FSW populations ? Authors’ response: We thank the reviewer for this comment There are differences in socio-demographic characteristics between the two cities. Indeed, Abidjan is the economic capital of the country, in full expansion compared to San Pedro where the level of poverty, literacy or education is lower. San Pedro was in the recent past the largest slum in West Africa and most of the FSWs sites are within the perimeters of this slum. (lines 242-246). -------------------------------------------------------------------------------- ------------------------------------------- Reviewer: 3 The manuscript is very well-written, and speaks to an extremely important topic for improving HIV programming. A few issues require addressing in order to strengthen conclusions and improve clarity. 1\) Line 170: “Price of the pass” means what? 2\) Lines 176-181: over the past several years, countries have been utilizing various assays for recent infection surveillance, with varying MDRI and FRR values. The specifics of the assay and algorithm used in this study need to be described in this section. 3\) Lines 311-317: given that 64.7% (per Table 1) of participants reported having had an STI in the past 12 months, while 86.9% reported “always” using condoms, perhaps the authors should comment on the reliability of reported condom use. 4\) Lines 318-321: it warrants mentioning that recent infection surveillance is a powerful tool with which Cote d’Ivoire’s national HIV/AIDS program may identify geographic areas and demographic groups within which HIV transmission is ongoing. This study is an excellent example of that, and similar such explorations more broadly would lend important insight into transmission dynamics in a high-stigma environment. 5\) Lines 322-327: if viral load data is not integrated into the RITA, this constitutes a major limitation of the study. Without VL data, reported recent infections may include individuals who did not disclose previously known HIV+ status, who retested having been on ART, and who are therefore misclassified. The possibility of misclassification has not been discussed in this paper and needs to be addressed. Moreover, quantifying re-testers (those who test positive on recency assays, but are actually virally suppressed because of ART) would be programmatically relevant in a highly stigmatized population and environment such as this one. Authors’ response: We thank the reviewer for this relevant and important remark 1\) « Price of the pass » mean the price of a single sexual encounter with a client. This clarification has been made in the text. (line 174) 2\) In order to be more precise, we have completed the text : See section “Assessment of HIV incidence” for more details. (lines 185-186) 3\) The point is well made. Indeed, the high percentage of STIs reported in the last twelve months contrasts with the proportion of systematic condom use. This could be the effect of temporality. Indeed, STIs were reported in the last 12 months while responses on condom use during sex were related to current use, in the week or month before the survey. Also, the presence of STIs could encourage and incite the FSW to systematically wear a condom during her sex work activity. The condom would be a means of preventing possible STIs. We have completed the discussion as follows : FSW’ responses to the question assessing condom use might refer to ‘typical use’ as opposed to specific circumstances \[ref 25\]. Also the difference between reported STI cases and the proportion of condom use could be the effect of temporality. Indeed, STIs were reported in the last 12 months while responses on condom use during sex were related to current use, in the week or month before the survey. (lines 331-335) 4\) We have included this remark in the manuscript: “Recent infection surveillance is a powerful tool with which Cote d’Ivoire’s national HIV/AIDS program may identify geographic areas and demographic groups within which HIV transmission is ongoing. More broadly, similar explorations would lend important insight into transmission dynamics in a high-stigma environment » (lines 342-346) 5\) Viral load data are taken into account by RITA tests. Indeed, there is substantial evidence that a proportion of people with long-standing HIV infection are misclassified as newly infected by currently available tests for recent HIV infection. Therefore, the false recent rate of the RITA that depend on these tests can never be considered as zero. The false recent rate takes into account one or more of the following characteristics : \- Advanced infection, defined by a diagnosis of AIDS or a low CD4+ T cell count ; \- Ongoing antiretroviral treatment ; \- « Elite controllers » who have a low or undetectable viral load (World Health Organization, editor. When and how to use assays for recent infection to estimate HIV incidence at a population level. Geneva, Switzerland: World Health Organization; 2011) The formula for calculating HIV incidence based on the results of a RITA therefore incorporates the false recent rate of the RITA into the incidence calculation. It is necessary to ensure that the false recent rate applied is relevant to the RITA and to the population for which the impact is estimated. This is why in our study we used the false recent rate from a sample of the Ivorian population. A sentence was added in the discussion : « As RITA is a biological assay, classification of infections as recent or not does not rely on self-reported information. Some people with long-standing HIV infection and on treatment may be misclassified as newly infected. However, these false recent cases are taken into account, as a false recent rate is applied when estimating incidence”. (lines 348-352) 10.1371/journal.pone.0271988.r003 Decision Letter 1 Sharifi Hamid Academic Editor 2022 Hamid Sharifi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 26 May 2022 PONE-D-22-02895R1Incidence of HIV infection and associated factors among female sex workers in Côte d’Ivoire, results of the ANRS 12361 PrEP-CI study using recent infection assaysPLOS ONE Dear Dr. N Marcellin Nouaman Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ==============================Dear Authors Thanks so much for submitting the revised file to PLOS ONE. The reviewers reviewed the revised file and they put a few more comments. Please revise the file based on the reviewers' comments. ============================== Please submit your revised manuscript by Jul 10 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Hamid Sharifi Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: (No Response) Reviewer \#3: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes Reviewer \#3: I Don't Know \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The data availability has restricted access and is not freely available. the manuscript needs to be edited - Table 1 and Table 2 -should state - 3 years or less and more than 3 years. no need to put more symbol as well as as state more. lines 158, 303, 308, 330, 348 need to be corrected Reviewer \#2: The authors have been responsive to reviewer comments made by myself and the other reviewers. I am mostly satisfied, however, I still strongly believe that it is inappropriate to provide confidence intervals for the HIV incidence rates because the data they are using comes from a convenience sample (not a probability-based sample). Formulas to calculate confidence intervals assume some kind of probability-based sample. I appreciate that the authors want to communicate a sense of uncertainty in their estimates but what is the benefit if the confidence intervals themselves are wrong and therefore ultimately uninformative? At the very least, this must be acknowledged in the limitations; although I think it would be more appropriate to remove the confidence intervals altogether (and explain why they are not included) or do a bootstrapping technique to estimate confidence intervals. Reviewer \#3: Thank you to the authors, once again, for this excellent manuscript. One of the initial questions was not addressed: 2)Lines 176-181: over the past several years, countries have been utilizing various assays for recent infection surveillance, with varying MDRI and FRR values. The specifics of the assay and algorithm used in this study need to be described in this section. In response to this question, the authors' response was: 2)In order to be more precise, we have completed the text : See section “Assessment of HIV incidence” for more details. (lines 185-186) However, the question was about the specifics of the recency assay utilized. The manuscript states (lines 179-182): "Then, a dried blood spot (DBS) sample was taken and transported to the laboratory of the University Hospital of Tours, France, to determine the window of infection (0.3 years) and false positive rate (13%) using a recent infection test adapted to the Ivorian context \[24\] and performed directly on plasma." Are there no additional details available regarding this assay? What is "a recent infection test adapted to the Ivorian context?" Is it a LAg-EIA? There are at least 10 different HIV recency assays currently and commercially available. Can the authors not shed any additional light on the assay that was utilized at University Hospital of Tours, akin to the specifics provided for the initial HIV rapid testing? \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0271988.r004 Author response to Decision Letter 1 4 Jul 2022 To the editors of the Journal of the PLOS ONE Abidjan, June 28th 2022 Response-to-reviewers letter and revised manuscript entitled “Incidence of HIV infection and associated factors among female sex workers in Côte d’Ivoire, results of the ANRS 12361 PrEP-CI study using recent infection assays” Dear Editors, This document provides point-by-point responses to the reviewers’ comments. Their comments, reproduced in italics, are followed by our responses and, finally by the additional text and changes to the manuscript second submitted. We thank the reviewers for their comments which have helped improve our paper. Reviewer : 1 The data availability has restricted access and is not freely available. The manuscript needs to be edited - Table 1 and Table 2 -should state - 3 years or less and more than 3 years. No need to put more symbol as well as as state more. lines 158, 303, 308, 330, 348 need to be corrected Authors’ response: We thank the reviewer for this comment. The data are available on [Zenodo.org](http://Zenodo.org), with the following link <https://zenodo.org/record/5948841>. Access to the data is available on request on the website [Zenodo.org](http://Zenodo.org). We mentioned this in the manuscript (line 212). In addition, we have attached in this new revision of the manuscript, the scripts and the analysis reports. Also, we have corrected tables 1 and 2 by removing the symbol at the level of 3 years or less and more than 3 which was unnecessary. Corrections have also been made to the lines 286, 287, 297, 300. -------------------------------------------------------------------------------- -------------------------------------- Reviewer : 2 The authors have been responsive to reviewer comments made by myself and the other reviewers. I am mostly satisfied, however, I still strongly believe that it is inappropriate to provide confidence intervals for the HIV incidence rates because the data they are using comes from a convenience sample (not a probability-based sample). Formulas to calculate confidence intervals assume some kind of probability-based sample. I appreciate that the authors want to communicate a sense of uncertainty in their estimates but what is the benefit if the confidence intervals themselves are wrong and therefore ultimately uninformative ? At the very least, this must be acknowledged in the limitations ; although I think it would be more appropriate to remove the confidence intervals altogether (and explain why they are not included) or do a bootstrapping technique to estimate confidence intervals. Authors’ response: We thank the reviewer for this relevant and important remark. Indeed, confidence intervals provide for the HIV incidence rates are inappropriate because our data become from a convenience sample. In order to find a more appropriate solution, we used a bootstrapping technique to estimate our confidence intervals as suggested by the reviewer. These confidence intervals were very wide, so we decided to remove them on the differents estimated incidences (see table 2) and we have mentioned this in the limitations of the study (lines 374 – 375). We have attached in this new revision of the manuscript, the scripts and the analysis reports for details. -------------------------------------------------------------------------------- ------------------------------------------- Reviewer : 3 Thank you to the authors, once again, for this excellent manuscript. One of the initial questions was not addressed: 2\) Lines 176-181: over the past several years, countries have been utilizing various assays for recent infection surveillance, with varying MDRI and FRR values. The specifics of the assay and algorithm used in this study need to be described in this section. In response to this question, the authors' response was: 2)In order to be more precise, we have completed the text : See section “Assessment of HIV incidence” for more details. (lines 185-186) However, the question was about the specifics of the recency assay utilized. The manuscript states (lines 179-182): "Then, a dried blood spot (DBS) sample was taken and transported to the laboratory of the University Hospital of Tours, France, to determine the window of infection (0.3 years) and false positive rate (13%) using a recent infection test adapted to the Ivorian context \[24\] and performed directly on plasma." Are there no additional details available regarding this assay ? What is "a recent infection test adapted to the Ivorian context ?" Is it a LAg-EIA ? There are at least 10 different HIV recency assays currently and commercially available. Can the authors not shed any additional light on the assay that was utilized at University Hospital of Tours, akin to the specifics provided for the initial HIV rapid testing? Authors’ response: We thank the reviewer for this relevant and important remark We have given further details by describing the principle of carrying out this test as follows : This recent infection test developed by Barin et al. is the Less-sensitive enzyme Immunodominant assay recent infection (EIA-RI/IDE-V3). This assay uses the enzyme immunoassay technique in a 96-well plate, based on the measurement of absorbances (OD) in one well sensitised with an equimolar peptide mixture TM (cons+D), corresponding to the immunodominant epitope of gp41 (consensus sequence envi - 1 group M and consensus sequence env - 1 subtype D), and in another well, sensitised with a V3 peptide solution (AE), corresponding to an equimolar mixture of the consensus sequences of the V3 region of gp120 of the HIV subtypes A, B, C, D and CRF01_AE. This test is an in-house test, applicable in the HIV-NRC virology laboratory, serology sector. The test can be performed on serum or plasma, as well as serum, plasma or whole blood on blotting paper or dried blood spots. The test uses a mathematical formula that combines the quantitative responses to gp41 antigens in each region to distinguish between recent and established infection. This in-house test has been the subject of preliminary studies using sequential serum samples from HIV-infected Ivorian patients with known dates of infection (the PRECO-CI ANRS 12277 and PRIMO-CI ANRS 1220 projects) and samples from patients at different stages of the disease (the Temprano ANRS 12136 and Trivacan ANRS 1269 trials) ; which allowed to distinguish a recent infection (≤180 days) from an established infection (\>180 days) with a window of infection (0.3 years) and false positive rate (13‰) for the Ivorian population studied. We have thus completed the paragraph “HIV screening and laboratory” in the manuscript (lines 185 - 202) 10.1371/journal.pone.0271988.r005 Decision Letter 2 Sharifi Hamid Academic Editor 2022 Hamid Sharifi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 Jul 2022 Incidence of HIV infection and associated factors among female sex workers in Côte d’Ivoire, results of the ANRS 12361 PrEP-CI study using recent infection assays PONE-D-22-02895R2 Dear Dr. N Marcellin Nouaman We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, Hamid Sharifi Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0271988.r006 Acceptance letter Sharifi Hamid Academic Editor 2022 Hamid Sharifi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 8 Nov 2022 PONE-D-22-02895R2 Incidence of HIV infection and associated factors among female sex workers in Côte d’Ivoire, results of the ANRS 12361 PrEP-CI study using recent infection assays Dear Dr. Nouaman: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Hamid Sharifi Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist. [^2]: ¶ The complete membership of the author group can be found in the Acknowledgments.

# Introduction More than 60 diseases are known to increase the risk of psychotic disorders in childhood and adolescence; they include genetic syndromes, inborn errors of metabolism, and autoimmune, neurologic, endocrinological, and nutrition disorders. As many as 12.5% of cases of childhood psychosis may have a medical (somatic) disorder contributing to the clinical presentation (ex: homocystinuria or intermittent porphyria) but their frequency in non-academic settings may be inferior given their more generalist patient recruitment. The involvement of such a disorder should be considered when the following signs are present: visual hallucinations, confusion, catatonia, fluctuating symptoms and intellectual deficiency, as well as when the course appears abnormal (early onset, sudden onset, progressive cognitive decline, or treatment resistance). Although relevant guidelines for organicity investigations are available, many of the disorders present without easily recognized phenotypes and are thus easily missed. Nevertheless, it is important to ensure these diagnoses are made, because even if most of psychoses of organic origins remain inaccessible to specific treatment, some can benefit from suitable treatment and significant clinical improvement. Diagnosing medical and genetic causes and risk factors of early psychosis is a challenge. This challenge includes both training of psychiatrists and implementing organicity investigations throughout all psychiatric units. Eliminating the research-to-practice gap is a challenge of implementation science. Eliminating the training-to-practice gap is another. Implementing changes in medical practice is a major matter that needs research. Collecting the professionals’ perspective on such matters is an important aspect of implementation research. Community-based centers, nonacademic hospital departments of child psychiatry, and one-stop youth-friendly medical services offer generalist care for a broad range of disorders. On the contrary, “expert centers” are at the forefront of research on psychosis and provide complete psychiatric assessments, including thorough organicity investigations. Most of them are part of university hospitals. How can organicity investigations provided in specialized expert centers be implemented effectively in nonacademic facilities? This study explores how practitioners trained in expert centers but working in nonacademic facilities use organicity investigations, including physical assessment and relevant laboratory workups, for their young patients with suspected early psychosis. We believe their personal perspectives will help identify the barriers to organicity investigations’ implementation and overcome them. # Material and methods This qualitative research was conducted in compliance with the COREQ guidelines. Participants were recruited via the mailing lists of three professional associations of psychiatrists working in the Paris region. The participants had training in organic causes of psychosis at expert centers but were currently working in nonacademic facilities in which they routinely treated adolescents and young adults. One-hour semi-structured face-to-face interviews were conducted by a male fourth-year resident in psychiatry (JLK) and took place in the participants’ office with no third-party present. They focused on their daily use of organicity investigations in their workplaces, their training and professional experience, the reactions of their colleagues, patients, and families to these organicity investigations, and their contacts with expert centers. The participants were not made aware of the interviewer’s objective and no relationship was established prior to the interview. The interviews were sound recorded, transcribed and analyzed with Grounded Theory, a standard methodology for social science research, which is particularly suited to understanding professional practices and organizational cultures. Grounded Theory links social structures with processes occurring at an individual level by focusing on themes that represent underlying interactions and their consequences. The transcripts were not returned to the participants for comment or correction. As in other inductive methods, no exact number of respondents was required before the research began. The data were coded to generate categories, which were then validated through constant comparisons as new interviews were performed. Thus, data analysis, further sampling, and theoretical development proceeded simultaneously until saturation was reached. To ensure reliability, two researchers (JLK, LB (female; PhD) independently coded and analyzed all data and their findings were discussed during research team meetings, a process called triangulation. No software was used during this process to manage the data. Consistency between the data and the findings was assessed by triangulation and participant feedback. No ethical approval was necessary since no patient data was used in this study. All participants gave informed consent. # Results Sixteen psychiatrists (4 male and 12 female) aged 28 to 60 years (mean: 37; SD: 9,38) were selected through snowball sampling and all of them agreed to participate in the present study. Four were residents, six were fellows and seven were senior practitioners. Most participants (14/16) had subspecialized in child and adolescent psychiatry during their medical studies. Four participants provided direct feedback to the researchers and agreed with the findings. The results of the qualitative analysis are summarized in. In this figure, major themes are presented with a blue background and minor themes with a withe background. For further details on the analysis, see. Organicity investigations were found to be useful in rationalizing psychiatric care for the young patient all the while building trust between the doctor and the patient’s parents. However, in their quest for implementation, recently trained psychiatrist faced challenges that left neither their practice nor their new workplace unchanged. ## A reassuring relational tool but a short-term avoidance of psychiatry ### Advantages of organicity investigations Organicity investigations were found to be a relational tool for families and a reassuring tool for doctors. Psychiatrists viewed organicity investigations as a way to initiate the medical relationship with the patient’s parents whom they expected to be anxious about their first contact with psychiatry. They announced the carrying out of organicity investigations at the earliest opportunity to reassure them and to bespeak their expertise to the patient’s family. Besides, organicity investigations postponed facing the possibility of a psychiatric label and its social stigma, a possibility the doctors did not expect families to welcome. When dealing with adolescents with early and unsettled symptoms, the concrete and scientific nature of organicity investigations helped in managing the medical uncertainty of *“unclear*, *imprecise*, *invisible”* psychiatry by providing a feeling of reassuring certainty of the absence of a somatic disorder. Once completed, they provided reliable results and legitimized a shift away from diagnosis and a focus on psychiatric care. Even though, as the saying goes, the absence of evidence isn’t the evidence of absence (these investigations only concern the somatic causes that we currently know about), they were therefore seen as a mandatory prerequisite to starting psychiatric care. ### Limitations of organicity investigations Organicity investigations are a pretext for short-term avoidance of psychiatry. Furthermore, they are unwelcomed for the patient and constraining for physicians. The initial relief provided by the organicity investigations was short-lived, however, because in the end, parents mostly found their child diagnosed with an emerging mental illness. Inversely, the young patients were described as either indifferent or opposed to organicity investigations. Psychiatrists attributed this opposition to psychiatric symptoms (such as delusions) rather than any reasoned decision. There are therefore two relational drawbacks identified by the participants. The first was the fact that the avoidance of dealing with families’ worries about their child’s long-term mental health, future schooling and social integration was only slightly delayed. Secondly, they sometimes found that trying to convince their patient at any cost generated tension. Lastly, organicity investigations were seen as constraining, given that they require clinical expertise to identify simultaneously atypical mental symptoms and signs of physical pathology. Participants found the identification of rare diseases a difficult body of knowledge to master and maintain; the skills they had acquired during their training in expert centers gradually fading. Furthermore, discouraging technical obstacles turned organicity investigations a meaningless and time-consuming procedure for psychiatrists working in nonacademic facilities. ## To transform: Implementing organicity investigations as a recently trained psychiatrist ### Young experts joining an aging institution Young experts joining an aging institution feel the duty to implement medical progress within the psychiatry department and must disseminate the knowledge tactfully. Recently trained participants considered themselves ethically bound to prescribe up-to-date medical investigations, such as organic and neurometabolic workups. The importance of organicity investigations seemed rooted in their belief that such investigations preserved their identity as medical doctors. Given their background in expert centers and their senior colleagues’ lack of knowledge about organicity investigations, residents found themselves in the position of experts in their new workplace and considered it a duty to be a driving force for their implementation. Some described a messianic view of their role in transmitting expert center know-how. However, the specificity of organicity investigations, on the edge of medicine and psychiatry, sometimes generated disinterest or rejection which encouraged some participants to tread lightly and disseminate their knowledge tactfully. Attempts to implement organicity investigations, despite colleagues’ *“somewhat hermetic attitude”* sometimes resulted in a segmentation of care or a delegation to practitioners of other specialties or expert centers, and sometimes aggravated political tensions between hospital departments, with the refusal to change practices resulting from institutional disagreements. ### Reshaping the use of organicity investigations Faced with the constraints of their new workplace, participants either persevered in their use of organicity investigations, renounced, or settled for a pragmatic patchwork of tests. Training in organicity investigations, however complete, did not appear to guarantee that the participants performed them in their nonacademic workplaces. None of the psychiatrists disputed the importance of organicity investigations in addressing emerging psychosis in adolescents. The constraints of their daily work reshaped their use, however: The practices of some participants remained modeled on that of the expert center they had been trained at. They inflexibly ordered the same set of examinations, thus conforming to the primacy of evidence-based medicine over the work organization of their non-academic facility. Other participants explained that the lack of support from their colleagues had exhausted their initial enthusiasm and led them to gradually halt organicity investigations orders within a few months. Meanwhile, the rarity of conclusive evidence in daily practice confirmed their abandonment without systematically referring patients to an expert center. Lastly, some adopted a more pragmatic method by taking into account the financial constraints of nonacademic facilities and prioritizing less costly explorations. Similarly, local technical possibilities sometimes dictated which investigations were done. Although only partially satisfactory, such strategies allowed a pragmatic compromise between medical requirements and the reality on the ground. ### Seeking external support Our participants feel they need external support in order to refer patients, partner up for complex cases and get accurate somatic assessment. Those without a professional network experience loneliness. The specific hindrances arising from the use of organicity investigations in non-academic workplaces and the reshaping it necessitates lead participants in a search for support and collaboration. For nonacademic psychiatrists, sharing common challenges with expert center colleagues enabled: - a collegial decision-making process about complex cases, therefore splitting the heavy medical responsibility of diagnosis and treatment choice. - mediation of the relationship with the patient and his or her family by providing external and expert validation of the proposed care, thereby strengthening the therapeutic alliance. Despite their training, some participants perceived former expert center colleagues as more competent and consulted them in order to refer their patients. Nonetheless, due the prestige of expert centers, their relations with their former colleagues sometimes generated feelings of misunderstanding, harsh judgments, and lack of dialogue between these two distinct worlds. Furthermore, the rarity and the diversity of diseases sought require a dialogue between psychiatrists and somatic specialists for accurate somatic assessment and optimal care. However, such cooperation was found to be difficult to establish. The psychiatrists’ enthusiasm for organicity investigations contrasted with the lack of commitment of the general practitioners working in nonacademic facilities. Therefore, participants were worried about the stigmatization of psychiatric patients would lead to less conscientious assessments. Practitioners who did not have a wide professional network experienced difficulty in finding specialists with whom to share their uncertainties. Consequently, the autonomy granted to the trained psychiatrists gave them the freedom to practice as they saw fit but generated loneliness and isolation. ## To be transformed: Changes in practices, as psychiatrists gain experience The use of organicity investigations evolves across the recently trained psychiatrists’ working lives. Thus, we can sketch a typical career path that often leads to the abandonment of the practices recommended by the expert center they trained in because of the constraints of their work in nonacademic departments. Each of these stages will be described by the ideal type defined by Weber, the aim of which is to highlight the most significant traits to enable a better understanding of the social action involved. ### The resident on a mission Young residents, driven by their professional ethics, are keen to spread the use of organicity investigations. Convinced by their teachers, young practitioners want to apply their expertise in their new environment. They feel they have a mission: to save their young patients long years of psychiatric care by diagnosing a curable disease. Accordingly, they are usually keen to spread this practice, they do not hesitate to manage complex clinical situations, or use their network in expert centers to seek help. In nonacademic settings where no protocol covers organicity investigations, professional ethics drive its pursuit. ### The rise of an inner conflict Over time, professional duties and local constraints generate doubt regarding the use of organicity investigations. An intensive work rhythm puts daily pressure on the initial enthusiasm of these recently trained practitioners. They come to doubt the possibility of pursuing the use of organicity investigations as they learned it while fulfilling their responsibilities in their nonacademic setting. Aware of the inconsistency between the time-consuming use of organicity investigations and their department’s inaction in regard to organizational planning, they begin to advocate a systematization of practice by developing departmental protocols. A compromise solution is then adopted to apply organicity investigations routinely: they no longer try to clinically identify atypical symptoms that justify carrying out targeted complementary examinations, but rather automate the use of organicity investigations to facilitate it. These explorations are then chosen based on financial constraints and technical possibilities rather than on clinical relevance. Thus, prescribing organicity investigations turns into a meaningless routine. ### Senior psychiatrists who give up organicity investigations Ultimately, trained psychiatrists renounce the use of organicity investigations, but trust expert centers will go on. Senior participants admitted that they were overwhelmed and *“saddened”* by the fact that their efforts at change had failed to move their teams out of the *’listlessness”* of nonacademic departments. They now referred their patients to expert centers. Their awareness of their renunciation resonated as a painful disavowal of their former commitments and as a betrayal of their teachers. # Discussion In this study, the obstacles to the implementation of organicity investigations came from the institution’s routine and overall inertia (leaving aside budgetary and technical problems). Psychiatrists and patients’ relatives find organicity investigations reassuring at a time when they are faced with the possibility of psychosis onset. In contrast, convincing the patient sometimes proved difficult, and psychiatrists tried to provide guidance and balanced information during these trying times. Their goal was to avoid the entanglement of care and coercion which affects caregivers, the patients and their families. Their feeling of exigency to implement organicity investigations may also originate from the relatively high frequency of psychoses of organic origin in the expert centers they trained at. Nonetheless, the generalist care nonacademic facilities are assigned to perform, the high level of expertise required for organicity investigations, and the rarity of the diseases investigated ultimately led to unsatisfactory use of organicity investigations or systematic referral of patients to expert centers. The participants also felt isolated from the institutions they had trained in and ultimately lost contact with colleagues to consult when needed. The science of implementation provides a relevant framework for the analysis of these findings and may help explain the gap between evidence-based practices and what is provided to consumers in routine care. Indeed, obstacles to change practices can arise at several levels in the healthcare system: the patient, the professional, team, organization, or environment. In the field of mental health, 15 to 20 years can separate the establishment of such practices and their widespread generalization, a gap that prevents patients from reaping the benefits of costly research. ## Relational advantages of organicity investigations Our results showed that organicity investigations were not only used as a technical medical tool but were especially considered as a way to manage the relationship and build trust with patients and their families. Indeed, despite the fact that these disorders are also associated with functional remission, psychiatrists envision psychosis as a disastrous life-long disorder and this viewpoint shapes their reluctance to disclose psychosis risks, a choice related to their belief in the self-fulfilling prophecy, and their knowledge of the stigma generated by a psychiatric label. Organicity investigations thus convey a reassuring message: it may be possible to avoid psychiatric diagnosis and find a curable illness instead. The prospect of early psychosis in a patient exposes practitioners to the uneasiness of medical uncertainty. Indeed, doctors face many uncertainties inherent in medical knowledge and are thus trained to remain in control in their daily practice. Psychiatry is especially subject to uncertainty because of its current technical and theoretical immaturity. The proper use of organicity investigations at the onset of psychosis is a good example: recommendations are fairly recent, not widely disseminated, and research is still embryonic. In completing academic courses in organicity investigations or training in expert centers, participants attempted to control medical uncertainty and as the same time used all of the latest medical tools available for their young patients’ care. They considered, as do we, that no distinction should be made between psychiatry and “somatic medicine”. Thus, organicity investigations are an integral part of a psychiatrist’s job. ## Institutional obstacles for organicity investigations implementation Our participants have experienced the difficulties of implementing alone a novel medical practice in a workplace with its own work culture and habits. Laypeople may think that medical knowledge and techniques are widely and rapidly adopted by the medical community when their superiority is established. This process is not nearly so straightforward, however. Behind the indisputable assertions of “finished science” lie many controversies and decisions not solely related to science. Studying science “in the making” makes it possible to pinpoint the moment when scientific discoveries might have gone in many other directions. Indeed, this study depicts the intricacies of the adoption of organicity investigations: rivalries between departments and professionals, budgetary and technical criteria, as well as institutions’ desire (or lack thereof) for change. This highlights the existence of segments in the medical profession. Behind its apparent homogeneity lie many disparities—in values, interests, and methodologies—that segment it, creating opposing dynamic forces that can lead to institutional change. They also produce an informal division of labor, with each segment delegating work to another. In our study, the segment of young experts urges the opposing segment—their nonacademic colleagues—to implement organicity investigations, with various outcomes. The multiplicity of parties involved in the decision-making process (psychiatrists, other specialists, administrators) and its centralized nature make any change long and its outcome uncertain. This “bureaucratic phenomenon”, characterized by its rigidity and inefficiency, becomes a source of tedious work (time-consuming paperwork and procedures) and frustration which clashes with participant’s professional ethics. Furthermore, various characteristics of research evidence affect its usability in clinical practice. For instance, change is difficult if the innovation requires complex changes (in the decision-making process or the acquisition of new skills), better collaboration between disciplines, or changes in the organization of care. Adherence to new guidelines also depends on the type of health issue: it is better for acute than chronic care. All of these findings help explain the difficulty the participants have had in implementing organicity investigations in their nonacademic facilities: it concerns a chronic health issue and necessitates complex clinical evaluations and close collaboration between practitioners of different specialties. Targeted interventions (small group interactive training, feedback on performance, computer assistance), which have been shown to be the most effective way to transfer evidence into practice, might usefully be applied to improve the dissemination of organicity investigations. ## Interactional obstacles for organicity investigations implementation The newly trained participants are the only ones with strategic knowledge on organicity investigations, they are the necessary link between two systems that otherwise have trouble communicating: the expert center and the nonacademic department. It has however been shown that better collaboration and recognizing the “felt” power differential between institutions are essential to improve relations in trying times such as psychosis onset. The newly trained participants therefore are at the center of a creative process that can lead to the adoption of organicity investigations. It is nonetheless a challenging position because they are not recognized as an integral part of either system, and they risk isolation when difficulties arise. Interactions between the participants and their colleagues led to what A. Strauss calls “self-appraisals” and ultimately to the adoption of new ethics, or alternatively to the abandonment of old ones. Furthermore, the lack of support from the organization’s leadership–experienced by some participants–is known to be a major barrier to implementation of new guidelines in routine practice. According to Becker, medical trainees never lose their idealistic view of medicine; they simply adapt realistically to the situations they face during their training. By interacting with their peers and teachers, they become “pragmatically idealistic”. We have shown this transformation by defining a “career path” in three characteristic stages, focusing not on *who* the participants are, but on *what* they are doing, and thus describing the social processes underlying their behavior. Indeed, daily interactions come into play over the years and lead young practitioners away from their mission by making them doubt the possibility of implementing organicity investigations in their nonacademic department. Ultimately, these processes can lead them to a heavyhearted abandonment or delegation of its use. # Conclusion Implementing organicity investigations in nonacademic departments is not simply a matter of access to training in expert centers but necessitates facing relational, institutional and interactional challenges which involve facing the medical uncertainty inherent to psychosis onset and an extensive reorganization of care. Surprisingly, we also found organicity investigations to be a tool to build trust between physicians and the patients’ parents. Further research might be helpful in quantifying the current benefit for patient care in regard to the time and effort conceded by professionals. Tackling these challenges is nonetheless essential to guarantee the best care available to young patients, especially because expert centers cannot shoulder the burden of these investigations alone. To that end, we suggest the following measures: # Supporting information 10.1371/journal.pone.0252610.r001 Decision Letter 0 Hutchinson Gerard Academic Editor 2021 Gerard Hutchinson This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Mar 2021 PONE-D-20-37522 Implementing organicity investigations in early psychosis: Spreading expertise PLOS ONE Dear Dr.Kurukgy  Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. One of the major areas of that must be addressed is the description, meaning  and perhaps renaming of 'youth psychiatrists' Kindly address all of the comments made by all of the reviewers. Please submit your revised manuscript by April 17th, 2021.  If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Gerard Hutchinson, MD Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <h ttps://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_b ody.pdf> and [https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf](about:blank) 2\. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see <http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data- access-restrictions>. In your revised cover letter, please address the following prompts: a\) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent. b\) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see <http://www.bmj.com/content/340/bmj.c181.long> for guidelines on how to de- identify and prepare clinical data for publication. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data- availability#loc-recommended-repositories> We will update your Data Availability statement on your behalf to reflect the information you provide. 3\. Your ethics statement should only appear in the Methods section of your manuscript. If your ethics statement is written in any section besides the Methods, please move it to the Methods section and delete it from any other section. Please ensure that your ethics statement is included in your manuscript, as the ethics statement entered into the online submission form will not be published alongside your manuscript. 4\. Please upload a copy of Figure 1, to which you refer in your text. If the figure is no longer to be included as part of the submission please remove all reference to it within the text. 5\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: [http://journals.plos.org/plosone/s/supporting-information](about:blank). Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. Additional Editor Comments (if provided): \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: N/A Reviewer \#2: N/A Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors were interested in a subject that has not yet been widely treated: the practical realization of organicity investigations (OI), particularly barriers and drivers to them, in adolescents with early psychoses treated in non-academic psychiatric centers, by interviewing French psychiatrists trained in expert centers about their practices. The authors identified several advantages to this OI: facilitating the initiation of a trusting relationship (especially with the patients' relatives), having a "technical" tool to hold on to, corresponding to a certain professional identity (notably professional requirement and conscience), and disadvantages: complexity of the practical realization, negativity in the vast majority of cases leading to a psychiatric diagnosis of difficult acceptance, high level of competence required, difficulties in changing practices and a consequent feeling of isolation The authors pointed out the evolution of practices between the training received in expert centers ("the ideal") and the daily life then implemented by the practitioners ("the reality"), with a certain diversity of the professionals' paths (compromise between these 2 visions, abandonment...). This research is very interesting because the subject is crucial given the potentially curable nature of certain diseases. The authors underline the extreme complexity of the subject given the absence of international consensual guidelines, the need for important competences and close links with paediatricians and expert centers given the large number of pathologies to be evoked, links that are not always well protocolized in practice. The discussion is well structured, and the authors propose measures to improve the situation as an opening remark. There are no major problems in this very well-written and enjoyable article. Some minor points: \- INTRODUCTION: Nuance the importance of recognizing psychoses of organic origin, most of which unfortunately remain inaccessible to specific treatment. This point could also be addressed in the discussion, since there seems to be a discrepancy between the perception of the frequency of organically-induced psychoses accessible to treatment (and which should therefore not be "missed") and the statistical reality of these very rare situations, a discrepancy that may be accentuated by the expert centers, which themselves only (or almost only) see organically-induced psychoses. This is perhaps partly responsible for the feeling of exigency of the newly trained practitioners, and may give rise to a "messianic" vision of their work in non-academic centers. \- RESULTS. 1.1 Caution about the impression of reassurance induced by the "elimination" of any somatic cause. "Absence of evidence is not evidence of absence". \- DISCUSSION: Clinical practice guidelines: consider a specific point concerning the improvement of collaboration with and training of potentially involved somaticians (pediatricians, general practitioners, neurologists, internists, etc.) Reviewer \#2: This is an interesting study that aimed to understand the challenges underlying the implementation of organicity investigations (OI) in non-academic facilities by practitioners trained in expert centers. I have the following comments: It is stated that providing medical (somatic) disorders can lead to suitable treatment and significant clinical improvement \[for psychosis\]. What is the evidence for this? Considering how strong this premiss is, it is important to review the evidence in support for this claim. It is suggested that organicity investigations in relation to psychosis diagnosis are already provided in specialized expert centres. It is important critically review how much these organicity investigations improve psychosis diagnosis and subsequent treatment outcomes in patients with psychoses. It is equally important describe how much extract time and effort organicity investigations will require from both clinicians and patients. It they entail more additional time and effort for almost no improvement in diagnosis precision and treatment outcomes in patients with psychosis then they need for implementation non-academic facilities may be questioned. When describing methods, it is stated that participants recruited in this study had training in organic causes of psychosis at expert centers but were currently working in non-academic facilities. Do the authors mean psychiatrists, trained psychologists or students? Please provide more detail on such an important aspect of the study. The description of the sample in the results section is also necessary including mean and SD of age, gender distribution years of experience and level of educational attainment. Also please explain what “youth psychiatrists” are. I assumed it was referring to psychiatrists who were young, but this may be incorrect interpretation. It is quite misleading and factually inaccurate to say that psychosis as a disastrous life-long disorder. Over 60% of people with a diagnosis of psychosis fully recover within the first 2-5 years The overall purpose of the study is not very clear. From the introduction it sounded like the aims of this work were to investigate how OI can be implemented in non-academic institutions. However, it appears that the actual aims were to identify personal perspectives of people with OI training on the benefits of OI and obstacles of their implementations. This disparity is rather confusing. The conclusions in this work are based 0n personal opinions of 16 youth psychiatrists, who may be rather naïve or inexperienced to make rational recommendations. To make the conclusions stronger psychiatrists with more years of experiences need to be included in this study. Additionally, evidence from research studies, such as CRT, in support to the conclusions would be helpful. Reviewer \#3: Very useful article! Althoug some wording revision should be applied: page 3, last sentence. replace semicolons by commas. Page 4: first sentence: Sixteen "youth" psychiatrists aged 28 to 60. I suggest to remove the word "youth" given the fact that the age range goes far beyong youth. Page 7: from third line: very long sentence. Try to improve wording by separating sentences. Page 8: "The practices of some participants remained modeled on that of the expert center they had \[been\] trained at." The verb on the sentence was missing. Page 9: wording suggestions: "Therefore, participants \[were\]worried \[about\] the stigmatization of psychiatric patients would lead to less conscientious assessments" Figure 1: I don't fully grasp the difference between the second ("to transform") and the third ("to be transformed") categories. I suggest to choose other words to highlight the core ideas \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: **Yes: **Paula Herrera-Gomez \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0252610.r002 Author response to Decision Letter 0 20 Apr 2021 Response to reviewers – Implementing organicity investigations in early psychosis: Spreading expertise Dear editorial team, Dear reviewers, It is with great pleasure that we submit for your consideration our revised manuscript. After careful examination of each point raised by the academic editor and the reviewers, the necessary changes have been made and are highlighted in the file, as requested. The term “youth psychiatrist”, used in our article to designate psychiatrists specialized in adolescents’ and young adults’ mental health, has been found to be confusing and unclear. This issue has been addressed by further detailing the participants’ status and specialty as well as their work settings. Following is a detailed list of the changes made to comply to the journal requirements: 1\. The manuscript now meets PLOS ONE's style requirements, including those for file naming, as detailed in the formatting guidelines. 2\. There are no ethical or legal restrictions on sharing de-identified data. Therefore, you will find in “S1_File” the minimal anonymized data set necessary to replicate the study findings 3\. The “ethics” statement has been moved to the required location and the “competing interest” statement has been removed from the manuscript 4\. Figure 1 has been uploaded in the manuscript. It can be found at the beginning of the Result section 5\. Captions for the Supporting Information file has been included at the end of the manuscript and the in-text citation has been updated to match accordingly Response to Reviewer 1 : • INTRODUCTION: Nuance the importance of recognizing psychoses of organic origin, most of which unfortunately remain inaccessible to specific treatment. This point could also be addressed in the discussion, since there seems to be a discrepancy between the perception of the frequency of organically-induced psychoses accessible to treatment (and which should therefore not be "missed") and the statistical reality of these very rare situations, a discrepancy that may be accentuated by the expert centers, which themselves only (or almost only) see organically-induced psychoses. This is perhaps partly responsible for the feeling of exigency of the newly trained practitioners, and may give rise to a "messianic" vision of their work in non-academic centers. • We agree with the reviewer and have modified the manuscript as follows: “As many as 12.5% of cases of childhood psychosis may have a medical (somatic) disorder contributing to the clinical presentation (ex: homocystinuria or intermittent porphyria) but their frequency in non-academic settings may be inferior given their more generalist patient recruitment (1).” And “Nevertheless, it is important to ensure these diagnoses are made, because even if most of psychoses of organic origins remain inaccessible to specific treatment, some can benefit from suitable treatment and significant clinical improvement (2).” And in the Discussion section: “Their messianic enthusiasm to implement organicity investigations may also originate from the relatively high frequency of psychoses of organic origin in the expert centers they trained at.” • RESULTS. 1.1 Caution about the impression of reassurance induced by the "elimination" of any somatic cause. "Absence of evidence is not evidence of absence". • We thank the reviewer for this comment and for providing us this adage which highlights the subjective nature of the reassuring feeling provided by organicity investigations. We have taken the liberty to insert it as is in the manuscript: “(…) the concrete and scientific nature of organicity investigations helped in managing the medical uncertainty of “unclear, imprecise, invisible” psychiatry by providing a feeling of reassuring certainty of the absence of a somatic disorder. Once completed, they provided reliable results and legitimized a shift away from diagnosis and a focus on psychiatric care. Even though, as the saying goes, the absence of evidence isn’t the evidence of absence (these investigations only concern the somatic causes that we currently know about), they were therefore seen as a mandatory prerequisite to starting psychiatric care.” • DISCUSSION: Clinical practice guidelines: consider a specific point concerning the improvement of collaboration with and training of potentially involved somaticians (pediatricians, general practitioners, neurologists, internists, etc.) • We agree with the reviewer and have added a specific point in the manuscript: “Sustained collaboration and training with somaticians (such as pediatricians, neurologists…) may improve the psychiatrists’ approach and knowledge of somatic diseases and strengthen the partnership in patient care”. Following the comment of Reviewer 2, the “Clinical Practice Guidelines” have been retitled “Proposed initiatives to facilitate organicity investigations’ implementation in nonacademic departments”. Response to Reviewer 2 : • It is stated that providing medical (somatic) disorders can lead to suitable treatment and significant clinical improvement \[for psychosis\]. What is the evidence for this? Considering how strong this premiss is, it is important to review the evidence in support for this claim. • We agree with the reviewer and have added a citation to back the statement that suitable treatment can lead to clinical improvement for psychosis of organic origins. (Merritt J, Tanguturi Y, Fuchs C, Cundiff AW. Medical Etiologies of Secondary Psychosis in Children and Adolescents. Child and Adolescent Psychiatric Clinics of North America. 2020 Jan 1;29(1):29–42). • It is suggested that organicity investigations in relation to psychosis diagnosis are already provided in specialized expert centres. It is important critically review how much these organicity investigations improve psychosis diagnosis and subsequent treatment outcomes in patients with psychoses. • We agree with the reviewer and have specified which citation illustrates the impact of organicity investigations on psychosis diagnosis and treatment (Giannitelli, M., Consoli, A., Raffin, M., Jardri, R., Levinson, D. F., Cohen, D., & Laurent-Levinson, C. (2018). An overview of medical risk factors for childhood psychosis: Implications for research and treatment. Schizophrenia Research, 192, 39–49. <https://doi.org/10.1016/j.schres.2017.05.011>) • It is equally important describe how much extract time and effort organicity investigations will require from both clinicians and patients. It they entail more additional time and effort for almost no improvement in diagnosis precision and treatment outcomes in patients with psychosis then they need for implementation non-academic facilities may be questioned. • We agree with the reviewer that the improvement in diagnosis and treatment has yet to be compared to the extra time and effort put into their carrying out in nonacademic settings. We have not found studies addressing this topic and this was one of our main motivation for conducting this research. However, our data suggests that the effort and time needed is currently perceived as a significant obstacle to sustained use of organicity investigations (cf. “Reshaping the use of organicity investigations” in the Results section). This limitation was highlighted in the Conclusion: “Further research might be helpful in quantifying the current benefit for patient care in regards to the time and effort conceded by professionals.” • When describing methods, it is stated that participants recruited in this study had training in organic causes of psychosis at expert centers but were currently working in non-academic facilities. Do the authors mean psychiatrists, trained psychologists or students? Please provide more detail on such an important aspect of the study. The description of the sample in the results section is also necessary including mean and SD of age, gender distribution years of experience and level of educational attainment. • We thank the reviewer for this comment and have provided the required information: “Sixteen youth psychiatrists (4 male and 12 female) aged 28 to 60 years (mean: 37 ; SD : 9,38) and working in a variety of settings, were selected through snowball sampling and all of them agreed to participate in the present study (see Table 1). Four were residents, six were fellows and seven were senior practitioners.” The type of education attainment regarding organicity investigations is listed for each participant in Fig. 1: “Residency in EC”, “Fellowship in EC” and “Academic courses”. • Also please explain what “youth psychiatrists” are. I assumed it was referring to psychiatrists who were young, but this may be incorrect interpretation. • We thank the reviewer for giving us the opportunity to clarify this point: the term “youth psychiatrist” has been removed as it seemed to be confusing. We have addressed this issue by specifying the participants’ work setting as well as their professional status and level of specialization: “The participants had training in organic causes of psychosis at expert centers but were currently working in nonacademic facilities in which they routinely treated adolescents and young adults. (…) The majority of participants (14/16) had subspecialized in child and adolescent psychiatry during their medical studies. » • It is quite misleading and factually inaccurate to say that psychosis as a disastrous life-long disorder. Over 60% of people with a diagnosis of psychosis fully recover within the first 2-5 years • We agree with the reviewer regarding the recovery of people with a diagnosis of psychosis. The identification of psychosis as a “life-long disaster” by psychiatrists originates from a qualitative study addressing the psychiatrists’ representation and attitude towards psychosis risk management To highlight the discrepancy between these representations and the reality of the evolution of such disorders, we have modified the manuscript as follows : “Indeed, despite the fact that these disorders are also associated with functional remission (25,26), psychiatrists envision psychosis as a disastrous life-long disorder (27) and this viewpoint shapes their reluctance to disclose psychosis risks, a choice related to their belief in the self-fulfilling prophecy (28), and their knowledge of the stigma generated by a psychiatric label (29–32). » • The overall purpose of the study is not very clear. From the introduction it sounded like the aims of this work were to investigate how OI can be implemented in non-academic institutions. However, it appears that the actual aims were to identify personal perspectives of people with OI training on the benefits of OI and obstacles of their implementations. This disparity is rather confusing. • We thank the reviewer for the opportunity to clarify the aim of our study. The overall aim of this study was to enquire to the professionals “on the field” about their usage of organicity investigations to determine the barriers to their implementation. In the framework of implementation science, the evidence- based practice is distinguished from its implementation process and the professionals’ perspectives play an important role. (Bauer, M.S., Damschroder, L., Hagedorn, H. et al. An introduction to implementation science for the non- specialist. BMC Psychol 3, 32 (2015). <https://doi.org/10.1186/s40359-015-0089-9>). To clarify the interconnexion between the professionals’ perspective and the implementation of organicity investigations, we have modified the Manuscript as follows : “Implementing changes in medical practice is a major matter that needs research. Collecting the professionals’ perspective on such matters is an important aspect of implementation research (10). (…) This study explores how practitioners trained in expert centers but working in nonacademic facilities use organicity investigations, including physical assessment and relevant laboratory workups, for their young patients with suspected early psychosis. We believe their personal perspectives will help identify the barriers to organicity investigations’ implementation and eliminate them.” To further highlight this, we have also changed the title of our table in the Conclusion: “Proposed initiatives to facilitate organicity investigations’ implementation in nonacademic departments”. • The conclusions in this work are based 0n personal opinions of 16 youth psychiatrists, who may be rather naïve or inexperienced to make rational recommendations. To make the conclusions stronger psychiatrists with more years of experiences need to be included in this study. Additionally, evidence from research studies, such as CRT, in support to the conclusions would be helpful. • We thank the reviewer for this comment. We believe our answers to the previous questions and subsequent Manuscript modifications address these issues. We agree with the reviewer regarding the usefulness of CRT in preventing relapse and improving cognitive efficacy in patients’ daily lives. (Mueller, D., & Roder, V. (2017). 101. Does Cognitive Remediation Therapy Prevent Relapses in Stabilized Schizophrenia Outpatients? A 1-Year RCT Follow-Up Study. Schizophrenia Bulletin, 43(Suppl 1), S53–S54. <https://doi.org/10.1093/schbul/sbx021.139> ; Amado, I., Moualla, M., Jouve, J., Brénugat-Herné, L., Attali, D., Willard, D.... & Yannick, M. (2020). Employment, Studies and Feelings: Two to Nine Years After a Personalized Program of Cognitive Remediation in Psychiatric Patients. Frontiers in Psychiatry, 11, 609 ; O’Reilly, K., Donohoe, G., O’Sullivan, D. et al. A randomized controlled trial of cognitive remediation for a national cohort of forensic patients with schizophrenia or schizoaffective disorder. BMC Psychiatry 19, 27 (2019). <https://doi.org/10.1186/s12888-019-2018-6>). However, as stated earlier, the aim of our study wasn’t to provide guidelines to improve the care of patients with psychosis, and therefore, we have not discussed the efficacy of psychological or psychosocial interventions. Response to Reviewer 3 : • page 3, last sentence. replace semicolons by commas: • We agree with the reviewer and have modified the manuscript as follows: “They focused on their daily use of organicity investigations in their workplaces, their training and professional experience, the reactions of their colleagues, patients, and families to these organicity investigations, and their contacts with expert centers.” • Page 4: first sentence: Sixteen "youth" psychiatrists aged 28 to 60. I suggest to remove the word "youth" given the fact that the age range goes far beyong youth. • We thank the reviewer for this comment which has allowed us to clarify this issue. I refer you to the answer made to Reviewer 2 on this subject. • Page 7: from third line: very long sentence. Try to improve wording by separating sentences. • We agree with the reviewer and have modified the manuscript as follows: “There are therefore two relational drawbacks identified by the participants. The first was the fact that the avoidance of dealing with families’ worries about their child's long-term mental health, future schooling and social integration was only slightly delayed. Secondly, they sometimes found that trying to convince their patient at any cost generated tension.” • Page 8: "The practices of some participants remained modeled on that of the expert center they had \[been\] trained at." The verb on the sentence was missing. • We agree with the reviewer and have modified the manuscript as follows: “The practices of some participants remained modeled on that of the expert center they had been trained at.” • Page 9: wording suggestions: "Therefore, participants \[were\]worried \[about\] the stigmatization of psychiatric patients would lead to less conscientious assessments" • We agree with the reviewer and have modified the manuscript as follows: “Therefore, participants were worried about the stigmatization of psychiatric patients would lead to less conscientious assessments.” • Figure 1: I don't fully grasp the difference between the second ("to transform") and the third ("to be transformed") categories. I suggest to choose other words to highlight the core ideas Thank you for your continued interest of this submission. I look forward to your reply. Best regards, Jean-Luc Kurukgy and Laelia Benoit 10.1371/journal.pone.0252610.r003 Decision Letter 1 Hutchinson Gerard Academic Editor 2021 Gerard Hutchinson This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 19 May 2021 Implementing organicity investigations in early psychosis: Spreading expertise PONE-D-20-37522R1 Dear Dr. Benoit, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, Gerard Hutchinson, MD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#2: All comments have been addressed Reviewer \#3: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#2: N/A Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#2: The authors have addressed all of my previous comments clearly and fully. The manuscript reads well and provides a clear outline of the challenges underlying the implementation of organicity investigations when diagnosing and treating individuals affected with psychosis. Reviewer \#3: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#2: No Reviewer \#3: **Yes: **Paula Herrera Gomez 10.1371/journal.pone.0252610.r004 Acceptance letter Hutchinson Gerard Academic Editor 2021 Gerard Hutchinson This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 2 Jun 2021 PONE-D-20-37522R1 Implementing organicity investigations in early psychosis: Spreading expertise Dear Dr. Benoit: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Gerard Hutchinson Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction While the morbidity and mortality from cancer are largely attributable to its metastatic dissemination, the integral components/features of the metastatic cascade are not well understood. The most widely accepted hypothesis underlying metastasis is that the primary tumor microenvironment (TME) induces an epithelial-to-mesenchymal transition (EMT) in a subset of epithelial cancer cells, that confers increased motility and invasiveness and facilitates their escape into the bloodstream. A number of studies lend support to this conjecture, for example studies that document EMT-related changes (and loss of EpCAM expression) in circulating tumor cells (CTCs). In spite of recognized shortcomings considerable evidence has accumulated showing that numbers of EpCAM+ CTCs in peripheral blood has prognostic significance for patients. However, the picture remains incomplete in a number of areas. One vexing question is which CTCs are the capable of initiating metastatic lesions (so called metastasis initiating cells, MICs) and another is how MICs find suitable landing places. With regard to the former, a corollary idea is that the EMT- altered cancer cells at the periphery of a primary tumor facilitate liberation of cancer stem cells with them, which would represent the MICs. Thus, the global level of the CTC population would stochastically represent a much smaller subset of MICs, which presumably arise from a competitive hierarchy of subpopulations of genetically diverse cancer stem cells. However, this story does not address the latter question, how MICs find suitable “niches” which allow them to establish metastases and proliferate. Certainly exosomes could play a part in preparing adjacent tissues (for example, sentinel lymph nodes;), but significant concentrations of exosomes at distant sites are more difficult to envision. An alternative theory for metastasis involves fusion of macrophages with tumor cells (macrophage-tumor cell fusions, MTFs). With some sort of recombination/reprogramming of genetic material, perhaps analogous to that being studied in stem cell fusions of genetic material, this could produce neoplastic cells which have acquired “professional grade” invasive properties characteristic of macrophages. Indeed, there are suggestions that the EMT might better be described as an epithelial-myeloid transition. There is considerable support for this notion from animal models, and some recent support from reports of human cancers, but how frequently this occurs is unknown and the basic premise seems to be at odds with the EMT/stem cell hypothesis. An intriguing synthesis of these ideas is that MTFs could facilitate development of TMEs at distant sites, potentially addressing the problem of how MICs find suitable niches for colonization. Here, we have isolated and cultured MTFs from peripheral blood from several patients with cutaneous melanomas, and describe their various properties, including their ability to disseminate and form metastatic lesions. We hypothesize that these MTFs play an early, integral part in the metastatic cascade. MTFs formed at the periphery of primary tumors could readily enter the bloodstream (or lymphatics) and invade distant tissues, subsequently releasing cytokines, particularly MIF (and potentially exosomes) which prepare these distant sites for colonization by MICs. The MTFs could themselves represent the metastasis initiating cells (MICs), recognizing the potential self-renewal capacities of macrophages. Alternatively, successful MICs could also be disseminating cancer stem cells which evolve within heterogeneous tumors, a scenario in which the MTFs would function as a stromal component to produce a TME suitable for metastatic growth. # Materials and Methods ## Isolation and Culturing of MTFs This Human Subjects Research was approved by the Pennsylvania State University/Hershey Medical Center IRB, and informed written consent was obtained from all participants. Peripheral blood was obtained from patients with cutaneous melanomas (or healthy volunteers). An initial CTC enrichment was performed using Oncoquick porous membrane gradients as previously described, or using Ficoll-Paque PLUS gradients (GE Healthcare). The CTC-enriched fractions were rinsed in PBS, and then plated onto standard culture dishes and cultured in RPMI 1640 + 10% bovine serum. After 24 hours, plates were carefully rinsed to remove non-adherent cells, and new medium was added and cultures were continued for various periods of time, with the medium changed 2X per week. Many of the cells appeared large and “epithelioid” at 24 h, and retained the same basic morphology throughout culturing; this was also noted in a previous publication which also examined CTCs captured on filters, where “typical” CTCs were also observed. After our initial experiences, the culture protocol was changed to use 25 cm dishes (instead of 75 cm), which seemed to allow better growth with no change in the characteristics of the cultures, and we were able to harvest greater numbers of cells (5x10⁵ cells vs. initial cultures which had yielded ≤ 10³−10⁴ cells). ## Subcutaneous implantation of cultured human melanoma MTFs into athymic mice 4 week old athymic nude mice (nu+/nu+)were obtained from the NCI stocks at Charles River Laboratories. After initial acclimation period (7 days), mice were anesthetized with 129 mg/kg Ketamine and 4 mg/kg Xylazine. Once properly anesthetized, mice (2) were injected with 5 x 10⁵ cultured human melanoma MTFs in a volume of 100 μl into a hind flank. Experiments used cultured MTFs from 2 separate melanoma patient samples, which had been grown in culture for \~4 weeks. Mice were monitored for any outward sign of tumor formation; after 47 days, mice were anesthetized and euthanized. A necropsy was performed, and tissues were harvested and fixed in 10% neutral buffered formalin. After 24 hours, tissues were transferred to 70% ethanol (v/v) and embedded in paraffin blocks, sectioned, and stained using immunohistochemistry as described below. ## BRAF Mutational Analyses Given the high prevalence (\~60%) of BRAF mutations in cutaneous melanomas, CTC- enriched fractions were analyzed for mutations in the kinase domain of BRAF. Enriched CTC populations from 11 patients were examined for specific activating BRAF V600 mutations, using specific primers. We also analyzed cultured MTF populations derived from 8 patients. Primers used were: Claw3461 BRAF Normal‐For = 5’‐ACAGGGCATGGATTACTTACA (for all reactions); Claw3462 BRAF Normal‐Rev = 5’‐GGACCCACTCCATCGAGATTTCA; Claw3463 V600E‐R = 5’‐GGACCCACTCCATCGAGATTTCT; Claw3464V600K‐R = 5’‐GGACCCACTCCATCGAGATTTCTT; Claw3465 V600R‐R = 5’‐GGACCCACTCCATCGAGATTTCCT; Claw3466 V600E2‐R = 5’‐GGACCCACTCCATCGAGATTTTT; Claw3467 V600D‐R = 5’‐GGACCCACTCCATCGAGATTTAT; Claw3468 V600E‐F2 = 5’‐GGTGATTTTGGTCTAGCTACAGA; and Claw3469V600E‐R2 = 5’‐TGCATTCTGATGACTTCTGGT. After primer-specific amplifications, products were verified by size, and where necessary by cloning/sequencing. ## Immunofluorescent Staining of MTF Cultures Immunochemical staining was performed with antibodies against a variety of different markers. Antibodies specific for human MLANA, CD204, and CD206 were used to unambiguously identify human cells in mouse tissues. Generally cells were stained for various combinations of 2 markers, as well as with DAPI, and examined by confocal microscopy. The human pancreatic ductal epithelial cell line HPDE (ATCC, CRL-4023) was used as a normal diploid control. Human melanoma cell lines SK-Mel-24 (ATCC, HTB-71, metastatic), SK-Mel-28 (CTRL HTB-72, primary), SK-Mel-31 (CTRL, HTB-73, primary), and the human pancreatic ductal adenocarcinoma cell line Panc-1 (ATCC, CRL-1469) cells were also examined in parallel. Cells were grown overnight in 8-well coated chamber slides (Lab-Tek II CC²) and immunostaining was performed as previously described. Blocking was performed in PBS + 1:50 dilution of serum (from the species the secondary antibody was produced in) for 1 h at RT. 200 μl/chamber well of primary antibody solutions (1:200 dilution) was used. The balance of the staining protocol was performed in the dark, or with slides protected from light. 200 μl of secondary antibody (1:500 dilution in blocking solution) was used. Controls routinely included no primary antibody. To counterstain nuclei, 200 μl of DAPI solution (1:30,000 in PBS) was added to each well, and incubated for 5 min at RT in the dark. Cells were again washed 3X for 3 min each in PBST, and then given a final rinse for 10 min at RT with PBST, and the final wash was removed by inversion wicking. Coverslips were mounted using 3 drops of ProLong Gold Antifade mounting medium (Invitrogen), pre-warmed to RT, and slides were stored in the dark at 4°C. The mounting medium minimizes the refractive index mismatch of the lens immersion liquid (Cargile oil, refractive index \~ 1.52). ## Immunohistochemical Staining Formalin-fixed paraffin-embedded (FFPE) tissue specimens from 6 primary melanomas, and 2 metastatic melanomas, were obtained under an IRB-approved protocol (without identifiers), and they were also stained for the macrophage, melanocyte, and epithelial markers (and DAPI) described above, and examined by standard light microscopy and/or confocal microscopy. Where appropriate, secondary antibodies and reagents for immunoperoxidase staining were purchased from Vector Laboratories. They were: ImmPress Anti-Rabbi Ig (peroxidase) (MP-7401) and ImmPress Anti-Mouse Ig (peroxidase) (MP-7402). Peroxidase substrate kits used were ImmPACT DAB Substrate (SK-4104) ## Nuclear Staining for Ploidy Analysis As described above, nuclei of the cultured MTFs from melanoma patients were fluorescently labeled with DAPI (4′,6-diamidino-2-phenylindole, from Invitrogen) and/or with TO-PRO-3 dye; although the latter dye has been reported to be a useful DNA stain, in our hands it exhibited even greater photobleaching than with DAPI, in spite of the ability to use pulsed excitation. For tissue sections, control diploid nuclei were imaged from normal, benign regions of the same slides. ## 3D Confocal Microscopy and Image Acquisition Confocal images of fluorescently labeled cells were acquired with a Leica AOBS SP8 laser scanning confocal microscope (Leica, Heidelberg, Germany) using a high resolution Leica 40X/1.3 Plan-Apochromat oil immersion objective. The laser lines used for excitation were continuous wave 405 (for Dapi), 80 MHz pulsed 499 nm (for Alexa 488), 80 MHz pulsed 591 (for Alexa 594) and 80 MHz pulsed 645 nm (for TO-PRO-3). These laser lines were produced by UV diode, 80 MHz white light laser (Leica AOBS SP8 module) respectively and the respective emission signals were collected sequentially using AOBS tunable filters as follow; 410–480 nm for DAPI (this exclude possible RNA bound DAPI emission which occurs above 500 nm), 504–571 nm for Alexa 488, 597–751 nm for Alexa 594 and 650–790 nm for TO-PRO-3. All images, spectral data and DNA ploidy measurement data were generated using the highly sensitive HyD detectors (with time gated option) in descanned mode and the photon counting mode was particularly used for collecting signals from DAPI and TO-PRO-3 for DNA ploidy measurements. The backscattered emission signals from the sample were delivered through the AOBS tunable filter (to remove irradiated laser), the detection pinhole set to 1 Airy unit (to obtain optimal lateral and axial resolutions), spectral dispersion prism, and finally to the HyD detectors. The width of the slits in front of each HyD could be software adjusted so that each HyD could detect spectral regions spanning from a 10-nm bandwidth up to the overall spectral capacity of the system (400–800 nm). Using this unique option, spectral scanning was performed on all the dyes to confirm signal specificity. ## Deconvolution *For 3D image data set acquisition*, the excitation beam was first focused at the maximum signal intensity focal position within the sample and the appropriate HyD gain level was then selected to obtain the pixel intensities within range of 0–255 (8-bit images) using a color gradient function. Later on, the beginning and end of the 3D stack (i.e. the top and the bottom optical sections) were set based on the signal level degradation. Series of 2D Images for a selected 3D stack volume were then acquired with 1024X1024 pixels and were line averaged 3–4 times depending on the noise level. The 3D stack images with optical section thickness (z-axis) of approximately 0.3 m were captured from cell volumes. For each cell volume reported here, z-section images were compiled and finally the 3-dimensional image restoration was performed using Imaris software (Bitplane). ## DNA Ploidy Quantification The DNA ploidy measurement from 3D rendered confocal image dataset is a well- established procedure. Specifically, the computation of DNA voxel intensities was performed on the 3D image data sets recorded from several areas of cell samples using IMARIS image processing software. Appropriate Gaussian noise removal filter was used depending upon the noise level. The lower threshold level in the histogram was set appropriately to exclude all possible background voxel values. Sum of all the voxel intensities above this threshold level was determined and was considered as the DNA content. We systematically then compared 3D image volume of cells generated using similar imaging conditions. In melanoma tissue sections, we found the cells both immunolabeled for macrophage markers and melanoma markers, as well as the cells immunolabeled either macrophage or melanoma markers. We used Cell Module within IMARIS to identify cells that immunolabeled for both macrophage markers and melanoma markers. The rest of the quantitation procedures were the same as described above. Deconvolution of 3D confocal image datasets was performed using Huygens software (SVI, Netherlands). For the purpose of deconvolution, the point spread function (PSF) of the optical system was measured for all the emissions using the sub- resolution Fluorescent Microspheres Size Kit (Life Technologies, USA) with identical instrumentation settings. Briefly, a series of confocal optical sections from 0.17 mm beads satisfying the Nyquist sampling criteria was sampled for each emission channel sequentially. The 3D images of these sub-resolution beads were then restored and the PSFs were measured using the Huygens Software. Compiled 3D confocal image data sets were deconvolved with these experimentally measured point spread functions using the iterative deconvolution method which is based on the Classical Maximum Likely Hood Algorithm (CMLE). ## Preparation of MTF cultures for TEM Cells were grown in culture as described, and then transferred to coverslips (Thermanox coverslips, 15 mm D, Cat#72275–01) and grown for 3 additional days. Cells were then washed with ice cold 0.1 M sodium cacodylate (NaCAC) buffer, pH 7.3 three times for 5 min. Cells were fixed for 1 h at 4 C in 0.5% glutaraldehyde and 4% paraformaldehyde buffered with 0.1 M NaCAC buffer. They were again rinsed 3X with 0.1 M NaCAC at 4 C, and then post-fixed in 1% osmium tetroxide/1.5% potassium ferrocyanide overnight at 4 C in a wrapped container. Preparations were then rinsed with buffer, dehydrated in a graded series of ethanol, and embedded in Embed 812 (Electron Microscopy Sciences). A diamond knife mounted in a Porter-Blum MT-2B ultramicrotome was used to cut 70–90 nm thin sections. Sections were mounted on 200-mesh copper grids and stained with 2% aqueous uranyl acetate + lead citrate. Sections were examined in a Joel Jem 1400 TEM. An Orius SC1000 bottom-mounted CCD camera was used to capture the images. ***s*** ## Live Cell Microscopy, Image Acquisition and Quantitation MTF cultures were grown in glass-bottom microwell dishes (MatTek, \#P35G-1.5-14-C) to \~ 50% confluence, then stained with NucBlue live cell stain (ReadyProbes, \#R37605) according to manufacturer’s instructions. The experimental protocol used basically followed that of Ohlund et al.. Cells were plated into glass-bottom wells which were either uncoated, or which had been pre-coated with type I collagen. For the live cell migration measurements under various conditions, DeltaVision Elite was utilized along with a heated live cell stage equipped with humidified CO₂ perfusion system (GE, USA). The live cell data mode integrated into the image acquisition software (Leica Confocal Software LAF) was used for the interactive imaging of live cells. The 3D live cell images were acquired using an Olympus 60X/ 1.4 NA high numerical aperture apochromat oil immersion objective. A sensitive cooled CCD camera was used for capturing images and the images (16 bit) were acquired every 30 seconds for 10 minutes with the appropriate pixel dimensions selected to satisfy the Nyquist sampling criteria. The time series images were compiled and and cell tracking and cell motility measurements were performed using a VOLOCITY (Perkin Elmer, USA) workstation. # Results ## Isolation and Culture of apparent MTFs Peripheral blood samples were obtained from patients with cutaneous melanomas under approved IRB protocols with informed consent. OncoQuick porous membrane gradient devices were used to obtain an “enriched CTC” fraction (eCTC; enrichment was \~4-500X vs. PBMCs) as described). In some cases, Ficoll-Paque PLUS gradients were used. The eCTC fractions were then plated onto standard tissue culture dishes and medium containing 10% serum was added. The cultures were rinsed carefully after 24 h (to remove non-adherent cells). At 24 h, many of the adherent cells generally appeared large and “epithelioid” as was reported previously (where we also captured CTCs on filters), and they retained the same basic morphology throughout culturing. Cells were cultured for 3–4 weeks, with medium changed every other day. In most instances, growth was initially quite slow. In initial experiments, cultures consisted of a few thousand cells. The cells seemed to grow better when plated in smaller (25 cm) culture dishes, which eventually provided \~ 5 x 10⁵ cells at \~ 4 weeks. No cells grew in cultures of peripheral blood obtained from normal volunteers. Upon initial isolation, the populations consisted of \~ 50% “typical” CTCs (i.e., staining for epithelial markers, and not staining for mesenchymal or macrophage markers, or CD45) and 50% apparent MTFs (dual-staining for epithelial and macrophage markers; see below). After 1–2 weeks in culture, the typical CTCs were lost and the populations uniformly were dual-staining. Morphologically, these cultured presumptive MTFs appeared large and spread out, often with diameters ≥ 50 μm. They were flattened (\~ 10 μm high), and generally showed abundant pseudopod extensions. ## Immunophenotypic Characterization of Cultured apparent MTFs Cultured MTFs were permeabilized and stained with various fluorescent reagents. They stained for standard epithelial markers including pan-cytokeratins (pan- KRTs) and EpCAM (even though melanomas are not derived from epithelium, they routinely express epithelial markers). Interestingly, staining for the pan-KRT marker often showed a preferential nuclear localization (: this was also true for the various cell lines examined, and for some MTFs in vivo; see), although EpCAM staining *showed the standard staining pattern*. MTFs stained strongly for markers specific for melanocyte lineage, including MLANA, and for cell surface markers often found in melanomas, such as ALCAM. The cultured MTFs also routinely stained positive for pan-macrophage markers, including CD14 and CD68. They also stained for a variety of markers indicative of M2 polarized macrophages, including CD204, CD206, and CD163. Most of the staining for CD206 was consistently nuclear. The human pancreatic ductal epithelial cell line HPDE was used as a normal diploid control for DNA analyses; surprisingly, we found that these “normal” HPDE cells also stained for the M2-polarization markers CD163, CD204 and CD206, although they did not express the pan-macrophage markers CD68 or CD14. Given this surprising finding, we then examined 3 human melanoma cell lines, as well as the human pancreatic ductal adenocarcinoma (PDAC) cell line Panc-1. The SK-Mel-24, SK-Mel-28, and SK-Mel-31 cell lines showed uniform staining for melanocyte, M2 macrophage, and epithelial markers, with staining patterns analogous to those seen with cultured MTFs. For example, CD206 staining was nuclear, as was seen with cultured MTFs, and pan-KRT staining was largely nuclear. Panc-1 cells did not stain for melanocyte markers, but they strongly expressed the epithelial markers pan-KRT (which again was largely nuclear, similar to what was observed in cultured MTFs and SK-Mel cells) and EpCAM. Panc-1 cells also expressed the M2 macrophage markers CD204, CD206, and CD163. As with cultured MTFs and SK-MEL cells, most of the CD206 immunostaining was nuclear in Panc-1 cells. Like the HPDE cells, Panc-1 cells were also negative for the pan-macrophage marker CD68. We examined cultured MTFs for expression of the pro-carcinogenic cytokine MIF, because of MIF’s prominent roles in M2 polarization of macrophages, the tumor microenvironment (TME), and cancer progression. The cultured MTFs routinely stained positively for the MIF, with some very intriguing patterns of staining noted. Many of the individual nuclei appeared to have “tunnels” through them; these tunnels (invaginations) were lined by an intact nuclear envelope, and often contained apparently normal cytoplasmic organelles such as ER, golgi, mitochondria, etc. The interior (cytoplasm) within these tunnels stained strongly for MIF, as was determined using 3D reconstructions and deconvolution of confocal images. Such tunnels were also observed within melanomas in situ (see below), and they were evident in HPDE cells (although we did not stain HPDE cells for MIF). Given the robust immunostaining for MIF, we also examined the functionally related stem cell markers CXCR4 and CD44. CXCR4 is a non-cognate receptor for MIF and CD44 represents the signaling component of the MIF:CD74 receptor complex. As with MIF, we observed strong expression of CXCR4 and CD44, indicative of pro-carcinogenic activities of these stem cell markers/pathways. ## MTF Ultrastructural Features Cultured MTFs were also examined ultrastructurally using transmission electron microscopy (TEM;). Essentially all cultured MTFs showed features characteristic of macrophages. The MTFs were generally elongated, large cells (50 μm diameter or larger), which characteristically showed exuberant pseudopod extensions, lamellipodia and exocytosis. They contained large numbers of mitochondria, lysosomes, autophagic vacuoles, and myriad stages of autolysomal breakdown products, including laminated bodies structurally comparable to lysosomes, and various structural remnants. A prominent component in essentially all cells were heterogeneously-sized autophagic vacuoles containing chromatin, and often micronuclei. In addition, most MTFs also prominently contained apparent melanosomes and/or premelanosomes (;). All had about the same density; longitudinally sectioned, they often showed tubular membranous elements containing substructures analogous to those in mature melanosomes. In many MTFs, cross-sections through prominent nuclear “tunnels” were evident: These tunnels (or invaginations) contained many organelles and membranous structures in their cytoplasmic component. ## MTFs contain DNA derived from Melanomas Given the prevalence of somatic activating mutations in cutaneous melanomas, we first assessed BRAF mutational status in the initial eCTC populations from 11 patients: V600 activating mutations in BRAF were determined using primer- specific RT/PCR and size analysis. 8 of the 11 eCTC populations contained activating BRAF mutations: Six patient samples contained the V600E mutation, 2 patient eCTC samples contained both V600E + V600R mutations, and one additional sample contained the V600K mutation. The cultured MTF populations were also interrogated for the characteristic activating mutations in BRAF. Of 8 MTF cultured populations examined, 5 contained activating BRAF mutations (all 5 had the V600K mutation, 2 samples contained V600K and V600E mutations, and 1 also contained the V600E2 mutation (this percentage reflects the prevalence of BRAF activating mutations in cutaneous melanoma patients). This establishes that these apparent MTFs contain melanoma-derived DNA. ## DNA Content Analyses After immunophenotypic examination, DNA ploidy was determined using confocal microscopy as previously described. Our basic premise was that if these cultured MTFs may have arisen by fusion, they should exhibit altered ploidy. SK-MEL-24, -28, and -31 melanoma cell lines, as well as Panc-1 and HPDE cells, were also examined, with HPDE cells serving as a standard diploid control. In contrast to their relatively homogeneous immunostaining, the cultured MTF populations were very heterogeneous with respect to DNA content (; similar results were observed with DAPI and TO-PRO-3 staining). Occasional large cells were binucleate, containing 2 diploid nuclei, and there was considerable heterogeneity in the size and shape of individual nuclei. The populations of cultured MTFs (from 3 separate patients) showed cells with DNA distribution peaks corresponding to \~ diploid and \~ tetraploid, but with many aneuploid cells distributed throughout the range, and some with DNA contents ranging up \~ octaploid or even higher. Surprisingly, SK-MEL-24, -28, and -31 cells showed similar DNA content distributions, with peaks at 4X and most cells showing \> diploid contents; occasional cells with \~8X DNA content were also observed. Panc-1 cells also showed very heterogeneous nuclear size and DNA contents, which resembled that observed in the MTF populations. HPDE cells showed the expected diploid DNA content (**Figs and**). Even more striking was an apparent “shedding” of DNA from the nucleus into the cytoplasm, which was evident in essentially all of the cultured MTFs. In many cases, this appeared as “sheets” of chromatin. This poorly understood process of DNA handling or reconciliation has been referred to as “symphiliosis. In many cells, smaller “clumps” of condensed chromatin were observed in the cytoplasm. This cytoplasmic DNA was contained within autophagosomes, often appearing as micronuclei. Chromatin texture analysis was then performed, which basically quantifies the DNA signal from the DAPI (or TO-PRO-3) DNA dye. The nuclear chromatin showed obvious signs of organization, with many regions of highly condensed (DAPI- intense) chromatin. Such regions have recently been described in fusions between stem cells and somatic cells, and have been linked to malignancy in prostate cancer. In most (if not all) cases, the extranuclear chromatin apparently being extruded showed uniform staining indicative of non-condensed chromatin (within autophagosomes). Since MTFs are present in peripheral blood of melanoma patients, we examined a number of human melanoma FFPE tissue specimens for apparent MTFs (6 primary melanomas and 2 metastatic melanomas were examined). In all primary melanomas, we routinely observed many apparent MTFs, which stained for macrophage (pan- and M2-polarization) markers, as well as for melanocyte-specific (MLANA) and epithelial (EpCAM and pan-KRT) markers. In these primary (and metastatic) melanomas, MTFs were clearly identifiable at the periphery of the lesions within nests of melanoma cells. These cells generally appeared irregular and large, with characteristically large nuclei (often multinuclear) and cytoplasmic extensions into the surrounding tissue. In addition to the MTFs, there were also subpopulations of macrophages and melanoma/melanocytes present in the specimens, which did not dual-stain for macrophage/melanocyte markers. We then assessed the DNA content specifically in the dual-staining apparent MTFs in primary melanomas. Their DNA content distribution also showed a very heterogeneous pattern, with many cells in the 4-9n ploidy range, and many additional cells containing much higher DNA contents (up to 18n). It is of interest that the DNA distribution for SK-MEL-24 cells, which were derived from a lymph node metastasis, closely resembles that found in apparent MTFs in situ. An important question was whether the cultured MTFs were capable of metastasis. First, using live cell microscopy, we found that the MTFs were highly motile in vitro (motility coefficient of 0.13 μm²/sec). To address metastasis capability, cultured MTFs (5x10⁵) were transplanted subcutaneously in the hind flanks of nude mice, and mice were sacrificed 6–7 weeks later. There was no obvious residual tumor at the injection site. However, there were human MTFs present in subcutaneous tissues in the adjacent skin sections in both mice, and more importantly there were metastatic foci present in the pancreas. These foci contained obviously pigmented cells, and were generally characterized by relatively “smooth”, well-defined borders, without obvious aggressive invasion into the surrounding adjacent pancreatic parenchyma (which appeared normal). In addition, there were also often single cells (or nests/aggregates of a few cells) found within stroma in various locations (we see similar features with MTFs from pancreatic adenocarcinoma patients; a manuscript is in preparation). These foci (or cells) dual-stained for human melanocytic markers (MLANA, ALCAM) and human M2-polarized macrophage markers CD206, and CD204, clearly documenting their human origin. # Discussion There are a number of findings here which merit discussion: These include; a) the morphologic, ultrastructural, and immunophenotypic (co-expression of epithelial, melanocytic, and macrophage biomarkers) characteristics of the MTF populations, and the presence of melanoma-derived DNA in them; b) potential importance of MIF, and related stem cell markers CXCR4 and CD44 produced by MTFs, in establishing “niches” at distant sites; and c) potential role of ploidy and DNA handling in progression and metastasis. With regard to the co-expression of markers characteristic of epithelial, melanocytic, and macrophage lineages in the cultured MTFs, all MTFs expressed the standard “epithelial” markers EpCAM and pan-cytokeratin (pan-KRT). The distribution of the expressed EpCAM was that typical of epithelial cells. However, much of the expressed pan-KRT we observed was localized to the nucleus. Although the meaning of this is not clear, distinct molecular forms of KRT have been described in the nucleus, which do not possess a filamentous structure. This largely nuclear localization of KRT was also observed in SK-MEL cell lines as well as Panc-1 cells. Nuclear pan-KRT localization was also occasionally observed in melanomas in situ. The MTFs expressed a number of markers characteristic of melanocytic differentiation or melanoma. For example, they expressed MLANA (Gene ID#2315, Melan A), which is involved in melanin biogenesis. Ultrastructurally, the MTFs prominently contained apparent melanosomes (and/or premalanosomes), based on previous detailed descriptions. We also note than MLANA mRNA serves as an excellent positive biomarker in peripheral blood from early stage melanoma patients, where it could represent circulating MTFs. MLANA expression is also useful for diagnosis of invasive cutaneous melanomas. The MTFs also expressed ALCAM (Gene ID#214), a marker which is characteristically expressed in melanomas. Expression of these markers was consistent across all cells in the populations, and does not seem to reflect promiscuous expression. Given the prevalence of activating BRAF mutations in cutaneous melanomas, we interrogated eCTC and MTF cultures for somatic activating mutations in BRAF. 8 of 11 of the initial eCTC populations contained activating mutations in the kinase domain of BRAF, a somatic mutation characteristic of many primary melanomas. When 8 additional MTF cultures were interrogated after 4 weeks in culture, 5 of them also contained the activating V600E mutations. This establishes that what appear to be macrophages contain DNA derived from melanomas. The MTFs uniformly expressed a number of macrophage markers, many of which are characteristic of M2-polarized macrophages, such as CD163, CD204, and CD206. CD163 (Gene ID#9332) is a member of the scavenger receptor cysteine-rich superfamily, which may reflect proinflammatory cytokine production, and there are various reports linking its expression with poor prognosis in various cancers. CD204 (Gene ID#4481) is officially known as MSR1, the class A macrophage scavenger receptor type 1. It is a functional receptor which mediates the endocytosis of low density lipoproteins, implying lipid metabolism, and its expression has also been linked with various cancers as well as with intralymphatic metastasis CD206 (Gene ID#4360) is MRC1, the mannose receptor, C type 1; it is involved in glycoprotein metabolism, and curiously has also been shown to be involved with CD44 in lymphatic trafficking. We believe that expression of these receptors may indicate use of alternative energy sources by transformed cells. It is also of interest that while Panc-1 and HPDE cells strongly express CD204 and CD206, they do not express pan-macrophage markers like CD68 (Gene ID#968) or CD14 (Gene ID#929): CD68 and CD14 are unlikely to contribute to altered metabolic requirements, although it is of interest that activating BRAF mutations do alter metabolic pathways in melanoma. Given that MTFs were found in peripheral blood of melanoma patients, we also examined human melanoma specimens for apparent MTFs. MTFs were readily identifiable in primary melanomas, with strong staining for melanocyte (MLANA and ALCAM), macrophage (CD204, CD206), and epithelial (pan-KRT, EpCAM) markers. It is of interest that CD206 expression in MTFs in primary melanomas generally did not show the nuclear localization observed in cultured MTFs or cell lines (compare with). There is a report which described “stealth” melanoma cells in histology negative sentinel lymph nodes; these may well have represented MTFs. The M2-polarization of cultured MTFs may have significant ramifications. M2-like macrophages are responsible for collagen degradation through a CD206-mediated pathway, and tumor associated macrophages (TAMs) generally acquire an M2-like phenotype that plays important roles in many aspects of tumor growth and progression. M2-polarized TAMs have also been found to promote the EMT in various carcinomas. In fact, there are a growing number of reports of expression of macrophage markers on various types of cancer cells; this does not simply reflect “plasticity” in gene expression, but rather a characteristic acquisition of macrophage systems. For example, CD163 expression on rectal cancer cells is associated with early local recurrence and reduced survival time. CD163 expression by breast cancer cells is related to early distant recurrence and reduced survival time, and breast cancer cells expressed CD68. In this regard, Shabo & Svanvik reported that 48% of breast cancer cells expressed CD163, and that 31% of rectal cancer cells expressed it. CD163 was again associated with early distant recurrence in breast cancer, and with local recurrence in rectal cancer, and with reduced survival times in both. Expression of DAP12, a macrophage fusion receptor, was also associated with advanced tumor grade and higher rates of skeletal and liver metastasis, and overall shorter distant recurrence-free survival. ## Macrophage Tumor Cell Fusions Pawelek and co-workers have long espoused the concept of MTFs (or more generally myeloid cell-tumor cell fusions) and their potential role in metastasis, and they recently described a CNS melanoma which may have represented a metastatic MTF lesion. Fusion between intestinal epithelial cells and macrophages in a cancer context results in nuclear reprogramming: MTFs reportedly retain a transcriptome identity characteristic of both parental derivatives, while also expressing a unique subset of transcripts. Ding et al. reported that MTFs may even acquire cancer stem cell properties in breast cancer cells. They observed aberrant CD163 expression in breast cancers, and created fusions between M2-polarized macrophages and breast cancer cell lines. The MTFs showed a CD44⁺CD24^-/low phenotype, and demonstrated increased migration, invasion, and tumorigenicity (but reduced proliferative ability) and they over-expressed EMT genes. In a related vein, other have described in vivo fusion of human cancer cells with hamster stromal cells, resulting in tumors which express the tumors oncogenic driver mutations as well as stromal characteristics. Macrophage-fusion cells are not exclusive to cancers; fusion between hematopoietic and epithelial cells in adult human intestine has also been described. ## MIF in the TME and Cancer Progre*ssion* Another noteworthy observation was the expression of MIF by MTFs, since MIF plays critical roles with M2 polarized macrophages, the TME, and it has been implicated in diverse pathways involved in cancer progression. MIF levels are associated with an increased incidence of prostate cancer, non small cell lung cancer, squamous cell carcinomas of the nasopharynx and esophagus, breast cancer, colorectal cancer, hepatocellular carcinoma and several other cancers. MIF produced in the TME also regulates angiogenesis in a melanoma model. MIF serves as a non-cognate ligand for CXCR2 and CXCR4: MIF (and CXCR4) in the TME are adverse prognostic indicators in esophageal cancer, and it can induce CXCR ligand and regulators of macrophage infiltration like CD44. Here we found that cultured MTFs also expression CXCR4 and CD44 in addition to MIF. In PDAC, MIF has been shown to induce the EMT, enhance tumor aggressiveness, and predict clinical outcome in resected PDACs. M2 polarized TAMs have specifically been shown to have prognostic importance in pancreatic ductal adenocarcinoma (PDAC), and targeting TAMs decreases tumor-initiating cells, relieves immunosuppression, and improves chemotherapeutic responses. MIF controls alternative activation of TAMs to M2-polarization; In turn, co- culturing M2-polarized TAMs with PDAC cells strongly induces the EMT. MIF expression is up-regulated during hypoxia generally associated with the TME, via an HRE found in the 5’-UTR of the gene. Proteomic and tissue array profiling has identified elevated hypoxia-regulated proteins in microdissected PDAC nests vs. normal ducts, prominently including MIF, which showed excellent ROC curves in discriminating PDACs. MIF is a direct transcriptional target of HIF1, and loss of MIF results in inefficient HIF1 stabilization induced by hypoxia. Intracellularly, MIF is also stabilized by complexing with HSP90 chaperone. Cancer cells contain constitutive endogenous MIF-HSP90 complexes, and inhibition of HSP90 function results in apoptosis, which can be overridden by ectopic MIF expression In fact, the metastasis-promoting CD44 in PDACs is actually the signaling component of the MIF-CD74 receptor complex. MIF signaling through CD74 promotes sustained ERK activation, <u>which corresponds to the main outcome from Ras mutations, mutations which figure so prominently in PDAC</u>. Inhibition of MIF using siRNAs leads to apoptosis in PDAC cells. Long et al. developed a unique mouse model for PDAC lymphatic metastasis. They developed a subline from the BxPC-3 PDAC cell line via serial passages in nude mice. The subline showed increased migration, invasion, and invasive ultrastructural characteristics. Metastasis-related gene alterations found in the subline were quite limited but included up-regulation of MIF. Here, MIF was characteristically found within “tunnels” in MTF nuclei, both in cultured MTFs as well as in apparent MTFs in primary melanomas, and in the SK- MEL cells lines. These “tunnels” often contain intact cytoplasmic organelles, and they are lined by an intact nuclear envelope. We are not aware of any previous reports describing similar nuclear tunnels, or of any focal, peri- nuclear localization of MIF, although an interaction between NME1 and MIF has been reported; NME1 was found to alleviate suppression of p53 activity by MIF, by disrupting binding of p53 by MIF. Another prominent finding was the heterogeneous polyploidy DNA content found in cultured MTF preparations, with a significant proportion of the MTFs containing 4n to 8n (or higher) DNA content. On the one hand, there is clear evidence linking DNA index to prognosis of several cancers. Using PDAC as an example, DNA index has been shown to be a strong prognostic factor in PDAC patients where 50–75% of patients showed non-diploid DNA contents. There is also a clear relationship between DNA content and survival in PDAC patients. Lymph node involvement was seen in 36% of patients with diploid tumors, vs. 79% of those with aneuploid tumors. 32% of patients with a diploid tumor survived at least 1 year, whereas none of the aneuploidy patients did, and aneuploidy showed a significant association with decreased cumulative survival. Tsavaris et al found that PDAC patients with ploidy score \> 3.6 had 5X higher probability of death compared with patients with ploidy score \< 2.2, and those with an intermediate ploidy score 2.2–3.6 had a 6.3X higher probability of death compared with patients with ploidy score \< 2.2. A similar relationship was found for patients with late stage colorectal cancers. On the other hand, the route by which polyploidy occurs may have a major impact on its consequences, as has recently been hypothesized based on lessons from plants. There, polyploid cells arising as “Allopolyploids” have far different characteristics than those arising as “autopolyploids”. When fusion of 2 different cell types occurs, two distinct cellular programs need to be merged somehow. This process has been referred to symphiliosis, the process of intracellular reconciliation. This suddenly produces new clones with emergent phenotypes. Cancer cells can transduce adjacent TME cells in vivo, and it has been suggested that in vivo fusion discloses genes implicated in tumor progression, as well as gene families coding for the organoid phenotype. Here, we note that the MTFs present a surprisingly uniform immunophenotypic profile, in spite of the often huge differences in DNA content. They appear to be undergoing cellular reconciliation, with apparent shedding of massive sheets (and/or clumps) of cytoplasmic chromatin, with the extranuclear DNA being handled by nucleophagy within autophagosomes which have sequestered chromatin and even micronuclei. Many aspects of micronuclei formation have been detailed, and degradation of micronuclei via autophagosomes has previously been reported, where it was speculated that removal of micronuclei may contribute to the genome-stabilizing effects of autophagy. Nucleophagy has been reported in various laminopathies and seems to be beneficial for cell survival. Chromatin texture analysis also identified focal areas of condensed “DAPI- intense” chromatin staining. Similar regions have been reported in fusions between embryonic stem cells and somatic cells, a setting which is currently being examined in the context of “reprogramming” There are many examples of naturally occurring polyploidization (e.g., liver and skin;), and polyploidization and cell fusion appear to contribute to wound healing in the adult Drosophila epithelium. However, senescence and autophagy are also thought to be intimately involved in the emergence of self-renewal potential in surviving cells that result from a process termed “depolyploidization”; Erenpreisa and co-workers have suggested that genotoxic resistance is afforded through a programmed life-cycle-like process which intimately unites senescence, polyploidy, and “stemness” (self-renewal) as steps to “immortality” for cancer cells. The process seems to involve macroautophagy- aided elimination of chromatin, which somehow entails sorting out what will be eliminated. While we find a “diploid” peak in DNA content in the MTF populations, it seems more appropriate to term it “paradiploid”, since it is known that polyploid tumor cells elicit paradiploid progeny through depolyploidizing divisions and regulated autophagic degradation. The process appears to involve substantial nuclear volume increases with spatial shifts of chromosome territories in nuclei of radiation-induced polyploidizing tumor cells; this reflected generation of large intra-nuclear chromosome territories and their repositioning prior to genome reduction (an interesting question is whether radiotherapy might influence fusion frequency in patients). Although many aspects of these various processes are not well-defined, it seems clear that here they ultimately produce a population of MTFs which are capable of sustained growth and metastases after implantation into nude mice. As noted with plants (allopolyploids vs. autopolyploids), the way in which polyploidy arose in these apparent MTFs may be an important determinant for how they handle their DNA, so that the typical “depolyploidization” process does not occur. We suggest that MTFs participate in the earliest stages of the metastatic cascade. We hypothesize that they form very early on at the periphery of melanomas, locally induce the EMT, and readily enter the circulation, subsequently colonizing distant sites and secreting cytokines such as MIF. This produces “niches” (mini-TMEs) suitable for establishment of metastatic lesions by MICs, which may be liberated cancer stem cells. # Supporting Information Confocal images were generated using a Leica SP8 located within the Pennsylvania State University Microscopy Imaging Core Facility. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: GAC GLM DMT RIN. Performed the experiments: PX YIK KFH TA ZD. Analyzed the data: GAC PX YIK KFH TA. Contributed reagents/materials/analysis tools: KFH TA RIN. Wrote the paper: GAC GLM PX YIK DMT KFH RIN TA.

# Introduction Bacteremia is an increasingly prevalent and life-threatening condition with a reported 30-day mortality above 15% in studies from industrialized countries. In addition to increased risk of infections, low socioeconomic status (SES) may also worsen infection outcomes. However, few studies have examined the association between SES and mortality after a severe infection, including after bacteremia. A recent US study of patients hospitalized for sepsis adjusted for demographic factors and pre-existing comorbidity and found that, compared with married patients, widowed, single, and legally separated patients had greater odds of in-hospital death. In another US study, Mendu et al. found an unadjusted relationship between neighborhood poverty rate and mortality within 1 year after bacteremia in patients admitted to intensive care units. In contrast, in a population-based study from New Zealand, Huggan et al. found no relationship between an area-based measure of SES and mortality after *Staphylococcus aureus* bacteremia. Thus, these previous studies have reached conflicting conclusions and used marital status as a proxy for SES or area-based measures of SES, with no data on detailed individual-level measures of SES. Moreover, none of them examined which prognostic factors may mediate socioeconomic disparities in mortality. Compared with patients of higher SES, patients of lower SES tend to experience less social support, which may lead to more severe infection at admission and a more severe prognosis. Several studies have documented the adverse impact of pre-existing comorbidity and conditions related to substance abuse on survival after bacteremia. Furthermore, treatment in hospitals with high patient volume and teaching status may be associated with improved outcome and patients of high SES may have a better chance of being admitted to large university hospitals. To examine the effect of SES on mortality after bacteremia, we designed a population-based cohort study. We used two different individual-level indicators of SES, educational level and personal income, to capture different aspects of socioeconomic stratification. We further evaluated if differences in social support, pre-existing comorbidity, substance abuse, infection characteristics, and characteristics of the admitting hospital could explain socioeconomic differences in mortality after bacteremia. # Methods ## Study Design We conducted this study as a population-based cohort study. The geographic area included two Danish regions (North Denmark Region and Capital Region) with a total population of 1.7 million persons. All hospitalized patients aged 30 to 65 years with first time bacteremia from 2000 through 2008 were included in the cohort. ## Setting The Danish tax-funded welfare system provides free access to health care, education, and benefits such as pensions and unemployment coverage. All citizens are granted free services at general practitioners and public hospitals. Only 1% of hospital beds are in the private sector. Since April 1, 1968, all citizens in Denmark have been registered in the Civil Registration System. A unique personal identification number (civil registration number) allows accurate linkage of information among national registries, including medical registries. ## Identification of Patients with Bacteremia We obtained data from the Danish Collaborative Bacteremia Network (DACOBAN). This network includes the Departments of Clinical Microbiology in the North Denmark Region (Aalborg Hospital) and the greater Copenhagen area (Hvidovre Hospital and Herlev Hospital). DACOBAN was established to enable coordinated surveillance of all cases of bacteremia in the two regions and to study risk factors and prognostic factors for bacteremia. The three departments that serve these regions record data on all microbiological specimens, including blood cultures, in an electronic laboratory information system (ADBakt, Autonik, Ramsta, Sweden). A research database that consists of all patients with a first- time diagnosis of bacteremia between January 1, 2000 and December 31, 2008 has been established from data from these laboratory information systems. Bacteremia was defined as bacterial or fungal growth in blood cultures where contamination had been ruled out. Coagulase-negative staphylococci, Corynebacterium spp., Bacillus spp. and Propionibacterium acnes were regarded as contaminants unless they were isolated from two or more separate blood culture sets within a 5-day period. We only included patients in the age group of 30 to 65 years because we assumed that most of them had completed their education and were of working age. The civil registration number, age, sex, the date on which the first positive blood culture was drawn (date of bacteremia diagnosis), hospital and specialty, microbial agent, and place of acquisition of infection (community-acquired or nosocomial) were included in the database for all patients. Bacteremia was defined as community-acquired if the first positive blood culture was obtained within 48 hours after hospital admission and as nosocomial if it was drawn more than 48 hours after hospital admission. ## Socioeconomic Status SES was based on patients’ educational level and personal income. Although the two indicators are related they measure different aspects of socioeconomic stratification. Formal education is normally completed in young adulthood and will therefore to some extent measure early life SES. In contrast, income can change over a life course, but may better capture aspects of SES later in life. To assess the effect of both early life SES and later life SES on mortality after bacteremia, we examined both SES indicators. SES data were obtained for all patients through registries maintained by the government agency Statistics Denmark. These registries contain detailed individual-level socioeconomic data, updated yearly, for all Danish citizens. Information on patients’ highest completed education was obtained from the Population’s Education Register, which consists of data generated from surveys and from the administrative records of educational institutions. In 2008 the register contained valid information on education for 97% of the Danish population born from 1945 to 1990. We categorized educational level into primary/lower secondary education (low), upper secondary education (medium), and tertiary education (high) according to the International Standard Classification of Education (1997). Patients’ personal annual income was all income subject to income taxation (wages and salaries, and all types of benefits and pensions). Income data was obtained from the Income Statistics Register. Data in the register are primarily supplied by tax authorities and the income data are assumed to equal the real income. Personal annual income was adjusted for inflation according to the year 2000 value of the Danish crown (DKK) and was grouped into tertiles: low-income (1^st tertile), middle-income (2^nd tertile) and high-income (3^rd tertile). We also obtained data on patients’ employment status, nationality, cohabitation status, and marital status. Employment status was grouped into employed/self- employed, unemployed/employment subsidized by labor market arrangement and early retirement pensioners. We used cohabitation status and marital status as markers for social support. For all variables, we used data from the year preceding the index date of the bacteremia diagnosis. ## Pre-existing Comorbidity We obtained data from the Danish National Patient Registry for all diagnoses recorded from the start of the registry until the date of bacteremia diagnosis. The registry contains information for almost 100% of all inpatient admissions to public and private non-psychiatric hospitals in Denmark since 1977 and from outpatient and emergency room visits since 1995. Each record includes one primary diagnosis and up to 20 secondary diagnoses, which have been classified according to the International Classification of Diseases. Pre-existing comorbidity was summarized according to the Charlson Comorbidity Index, which was originally developed to predict 1-year mortality in hospitalized medical patients. Since then, the index has been adapted and validated for use with hospital diagnoses and has been used in previous studies of the association between comorbidity and survival after bacteremia. The index consists of 19 disease groups and each disease group is assigned a specific weight depending on the severity of the pre-existing condition. Based on the Charlson index scores three levels of comorbidity were defined: 0 (low), corresponding to patients with no recorded pre-existing comorbidity; 1–2 (medium), and \>2 (high). Diagnoses related to substance abuse (alcohol and drug abuse) are not included in the Charlson index and may influence prognosis after bacteremia. Therefore, we also collected data on previous alcohol- and drug-abuse-related diagnoses from the National Patient Registry. ## Hospital Characteristics Patients were treated in 1 of 16 public hospitals. We characterized these hospitals according to number of hospital beds, hospital volume, and medical school affiliation. We categorized hospital beds, setup and staffed for use, as less than 300 beds or 300 beds or more. Hospital volume was defined as the annual number of bacteremia patients treated at the institution and categorized as low-volume (≤99 bacteremia patients treated per year), medium-volume (100–299 per year), and high-volume (≥300 per year) hospitals. Teaching hospitals were defined as hospitals directly affiliated with a medical school. ## Follow-up and Mortality Our outcome was all-cause mortality within 30 days after the bacteremia diagnosis. We obtained data on each patient’s vital status from the Danish Civil Registration System. This registry contains daily updated information on all changes in vital status and migration for all Danish citizens. Patients were followed from the date of their diagnosis to the time of death, date of emigration, or completion of 30 days of follow-up, whichever occurred first. ## Statistical Analysis We first constructed contingency tables to provide information on baseline characteristics and crude outcomes according to SES, which were inferred on the basis of patients’ educational level and personal income. Using Kaplan-Meier plots we examined mortality within 30 days after bacteremia according to SES. A Cox proportional hazards model was constructed to determine the association between SES and 30-day mortality. We performed a sequential cumulative adjustment analysis to assess whether differences in social support (cohabitation and marital status), pre-existing comorbidity (comorbidity included in the Charlson Comorbidity Index and conditions related to substance abuse), infection characteristics (place of acquisition, microbial agent, and admitting specialty) or hospital characteristics (number of hospital beds, hospital volume, and medical school affiliation) accounted for socioeconomic differences in mortality. We included the potential mediators in our analyses in a sequence that reflected the temporal relation of the potential mediators (e.g., we assumed that SES would normally precede comorbidity existing at the time of the bacteremia diagnosis). Log-minus-log plots were used to confirm that the proportional hazards assumption was not violated. All analyses were performed with Stata statistical software, version 11.2 (StataCorp, College Station, TX). The study was approved by the Danish Data Protection Agency (Record no. 2010-41-5650). Informed consent was not required by Danish law. # Results ## Patient Characteristics The cohort consisted of 8,653 hospitalized patients aged 30 to 65 years with a first time diagnosis of bacteremia, which corresponded to an incidence of 114 episodes per 100,000 person-years in our study population. The median age of the cohort was 55 years (interquartile range, 47 to 61), and 3,853 (44.5%) were women. shows baseline characteristics according to educational level. Only small differences with respect to age and gender were seen. Patients with higher education were slightly younger. The medium-educated patients were more likely to be male. On average, patients with lower education were much less affluent than those with higher education (i.e., 44.7% vs. 17.4% in the lowest income tertile). They were more likely to live alone and be unmarried, and were substantially more likely to be out of the workforce (70.6%) than the patients with higher education (34.6%). Virtually all pre-existing comorbidities were more prevalent among patients with lower education. Only solid cancer, leukaemia, and lymphoma were more prevalent among those with a higher education. Similarly, the prevalence of conditions related to alcohol and drug abuse were substantially increased among those with lower education. Patients with lower education were also more likely to have *Staphylococcus aureus* bacteremia, to have a nosocomial infection, and to receive intensive care. In addition, patients with lower education were more likely to be admitted to small, low- volume, and nonteaching hospitals. A similar but more extreme pattern of differences was seen for income categories: bacteremia patients with low versus high income were 1.5 times more likely to live alone and be unmarried, had a 1.5 to 4 times higher prevalence of many comorbidities, and had a more than 4-fold higher risk of substance abuse. ## Mortality Overall, 1,374 patients (15.9%) died within 30 days of follow-up. There was a substantial gradient in mortality according to both educational level and income categories. Survival curves for the different levels of education and income diverged early after the bacteremia diagnosis and the differences in mortality persisted throughout the 30 days of follow-up. The 30-day mortality among the lower educated patients was 17.4% compared with 13.0% for those with higher education, which was an absolute difference in 30-day mortality of 4.5% \[95% confidence interval (CI): 2.4–6.5%\] and corresponded to a crude hazard ratio of 1.38 \[95% CI: 1.18–1.61\]. The difference in 30-day mortality was 6.7% \[95% CI: 4.8–8.6%\] when patients in the low-income tertile (30-day mortality, 19.7%) were compared with patients in the high-income tertile (30-day mortality, 12.9%). This corresponded to a crude hazard ratio of 1.58 \[95% CI: 1.39–1.80\]. Sequential adjustment for social support, pre-existing comorbidity, substance abuse, and infection characteristics attenuated the effect of SES on mortality. Further adjustment for differences in characteristics of the admitting hospital had only a marginal impact on the adjusted 30-day mortality hazard ratios. The fully adjusted risk estimates showed that there was still a residual difference in mortality according to both educational level (low vs. high education, 1.14 \[95% CI: 0.97–1.35\] and income categories (low vs. high income, 1.30 \[95% CI: 1.13–1.49\] after adjustment for a range of known prognostic factors. To examine whether income mediated the effect of education, we also included income as a covariate in our statistical model examining the effect of education on mortality. The effect of education on mortality after bacteremia was further attenuated after inclusion of income as a covariate (low vs. high education, 1.08 \[95% CI: 0.91–1.28\]), showing that income in part mediated the effect of education. # Discussion In this population-based cohort study we found that patients of lower SES, inferred on the basis of educational level and income, had higher 30-day mortality after bacteremia than those of higher SES. The association between SES and mortality was most pronounced when we used income as SES indicator. More than half of the socioeconomic differences in mortality could be explained by differences in social support, pre-existing comorbidity, alcohol and drug abuse, and characteristics of the infection. In contrast, characteristics of the admitting hospital had only a marginal explanatory effect on the socioeconomic mortality differences. Our findings thus suggest that socioeconomic disparities in mortality after bacteremia are to a large extent explained by a range of adverse prognostic factors that are present before hospital admission and include severe pre-existing comorbidity, unhealthy lifestyle, and lack of social support. For the clinician it is important to know if certain groups of patients with severe infection have a poorer prognosis than others and our study therefore has clinical implications. Worse prognosis among bacteremia patients of lower SES imply that these patients would benefit from increased clinical attention. Our findings of a much higher prevalence of comorbidities among bacteremia patients of lower SES suggest that special attention should be paid to improved management of these patients’ comorbidities, which may reduce their excess mortality risk. We are aware of only three studies that have examined socioeconomic disparities in mortality in cohorts of patients with sepsis or bacteremia. Seymour et al. conducted a population-based cohort study of 37,524 hospitalizations for sepsis in New Jersey, US. After adjusting for demographic factors and comorbid conditions, they found that single and divorced men and single women had greater odds of in-hospital mortality than married persons. In contrast to our study, this study used marital status as a proxy for social factors and did not include data on more precise measures of SES, such as educational level and income. Furthermore, identification of patients with infections was based on administrative ICD-9-CM codes and the infections were not microbiologically verified. Mendu et al. analyzed data from 2,435 patients with bacteremia who were admitted to intensive care units at two hospitals in Boston, Massachusetts, US. This study reported an unadjusted ‘dose-response’ relationship between neighborhood poverty rate and mortality within one year after bacteremia among patients receiving critical care. Adjustment for demographic factors, patient type, comorbidity, laboratory data, and severity of illness attenuated this association substantially and the authors concluded that neighborhood poverty was not associated with mortality after bacteremia. However, their use of an aggregate area-based measure of SES may have led to some inaccuracy due to misclassification of individual SES. A third study by Huggan et al. included 779 patients with *Staphylococcus aureus* bacteremia admitted to hospitals in Canterbury, New Zealand. The authors reported that there was no relationship between an address-based measure of deprivation and mortality but did not present any estimates. This study also used an area-based measure of SES, which may have resulted in misclassification of individual SES. In contrast with previous studies we used a sequential cumulative adjustment analysis to evaluate a range of recognized prognostic factors as potential mediators of the socioeconomic mortality differences. Even after full adjustment for all the potential mediators included in our study, we still found a residual difference in mortality. It is likely that the residual difference may be explained partly by greater levels of undiagnosed comorbidity among patients of lower SES, which again may be due to poor self-care and late presentation of clinical disease. However, we speculate that unmeasured variation in severity of infection at admission may also explain some of the residual mortality differences. Furthermore, since we did not have precise data on bacteremia patients’ care and treatment, we cannot rule out that difference in the quality of care contributed to the residual mortality differences. Our study has several important limitations. First, we used information collected during routine clinical work and data from administrative registries, which limited clinical detail in our study. More information on clinical parameters would have enabled us to better characterize severity of the infection and to assess any differences in severity according to SES. On the other hand, the use of routinely collected microbiological data and accurate linkage of high-quality registries enabled us to avoid some major methodological problems, such as selection and surveillance bias. Use of prospectively recorded individual data on SES indicators, which were collected independently, also reduced misclassification of patients SES. Second, our outcome measure was all- cause mortality, and we did not have any data on other important outcomes, such as recovery and functional status. When interpreting all-cause mortality, it must be considered that patients may have died from causes unrelated to their infection. Nevertheless, by including only deaths that occurred within 30 days after the date of bacteremia diagnosis we assume that most deaths would be at least to some extent related to the infection. Third, even though we studied a large cohort of bacteremia patients statistical precision was still limited. Fourth, we included bacteremia patients with a microbiologically confirmed infection, which allowed us to assess the mediating effect of the microbial agent on the association between SES and mortality. We can assume that the vast majority of the bacteremia patients in our cohort fulfilled the criteria for sepsis. However, since less than 50% of patients with sepsis have documented bacteremia, our findings cannot necessarily be generalised to sepsis patients. In conclusion, we found that patients of low SES had higher mortality within 30 days after bacteremia than those of high SES. Differences in social support, pre-existing comorbidity, substance abuse, and characteristics of the infection were important explanatory factors for these SES-mortality gradients. In contrast, characteristics of the admitting hospital seemed to have a negligible role in explaining disparities in mortality. The residual impact of SES on mortality might be explained by differences in bacteremia severity and treatment of the infection. Future studies should assess the importance of potential differences in severity of illness and quality of care in explaining socioeconomic mortality differences after bacteremia. **Contributing members of DACOBAN** Christian Østergaard Andersen (Department of Clinical Microbiology, Copenhagen University Hospital, Hvidovre Hospital, Copenhagen, Denmark), Magnus Arpi (Department of Clinical Microbiology, Copenhagen University Hospital, Herlev Hospital, Copenhagen, Denmark), Kim Oren Gradel (Center for National Clinical Databases - South, Odense University Hospital, Odense, Denmark), Ulrich Stab Jensen (Department of Clinical Microbiology, Copenhagen University Hospital, Rigshospitalet, Copenhagen, Denmark), Jenny Dahl Knudsen (Department of Clinical Microbiology, Copenhagen University Hospital, Hvidovre Hospital, Copenhagen, Denmark), Kristoffer Koch (Department of Clinical Microbiology, Aalborg University Hospital, Aalborg, Denmark), Mette Pinholt (Department of Clinical Microbiology, Copenhagen University Hospital, Herlev Hospital, Copenhagen, Denmark), Henrik Carl Schønheyder (Department of Clinical Microbiology, Aalborg University Hospital, Aalborg, Denmark), Mette Søgaard (Department of Clinical Microbiology, Aalborg University Hospital, Aalborg, Denmark). [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KK MN HCS MS. Performed the experiments: KK MN HCS RWT MS. Analyzed the data: KK. Contributed reagents/materials/analysis tools: KK MN HCS RWT MS. Wrote the paper: KK. Data management: KK. Critical revision of the manuscript: KK MN HCS RWT MS. Final approval of the submitted manuscript: KK MN HCS RWT MS.

# Introduction Though the idea of using satellite imagery to map penguin colonies is over thirty-years old, the use of these approaches for ecological research and long- term monitoring has been in active development only in the last few years. Recent data policy changes, such as the free availability of archival Landsat imagery, have opened access to vast amounts of medium resolution satellite imagery over Antarctica, and while access to very high-resolution (VHR) imagery is hampered by licensing and cost, increased tasking of satellites for imagery in the Antarctic have rapidly increased the quantity of imagery available. Recent studies demonstrate that penguin colony presence and abundance can be detected on a continental scale by manual interpretation of very high resolution (VHR) satellite imagery and that automated algorithms that exploit the spectral signature of penguin guano can be used to detect the presence of penguin colonies in medium-resolution Landsat data. Taken together, these new analytic methods—and the Terabytes of data to which they can be applied—afford the possibility of routine, continent-wide surveys of Adélie penguin presence and abundance, although to date no attempt has been made to use Landsat for regional abundance estimation or to compare the detection and abundance information derived from these two types of satellite images. Our analysis represents a bridge between abundance estimates based on very high-resolution imagery and the medium-resolution data record available from Landsats-4, -5 and -7. Forging this connection will facilitate integrated abundance estimates in the era when both of these satellite data types overlap, and will advance the development of a continuous time series of abundance that leverages a Landsat data record dating back to the 1980s. This paper presents the results of an automated retrieval of Adélie penguin colonies along the entire continent of Antarctica from medium-resolution Landsat-7 imagery (30 m pixel size), and its comparison with data on presence and abundance based on manual interpretation of VHR satellite imagery (∼0.60 m pixel size). Landsat-7 launched on April 15, 1999 and has collected imagery over Antarctica continuously since that time. Unfortunately, the ETM+ Scan Line Corrector failure in 2003 occurred just as some of the first very-high resolution images became available for the Antarctic, precluding direct contemporaneous comparison of abundance estimates from these two types of sensors. We leverage the recently published global Adélie survey to assess whether Landsat-7 can be used to estimate abundance as well as occupancy, and to estimate the lower detection threshold for Landsat-7. Since the uncorrupted Landsat-7 imagery dates from 1999–2003 and the global Adélie survey estimates predominately stem from 2010–2013, the correlation between the abundance estimates will be highly conservative, since the correlation is degraded by changes in true abundance in the intervening decade. Nevertheless, the relatively slow dynamics of penguin colonies means that abundances between these two periods should be fairly consistent within the margin of error of the abundance estimation itself, allowing us to investigate whether the number of pixels retrieved in the earlier period scales with the abundance reported by for the latter period. This cross-validation approach is the best currently available to quantify the accuracy and identify sources of error in using medium-resolution Landsat imagery for occupancy (presence vs. absence) and abundance. Controlling or correcting these errors is a key consideration for future automated surveys of Adélie penguins in Antarctica. In our analysis and discussion, we explicitly distinguish between the probability of detection and the estimation of abundance conditional on the detection of a colony. The detection of a colony based on the spectral characteristics of its guano stain is a binary classification subject to errors of omission (false negative) and commission (false positive). Abundance estimation is a separate challenge in which success is judged by the accuracy (bias and variance) of estimates of population size. We use the detection probability model along with a probability model for colony size to estimate the total abundance of penguins missing given the results of a Landsat survey. This methodology allows us to correct future Landsat surveys for detection failures and paves the way for achieving regular unbiased estimates of the global Adélie population. We considered continental Antarctica (in which the Adélie penguin is the only breeding *Pygoscelis* spp. penguin) separate from the Antarctic Peninsula in our analysis. Guano stains on the Antarctic Peninsula are ambiguous with respect to species in Landsat imagery and the three *Pygoscelis* spp often nest in colonies mixed at Landsat's larger spatial resolution. Additionally, as described in the Discussion, these two regions have different ‘failure modes’ with respect to detection and imagery interpretation. Our goals for this analysis were three-fold: (1) To estimate the lower detection threshold for Landsat-7 imagery, (2) to demonstrate that the number of pixels classified as penguin guano in Landsat-7 scales with abundance, and (3) to estimate variance in abundance estimates at varying scales of spatial aggregation. Our survey includes both continental Antarctica, where Landsat imagery has been used for detection of Adélie penguin colonies previously, as well as the Antarctic Peninsula, where its use is novel. The comparison and cross-validation of multiple data streams is a necessary step in the full integration of these data into a single remote sensing retrieval and analysis pipeline for monitoring Adélie penguin populations. This approach represents a significant improvement over direct field counts alone, which lack the scalability of remote sensing surveys, and are subject to underestimation bias in regional or global estimates due to the remoteness and inaccessibility of sections of the Antarctic coastline. Population estimates that scale in space and time are expected to more seamlessly interoperate with trophic and climate models, so that changes in Adélie penguin populations can be clearly tied to harvesting in the Southern Ocean fisheries and climate change. # Methods ## Automated Landsat retrievals Retrieval of Adélie penguin colony location and spatial extent over the Antarctic Peninsula was conducted using Landsat-7 Enhanced Thematic Mapper-Plus ETM+ data. The Landsat data used in this study were obtained as a Climate Data Record (CDR), which was generated using the methods developed by, and distributed by the U.S. Geological Survey Earth Resources Observation and Science Center (<https://espa.cr.usgs.gov>). This CDR applies a uniform processing to the entire dataset, and includes a host of products including top of atmosphere (TOA) reflectance, surface reflectance, brightness temperature, as well as masks for clouds, cloud shadows, adjacent clouds, land and water. The Landsat CDR TOA product was used for the automated retrieval of Adélie penguin colonies following the methods described in. One difference between our Landsat survey and that described in is that additional exposed soil and rock surfaces were sampled and included in the training dataset for the AP retrieval algorithm because the AP has a proportionally larger expanse of exposed soil and a different suite of soil types than the rest of the continent. The training dataset consisted of 473 Landsat-7 pixels selected from known Adélie colonies plus an additional 10,688 pixels selected from areas of exposed rock, soil and vegetation. The retrieval algorithm used this training dataset to define an ellipsoid in the spectral data space that surrounds pixels from Adélie colonies, with the assumption that the surface of the ellipsoid separates the pixels within its bounds from other surface types. The equation of the ellipsoid is captured in a “transition matrix” which is used to build a decision rule for determining whether or not new pixels belong to the Adélie colony class. All of the Landsat-7 ETM+ reflective bands (bands 1–5 plus band 7) were used in the AP retrieval. The retrieval on the AP and surrounding islands was conducted on 62 scenes covering this region but did not include the South Shetlands or the South Orkney Islands because cloud-free imagery was unavailable for the era investigated. The methods described, however, can be directly applied to imagery from these areas in future surveys. Additional details on the AP retrieval algorithm, including the transition matrix and decision rules used in this study, can be found in the additional information within the file. An Antarctic continent-wide retrieval of Adélie penguin colonies was obtained by combining the AP retrievals with those previously generated by. In all cases, imagery was selected from the Landsat-7 era in the years from 1999–2003, prior to the failure of the ETM+ Scan Line Corrector. ## Abundance and distribution from high-resolution commercial satellite imagery We used the Adélie penguin survey by, which was based on VHR imagery supplemented with recent field surveys, as the basis for studying the performance characteristics of Landsat retrievals. Details of the detection and abundance estimation of Adélie penguins in VHR imagery may be found in and. From their global survey of Adélie penguins, produced a detailed list of all Adélie penguin colonies (location and estimated abundance) known to exist at the time of the survey, which we used as a baseline for assessing successful retrieval and abundance estimation using the lower resolution Landsat survey method. The comparison of our automated detection algorithm to the Adélie survey results from yields four possible outcomes for each potential Adélie penguin breeding site. 1. A penguin colony is detected using both methods. 2. No penguin colony is detected by either approach. This situation encompasses the vast majority of the Antarctic coastline. 3. A penguin colony is identified in but not identified by our automated detection algorithm applied to Landsat imagery. These incidents are considered errors of omission for the Landsat detection algorithm because the colonies identified by were, with a few exceptions, validated by direct field surveys. In these cases, we consulted the available imagery to determine what characteristics may have led to this error of omission. 4. An area is flagged as containing a colony of breeding penguins by the automated detection algorithm but was not reported in. These cases may represent an error of omission for or an error of commission by the automated detection algorithm. In these cases, we searched for additional VHR imagery that might resolve these two possible scenarios. If no additional imagery was available, we were not able to determine the nature of the discrepancy. All such cases are reported in the hopes that future field surveys can validate the true occupancy status of these locations. The conversion between sensor pixels flagged as guano and the abundance of penguins breeding within that area is determined by an ‘apparent density’ that is not necessarily equal to the actual density of breeding pairs as measured on the ground. Nevertheless, it has been demonstrated that the closely packed nature of Adélie penguin colonies allows for the estimation of abundance based on the area of guano staining in very high-resolution imagery. As a first pass, we used Kendall's rank correlation coefficient to investigate whether the number of pixels retrieved by Landsat-7 scales with penguin abundance as reported by. In other words, we ask whether the colony with the largest number of pixels retrieved by Landsat-7 is also the largest colony as reported in, and so forth. We then use the abundance estimates of to develop a predictive model for estimating the total abundance of Adélie penguins from medium-resolution Landsat imagery, recognizing that the ∼10 year time lag between the abundance estimates reported in and the older Landsat imagery significantly increases the uncertainty of these estimates. Nevertheless, such ‘ballpark’ estimates can be invaluable in the near term, particularly for estimating approximate abundance for previously undiscovered colonies, and provide a road map for re-analysis when matched-pairs of Landsat-8 imagery and field counts become available in the future. This methodology, applied here to Landsat-7 imagery and in the future to Landsat-8 or other medium resolution sensors, can be used to automate future continental scale surveys yielding global estimates of Adélie distribution and abundance with little to no manual interpretation. This abundance model involves two components: (1) the estimation of breeding abundance represented by the pixels identified as guano in the Landsat survey, and (2) the estimation of the abundance of breeding Adélie penguins not detected. ## Abundance estimation for detected colonies For the first component of total abundance, involving the estimation of abundance conditional on detection, we considered all breeding populations found by the Landsat-7 automated retrieval process whose population had been estimated previously by. The data used for our abundance estimates are detailed in. Seven colonies were not included in the development or validation of the abundance model because they either included a significant number of flying birds (which inflated the estimated size of the Adélie colony) or because a discrepancy between the Landsat retrieval and the location reported in made it impossible to know whether the same breeding population was being considered in each method. We used a Poisson regression model for abundance as a function of guano area where is the expected abundance of the *i*^th colony and e can be interpreted as an “apparent density” (nests per ‘detected guano area’). Since many applications, particularly within the context of the Convention on the Conservation of Antarctic Marine Living Resources, require abundance estimates at scales larger than an individual breeding colony (e.g., Convention Subareas), we also quantified predictive accuracy of our final abundance model for spatial units of increasing scale, from individual colonies to the entire continent. ## Estimation of abundance in colonies not detected by Landsat The total abundance of Adélie penguins not detected in a Landsat survey is a convolution of the probability distribution of colony abundance and the probability of detecting a colony given (i.e. conditional on) its size. The expected size of a colony that is not detected by Landsat imagery can be calculated as the following conditional expectationwhere *N* is the number of breeding pairs, and *M* represents the binary probability that a colony is missing (M = 0) from a Landsat survey. The probability that a colony is missing conditional on its size (i.e., *P(M = 0\|N = n)  =  1-detection probability*) can be estimated with a logistic regression model fit using the abundances of each colony as reported in. In the absence of future matched VHR-Landsat surveys, the detection probability can be assumed a function of the sensor that would not change for future surveys using Landsat-7 imagery. The probability distribution of colony sizes *P*(*N*) is a long-tailed distribution best fit by a log-normal distribution. In theory, we could estimate the best-fit parameters for the colony size probability distribution *P*(*N*) using the partially- censored Landsat survey detections by inverting Eq. 2 for *P*(*N*); however, the low rates of detection for very small colonies make it difficult to establish the shape of the colony size distribution from the Landsat detections alone. We therefore fit a log-normal distribution to the abundance data in and used a discretized version of this best-fitting model for *P*(*N*) in Eq. 2 to estimate the total abundance of Adélie penguins missing from the Landsat survey due to non-detection. The sum in Eq. 2 can be calculated using Markov Chain Monte Carlo integration, and the resulting expectation can be multiplied by expected number of missing colonies to estimate the total missing abundance. We modeled the total number of undetected Adélie colonies in the surveyed area as a sample from a Negative Binomial distribution where the overall probability of detection was estimated using the survey data in as a basis for comparison. ## Estimation of abundance at aggregated spatial scales To estimate the improvement in predictive accuracy at regional scales containing multiple penguin colonies, we compared total abundance estimated by VHR and Landsat for subsets of the original data. Our goal was to determine how many colonies would have to be enclosed within an area for the estimate of total abundance by Landsat-7 to be statistically indistinguishable from the corresponding total abundance estimated by VHR. For example, to compare total abundance estimates for a (hypothetical) area containing *k* = 5 colonies, we would draw 5 colonies at random from the total set of colonies in our study (continental and Peninsula areas combined) and use Eq. 1 (with best-fit parameters; see) to estimate the total abundance that would be estimated for those 5 colonies. The total abundance would then be compared to the expected total abundance from the VHR survey of, and the scaled difference recorded as Repeating this procedure 1000 times, we generated a distribution of scaled differences between the two methods. The null distribution consisted of calculating the same scaled difference but with the Landsat abundance replaced by a second independent draw from the VHR abundance distribution. (The abundance of each colony is associated with a sampling distribution, so independent draws from that distribution yield different estimates of abundance. This variability represents our uncertainty in the true abundance of the colony.) In other words, we wanted to differentiate between differences due to Landsat's relatively poorer performance characteristics, and differences due to inherent uncertainty for the VHR counts used as “true” abundance. The empirical cumulative distribution functions (VHR-Landsat vs. VHR-VHR) were compared using a one-sided Komolgorov-Smirnov test. # Results Using Landsat data over the Antarctic Peninsula, the automated retrieval flagged 143 areas as belonging to the “Adélie penguin colony” class. Based on the site- by-site comparisons described above, these retrievals included 16 confirmed Adélie penguin colonies. Among these retrieved areas were 17 additional areas identified as likely Adélie penguin colonies that could not be confirmed because VHR data were not available and the sites were not previously mentioned in the literature. Using Landsat data over continental Antarctica, the automated retrieval flagged 187 areas as belonging to the “Adélie penguin colony” class, within which were 91 confirmed Adélie penguin colonies, and 8 areas identified as likely, but unconfirmed, Adélie penguin colonies. The Landsat retrieval of Adélie penguin colonies over the entire continent of Antarctica can be quickly visualized via the gzip-compressed Keyhole Markup Language (kmz) formatted file provided (see additional information within the file). ## Probability of detection The detection performance of the automated Landsat retrieval improved with increasing Adélie penguin colony population size. The probability of detecting an Adélie penguin colony using Landsat (*p_detection*) was modelled using the logistic transformationwhere N is the abundance of breeding pairs,  = −1.00 (SE = 0.32) and  = 0.0004 (SE = 0.0001) for the continental colonies and  = −1.72 (SE = 0.45) and  = 0.0003 (SE = 0.0001) for the Peninsula colonies. The smallest colony detected in the Landsat retrieval had 322 pairs (Clark Island in Marie Byrd Land), but the threshold for 50% detection probability (as estimated by the logistic model) was 2240 breeding pairs for continental colonies and 6405 breeding pairs for Peninsula colonies. While nearly all of the smallest colonies go undetected using the automated interpretation of Landsat imagery, small Adélie penguin colonies (1 to 3000 pairs) contribute only ∼2% to the total Adélie population. Along the coastline of continental Antarctica, the Landsat method successfully identified Adélie penguin colonies making up 98.3% of the population total for this region. The AP has a significantly larger proportion of small colonies when compared to the rest of the continent, and consequently the Landsat method missed a larger number of these colonies in both absolute numbers and in proportion to the total . On the AP, the colonies found by the Landsat method account for 90.2% of the sample population. When all data are pooled together, the Landsat method retrieved those colonies that account for 97.3% of the population total. Furthermore, the Landsat retrieval algorithm detected a number of potential Adélie colonies not captured by the manual imagery search of, see. There were three Adélie colonies with populations greater than 4000 pairs that were missed in the Landsat retrievals: on Haswell Island (4011 pairs), Coulman South (14786 pairs), and Penguin Point (19377). As described in it is likely that Haswell Island was abandoned during the period when the Landsat imagery were collected (2001 and 2002). The Coulman South colony was covered by deep shadow in the Landsat imagery and is therefore a true error of omission. The only useable Landsat imagery over Penguin Point available during the survey era was collected on November 24, 2001, which is very early in the Adélie penguin breeding season. A second cloud-free image from January 8, 2001 was available but it was unusable due to band saturation. It is likely that cloud-free Landsat imagery from later in the breeding season would resolve the Penguin Point colony. ## Abundance estimation for detected colonies The number of Landsat-7 pixels retrieved as penguin guano is significantly correlated to the abundance of breeding penguins as reported in (Kendall's  = 0.73 \[n = 88; p\<0.001\]). The Poisson regression model in Eq. 1 yielded estimates of  = −1.0844 (SE = 0.0007) and  = −0.661 (SE = 0.002), for the continental and Peninsula data, respectively, corresponding to estimates of apparent density of 0.34 nests/m² and 0.52 nests/m², respectively. As expected from a Poisson regression, the predictive envelope grows with colony size. ## Estimation of abundance of colonies not detected by Landsat We used the distribution of colony abundances reported in, to find the best estimates of the log-normal distribution parameters for *P*(*N*) and found  = 8.06 (SE = 0.21) and  = 2.34 (SE = 0.15) for the continental colonies and  = 6.64 (SE = 0.32) and  = 2.19 (SE = 0.23) for the Peninsula colonies. Using Eq. 2, we estimate the total (Peninsula + continent) number of Adélie breeding pairs missing from the Landsat survey to be 119738 (95^th percentile CI: 70364–188070), which is higher but generally consistent with the empirical value of 75582 breeding pairs found in the VHR survey but not found by Landsat-7. ## Estimation of abundance in colonies with no prior estimates The Landsat survey found eight colonies that were not detected in the VHR survey reported by but which are outside the breeding range of either gentoo penguins (*Pygoscelis papua*) or chinstrap penguins (*P. antarcticus*), which can appear similar in satellite imagery. While other errors of commission are possible, we can be confident that these are not conflated with mis-identification of penguin species. Using Eq. 1, we estimate the total abundance of Adélie penguins in these eight colonies as ∼38000 breeding pairs, although these estimates, and in fact the existence of Adélie penguins at these eight locations, needs to be confirmed by higher-resolution satellite imagery or direct observation. lists 17 additional locations flagged as penguin colonies by the Landsat method that are within the breeding range of all three *Pygoscelis* species and thus cannot be identified to species, three of which are large penguin colonies in the Danger Islands. The Danger Islands, a collection of islands off the tip of Joinville Island at the end of the Antarctic Peninsula, are poorly surveyed due to frequent heavy ice cover and steep terrain that preclude easy access, but are known to contain extremely large populations of Adélie and gentoo penguins. Within the Danger Islands, the Landsat survey identified three major penguin populations at Brash Island, Earle Island, and Darwin Island, containing an estimated 166078 (95^th percentile CI: 123666–228268), 23649 (95^th percentile CI: 17361–32163), and 7419 (95^th percentile CI: 5384–9931) breeding pairs of penguins, respectively. Based on prior surveys of the area, these are likely to be overwhelmingly composed of Adélie penguins, making Brash Island potentially one of the largest Adélie penguin colonies in the world. Jorge Island was identified by as one location for which no census information existed, and was another location identified by the Landsat survey as containing breeding penguins. We estimate 3733 (95^th percentile CI: 2710–5064) breeding pairs of penguins (of unknown species) breeding at this location. ## Errors of commission An earlier study of Adélie penguin detection in continental Antarctica reported commission errors for the Landsat method on the order of 1% or less in the retrieval of Adélie colonies along the southern coastline of Antarctica. In a similar analysis for the retrievals on the Antarctic Peninsula, we used required prior knowledge of these locations and, where possible, confirmation by high- resolution satellite imagery. We find that commission errors for the Antarctic Peninsula fall into several categories including: pixels that identify potentially new colonies as determined by visual examination of high-resolution satellite imagery but which have not yet been confirmed by other methods, pixels that identify other penguin colonies or those with mixed species including Adélie penguins, pixels that identify flying bird breeding areas (known to exist, or previously unknown to exist), and pixels that are “pure errors” where various soil types are incorrectly categorized as Adélie penguin colonies. The commission error for soils incorrectly classified as penguin colony was calculated to be 3% for the Antarctic Peninsula, and 2% for the entire continent. ## Aggregated estimates of abundance While abundance estimates of individual colonies are variable using Landsat retrievals, our model for estimating abundance from the number of pixels identified as guano appears unbiased and thus neither consistently overpredicts or underpredicts abundance. Consequently, abundance estimates using Landsat retrievals improve at larger scales of aggregation. We find that abundance estimates using Landsat retrievals are statistically indistinguishable from those obtained using very-high resolution methods for aggregations containing \>c. 40 colonies. # Discussion As described below, the retrievals of Adélie colony extent from medium- resolution imagery correlate well with comparable datasets, and exhibit relatively low errors of omission and commission. As has been demonstrated also for high-resolution imagery, the retrievals of colony extent (i.e. guano area) as captured by automated classification of Landsat imagery correlate well with abundance. As a result, this area-abundance relationship can be used as the basis for estimating abundance in areas where the population is otherwise unknown. As illustrated in, the performance of the Landsat retrieval for colony detection improved with increasing colony population size. While nearly 100% of the smallest Adélie penguin colonies were undetected by the Landsat method, these contribute very little to regional and continental aggregated abundance and the Landsat retrievals are thus able to capture the vast majority of the Adélie penguin population. The Antarctic Peninsula has a significantly larger proportion of small colonies when compared to the rest of the continent and a higher threshold for detection, and consequently the Landsat method missed a larger number of these colonies in both absolute numbers and in proportion to the total. ## The Antarctic Peninsula remains a challenge The higher threshold for colony detection on the Antarctic Peninsula likely stems from several factors. These include a larger ratio of exposed surface material compared to snow cover, a greater variability of rock and soil types compared to the rest of the continent, and increased surface weathering. The AP also experiences significantly more rainfall than the continental portions of Antarctica. Weathering may wash away guano making it more difficult to detect smaller colonies, and it may contribute to estimation errors as the guano stain grows or shrinks within a given breeding season. At the same time, the topography of the Antarctic Peninsula is highly complex, with steep terrain that may obscure smaller colonies. Finally, higher cloud cover results in fewer usable scenes each year. One of the major challenges of using satellite imagery to map Adélie penguin colonies on the Antarctic Peninsula is the existence of two other, closely related, species of penguins (gentoo and chinstrap) whose guano displays similar spectral signatures to that of the Adélie. Moreover, these three species frequently nest in mixed-species colonies, and while differentiation of species has been shown possible in high-resolution imagery, the spatial resolution of Landsat is larger than the scale at which Adélie penguins may cluster within a mixed-species colony. In other words, a single Landsat pixel of a mixed-species colony likely contains multiple species that cannot be distinguished. Other *Pygoscelis* spp. penguins are not the only source of errors of commission. Flying bird species, such as the blue-eyed shag (*Phalacrocorax atriceps*), also produce guano that can be incorrectly identified as penguin guano and can cause errors of commission in both Landsat and VHR imagery. ## Errors of commission vary across regions and sensors Our experience with both VHR and Landsat imagery interpretation has shown that the causes of false positives (errors of commission) are both region and sensor specific. In the Antarctic Peninsula region, errors of commission can stem from breeding blue-eyed shags, as well as from large areas of snow algae, which is often red or pink in color, and iron-tinted scree on steep sloping ridges (Lynch Unpublished data). As well, areas used by penguins for molting can contain significant accumulations of guano and confuse mapping of breeding territories. Landsat imagery retrievals may contain areas used by flying birds either alone or with breeding Adélie penguins. We have also found abiotically-driven errors of commission in Landsat retrievals due to large wet alluvial fans, and others apparently due to sediment runoff or chlorophyll in the water just offshore. ## Future research Increased weathering, complex terrain, and the existence of congeneric species make it difficult to estimate omission and commission error rates for the Antarctic Peninsula. While a Landsat retrieval of a gentoo colony is an error of commission in the context of an Adélie survey, it still has biological significance, particularly if it identifies a colony that was previously undiscovered. The use of prior information regarding the distribution of penguins and the Antarctic's flying bird species may allow us to use adaptive classification methods that minimize both errors of commission and omission by essentially ‘raising the bar’ for the detection of previously unknown Adélie penguin colonies and simultaneously ‘lowering the bar’ for the re-detection of known Adélie penguin colonies. In the same vein, prior estimates (e.g., from direct field surveys) of the fraction of penguins that are of each species (Adélie, gentoo, and chinstrap) at mixed colonies may allow us to estimate Adélie abundance to produce less biased estimates at the regional of continental scale. While Landsat imagery lacks the sensitivity to detect small colonies, and is therefore inappropriate for studying processes such as colonization and range expansion, we have demonstrated that it can be used to obtain estimates of Adélie abundance over regional scales, and that bias in regional estimates can be corrected for non-detection. Corrections for non-detection will be made more accurate by better statistical models for colony size, but the existing precision of regional and global Adélie abundance is minimally affected by non- detection of the smallest colonies. # Supporting Information We thank Claire Porter for her expertise and tireless efforts processing the satellite imagery required for this analysis. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HJL MRS. Performed the experiments: HJL MRS. Analyzed the data: HJL MRS. Wrote the paper: HJL MRS.

# Introduction *Mycoplasma pneumoniae* (*M*. *pneumoniae*), a small prokaryote devoid of a cell wall, is one of the most common etiological agents of human respiratory tract infections. About 10–40% of all cases of community acquired pneumonia can be attributed to *M*. *pneumoniae*, which occurs in epidemic peaks at intervals of 3–7 years. Occasionally, severe cases with extrapulmonary involvement can result in hospitalization and death due to neurological disease, such as encephalitis. The high rate of morbidity and the occasional mortality reinforces the need for timely diagnosis. While the culturing of *M*. *pneumoniae* is time-consuming (2–8 weeks) and unreliable, serological detections require the availability of paired sera exhibiting IgG antibodies, which is age- and time-dependent. Molecular approaches, especially real-time PCR methods, are currently used for rapid, sensitive, and specific detection of *M*. *pneumoniae* in clinical specimens. In addition to detection, typing of clinical isolates is practically important for understanding the epidemiology of *M*. *pneumoniae* infections and for analysis of endemic outbreaks. In general, *M*. *pneumoniae* is regarded as a genetically highly uniform species. Previous investigations have revealed 2 types of clinical isolates (type 1 and type 2) that significantly differ in their sequences of the P1gene, *mpn 141*. This P1 gene, which codes for an important 170-kDa adhesin and antigenic factor, is the most frequently used typing target for *M*. *pneumoniae*. Typing schemes based on conventional PCR, restriction fragment length polymorphism, real-time PCR with high-resolution melt-analysis, and sequencing have been developed to differentiate the P1 genotypes of *M*. *pneumoniae*. About 8% of the *M*. *pneumoniae* genome consists of multiple copies of four RepMP elements, two of which (RepMP 2/3 and RepMP 4) are within *mpn 141*. Besides the two major *M*. *pneumoniae* types, in-depth analysis of the RepMp2/3 and RepMp4 repeat sequences within the P1 gene has revealed one sequence variant of type 1 (i.e., variant 1 or V1) and four sequence variants of type 2 (i.e., V2a, b, c, and d). This variation can be generated by homologous recombination of the RepMP elements located both inside and outside the P1 gene. Preservation of these repetitive sequences during presumed genome minimalization reinforces their importance, as they form pools of sequences for homologous recombinations that yield antigenic variations of *M*. *pneumoniae* proteins. It is therefore highly likely that novel variants of the P1 gene will continue to appear with homologous recombination, thereby increasing the ambiguity of the traditional P1 typing approach. Furthermore, some studies have indicated that typing targets other than the P1 gene may exist. Musatovova et al. revealed that the sequence divergence involving the RepMP 1-containing genes (*mpn 130*, *mpn 137*, and *mpn 138*), is strictly two types specific. Spuesens et al. and Catrein et al. also confirmed two highly conserved groups of *M*. *pneumoniae* strains based on sequence divergence of the RepMP5-containing gene, *mpn 142*. In recent years, attempts have been made to find additional proteomic or genomic markers for typing of *M*. *pneumoniae*. Pereyre et al. and our group classified *M*. *pneumoniae* strains into two distinct groups, based on polypeptide levels, by matrix-assisted laser desorption ionization–time of flight mass spectrometry (MALDI-TOF MS). Genomic analyses have also demonstrated that *M*. *pneumoniae* strains can be classified into two distinct types and further classified into sequence variants of each type. This notion was supported by the analysis of phylogenetic trees generated from all RepMP2/3 and RepMP4 sequences. In each of these trees, clear separation was observed between the type 1 and type 2 strains. In 2015, two reports of the whole-genome analysis of 38 *M*. *pneumoniae* were published. Whole-genome sequences were obtained by sequencing 20 clinically isolated strains, including type 1, type 2, V2a, and V2c strains, at our laboratory (unpublished data). We compared our genome sequences with the following *M*. *pneumoniae* whole-genome sequences reported on the National Center for Biotechnology Information (NCBI) website: M129 (GenBank: U00089.2), FH (GenBank: CP002077.1), resequenced FH (GenBank: CP010546.1), 309 (GenBank: AP012303.1), M29 (GenBank: CP008895.1), M129-B7 (GenBank: CP003913.2), PO1 (GenBank: CP010551.1), 19294 (GenBank: CP010539.1), 39443 (GenBank: CP010540.1), 85138 (GenBank: CP010545.1), 85084 (GenBank: CP010544.1), 54524 (CP010543.1), 54089 (CP010542.1), 51494 (CP010541.1), MAC (CP010550.1), M2592 (GenBank: CP010549.1), M2192 (GenBank: CP010548.1), M1139 (GenBank: CP010547.1), and PI1428 (GenBank: CP010538.1). We found that the genome sequences exhibited distinct characteristics consistent with two biological types of *M*. *pneumoniae*. Taken together, these previous reports confirm the existence of two highly conserved types of *M*. *pneumoniae* strains, each with respective variants based on minor sequence variations in the P1 gene. Nevertheless, broad application of the aforementioned conventional bacterial typing schemes is hampered by the time-consuming and unreliable cultivation of *M*. *pneumoniae* from clinical specimens. Thus, these lengthy typing methods have low sensitivity compared to real-time PCR performed directly in clinical specimens without cultivation of *M*. *pneumoniae*. To simplify the detection/typing procedure and increase sensitivity, our aims were to develop a culture-independent, duplex real-time PCR method for detecting and typing *M*. *pneumoniae* simultaneously in clinical specimens and to evaluate the detection/typing performance of this method by comparison to that of conventional methods. # Materials and Methods ## Ethics statement The study was approved by the Research Ethics Committee of the National Institute for Communicable Disease Control and Prevention (ICDC). No specimens were collected for this study. We used 1344 preserved throat swab specimens that were obtained in our laboratory during previous studies. The consent procedure for obtaining the throat samples was approved by the ethics board committee of Beijing Chao-Yang Hospital, Beijing Children’s Hospital, and Beijing Centers for Disease Control and Prevention, separately. Because throat swab is noninvasive, verbal consent was obtained from all patients or their guardians (for minor patients). ## Clinical specimens and strains From August 2008 to December 2014, a total 1344 throat swab specimens from 388 pediatric (\< 14 years of age) and 956 adolescent and adult patients were collected from Beijing Chao-Yang Hospital, Beijing Children’s Hospital, and Beijing Centers for Disease Control and Prevention. All clinical specimens were obtained from respiratory tract infection patients according to clinical symptoms. The following American Type Culture Collection (ATCC) reference and clinically isolated strains were used to assess the sensitivity, specificity, lowest detectable limit (LDL), and typing ability of the duplex real-time PCR method: *M*. *pneumoniae* (ATCC 29342, 15531, 39505, 29343, 29085, 15377, 15492, 15293, and 49894; ATCC, Manassas, VA), *Mycoplasma salivarium* (ATCC 23064), *Mycoplasma orale* (ATCC 23714), *Mycoplasma faucium* (ATCC 25293), *Mycoplasma genitalium* (ATCC 33530), *Ureaplasma urealyticum* (ATCC 27618), *Mycoplasma fermentans* (ATCC 19989), *Mycoplasma hominis* (ATCC 23114), *Mycoplasma penetrans* (ATCC 55252), *Mycoplasma hyorhinis* (ATCC 17981), *Mycoplasma pirum* (ATCC 25960), *Escherichia coli* (ATCC 11229), *Streptococcus pneumoniae* (ATCC 49619), *Staphylococcus aureus* (ATCC 29213), *Staphylococcus epidermidis* (ATCC 35984), *Pseudomonas aeruginosa* (clinical isolate), *Chlamydophila pneumoniae* (clinical isolate), *Mycobacterium tuberculosis* (clinical isolate), *Legionella pneumophila* (clinical isolate), *Haemophilus influenzae* (clinical isolate), *Neisseria meningitidis* (clinical isolate), and 56 other *M*. *pneumoniae* clinical isolates. ## Duplex real-time PCR assay By comparing the genome sequences of 20 *M*. *pneumoniae* clinical isolates with those of the reference strains, we observed remarkable differences in large fragment insertions or deletions, mainly between genes *mpn* 459 (corresponding to the type 1 strain, M129) and *mpna* 5864 (corresponding to the type 2 strain, 309). The type 1 strains had an approximately 3-kb insertion encoding three genes, *mpn* 457–459, whereas the type 2 strains had an approximately 5-kb insertion encoding five genes, *mpna* 5861–5865. Therefore, the primers and probes, designed with Primer Express software (version 3.0; Life Technologies- Applied Biosystems, Grand Island, NY, USA), were based on the conserved sequences of the 3kb and 5kb insertion fragments within the type 1 and type 2 strain, respectively. These primers and probes are listed in. Each PCR mixture was prepared in a total volume of 25 μl and contained the following per reaction: 12.5 μl Platinum Quantitative PCR SuperMix-UDG (Life Technologies- Invitrogen), 1.5 μl MgCl₂ (50 mM), 0.5 μM final concentration of each primer, 0.2 μM final concentration of each probe, 1.25 U Platinum Taq DNA polymerase (5 U/μl; Life Technologies-Invitrogen), 1 μl PCR nucleotide mix (10 mM), 5 μl nucleic acid extracted from each specimen, and nuclease-free water to achieve a 25 μl final volume. Real-time PCR for each target was performed in the CFX96 Real-time PCR Detection System (Bio-Rad, Hercules, CA, USA) under the following conditions: predenaturation at 95°C for 2 min, followed by 45 cycles at 95°C for 15 s and 56°C for 15 s. The data were analyzed with the CFX Manager Software (version 2.1; Bio-Rad). ## LDL, sensitivity, and specificity of duplex real-time PCR assay DNA was extracted from the reference strains and clinical isolates mentioned above with the QIAamp DNA Blood Mini Kit (QIAGEN, Hilden, Germany). The standard curves and LDL for each *M*. *pneumoniae* type were determined by analyzing serial 10-fold dilutions of Tris-buffered DNA extracted from *M*. *pneumoniae* type 1 (M129 strain \[ATCC29342\], 0.162 fg/μl to 0.162 ng/μl) and *M*. *pneumoniae* type 2 (FH strain \[ATCC 15531\] 0.183 fg/μl to 0.183 ng/μl). *M*. *pneumoniae* type 1 DNA was detected with FAM fluorescence, and type 2 DNA was detected with VIC fluorescence. Each dilution of DNA was assayed in triplicate. ## DNA mixtures for duplex real-time PCR detection analysis The detection ability was determined by analyzing the following four mixtures comprised of different compositions of ATCC 29342 and ATCC 15531 DNA: mixture A (1.62 pg/μl ATCC 29342 and 1.83 fg/μl ATCC 15531), mixture B (1.62 fg/μl ATCC 29342 and 1.83 pg/μl ATCC 15531), mixture C (1.62 pg/μl ATCC 29342 and 1.83 pg/μl ATCC 15531), and mixture D (1.62 fg/μl ATCC 29342 and 1.83 fg/μl ATCC 15531). Each DNA mixture was assayed in triplicate. ## *M*. *pneumoniae* strains for comparative typing analysis by culture-based and duplex real-time PCR methods DNA extracted from 9 *M*. *pneumoniae* ATCC strains and 56 clinical isolates were amplified with the duplex real-time PCR method for typing analysis. All 65 *M*. *pneumoniae* strains were genotyped by full-length sequencing of the P1 gene with the amplification primers SeqP1-F (`5’-ATGCACCAAACCAAAAAAACTGCCT-3’`) and SeqP1-R (`5’-CTAAGCGGGTTTTTTAGGTGGTTGC-3’`). The same *M*. *pneumoniae* strains were also genotyped by MALDI-TOF MS, based on their peptide mass fingerprints, as described in our previous report. ## Analysis of *M*. *pneumoniae* detection and typing in clinical specimens A total of 1344 throat swab specimens were cultured with selective Mycoplasma Broth (Thermo Scientific-Oxoid Limited, Basingstoke Hampshire, UK). Each culture-positive specimen was subcultured and purified by the filtration-cloning technique for *M*. *pneumoniae* clinical isolates. Genomic DNA of each isolate was then extracted with the QIAamp DNA Blood Mini Kit, and subsequently genotyped with the P1 gene-based PCR method as previously reported. The duplex real-time PCR assay and the previously reported MpP1 real-time PCR assay were used simultaneously to detect *M*. *pneumoniae* DNA in the throat swab specimen extracts. ## Statistical analysis Quantitative data are presented as the mean ± standard deviation (SD). The results were calculated in the Statistical Analysis System software package (version 9.3; Cary, NC, USA). Differences in number of positive results between each two of the three detection/typing methods were tested with the paired chi- square test. Differences in *M*. *pneumoniae* typing differentiation ability and in number of positive typing results between the P1 gene-based PCR method and the duplex PCR method were tested with the chi-square test and paired chi-square test, respectively, with a *P* value \< 0.01 considered statistically significant. # Results ## LDL, sensitivity, and specificity of duplex real-time PCR assay Based on the crossing threshold (CT) values for the serial 10-fold dilutions of DNA, the LDLs of *M*. *pneumoniae* type 1 and type 2 were approximately 8.1 fg and 9.3 fg DNA, respectively. Both these amounts are equivalent to approximately 3 colony forming units (CFU). The standard curves (CT vs. log CFU) for *M*. *pneumoniae* type 1 and type 2, generated by the serial 10-fold dilutions, yielded coefficient of determination (r²) values of 0.996 and 0.997, respectively. The duplex real-time PCR assay amplified DNA from all *M*. *pneumoniae* species tested (9 ATCC strains and 56 clinical isolates). This assay did not amplify DNA from any of the other Mycoplasma species, other respiratory bacteria (10 mollicute species and 10 common respiratory bacteria, listed in Materials and Methods), or human DNA, even when tested at concentrations (at ng level) markedly higher than the LDL. ## Duplex real-time PCR detection analysis with DNA mixtures Mixture C, with higher (pg) concentrations of *M*. *pneumoniae* type 1 and type 2 DNA, detected positive for both types, but the CT for each fluorescent (FAM and VIC) signal was delayed about 1–2 cycles compared to that for each signal of amplified *M*. *pneumoniae* DNA detected individually at the same concentration. Mixture D, with lower (fg) concentrations of type 1 and type 2 DNA, detected positive for both types only once. In mixtures A and B, with different concentrations of type 1 and type 2 DNA, only the type with the higher concentration could be detected. ## Comparative *M*. *pneumoniae* typing analysis by culture-based and duplex real-time PCR methods Duplex real-time PCR typing of 9 *M*. *pneumoniae* ATCC strains and 56 clinical isolates revealed 44 type 1 strains and 21 type 2 strains. These typing results were completely consistent with those produced by MALDI-TOF MS and P1 gene-based typing. The 21 type 2 strains revealed by the P1 gene-based method comprised of three traditional type 2, two V2a, and 16 V2c strains, based on the minor variations in the P1 gene. No variant 1 strain was detected. ## *M*. *pneumoniae* detection and typing analysis in clinical specimens Of the 1344 throat swab specimens, the *M*. *pneumoniae* positive detection rates of the culture method, MpP1 real-time PCR assay, and our duplex real-time PCR assay were 26.9% (362/1344), 34.4% (462/1344), and 33.7% (453/1344), respectively. The positive detection rates of the MpP1 and duplex real-time PCR assays were not significantly different (paired χ², *P* = 0.0389). Differences were observed between those of the culture method and each of the PCR methods (both paired χ², *P* \< 0.01). A total of 356 specimens tested positive by all three methods, 92 specimens tested positive by the two real-time PCR methods, and only 4 specimens tested positive by the culture method alone. Of the 362 *M*. *pneumoniae* isolates cultured from the throat swab specimens, 98 were genotyped by full-length sequencing of the P1 gene, and the other 264 isolates were genotyped by the P1 gene-based PCR method. Based on genotyping analysis by the culture-based methods, 85.4% (309/362) of the isolates were traditional type 1 (no variant 1 detected), and 14.6% (53/362) were type 2 (2 isolates were variant 2a, and 51 isolates were variant 2c). Compared to the aforementioned culture-based typing methods (362/1344), our duplex real-time PCR (453/1344) analysis of the throat swabs identified a greater number of *M*. *pneumoniae* genotypes (paired χ², *P* \< 0.01). Among the 453 clinical specimens that tested positive by duplex real-time PCR, both *M*. *pneumoniae* types were identified in 3 specimens (0.7%). MpP1 real-time PCR identified only one of the 3 specimens as positive, and each of these 3 specimens was negative by culture and P1 gene-based typing. Based on genotyping analysis by the duplex real-time PCR assay, 84.4% (385/456 identified genotypes) were type 1, and 15.6% (71/456 identified genotypes) were type 2. A total of 97 culture negative specimens without a P1 gene-based typing result were successfully genotyped with the duplex PCR assay, and only 6 culture- positive specimens genotyped with the P1 gene-based typing method were not detected positive and typed by the duplex real-time PCR assay. From 2008 to 2014, the yearly percentages of *M*. *pneumoniae* type 2 genotypes identified by the P1 gene-based typing method were 11.5% (7/61), 20.0% (4/20), 3.1% (2/64), 9.9% (13/131), 15.2% (5/33), 42.3% (11/26), and 40.7% (11/27), respectively. During the same period, the yearly percentages of *M*. *pneumoniae* type 2 genotypes identified by the duplex real-time PCR assay were 10.8% (8/74), 12.9% (4/31), 5.1% (4/79), 10.9% (18/165), 19.5% (8/41), 45.5% (15/33), and 42.4% (14/33), respectively. The total percentages of *M*. *pneumoniae* type 2 genotypes identified by the P1 gene-based typing and duplex real-time PCR methods during this period were 14.6% (53/362) and 15.6% (71/456), respectively. These genotype percentages were not different from those generated by the culture-based typing methods mentioned above (χ², *P* = 0.7128). Between 2008 and 2012, the percentage of identified *M*. *pneumoniae* type 2 genotypes in clinical specimens from Beijing did not exceed 20%. However, the percentage of type 2 genotypes rapidly increased to over 40% in2013 and 2014. # Discussion In this study, we demonstrated that our duplex real-time PCR assay can simultaneously detect and type *M*. *pneumoniae* directly from extracted DNA in clinical specimens, without cultivation of the pathogen. The design of our primers/probes was based on the conserved sequences of the type 1-specific 3kb and type 2-specific 5kb insertion fragments. In the NCBI database, the *M*. *pneumoniae* FH strain (GenBank: CP002077.1) is the only type 2 strain that does not have the 5-kb insertion fragment. Unexpectedly, the resequenced FH strains (GenBank: CP010546.1) had the 5-kb insertion at position *mpna* 5861–5865. Furthermore, the ATCC 15531 (FH) strain that we acquired and used in our analysis still had the 5-kb insertion. This finding was also reported in a previous study by Kenri et al.. Whether the ATCC FH strain in our laboratory and the genome-sequenced FH strain (GenBank: CP002077.1) in the NCBI database are in fact different is unclear. Results from the paired chi-square analyses of the positive detection rates generated by the culture method and the two real-time PCR methods indicate that the detectability of our duplex PCR assay was similar to that of the MpP1 PCR assay and higher than that of the culture method. Only 6 of the 362 culture- positive throat swabs were detected negative by the duplex real-time PCR assay. After further purification of the *M*. *pneumoniae* DNA from these specimens, we repeated the assay and found that all 6 specimens tested positive and were correctly genotyped. It is likely that the DNA extracts from these clinical specimens initially contained unknown substances that inhibited PCR amplification. Our assessment of the duplex real-time PCR typing ability revealed that the ATCC strain and clinical isolate typing results from the duplex real-time PCR method were in perfect agreement with those from the culture-based methods. Furthermore, duplex real-time PCR analysis (453/1344) of the clinical specimens identified a greater number of *M*. *pneumoniae* genotypes compared to the culture-based typing methods (362/1344), and there were no significant differences in the percentages of each *M*. *pneumoniae* type between the duplex PCR and the culture-based methods. Taken together, these findings demonstrate that this novel duplex real-time PCR assay has high accuracy for genotyping as well as detecting *M*. *pneumoniae* in clinical specimens. Coinfection with both types of *M*. *pneumoniae* in humans has never been previously reported. In this study, the duplex real-time PCR assay yielded a coinfection rate of 0.7%. However, this rate is likely inaccurate, because analysis of the *M*. *pneumoniae* type 1 and type 2 DNA mixtures revealed that the fluorescent signal corresponding to the DNA type of lower concentration could be quenched by that corresponding to the DNA type of higher concentration. Furthermore, rare strains may exist that contain a single P1 gene, as well as both the 3- and 5-kb insertions. Although our findings do not prove their existence, the possibility of such rare strains cannot be excluded. Unfortunately, *M*. *pneumoniae* strains could not be isolated from 3 specimens. Therefore, the clinical significance of coinfection with both types and the possible existence of a rare strain of *M*. *pneumoniae* remain unclear and require further study. The annual rates of *M*. *pneumoniae* genotypes in clinical specimens from Beijing revealed that type 1 was the predominate genotype from 2008 to 2012. This trend continued for about 5 years, and then a shift toward type 2 began in 2013 and continued through 2014. In previous years, type 1 was always the predominate genotype in China. Therefore to our knowledge, this is the first report of *M*. *pneumoniae* type shifting in China. Of the 53 type 2 strains detected by the culture-based methods, the majority of them were V2c. Unfortunately, the type 2 variants could not detected by duplex real-time PCR. Previous reports have indicated that a certain strain of *M*. *pneumoniae* could confer a type-specific immunity. Although type-specific protection against *M*. *pneumoniae* in response to previous infections in the human population has been reported, the relationship between the *M*. *pneumoniae* type shift phenomenon and pneumonia outbreaks needs to be further verified with more extensive surveillance data. This type shift phenomenon has already been reported in Japan and France. Given that, as reported by Kenri et al., a type shift phenomenon occurs every 8–10 years, and a shift from one type to another requires 2–3 years, we can deduce that a shift from *M*. *pneumoniae* type 1 to type 2 in Japan would have occurred around 2012–2013. A recent study by Kubota et al. found that type 1 was still the predominate genotype among *M*. *pneumoniae* strains isolated in Japan between 2011 and 2012. Another report by Ishiguro et al. revealed that the percentage of *M*. *pneumoniae* type 2 isolates in Japan increased from December 1, 2012 to July 31, 2014. To a certain extent, these results support our deduction. Given that Beijing and Japan are located in the same latitude of the East Asia region where there is frequent population migration, it is possible that an *M*. *pneumoniae* type shift phenomenon occurred in these two regions simultaneously. Based on our own findings and inference from the Kenri et al. report, we hypothesize that an *M*. *pneumoniae* type shift phenomenon began in Beijing in 2013 and will last for a few years. Currently, *M*. *pneumoniae* in our population is still in the process of type shifting. Type 2c (V2c) will most likely become the predominate genotype in our population over the next few years. Although *M*. *pneumoniae* type 2 variants have been detected in many regions, they comprised a very small percentage of the *M*. *pneumoniae* isolates and never became the predominant genotype. In this study, we found that *M*. *pneumoniae* in our population started shifting from type 1 to 2c at about the same time that this type 2 variant was becoming increasingly prevalent in Japan. Based on these findings, *M*. *pneumoniae* type 2c will likely become the first predominant type 2 variant in the world, perhaps as a result of selection pressure on the human population. Interestingly, a recent publication failed to find a clear increasing or decreasing trend in P1 type 1, P1 type 2, or their proportion from 2003 to 2012 in Germany. Rather than supporting the hypothesis of type shifts before or during epidemic phases, this result favors the influence of a decreasing level of specific antibodies within the human population as the reason for the start of epidemics. Exactly why these results were different is not clear and requires further study. In any event, we will continue to monitor the status of V2c in Beijing over the next decade. Although the relationship between the type shift phenomenon and pneumonia outbreaks needs to be further verified, increased attention on *M*. *pneumoniae* surveillance in Beijing is warranted. # Conclusion Compared to the conventional, culture-based *M*. *pneumoniae* typing methods, our duplex real-time PCR assay has distinct advantages, including time saving and higher typing sensitivity. Although this assay can rapidly differentiate the two *M*. *pneumoniae* types, it is unable to differentiate the variant strains. Based on the findings from this study, this duplex PCR assay is most useful as a rapid surveillance tool for *M*. *pneumoniae* rather than a refined scientific research method. The sensitivity and specificity of the assay should be validated by testing more known *M*. *pneumoniae* isolates in future studies. In conclusion, this novel duplex real-time PCR assay can detect and type *M*. *pneumoniae* in clinical specimens with high sensitivity and accuracy. Our genotyping results provide important information for understanding recent type shifts that can change the epidemiological characteristics of *M*. *pneumoniae* in Beijing. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FZ JZ. Performed the experiments: FZ LL XT LH FM. Analyzed the data: FZ JZ. Contributed reagents/materials/analysis tools: FZ. Wrote the paper: FZ JZ.

# Introduction Large intergenic non-coding RNAs (lincRNAs) are recently emerging as a novel class of functional non-coding RNAs, which are more than 200 nucleotides in length, derive from the intervals between protein-coding genes, have similar exon-intro-exon structure, but lack of protein-coding capacity. As yet, the quantity of discriminated human lincRNA transcripts continue to increase, and many of them have been found to play important roles in multiple biological processes, including epigenetic regulation of protein-coding gene expression – and crucial action in development process. Emerging evidence has also demonstrated that numerous lincRNAs were associated with a wide range of human diseases. Recently, several profiling studies have revealed that dysregulated expression of lincRNAs was involved in several forms of human cancer. For example, a study has reported that the expression level of lincRNA *PCGEM1* was higher in prostate tumor specimens than in matched normal tissues. LincRNA *HOTAIR* (HOX antisense intergenic RNA) can be regard as an independent cancer prognostic marker due to its significantly overexpression in breast cancer, hepatocellular cancer, colorectal cancer and laryngeal squamous cell carcinoma. Another highly abundant lincRNA *MALAT1* (also known as *NEAT2*) is originally identified as a marker for lung cancer metastasis; its expression is strongly regulated in many tumor entities including lung adenocarcinoma and hepatocellular carcinoma. In addition, it has been demonstrated that up-regulation of a lincRNA *HULC* is highly associated with the incidence of hepatitis B virus (HBV) infection. However, despite a number of lincRNAs having aberrant expression in disease states, the causality that affects the expression abundance of lincRNAs has yet to be completely understood. Previous studies have shown that single nucleotide polymorphisms (SNPs) in transcription factor binding sites (TFBSs) of protein-coding genes could affect gene expression by altering transcription factor binding, and participated in human diseases. A recent study on a tumor suppressor lincRNA has also demonstrated that a SNP (rs944289) could predispose to papillary thyroid carcinoma through dysregulating lincRNA (PTCSC3) expression by decreasing the binding activity of both C/EBPα and C/EBPβ. Thus, SNPs in the human lincRNA TFBSs can act as a set of functional variants, which may disrupt transcription factor binding, resulting in the diversity of lincRNA expression and, potentially, diverse diseases. Furthermore, with the advent of high-throughput technologies, large-scale lincRNA annotation data, SNP data, predicted and experimentally supported TFBSs data have been generated. This provides a great opportunity to systematically identify SNPs in the human lincRNA TFBSs. For example, in the new update of NONCODE database, the lincRNA data set were expanded by collection of newly identified lincRNAs from published literatures and integration of the latest version of RefSeq and Ensembl. LncRNADisease database collected experimentally supported lncRNA-disease associations and lncRNA interacting partners at various molecular levels. ChIPBase database was developed to annotate and identify TFBSs and transcriptional regulatory relationships of lncRNAs and miRNAs from ChIP-Seq data. In addition, the ENCODE project has compiled a large number of ChIP-Seq experiments for many human TFs in different cell lines and tissues. Enriched peak regions of these ChIP-Seq data can be mapped to the promoter regions of lincRNAs, which facilitate the discovery of experimentally supported TFBSs of human lincRNAs in different cell lines and tissues, and also give us a better opportunity to identify SNPs in lincRNA TFBSs for a cell line of interest. Therefore, to provide a beneficial annotation of these potential functional variants in human TFBSs, we developed a SNP@lincTFBS database for integrating and annotating functional SNPs in predicted lincRNA TFBSs. We identified 6,665 SNPs occurring in 6,614 TFBSs of 2,423 human lincRNAs, and provided a comprehensive and useful resource of candidate SNPs relevant to the aberrant expression of lincRNAs. The SNP@lincTFBS database will be helpful to identify functional SNPs of lincRNAs in the level of transcription and contribute to profound complex disease study. # Materials and Methods ## Human lincRNA data We obtained 6,631 human lincRNAs with genomic coordinates from the lincRNA list of GENCODE project (version 16), and removed lincRNAs without unique determinate chromosomal location. Finally, 5,835 lincRNAs were contained in SNP@lincTFBS. ## Identifying conserved TFBSs of human lincRNAs We downloaded the locations and scores of conserved TFBSs from the UCSC genome browser. These data were obtained by running the program tfloc (Transcription Factor binding site LOCater) on multiz46way alignments, restricting only to the July 2007 (mm9) mouse genome assembly, the November 2004 rat assembly (rn4), and the February 2009 human genome assembly (hg19). A binding site is considered to be conserved across the alignment if its score meets the threshold score for its binding matrix in all 3 species (human, mouse and rat). Transcription factor information was downloaded from the Transfac Factor database, and the score and threshold were computed with the Transfac Matrix Database (v7.0) created by Biobase. Then, We defined 5 kb upstream to 1 kb downstream region of the transcription start site of each lincRNA as its promoter region refer to previous study. We identified the conserved TFBSs of human lincRNAs in these regions; as a result, we identified 33,181 TFBSs in defined promoter regions of 3,839 human lincRNAs. ## Identifying TFBSs of lincRNA using genome-wide ChIP-Seq data We downloaded 690 ChIP-Seq datasets for 169 human transcription factors in different cell lines and tissues from ENCODE project. These peak datasets were computed by a peak calling method (PeakSeq), which identified enriched peaks through comparing each ChIP-Seq dataset to corresponding control experiment. Then, we identified the peaks that were located in the promoter regions of human lincRNAs (5 kb upstream to 1 kb downstream region of the transcription start site for each lincRNA). In total, we identified 323,256 transcription factor peaks of different transcription factors in 4,831 lincRNA promoter regions. ## Identifying SNPs in the TFBSs of human lincRNA We downloaded SNPs (common and rare variants) in public dbSNP database (build ver. 137) and identified 6,665 SNPs within 6,614 putative TFBSs of 2,423 human lincRNAs. In addition, with ChIPSeq dataset, we identified 139,576 SNPs in 304,517 transcription factor peaks of 4,813 lincRNAs. Then, we downloaded the annotation information of minor allele frequencies and others from 1000 Genomes Project (release of July 2012) datasets across 11 populations, and performed comprehensive annotation for these SNPs in lincRNA TFBSs. For each SNP in a lincRNA TFBS, we also extracted the flanking sequence of 30 nt up-/down-stream of the SNP position from RefSeq reference genomic sequence. ## Collecting experimentally supported disease-associated SNPs in lincRNA TFBSs We manually collected known disease-associated SNPs in lincRNAs TFBSs using PubMed to search the previous studies. We also annotated lincRNAs in SNP@lincTFBS that have been reported to be associated with diseases, and identified SNPs within their putative TFBSs. In addition, we integrated recently well-known disease-associated SNPs and disease lincRNAs into SNP@lincTFBS database. ## Database implementation SNP@lincTFBS is an online query tool developed utilizing ECLIPSE platform as the frontend, and MySQL as the backend database. The web engine was implemented using JSP technology, Struts framework and the Java connection pool Proxool, and web server was built using Apache Tomcat. # Results ## Overview of the SNP@lincTFBS Database We developed a novel integrated database named SNP@lincTFBS that allows users to perform SNP and TFBS searches in human lincRNAs. In this database, we: 1) obtained human lincRNAs, 2) identified conversed TFBSs and transcription factor peaks in defined promoter regions of human lincRNAs, 3) identified SNPs in the TFBS of lincRNA and collected experimentally supported disease-associated SNPs in lincRNA TFBSs, 4) integrated annotation information of SNP, TFBS and lincRNA. The architecture of identifying SNPs in lincRNA TFBSs is shown in. Currently, SNP@lincTFBS contains 8,290 entries of annotated SNP-TFBS-lincRNA associations, including 3,839 lincRNAs, 33,181 conserved TFBSs, 6,665 SNPs and 165 transcription factors. In addition, 19,878,236 entries of SNP-peak-lincRNA associations were stored in SNP@lincTFBS, including 4,831 lincRNAs, 323,256 transcription factor peaks, 139,576 SNPs and 169 transcription factors. We identified a large number of conserved TFBSs in the promoter regions of human lincRNAs and found that the distribution of SNPs in these lincRNA TFBSs was extensive. Previous studies have shown that each transcription factor can bind to several TFBSs in the promoter regions of protein-coding genes, thereby controlling the transcription of genetic information from DNA to messenger RNA. We also found a similar phenomenon in human lincRNA and a transcription factor could bind to many conserved lincRNA TFBSs (∼247 lincRNA), whereas ∼20 TFBSs that have been identified SNPs within them, and every 5.3 TFBSs had a SNP for each transcription factor. In addition, we observed that high frequencies of SNPs within lincRNA TFBSs to be located around lincRNA start site, suggesting that these SNPs within lincRNA TFBSs might greatly affect the expression of lincRNAs. ## Web interface The SNP@lincTFBS database website includes seven modules: home, search, overview, disease lincRNA, GWAS SNP, download and help (available at <http://bioinfo.hrbmu.edu.cn/SNP_lincTFBS>). HOME page provides a brief description of the SNP@lincTFBS database, users can browse the high-resolution flowchart of this work to get the main idea of this database. SEARCH page provides a quick search by query of three kinds of entries: 1) a lincRNA name (Ensembl ID), 2) an SNP identifier (rs number from dbSNP), and 3) a transcription factor name. Statistic of dataset contained in the database is introduced. Search result shows lincRNA summary information and all identified TFBSs and TF peaks in promoter region of this lincRNA. SNPs in these TFBSs and TF peaks are listed below. OVERVIEW page provides a general overview of transcription factors stored in SNP@lincTFBS. Disease lincRNA page shows existing experimentally supported disease-associated lincRNAs with their annotations and internal links for their TFBSs and SNPs mapped within them. GWAS SNP page shows disease-associated SNPs from GWAS researches that can be mapped to the lincRNAs TFBSs, whole annotations about lincRNA and TFBS are also available by internal link. PubMed external link for relevant literature is provided. DOWNLOAD page allows users to download all data we provided at present, including TFBSs and TF peaks of lincRNA promoter regions and SNPs mapped within lincRNA TFBSs and TF peaks in the TXT format. HELP page provides detailed column label description of SNP@lincTFBS. Instruction and contact information are also obtained. ## Known disease SNPs in lincRNA TFBSs The SNP@lincTFBS database was developed not only as a resource for identifying SNPs in putative TFBSs of human lincRNAs, but also as a direction for further confirmation of predicting novel disease-associated SNPs and lincRNAs. Previous studies have found that lincRNAs may tend to associated with the same diseases with the disease-associated SNPs within their TFBSs by affecting the expression of lincRNAs. We found 22 known disease-associated SNPs in lincRNAs TFBSs using PubMed to search the previous studies. For example, we found two SNPs, rs2001844 and rs6982502 in two predicted TFBSs of a lincRNA *ENSG00000253111*. These two SNPs were identified to be associated with the variation in the magnitude of statin-mediated reduction in total and LDL-cholesterol based on a genome-wide association study, thus this lincRNA might have a relationship with cholesterol- associated diseases. Further experimental validation of the role of these disease-associated SNPs in lincRNA TFBSs might provide new insights into mechanisms underlying human diseases. We also found several lincRNAs in SNP@lincTFBS that have been reported to be associated with human diseases, and these lincRNAs had SNPs within their putative TFBSs. For example, we found human lincRNAs NAG7, MEG3, PCAT1, CASC2 and LINC00032, which were involved in nasopharyngeal carcinoma, glioma and bladder cancer, prostate cancer, endometrial cancer and melanoma. We identified several SNPs in the TFBSs of these disease-associated lincRNAs. These SNPs might be potential risk SNPs for diverse diseases by regulating the expression of disease-associated lincRNAs. For example, the research on *NAG7* gene involved in human nasopharyngeal carcinoma (NPC) susceptibility can be traced to more than a decade, and previous studies have found that *NAG7* played a key role by means of both expression and interaction, it could inhibit proliferation and induce apoptosis in NPC cell but also stimulate NPC cell invasion. Soon after, *NAG7* gene was provided as a long intergenic non-protein coding RNA 312 (*LINC00312*) in HGNC (HUGO Gene Nomenclature Committee). Recently, an investigation aiming to assess the possible correlations of *LINC00312* expression with NPC progression based on microarray technology has indicated that *LINC00312* was significantly down-regulated in NPC tissues and it could represent a potential biomarker for metastasis, progression and prognosis in NPC. In the SNP@lincTFBS database, we found a SNP (rs112175570) located within the TFBS for the transcription factor NF-κB and RelA in the promoter of *LINC00312* gene (Ensembl ID: ENSG00000237697), and rs112175570 might be a potential risk SNP for nasopharyngeal carcinoma by regulating the expression of *LINC00312*. Besides cancer, we also found several neurological or psychiatric disorder associated SNP in human lincRNA TFBSs. For example, we found three SNPs (rs141600967, rs111946796, rs147394431) in the TFBSs of a lincRNA, ENSG00000214548 (also known as MEG3), ENSG00000214548 has been demonstrated to be associated with multiple human diseases, including glioma and neuroblastoma. We found three SNPs (rs2973034, rs2973034, rs78670708) in the TFBSs of a lincRNA, ENSG00000248587 (also known as GDNF-AS1), ENSG00000248587 has been demonstrated to be associated with Alzheimer disease. In addition, we found a Alzheimer's disease risk SNPs (rs6472116, p = 9.59×10⁻⁵) in a lincRNA TFBS (ENSG00000253583). Therefore, further experimental verification of this SNP might provide novel insights and lead to new treatments. Taking advantage of our database, it is possible to further investigate the mechanism of lincRNA involved in human diseases. # Discussion Accumulating studies of dysregulated lincRNA expression in diverse cancers have suggested that lincRNAs might act as potential tumor suppressor genes and novel prospective therapeutic targets in cancer treatments. SNP@lincTFBS is designed to serve as a practical resource of SNPs in the TFBSs that dysregulate the expression of human lincRNAs. The database provides available genomic informations and annotations of SNPs in the TFBSs in putative promoter regions of human lincRNAs, and also a web-based interface allowed easy access to query and download flexibly. Most human lincRNAs have TFBSs in their promoter regions and the distribution of SNPs in these TFBSs of lincRNAs is widespread. Previous studies have demonstrated that the genetic variants in the TFBSs of human lincRNA regulatory regions may change lincRNA expression, and thereby affecting the susceptibility to human diseases. Thus we developed the SNP@lincTFBS database, which is devoted to the exploration and annotation of SNPs in potential TFBSs of human lincRNAs. One of the distinctive features of SNP@lincTFBS is that all SNPs that can be mapped to human lincRNA TFBSs are identified and annotated. The other databases that are related to transcriptional regulation for lncRNAs, such ChIPBase, only collect TF-lncRNA regulatory relationships that have been identified from ChIP-Seq data. In SNP@lincTFBS, we considered not only transcription factor of lincRNAs (like ChIPBase), but also the SNPs that affect the capability of binding to the lincRNA promoter regions of each transcription factor. Our database has the potential to become an available resource for further studies of lincRNA function and complex disease. For example, we found several disease-associated SNPs and lincRNAs in SNP@lincTFBS, suggested the potential application of the SNP@lincTFBS in the field of disease-associated lincRNA variants. We found multiple SNPs in the TFBSs of cancer-associated lincRNAs, further experimental verification of these disease candidates might yield novel insights into disease pathophysiology. In addition, we also found multiple SNPs in the TFBSs of neurological or psychiatric disorder associated lincRNAs, this finding was consistent with previous studies, which revealed that lincRNAs played important roles in brain and neuropsychiatric disorders. Although the current number is limited, with the growth of interest in human lincRNAs and the availability of high-throughput technologies, the total number of disease- associated lincRNAs and SNPs will undoubtedly continue to grow, SNP@lincTFBS will become increasingly useful in future studies. In the future, we envisage the database to be available as a semantically linked interoperable data resource. We hope that SNP@lincTFBS will be a useful tool for researchers in pertinent fields, and will benefit the functional study of human lincRNAs. With the increasing availability of genome-wide transcriptome identification and functional annotation of human lincRNAs in the public domain, we would enrich the database with this information. We will update the disease- associated lincRNAs with their annotations and disease-associated SNPs mapped to the TFBSs of lincRNAs every 4 months. SNP@lincTFBS may act as an advance resource that can provide great convenience for the research on identification of disease-associated lincRNAs or risk SNPs and the discovery of responsibility for discrepant expression abundance of lincRNAs. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: LW XL. Performed the experiments: SN ZZ JY. Analyzed the data: SN PW HZ RL TW JW. Wrote the paper: SN ZZ JY XL.

# Introduction Successful treatment of neurological disorders requires effective techniques to deliver drugs to the central nervous system (CNS). Systemic delivery can be problematic because of the blood brain barrier (BBB) which prevents passive passage of the majority of large molecules from the bloodstream into the extracellular space. Convection-enhanced delivery (CED) is a technique to deliver therapeutic macromolecules directly to the CNS and bypass the BBB. In CED, a cannula (needle) is inserted directly into the parenchyma, and drug infusate is delivered at controlled flow rates into the extracellular space. With convection (advection) as the dominant transport mechanism, this technique can improve local delivery by providing larger distribution volumes than diffusion-dependent methods. CED requires the surgical insertion of a needle into brain tissue, and this raises specific concerns about minimizing penetration injury and ensuring proper targeting. Even with stereotactic positioning, backflow (leakback) along the needle track continues to be a major source of non-specific targeting, especially at high flow rates. With backflow, infusate flows back along the outer needle wall instead of penetrating into tissue at the needle tip. This backflow phenomenon is normally undesirable since large tissue distributions are not achieved, and drugs can reach regions of the brain where they are unintended, not effective, or toxic, resulting in unwarranted side effects. Indeed, improper targeting has been implicated as one of the most significant barriers to successful implementation of CED in previous clinical brain tumor trials. In these trials, high rates of ineffective delivery resulted in the majority of infusate leaking into CSF spaces. These studies employed simple open-ended cannulas that are prone to backflow. To reduce backflow, research has focused on developing new cannula designs, imaging to identify backflow in real- time, and on modeling to predict the extent of backflow. Backflow can be produced by local tissue injury, fluid-tissue interactions, or by a combination of these effects. With needle insertion, tissue disruption and injury may create a fluid-filled gap between the needle and surrounding tissue through which infusate can easily flow with little fluid resistance. Edema may also develop in response to cellular injury or vascular damage, and this abnormal accumulation of fluid can result in swelling and volumetric tissue changes. Expansion of extracellular space may introduce a high hydraulic conductivity (permeability) layer of traumatized tissue around the cannula through which infusate can easily flow. Alternatively, cellular swelling may reduce extracellular space and help to close gaps at the needle-tissue interface. Even when no fluid-filled gap is present at the needle interface, fluid pressure at the needle tip is increased with an increasing infusion flow rate, and this elevated pressure can push soft tissue away from the cannula surface to create a fluid-filled gap (intrinsic backflow). Related to this mechanism, tissue coring, defined as tissue going inside the cannula during the insertion process, has being implicated as another cause of increased backflow. Peaks in needle fluid pressure which may be required to clear the obstruction also result in a peak in fluid pressure at the needle tip potentially creating a bolus infusion and backflow. Tissue stresses which are introduced or are pre-existing may also influence the mechanics of backflow. We define pre-stress as the stress that exists around the needle after insertion and before the start of CED infusion. We assume this stress is radially-directed along the needle-tissue interface, and it is created by tissue disruption, displacement, and subsequent compaction around the needle with insertion. The introduction of this stress may reduce backflow since fluid pressure would have to overcome it to separate the tissue from the needle surface. In our previous studies, we have evaluated the effect of pre-stress on CED backflow in hydrogel tissue phantoms. The presence of pre-stress resulted in reduced backflow. However, experimental measures or mechanical analysis of this stress is currently missing for CNS tissues. In CNS tissue, local tissue swelling around the needle can affect backflow if it produces changes in stress distribution before or during infusions. Therefore, local tissue swelling in response to needle injury can result in time-dependent pre-stress and backflow outcomes. In addition, it should be noted that residual stresses already exist in brain tissues even before needle insertion. Such residual stresses may also influence backflow by altering the mechanics of tissue disruption, changing fluid-gap dimensions, or changing the magnitude of pre-stress. Studies of local CNS tissue penetration injury have been previously conducted for insertion of electrodes and needles and for ballistic tests. Blood vessel rupture and hemorrhage have been previously reported as evidence of tissue damage during electrode insertion in cortical tissues. Previous CED studies by White et al. observed regions of increased tissue injury with hemorrhage reported mainly within white matter regions, and pressure-induced tissue fracture near the needle tip. Regions of injury were also found to be coincident with backflow. While there are no needle insertion studies measuring acute tissue swelling, early tissue swelling has been detected due to other kinds of brain trauma. Few studies have focused on evaluating insertion speed on CNS tissue damage. Previous electrode studies have been conducted in which fast insertions of sharp electrodes and electrode arrays into cortical tissue were shown to cause less vascular rupture and bleeding. In our previous study, we have evaluated the influence of needle insertion speed on backflow in a hydrogel tissue phantom. Backflow was considerably reduced if the needle was inserted fast (1.8 mm/s). A pre-stress was estimated, and the existence of this compressive interfacial stress was postulated to contribute to reduced backflow. At a lower speed (0.2 mm/s), hydrogel accumulation at the needle tip produced a damage gap that resulted in backflow. Thus, rate-dependent damage had a great influence on the extent of backflow. In the CNS, we assume that surgical placement of the infusion cannula creates tissue damage and a volumetric loss of tissue. For example, as cell membranes are irreparably torn by needle penetration, the fluid content of each cell will disperse or be squeezed out into the interstitial space through which it can flow easily to other tissue regions. Also, large strains introduced by needle insertion may result in deformation that results in a permanent track observed along the needle pathway. In this study, the effect of the needle insertion speed on such tissue damage and CED backflow was evaluated *in vivo* in rat brain tissue. The caudate putamen, a gray matter region, was targeted with the needle track passing through both the cortex and external capsule, a white matter structure, before reaching target tissue. Local tissue damage due to needle coring, needle disruption/displacement, and time-dependent swelling was evaluated. The extent of tissue coring was evaluated by monitoring CED infusion pressures. Local tissue damage and time-dependent swelling in the vicinity of the needle were evaluated by histological hole measurements at two time points after needle retraction. Hole measurements were used to reconstruct the tissue penetration damage along the length of the needle track. Hole diameters were also used as an input into a mechanics model to estimate *in vivo* pre-stresses assuming brain tissue to behave as a neo-Hookean material. Finally, these damage assessments were used to understand the influence of rate-dependent injury and pre-stress on backflow. CED experiments infused visible Evans blue albumin (EBA) tracer at varying flow rates and insertion speeds. Backflow was quantified in two ways: percentage of tracer infusate outside the targeted caudate putamen and distance back along the needle track. This study provides a visual way to evaluate mechanical damage along the needle track. Also, to the best of our knowledge it is the first to quantify pre-stress in CNS tissues and study its relation with insertion rate and backflow. The results of this study may be used to better understand and improve the insertion processes of needles, electrodes, probes or other devices in the brain to reduce tissue injury and improve functional outcome. # Methodology Three series of *in vivo* rat brain experiments were conducted: (1) needle insertion experiments to measure local tissue damage along the needle track (infusion pressure measurements and hole measurements), *n* = 12 rats, (2) additional histological assessment, *n* = 3 rats, and (3) CED backflow experiments, *n* = 23 rats. ## Animal preparation and surgical procedures All experiments were performed on male Sprague-Dawley rats (300–350 g) using protocols and procedures approved by the University of Florida Institutional Animal Care and Use Committee. Anesthesia was initiated with xylazine (10 mg/kg, SQ) and isoflurane (4%) in oxygen delivered at 1.0 L/min. The head was shaven and disinfected with iodine/alcohol. Then animals were placed in a stereotaxic frame (model 900, David Kopf Instruments, Tujunga, CA), and inhalation anesthesia (1.5% in 0.5 L/min of oxygen) was delivered via a nose mask. Body temperature was maintained (∼37°C) with a heating pad during the entire procedure. The skull was exposed by a mid-sagittal incision that began between the eyes and extended caudally to the level of the ears to expose the bregma and lambda. Two holes of 2 mm diameter were drilled by hand into the skull above the caudate putamen (CPu). Dura mater was carefully taken off and any residual blood was cleaned with phosphate-buffered saline (1× PBS). The exposed surface of the brain was kept wet with PBS during the course of the experiment. The CPu was chosen as the target because it is a relatively homogeneous region composed mainly of gray matter and has been previously used for CED backflow studies,. To reach the CPu, the needle must pass through the cortex, which is mainly gray matter and the external capsule (ec) which is a white matter region. The needle was stereotactically inserted at a 5 mm depth within the CPu. Bilateral infusions were conducted on the right and the left sides (AP = 0.5, ML = ±3, DV = −5). After needle insertion, two minutes were allowed for tissue relaxation before infusions. The needle was retracted three minutes after the end of all infusions. The needle was retracted at 0.2 mm/s, and cleaned with 3% hydrogen peroxide followed by 1×PBS. In needle insertion experiments and histology experiments, brain tissues were fixed to avoid further swelling after rat death and to avoid additional distortion and damage during removal and slicing. After needle retraction, rats were immediately euthanized by perfusion fixation using 10% formalin. Brains were removed and put in 10% formalin overnight and then moved to 30% sucrose for 2 days before tissue slicing. To be consistent with previous backflow studies, brains used for CED backflow experiments were not fixed, and rats were euthanized by decapitation immediately after needle retraction. ## Needle insertion and CED system For all experiments, a 32 gauge (0.235 mm diameter) blunt tip stainless steel needle (Hamilton, Reno, NV) was used. Blunt tip needles were used because needles of this kind have been used in clinical trials, are readily commercially available, and have been used by our lab in previous CED studies. Moreover, even though new cannulas to minimize backflow have being designed, the tip geometry of many of these new designs is often a basic blunt tip. shows a schematic of the experimental set-up used for needle insertion and CED. Needle insertion (*z*-direction) was controlled using a linear stage system (model LP28T, Applied Motion Products, Watsonville, CA) mounted on a machine frame. The needle was coupled to the linear stage using an electrode holder (model 900, David Kopf Instruments, Tujunga, CA) which also provided alignment. The stereotaxic frame and rat were placed on a *x-y* table which was used to position the rat under the needle such that the drilled holes in the rat skull at target AP and ML coordinates was under the needle tip. The infusion system consisted of a syringe pump driving a 100 µL gas-tight syringe (Hamilton, Reno, NV) connected to 40 cm of minimally compliant polyetheretherketone (PEEK) tubing (1 mm inner diameter and 1.58 mm outer diameter). This infusion line was coupled to the needle via a reducing union (Valco Instruments, Houston, TX). ## Macromolecular tracer Evans blue labeled albumin (EBA) was used as the visible macromolecular tracer. EBA was prepared with 20 mg/ml of bovine serum albumin (Sigma-Aldrich, St. Louis, MO) and 1.25 mg/ml of Evans blue (Sigma-Aldrich, St. Louis, MO) in 1× PBS. The concentration of Evans blue was low enough that Evans blue was completely bound to the albumin. ## Needle insertion experiment Experiments were conducted at three insertion speeds: 0.2, 2, and 10 mm/s. Four rats were tested at each insertion speed (12 total rats). The first two speed levels were chosen because they are in the range of a previous electrode insertion study and similar to the speed levels of our previous hydrogel CED study. The highest insertion speed was chosen based on a previous microelectrode study. 4 µL of tracer was infused at 2 µL/min to reproduce conditions of backflow and to provide easier identification for hole measurements in fixed tissue sections. To evaluate changes in our damage parameter, hole diameter, due to acute injury swelling, tissues were fixed at two different time points after needle retraction, 10 and 25 min. Bilateral insertion experiments were conducted on each rat. The second retraction was 15 min after the first one. Then, the rat was immediately euthanized by perfusion fixation using 10% formalin which took ∼10 min. ### Infusion pressure measurement In-line infusion pressure provided a separate measure of needle-tissue interaction. In previous studies, spikes in infusion pressures have indicated the occurrence of tissue coring within the needle tip. In-line pressure measurements were performed for all needle insertion experiments to monitor pressure changes during insertion and CED at 2 µL/min. Pressures were measured using a custom-designed fiber optic pressure transducer (model FOP-MIV-NS663, FISO Technologies, Quebec, CA) that was connected in-line to the infusion system between the syringe and the needle. Pressure was monitored at 1 Hz during the insertion, infusion, and retraction process. Prior to each brain infusion, a reference pressure was recorded outside the brain with the infusion pump running at the test infusion rate and with the needle at the same stereotaxic level as within the rat brain during CED. This value was a measure of the pressure drop within the infusion line and was subtracted from the *in vivo* pressure measured during infusion in brain to determine pressure at the needle tip. ### Hole measurements along the needle track Fixed brains were sectioned into 100 µm slices in the horizontal plane on a cryostat (∼−20°C). Approximately 45 slices were obtained along each needle track. Slices were put in 1× PBS at room temperature for 30 min and then mounted on microscope slides for imaging. Needle holes on each slice were imaged on a microscope (model IX-71, Olympus America Inc., Center Valley, PA) and a digital camera (model SPOT RT3, Diagnostic Instruments, Inc., Sterling Heights, MI) using bright light. Color RGB images of the hole were used to measure the area and the perimeter using a MATLAB script. Images were converted to gray scale and the hole was detected by considering only pixels with intensity higher than a threshold value determined from surrounding tissue in each slice using Otsu's method. Since the holes were not necessarily circular, the measured areas were used to calculate the equivalent diameter, which was defined as the diameter of a circle with the same hole area. The aspect ratio, defined as the ratio between the equivalent perimeter (calculated using the equivalent diameter) and the measured perimeter, was also used to evaluate the shape of the hole. The aspect ratio is 1 for a circular hole, and the aspect ratio goes to zero as the shape becomes more irregular. In some slices, especially those close to the brain surface, the hole was difficult to measure because of excessive bleeding produced during dura mater removal or because the hole was closed. These holes were not included in the data analysis. ## Additional histological assessment In an additional test group (*n* = 3 rats), needles were inserted at the three insertion speeds and 4 µl of albumin in solution with 1×PBS (20 mg/ml) was infused at 2 µL/min. Evans blue was not infused to allow H&E staining. Fixed brain tissue was sliced into 50 µm thick slices in horizontal planes and standard procedures for H&E staining were followed. ## Pre-stress estimation Hole diameters were input into a tissue mechanics model to estimate pre-stresses along the length of the needle track. The model assumed that during insertion, tissue was radially displaced and conformed around the inserted needle. This is based our assumption that frictional drag of tissue in the z-direction is small along the main length of the needle and that shear stresses become negligible as the tissue equilibrates around the needle before the start of CED. Also, a plane strain condition was assumed because of the relatively large length of the needle with respect to the diameter. In this way, changes in hole diameters can be related to corresponding radially directed tissue stresses at the needle surface. Axial symmetry and a semi-infinite medium were considered because of the relatively small diameter of the needle compared with the surrounding tissue. Tissue stress was determined at equilibrium after needle insertion; therefore viscoelastic effects were not considered. Pre-stress was calculated at the most immediate tissue fixation time point (10 min) when swelling was minimized. Since swelling and pre-existing residual stresses can change the hole size, the approach to estimating pre-stress considers the net effect of pre- existent stresses and stress generated during needle insertion. Pre-stress was calculated as the compression load on the surface of the hole necessary to re-expand the equivalent hole diameter to the diameter of the inserted needle (*a*) assuming the solid phase of tissue behaves as a neo- Hookean material. By assuming incompressibility of the tissue, the Cauchy stress tensor was calculated using where *s* is a material constant, is the gradient deformation tensor, and is the Kronecker delta tensor. The simplified energy function waswhere *μ* is the shear modulus. By considering plane strain conditions, stretch ratios were defined as The constant *s* was calculated far away from the needle where stress and displacements go to zero, and the stretch ratio was equal to unity. From, *s* was equal to *μ* After calculating the derivative of *W*, the stress tensor becomes The pre-stress is the normal stress in the radial direction Pre-stress was calculated using equilibrium shear moduli values of 485 Pa for the cortex and 238 Pa for the corpus callosum using values reported by Elkin et al.. In this case, the external capsule was assumed to have similar modulus to the corpus callosum which is a predominately white matter region. This white matter shear modulus was assigned at two points, 2.8 and 2.9 mm depths, corresponding to the ∼0.2 mm thick layer of external capsule according to a rat brain atlas. All other points were calculated with the modulus of cortex gray matter. ## Backflow experiments A separate set of experiments was conducted to measure backflow at each insertion speed (0.2, 2, and 10 mm/s). 4 µL of EBA was injected at three flow rates (0.5, 1, and 2 µL/min). Five experiments were performed for each experimental condition for a total of 45 experiments (23 rats with bilateral injections). Following CED experiments, rats were immediately euthanized and freshly excised brains were sectioned into 100 µm slices on the cryostat in the coronal plane. Slices within infused brain regions were mounted on microscope slides and placed on a white surface. White light was projected on each slice, then slices were imaged using a CMOS camera and saved as a RGB file for analysis. Backflow was quantified by using two different measures: (1) the percentage of tracer infusate outside of the targeted CPu region and (2) the distance from the needle tip to the point of maximum tracer penetration back along the needle track. ### Backflow as percentage of tracer infusate outside the target Previous CED studies have calculated the infused volume using a threshold intensity value and assuming constant concentration in regions with intensity above this threshold. This approach can introduce some error since the concentration of infusate is not necessarily constant throughout infused brain tissues. To avoid this, slice images were converted into concentration maps of Evans blue, which was the visible component of the tracer. The distributed tracer regions were used to calculate the mass of tracer outside the target CPu. To obtain a gray scale image of each slice and calculate a pixel intensity, the red value (*R*) at every pixel was normalized by the total pixel intensity, where *G* is the green component and *B* is the blue component of the white light. Images were analyzed using a Matlab subroutine, and pixel intensity was converted to dye concentration using a modified Beer's law where *L* is the path length of light inside the specimen, is the Evans blue concentration, is the absorption or extinction coefficient, and *I_o* is the incident light energy. Beer's law was modified to account for non-transparent tissue such that *M* is an introduced factor that accounts for the increase in path length of photons that are detected but take a non-straight path, and *N* is a factor which accounts for the scattering of light. The modified law can then be written as where, and is the thickness of the slice, because the light was assumed to cross the specimen then reflect back to the camera. The constant is thus the intensity of the background and was obtained by applying to a reference tissue region without any tracer. Next, the mass of Evans blue in a given tissue region was determined in terms of image intensity using mass conservationwhere, *A*, and are volume, differential of volume, area, differential of area, and number of slices, respectively. This relation assumes no capillary clearance or metabolism of EBA over the course of the experiment. To obtain the constant, the total mass of Evans blue introduced into the brain by CED ( 1.25 mg/mm³×4 µL = 5 mg) was compared with intensity data from all slices within the infused region. Once the constants and were obtained, and were used to estimate EB concentration and mass accumulated in tissues. An example of a tissue slice image and corresponding Evans blue concentration map within the infused region are shown in. CPu and white matter regions where easily delineated on each slice as shown in. It can be seen that concentration gradients existed inside the CPu and concentrations were not constant. The backflow percentage (*BF*) was calculated aswhere is the mass of Evans blue inside the target CPu. The region with EBA inside the CPu was isolated on each slice using ImageJ software (National Institutes of Health). The mass of Evans blue outside the CPu was not calculated directly from the images because blood was frequently observed outside the CPu, especially in white matter regions. ### Backflow distances Backflow distances from the needle tip to the point of maximum tracer penetration back along the needle track were also determined. For each infusion site, a slice containing the needle track or as close as possible to it was used. Because the position of the needle tip was difficult to identify in the brain slices, backflow distances were calculated as the difference between the total insertion depth (5 mm) and the distance (*d*) from the point of maximum dye penetration along the needle track to the surface of the brain, see. ## Statistical analysis Analysis of variance (ANOVA) was used in the analysis of experimental data to test whether or not the means of several groups are all equal. One-way ANOVA was used to examine the influence of one factor on the experimental data and two-way ANOVA was used to evaluate the effect of two or more factors. The results of ANOVA can be considered reliable as long as the data are normally distributed. If normal distribution could not be guaranteed, Kruskal-Wallis test was used. Average infusion peak pressure and steady state pressure was compared for the three insertion speeds by using one-way ANOVA. The total average hole areas, diameters, and pre-stresses, were compared for the three insertion speeds and the two time points by using the Kruskal-Wallis test because data was not normally distributed. Also, the Kruskal-Wallis test was used to compare averages areas, diameters, and pre-stresses among regions in the brain along the needle track (cortex, external capsule, and CPu). Average hole area for each point along the insertion track was compared for the three insertion speeds and for the two time points by using two-way ANOVA. At some depths along the needle track, the hole was difficult to see in images and was not measurable. Therefore statistical analysis was performed only for those depths where the complete data set which consisted of four repetitions was obtained. Average backflow percentage and backflow distances were compared for the three insertion speeds and the three flow rates by using two-way ANOVA. Data are presented as mean ± 1 standard deviation. All p-values of \<0.05 were considered significant. # Results ## Damage assessment ### Infusion pressure Small variations in infusion pressure were observed during needle insertion. Once infusion started, pressure increased to a peak value between 14 and 25 kPa and then decreased to an approximately steady-state value between 2.7 and 7.4 kPa. The peak pressure may be indicative of tissue coring or some other initial tissue resistance which the infusate encounters. After infusion was stopped, infusion pressure decreased to a value just above zero and then went to zero after the needle was fully retracted. For infusions performed with needle insertion speeds of 2 and 10 mm/s, a second small peak (*P_B*) between 0.2 and 2.46 kPa greater than the local minimum value (*P_A*) was also observed. shows the average values of peak pressure, steady state pressure, and the second peak of pressure (*P_B*) for the different insertion rates. The difference between average peak and steady state values was not significant with respect to insertion speed (p-value = 0.62 for peak pressure; p-value = 0.29 for steady state pressure). That peak pressure did not depend on insertion speed indicated that tissue coring or tissue resistance due to tissue separation was not insertion rate dependent. The difference in the second peak between insertion speed 2 and 10 mm/s was not significant (p-value = 0.6). Magnitude of the second peak of pressure was given by the difference. The average of that difference was 0.54 and 0.84 kPa for 2 and 10 mm/s insertion speeds, respectively. The averages of these magnitudes were also not significantly different (p-value = 0.39). ### Hole measurements In the majority of tissue slices, the hole left by the needle appeared as empty space, but in other slices (∼5% of slices), especially those close to the surface on the cortex, the hole appeared as a narrow crack surrounded by EBA. In other cases (∼20% of slices), only a region with accumulation of red blood cells was observed where it was difficult to distinguish the actual track of the needle. These corresponded mainly to locations close to the surface of the brain at insertion depths shallower than 1.4 mm. Data from slices like those shown in were not considered in any data analysis. Therefore, ∼36 hole measurements per needle track were considered. Measured hole areas were averaged at each spatial location along the needle track for each insertion speed and fixation time point. Hole area varied spatially with insertion depth. Track holes increased in size with penetration in the cortex. Between 2 and 3.5 mm depth, hole size reached a maximum, and decreased again for higher depths. The region of maximum hole area corresponded to the white matter region (external capsule). For two experiments performed at 10 mm/s insertion speed, the area of the hole within white matter regions was practically the same as the cross area of the needle (0.0434 mm²). For the rest of the measured slices, the area of the hole was always considerably smaller than the area of the needle. shows the aspect ratios of the holes along the needle track for the 0.2 mm/s insertion speed. Holes were more circular (higher average value and lower standard deviation) between depths of 2 and 3 mm. Holes were found to be less circular at the CPu target site and near the surface of the brain. The aspect ratio profiles for other insertion speeds were found to be similar and are not shown. Areas of all the holes along the needle track for each insertion speed and time point were averaged and compared by using the Kruskal-Wallis test. As shown in, the average area increased for increasing insertion speed, see. Also, average hole area was larger for the earlier fixation, 10 min time point. The total average area for the 10 min fixation time point (0.01624±0.010 mm²) and for the 25 min time point (0.0139±0.009 mm²) was also significantly different. Changes in hole area were more sensitive to changes in insertion speed than to fixation time point; the average area for 10 mm/s was 1.87-fold and 1.32-fold the average area for 0.2 mm/s and for 2 mm/s respectively, while the average area for 10 min before fixation was up to 1.17-fold the average area for 25 min after fixation. Analysis of variance was performed to compare the area of the holes at each point along the needle track, see. With respect to insertion speed, difference was significant for the majority of points between 2 and 4 mm depths. Here, increased insertion speed produced increasing hole areas except within a narrow region at ∼3 mm depth which likely corresponds to the white matter region. With respect to the fixation time point, hole differences were significant for 12.5% of depth locations. shows average hole areas for the three main brain regions along the needle track. Average values for the cortex (0.0133±0.011 mm²), external capsule (0.0218±0.009 mm²) and CPu (0.016±0.008 mm²) were significantly different from each other. The maximum difference in average area was between the external capsule and cortex where average area of the external capsule was 1.64-fold the average area in the cortex. This ratio is smaller than for the ratio between 10 mm/s and 0.2 mm/s insertion speeds. Therefore, tissue damage was more sensitive to insertion speed than brain region over the range of parameters tested. Average hole diameters and perimeters presented similar trends as the hole areas. These results are summarized in. ### Additional histology Several rupture mechanisms were observed in brain tissue slices. Tissue fracture was defined as a crack or narrow opening with bleeding in interior regions of the tissue slice. The presence of blood indicated that fracturing occurred *in vivo* during needle insertion or retraction and not during tissue processing, e.g. slicing and mounting. shows tissue fracture and tissue stained with blood close to the needle hole. Generally, this injury was observed mainly in white matter regions of the external capsule and in the CPu. More intensive bleeding and tissue fracturing was also often observed at higher speeds, 2 and 10 mm/s. Tissue fractures between 20 and 150 µm in length were found. This type of damage was observed in approximately 35% of the needle insertion experiments conducted at 2 or 10 mm/s. At the slower insertion speed, bleeding was only observed close to the hole, and tissue fracturing was not observed. Individual cell separation from the bulk of tissue was also frequently observed at the surface of holes. Narrow strips of tissue were also separated from the tissue surface (arrow). Layers of cells or fibers were occasionally observed to be curved at the hole boundary, especially at slower insertion speeds of 0.2 mm/s. This provided some evidence of radial tissue compaction. ## Pre-stress Pre-stress was calculated from equivalent hole diameters at the earliest fixation time point for each of the three insertion speeds, see. Obtained pre- stress values were between 0 and 485 Pa. Pre-stress in white matter was approximately zero at five locations corresponding to locations where the area of the measured holes were practically the same as the needle. Average pre- stress values along the entire needle track are presented in. The average pre- stress for 0.2 mm/s was found to be significantly higher than pre-stress averages at the other two insertion speeds. There was no significant difference between pre-stress averages at 2 and 10 mm/s insertions (p-value = 0.16). Pre-stress was not constant along the needle track. Maximum values were found in the cortex close to the surface, minimum values were found in the external capsule white matter region, and pre-stress increased with greater depths beyond the external capsule. Average pre-stress values for cortex, external capsule, and CPu regions are shown in. The average pre-stress value for the external capsule was significantly different from the other two tissue regions. There was no significant difference between average pre-stress in the cortex and CPu (p-value = 0.5). ## CED backflow distributions CED distributions in brain for varying needle insertion speeds are shown in. Tracer distributions in tissue were found to be relatively uniform over regions of convective transport with evidence of diffusion at the peripheral front. Backflow was observed as leakage of infusate into white matter regions of the external capsule and corpus callosum (cc) outside of the targeted CPu. Infusate was not observed to flow back into the cortex. In general, backflow increased with increasing insertion speed. In these experiments, evidence of bleeding was frequently found within the external capsule, especially for insertions at 10 mm/s. Leakage of infusate into the lateral stripe of the striatum (Lss), as shown at the bottom of, was also frequently found (∼65% of cases) independent of flow rate or insertion speed. Leakage to the lateral ventricles was observed in two cases performed at an insertion speed of 10 mm/s. No evidence of tracer transport directly between Lss and external capsule regions was observed in any of the slices. For that reason, tracer distributions in the Lss were not considered in backflow calculations. Average backflow percentages for varying insertion speed and infusion rate are shown in. Significant influence of the flow rate and the insertion speed was found. Average backflow percentages for the three insertion speeds were 16.7, 30.8, and 36.6% for flow rates of 0.5, 1, and 2 µL/min, respectively. Average backflow at 0.5 µL/min was significantly different from the average backflow at the other two infusion rates. Average backflow percentage was also found to increase for increasing insertion speed, see. The average percentage for the three flow rates were all significantly different at the three insertion speeds. Comparing backflow percentages obtained at 10 mm/s with respect to 0.2 mm/s insertion speeds, backflow increased by 2.46, 2.27, and 2.26-fold at 0.5, 1, and 2 µL/min flow rates, respectively. Since backflow was diverted away from the needle track into white matter regions, backflow distances along the needle track were relatively constant over the insertion speeds and flow rates tested. No significant differences between flow rates or insertion speeds for backflow distance was found (p-value = 0.52 for insertion speeds, and p-value = 0.65 for flow rates). # Discussion In this study, brain tissue damage and backflow during CED was evaluated for varying needle insertion speeds. Histological assessment and hole measurements were used to characterize local tissue damage along needle insertion tracks. This unique approach to characterizing tissue disruption was also used to estimate tissue stresses introduced by needle insertion (pre-stress). To the best of our knowledge, this is the first study to estimate pre-stress around an implant in brain tissue, and this information can be used to improve needle implantation surgical procedures or to improve prediction of backflow. Tissue damage, evaluated as the size of the hole left by the needle after retraction, bleeding, and tissue fracturing, was found to increase for increasing insertion speeds and was higher within white matter regions. A statistically significant difference in hole areas with respect to insertion speed was found. While there are no previous needle insertion speed studies with which to directly compare, previous electrode insertion studies have noted greater brain surface dimpling and insertion forces with increasing insertion speed. These higher deformation and force measures may indicate greater brain tissue damage which is in agreement with the present study. There are also studies which have found that fast insertion of sharp tip electrodes produced less blood vessel rupture and bleeding. These differences in rate dependent damage may be due to differences in tip geometry (diameter and tip) or tissue region, since these electrode studies focus mainly on the cortex. In the present study, hole measurements were small in the cortex, and no substantial bleeding was observed in the cortex except when it was produced during dura mater removal. Any hemorrhage was observed primarily in white matter regions of the external capsule and the CPu. ## Hole measurements The hole left in tissue by the needle provided a visual way to reconstruct tissue damage along the needle track. Holes area were relatively small in the cortex, maximum within the white matter region and decreased again in the CPu. The decrease in hole area in the CPu indicated that damage did not accumulate with depth, and that there was negligible tissue accumulation around the needle tip. Hole measurements were more circular in shape within inner cortex and white matter regions and more irregular close to the brain surface and within the CPu. The irregular shape of the hole may have resulted in non-uniform pre-stresses which may have increased the likelihood of backflow in less compacted regions at the needle periphery, i.e., regions of low pre-stress. The variation of the hole area along the needle track is likely because of differences in tissue composition, mechanical properties (modulus, friction, and rupture strength), as well as pre-existing residual stresses in the brain. In a previous study by Xu et al., white matter corpus collosum was reported to be under tensile residual stress that was up to 1.2 kPa. The larger diameter holes observed in the white matter regions of this study may be due to in part to this tensile residual stress. Local tissue damage was also assessed. More intensive bleeding around the needle was found in white matter regions which is in agreement with hemorrhage patterns seen by White et al. in their CED studies into the rat CPu. This indicates that vasogenic edema is likely. Also, widespread damage to both ECM and cells is possible. A corresponding observation was tissue fracturing observed primarily within white matter regions and the CPu. This is in partial agreement with the study by White et al. where they also reported tissue fracturing but only close to the needle tip (CPu) but not within white matter regions. Greater injury in white matter may be produced by higher rupture stresses as suggested by previous probe insertion force measurements where significant increases in insertion force where found along white matter tracks. Larger deformations associated with greater tissue strength may be the cause of more damage in these regions. With respect to the two fixation time points evaluated, significant difference in the hole size was found. Swelling, inflammation or other injury processes may be increasing tissue volume after initial injury resulting in the measured decreases in hole areas. These results are in agreement with previous studies where brain tissue swelling and increases in intracranial pressure were detected at early time points after injury. Therefore, hole measurements include the combined effects of tissue swelling, tissue tearing due to needle insertion, and residual stresses. Reported pre-stress values therefore account for swelling at these specific fixation time points and can be applied to tissue-needle interactions at these time points. Volumetric increases in tissue with swelling will increase pre-stresses with time. If these changes are large enough, they may reduce backflow. However, previous CED studies have not found additional wait time before infusion to be an important influence in preventing backflow. It should be noted that hole damage was found to be less sensitive to changes in fixation time point than to changes in insertion speed. ## Tissue coring If tissue coring occurs during needle insertion, pressure inside the needle increases at the beginning of the infusion in order to expel cored tissue and clear the needle. Infusion pressures were used to determine if higher insertion speeds produce greater coring as another cause of greater backflow; however, pressure measurements showed no significant difference in the peak expulsion pressures. This indicated that tissue coring was not the main cause of insertion rate dependent changes in backflow. The peak and steady state infusion pressures were greater than or similar to those reported in previous studies. Increased pressures may be due to the use of smaller cannulas which require greater pressure build up than larger cannulas to expel the same volume of tissue. Interestingly, incidence of backflow increased with the appearance of a second small peak in pressure observed at 2 and 10 mm/s insertion speeds. This small peak may be due to dynamic fluid-tissue interactions along the wall or with any cored tissue. Occurrence of a second peak may be a better indicator of backflow than having a high peak pressure value. Decreases in pressure over time may be due to viscoelastic properties of tissue. Slow separation of tissue from the needle (increasing surface area) may also explain decreasing infusion pressure. ## Pre-stress Volumetric tissue changes due to cellular and tissue disruption, tissue swelling and residual stresses ultimately influences internal tissue stresses surrounding the implanted needle. Improved understanding of theses stresses is needed for improved understanding of tissue-infusate interactions that affect backflow. This study provided measure of changes in pre-stress with changes in insertion speed and brain region. Calculated pre-stress was found to be significantly smaller for high insertion speeds because of the greater tissue damage produced at these speeds. Also, pre-stress was determined to be non-uniform along the needle track. It was smallest within the white matter region of the external capsule and it was predicted to be always present and compressive within the cortex and targeted CPu. Thus, the idealized model predicted no gaps between the needle and the tissue within the cortex and CPu. Predicted pre-stress values were low (\<500 Pa) in part due to the low shear modulus values of brain tissue and the large extent of tissue disruption introduced by the needle. These low pre-stress values explain the sensitivity of CED to backflow since there is less of a ‘seal’ at the tissue-needle interface. Consistent with this, calculated pre-stresses were smaller than measured infusion pressures. Applied to CED outcomes, regional variation in pre-stress also supports the observed backflow patterns where infusate moved back up the needle track in the CPu and diverted mainly into adjacent white matter where less of a ‘seal’ exists. Therefore, pre- stress in the CPu region is the most relevant value when predicting backflow because this is the region where infusion and backflow is initiated. Reported pre-stress was dependent on the mechanics model chosen and the mechanical properties of brain tissue used. These factors affect the magnitude of the reported pre-stress value. There exist some techniques to directly estimate stress at the tissue-needle interface using probes or microballoons as reported in previous studies. However those devices are too large to be used in the rat brain. Moreover, with these techniques only one average value for stress is usually obtained, and information about regional variation along the needle track is missing. Therefore, the combined experiment and modeling approach used in this study establishes a useful way of estimating pre-stresses variation surrounding needles and other brain implants. ## Sources of error Friction and adhesion which were not considered in this study are potentially important parameters for understanding mechanics between the needle and tissue during needle insertion. Further studies measuring insertion forces are needed to estimate these effects. While fixation was introduced to stop tissue swelling, fixation may also produce volumetric tissue contraction which may be a source of hole measurement and pre- stress error. Hole contraction due to the combined effect of fixation and swelling will result in pre-stress overestimation. We estimated the net effect of these sources of error by incorporating volumetric tissue changes of 8% measured in previous fixation studies and our own swelling analysis assuming uniform contraction in the radial direction. The effect of tissue swelling on hole measures was estimated by linear extrapolation of average hole area at 10 and 25 min time points to a zero time point. This adjusted hole area was resulted in a net decrease in pre-stress of approximately 5% which is small. Moreover, because our tissue analysis is a post-mortem technique, hole measures may be sensitive to tissue preparation, e.g. slicing and mounting. However, it is expected that these errors were uniformly spread along the needle track. ## CED Backflow In all CED experiments, backflow was observed. It was smallest for low flow rate and low insertion speed and significantly increased with higher flow rate and insertion speeds. To induce backflow, tracer infusate likely had to overcome pre-stresses in the CPu to separate tissue and form gap at the needle interface. Once backflow reached the white matter of the external capsule or corpus callosum, it was found to accumulate into these regions. The combination of low pre-stresses and the higher hydraulic conductivity of the white matter with respect to gray matter likely contributed to this diversion of infusate into white matter. There did not seem to be sufficient pressure to open up the tissue gap in the cortex region since no tracers were observed to reach the cortex or the surface of the brain during infusion. With increasing insertion speed, there was less delivery of EBA to the targeted CPu and greater accumulation in the external capsule and corpus callosum. This reflected greater backflow measured as percentage of the total tracer infused. While backflow distances were not significantly different this was to be expected since tracer was diverted from the needle track to the white matter region. As observed in this study, backflow distances were found to be dependent on the needle pathway and the anatomy of white matter intersection. For this needle pathway, backflow percentage was a better measure of the extent of backflow. With respect to flow rate, backflow increase with increasing flow rate which was in agreement with numerous previous studies. With respect to insertion speed, the results of the present study are opposite to that of our previous hydrogel study where material damage and backflow increased for a slow needle insertion speed. Since similar experimental parameters (test system, insertion rates, needle tip) were used, the different outcome is likely due to differences in the mechanical behavior of the materials. At the slow insertion speed in hydrogel, an accumulation of material was observed to form at the needle tip which created a larger zone of damage. In contrast, in brain tissue, no tissue accumulation was noted at the tip based on the tissue damage patterns. In both materials there was still correspondence between pre-stress and backflow. The main difference between both materials was the different failure behavior that resulted in different pre-stress states. Brain tissue has been found to fail after high plastic deformation with a fibrous like fracture. On the other hand, hydrogel fracture occurs with low plastic deformation. ## Clinical Implications Backflow has been a persistent problem for CED especially at high flow rates required for clinical applications due to a lack of infusate control. In clinical and primate studies, larger cannulas and designs with step changes in diameter have been used. These needles share the same blunt tip used in our studies which show tissue damage along the needle track to be an important factor affecting backflow. We expect similar dependence of local tissue damage on insertion rate and tissue region. Slow insertion rates that minimize damage should improve targeting. Pre-stresses likely increase with cannula diameter, and increases in tissue compression may partially explain success of step designs in reducing backflow. This effect may also reduce backflow using polymer coated cannulas that swell *in situ*. There are many sources for discrepencies between CED studies (cannula design, cannula alignment, fluidic components, and infusion rates, as well as animal model and infusion site). Tissue rupture and failure initiated at the needle tip is highly dependent on tip geometry, and previous studies have shown sharp or rounded needle tips may reduce the extent of damage and backflow along the needle track. Tissue coring also contributes to pressure build up at the needle tip. Designs that reduce these effects, e.g. side port needles or hollow fiber catheters, will also reduce backflow. Further studies looking at the effects of needle tip geometry on tissue damage and stress in the vicinity of the needle tip are needed to better understand infusate-tissue interactions that contribute to backflow. # Conclusions In this study, the effect of needle insertion speed on local tissue injury and backflow was evaluated *in vivo* along a track into the caudate putamen. Fast insertion speeds were found to produce more injury including tissue bleeding and disruption, lower pre-stresses, and greater backflow. New pre-stresses values provide important mechanics information at the tissue-needle interface needed to study fluid-solid interactions and may be used in the future as a predictor of backflow. In addition, patterns of tissue disruption and pre-stress were reported to vary along the length of needle track with difference found between white and gray matter regions. Thus, insertion speed dependent damage and changes in pre-stress were found to directly contribute to the extent and patterns of backflow measured. Overall, slower insertion rates resulting in less damage and improved targeting. Future studies should focus on cellular level injury along needle tracks as well as the influence of different CED parameters, e.g. needle tip geometry and needle pathways, on damage and backflow. We thank Dr. Svetlana Kantorovich, Tatiana Nobrega, K.N. Magdoom and Francisco Delgado for providing valuable discussions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FC PC MS. Performed the experiments: FC. Analyzed the data: FC PC MS. Contributed reagents/materials/analysis tools: PC MS. Wrote the paper: FC MS.

# Introduction Accelerometers are increasingly used for studying daily physical activity. A common technique to process accelerometer data is the so called ‘cut-points’ approach. This approach allows calculation of time spent with the acceleration registered by the accelerometer between certain thresholds to define physical activity intensity levels (sedentary, light, moderate, vigorous), at different bouts duration. The threshold values used in this approach are calibrated relative to an indirect calorimetry derived metabolic equivalent (MET) level which, in brief, is a proxy for energy expenditure relative to rest. Results from the cut-points approach are easy to report and reproduce. However, this approach comes with several challenges. Firstly, a known challenge is the complex relationships of acceleration with energy expenditure, activity types, study populations, and study designs, make that cut-points easily overfit to the experimental conditions under which they are derived. Secondly, the approach involves many parameters, such as bout length, that are often chosen without a clear exercise physiological motivation. Thirdly, the cut-points approach leads to collinearity between classes which partly result from the compositional nature of the data and partly from causal relations between behaviors. This collinearity complicates the study of interactions between behavioral categories. The cut-point approach traditionally used the magnitude of acceleration as its input. The orientation of the accelerometer under static conditions has also emerged as an additional informative metric to detect human posture, more recently for the detection of sleep, and in the Sedentary Sphere method for detecting sitting behavior and visualizing the data. Further, the data can also be explored using automated methods such as machine learning. Machine learning methods that use labelled data, referred to as supervised machine learning, have previously been used for activity type classification and energy expenditure estimation. Although such methods have shown potential for physical activity intensity assessment, they have disadvantages similar to the cut-points approach in that the trained classifier may overfit to the specific experimental conditions under which it was trained. Unsupervised machine learning on the other hand has received less attention in relation to physical activity intensity assessment. These methods are data-driven, allow identification of the characteristic states in the data, and can be applied to free-living data directly. Note that they are called states rather than categories, because they are defined by a Markov model rather than by absolute thresholds. As a result, they do not require time consuming and expensive calibration studies including a year of work to plan and conduct the study, they do not require costs related to exercise laboratory usage, and they may avoid arbitrary decisions in the design of the cut-point approach. We hypothesize that if such a data-driven approach to segment the data is only provided with input data that has a known physiological meaning like the magnitude of acceleration, it may be possible to learn physiologically meaningful segments in the data. If successful, this would overcome limitations of the methods requiring population sample calibration, such as the cut-point approach. Our primary aim is to implement HSMM to identify states characterized by the intensity of the activity undertaken. There is currently no gold standard method for categorising activity intensity. We therefore assess the comparability of our new approach with the traditional cut-points approach, by looking at collinearity between time spent across different activity intensity states and between cut-point categories. Large collinearity between intensity states or categories complicates modeling behavioral interactions. We will investigate this collinearity with correlation analysis and principal component analysis. Further, we examined the plausibility of the relation between resulting HSMM-defined activity intensity states, cut-points categories and time-use diary records. A supplementary aim was to assess the influence of adding accelerometer orientation metrics. In this study we used data from 500 fourteen-year-old participants. # Methods ## Participants Members of the Millennium Cohort Study (MCS), a longitudinal study managed by the Centre for Longitudinal Studies at the UCL Institute of Education participated. The MCS follows over 19,000 young people in the UK born in 2000/1, conducting home interviews every 2 to 4 years. Its sixth survey, at age 14, included the collection of physical activity data using wrist-worn activity monitors. The study obtained ethical approval from the National Research Ethics Service (NRES) Research Ethics Committee (REC)–London Central (ref: 13/LO/1786). Written consent was first required from a parent/guardian, and verbal consent from the cohort member. A random subset of 500 participants was selected for the present analysis. Millennium Cohort Sixth Sweep data, protocols and metadata are available to the scientific community, under DOI 10.5255/UKDA-SN-8156-3. ## Study protocol Interviewers placed tri-axial accelerometers (GENEActiv) with respondents during home visits and requested them to wear the device on their non-dominant wrist for two complete days; one during the week and one at the weekend, randomly selected at time of placement. Each day lasted for 24 hours: from 4 am in the morning to 4am the following morning. Participants received text messages reminding them to complete the tasks on the selected days. Upon completion of the second day of activity data collection respondents were required to return the accelerometer in a pre-paid envelope. The accelerometer sample frequency was set at 40 Hertz and the dynamic range was ±8*g*. The orientation of the acceleration axes, seen from the anatomical position, is as follows: the x-axis points in medio-lateral direction (direction of thumb), the y-axis in longitudinal direction (direction of middle finger), and the z-axis in dorsal- ventral direction (perpendicular to skin). For further details on the study design and protocol see. Additionally, participants were asked to record their categories of behavior in a time use diary. Participants were asked to provide a full record of what they did on the two days (activities), from 4am to 4am the next day, as well as where they were, who they were with, and how much they liked each activity—using pre-coded lists. Participants were offered the choice between an app and online version (with a paper version available for those unable or unwilling to use one of the other two). ## Accelerometer data pre-processing The raw data from the accelerometer is processed with R package GGIR which extracts the two days on which the accelerometer was supposed to be worn (days defined from 4am to 4am). Next, it estimates calibration error based on static periods in the data and corrected if necessary, and estimates accelerometer non wear time using a previously reported algorithm. The main metrics extracted from the data are five second average of the *vector magnitude of body acceleration* and the *orientation of each of the three acceleration sensors relative to the horizontal plane*. The vector magnitude of acceleration is calculated as Euclidian Norm Minus One (ENMO), which in formula corresponds to $max\left\{ {\sqrt{{{acc}_{x}}^{2} + {{acc}_{y}}^{2} + {{acc}_{z}}^{2}} - 1,0} \right\}$, with acc_x, acc_y, and acc_z referring to the three orthogonal acceleration axes pointing in the lateral, distal, and ventral directions, respectively. From here on, we will refer to the ENMO value as simply *acceleration*. The orientation angles of the acceleration axis relative to the horizontal plane are calculated as ${angle}_{z} = \left( {\text{tan}^{- 1}\frac{{acc}_{z}}{\sqrt{{{acc}_{x}}^{2} + {{acc}_{y}}^{2}}}} \right) \bullet \frac{180}{\pi},\mspace{720mu}{angle}_{y} = \left( {\text{tan}^{- 1}\frac{{acc}_{y}}{\sqrt{{{acc}_{x}}^{2} + {{acc}_{z}}^{2}}}} \right) \bullet \frac{180}{\pi}$, $and\mspace{720mu}{angle}_{x} = \left( {\text{tan}^{- 1}\frac{{acc}_{x}}{\sqrt{{{acc}_{y}}^{2} + {{acc}_{z}}^{2}}}} \right) \bullet \frac{180}{\pi}$. For all continuous time periods with no z-angle change of more than 5 degrees lasting at least 5 minutes, the acceleration values were set to zero to take out the possible influence of increased calibration error during sustained inactivity periods, e.g. as a result of temperature. To account for variation in sign of the signal as a result of wearing the accelerometer upside down, the angles were corrected as follows: If the value for the *x-angle* has a positive median during all time periods detected as active (calculation described in the next section) then the device is considered to be worn incorrectly, in that case, the *x-angle* and *y-angle* (flipped around zero) are negated to mirror the orientation. ## Conventional cut-points approach The acceleration magnitude and angle-z metric are used to assign each of the 5-second epochs to one of the following ten exclusive categories: - Sustained inactivity: all continuous time periods with no z-angle change of more than 5 degrees lasting at least 5 minutes; - Inactivity: defined, outside identified sustained inactivity, as acceleration below 40 m*g*, divided into periods of inactivity lasting less than 10 minutes, bouts of inactivity lasting between 10 and 29.9 minutes, and bouts of inactivity for at least 30 minutes; - Light physical activity (LPA): defined as acceleration between 40 and 120 m*g*, subdivided into spontaneous LPA lasting less than 1 minute, bouts LPA lasting between 1 and 9.9 minutes, and bouts LPA lasting at least 10 minutes; - Moderate or vigorous physical activity (MVPA): defined as acceleration above 120 m*g*, subdivided into spontaneous MVPA lasting less than 1 minute, bouts MVPA lasting between 1 and 9.9 minutes, and bouts MVPA lasting at least 10 minutes. In order to account for natural variations into acceleration values inside one activity bout, bouts were calculated so that short interruptions accumulate to no more than 20% of the bout length (MVPA) and no more than 10% (light and inactivity bouts). Further, interruptions in bouts are not allowed to last longer than a minute, and interruptions in bouts are counted towards the time spent in bouts over a day. The duration of all bouts was defined as the time elapsed between the start (first epoch meeting the threshold criteria) and the end (last epoch meeting the threshold criteria) of the episode. The bouts were computed with function g.getbout from R package GGIR, metric 4. In the absence of validated cut-points for our population, we used the reported acceleration values across activity types in children and adults from a study by Hildebrand and colleagues. The choice for three intensity levels, inactivity, light and MVPA, is widely used in the physical activity research community. The sub-classification of these levels in bout durations, was guided by the common practice to look for bouts of at least 10 minute of MVPA, and the common practice of looking for bouts of at least 30 minutes inactivity or sedentary behavior. The sustained inactivity category has been shown to be a proxy for sleeping time in adults, but could generally be interpreted as time segments without movement and rotation. ## Hidden semi-Markov models The goal of using an unsupervised method is to segment the data in time periods that can be clustered into segments with similar behavior. Hidden Markov Models (HMMs) as used by others for accelerometer data do not model the duration of the state. The related Hidden Semi-Markov Models (HSMM) have the advantage of modelling time distribution for the different behavioral states. HSMM have proven to be valuable for similar segmentation tasks in ubiquitous computing. In a Hidden semi-Markov model (HSMM) clustering is performed into *hidden states*. The word *hidden* is used because the states are not directly observed, but found by the algorithm. Further, the abstract word *states* is used for the data clusters because we do not know (yet) what physical activity intensity category they represent. However, in practice a state can be interpreted as a physical activity intensity category, with its own characteristic distributions of orientation and acceleration values (the observations that are not hidden) and a characteristic distribution of duration. A graphical representation of the HSMM is visualized in. The HSMM is an extension of the widely used Hidden Markov Model. The difference with traditional Hidden Markov models is an explicit distribution for duration of the state. This duration, sometimes called *sojourn time*, is the number of time steps that the model resides in one state before transitioning to the next state. The observations (acceleration and orientation values) are modelled as Multivariate Gaussian distributions, where each state holds its own mean and variance parameters. The durations are modelled as discrete Poisson distribution, where each state holds its own lambda parameter. In addition, there is a transition probability matrix that indicates how likely it is from each state to transition to each other state. The parameters of the model (observation distribution parameters, duration distribution parameters, transition matrix), are learned in a Bayesian manner, with a Hierarchical Dirichlet Process HSMM (HDP-HSMM) as presented in. In Bayesian parameter learning, all parameters are represented as prior distributions, and are updated based on the data (Bayesian inference). The specific inference method used for the HDP-HSMM is Gibbs sampling with weak limit approximate algorithm. In Gibbs sampling, each of the parameters is updated iteratively, by sampling from the conditional probabilities for that parameter, based on the current estimated distributions. In the HDP-HSMM, the transition probabilities between the states are represented as a Dirchlet Process for, in principle, an infinite number of states. However, the number of states is not actually going to be infinite for the following reasons: the Dirichlet Processes share a common parameter that causes the model to favor a small number of states; the weak limit approximate algorithm assumes a maximum number of states (needs to be provided by the user), and the actual number of states is inferred by the model with states being dropped if there are no transitions going in and out of them. Therefore, the number of states can be smaller than the maximum number of states defined by the user. The forward-backward algorithm calculates the distribution over states, conditioned on the observed data and all model parameters. It serves as a step in the Gibbs sampling, and is used to determine optimal state sequence for a given observation sequence. In the forward pass of this algorithm, the initial state assignment and duration is randomly sampled from the distribution. Note that this random sampling can result in slightly different state assignments in multiple runs of the algorithm. ### HSMM parameters There are several choices that influence the runtime of training the model. Since the forward-backward algorithm is used in each iteration of the training with Gibbs sampling, the complexity of this algorithm contributes to the total training time. The forward-backward algorithm has a runtime of *O*(*T* ∙ *N* ∙ *d*_*max* + *T* ∙ *N*²). In this formula, *T* is the length of the sequence, *N* is the number of states and *d*_*max* is a maximum chosen duration length. The maximum state duration *d*_*max* is a user defined input to control training time. We set *d*_*max* at 720 five second epochs, which corresponds to 60 minutes. The maximum number of states *N*_*max* is an input for the weak limit algorithm. Although the number of states is inferred by the algorithm, it can be useful to limit the number of states. It is more computationally efficient to have a small number of states, and easier to interpret the resulting states. *N*_*max* = 10 was evaluated, similar to the number of cut-points categories. The number of iterations of Gibbs sampling needs to be chosen so that the algorithm converges to a stable parameter set. Early stopping is introduced when the hamming distance between the assigned states of two consecutive iterations is smaller than 0.05. In other words, convergence is reached if not more than 5% of the time steps are assigned a different state than in the previous iteration. Further, we chose a maximum of 15 iterations. Lastly, the metrics that are used as observations in the model can be varied. We experimented with two models with different observations as input. The first model used only acceleration, the second model used acceleration together with *angle*_*x*, *angle*_*y* and *angle*_*z*. We will further refer to the resulting models as the *acceleration* model and the *acceleration+angles* model, respectively. The information used *acceleration* model is most comparable to the cut-points approach, since the cut-points approach only uses angle for one out of the ten categories. It is not possible in HSMM to instruct the model to use specific variables for only a single state. For the implementation of the Bayesian HSMMs the python package **pyhsmm (**<https://github.com/mattjj/pyhsmm>) is used, the code is available on Github ( and). ## Evaluation In the following sections, we use the term ‘categories’ to distinguish those slices for the cut-points approach and we use the term ‘states’ to discriminate between those slices for the HSMM approach. Time spent in each state per day and cut-points category was calculated for participants with full 24 hours of data. Principal component analysis was used on the time spent in states and cut-point categories separately. Next, the cumulative variance of principal components, starting at the first principal component, was used to quantify the number of principal components needed to explain at least 95% of the variance across all principal components. This number of required principal components was used as an indicator for the information dimensionality produced by the cut-points approach, HSMM approach with only *acceleration*, and the HSMM approach with *acceleration+angles*, in order to assess whether the angle variable added to the dimensionality of the activity pattern. The PCA statistics give an indication of collinearity between the variables: the more components are needed to explain the variance of the data, the less colinear the variables are. As an alternative way of looking at information dimensionality we also looked at the cross correlation between states and between cut-point categories. Next, to assess comparability of the cut-points and HSMM approaches, correlation coefficients were calculated between time spent in states and cut-points categories, grouped by acceleration level. We expect that the states come with a plausible variation with respect to the conventional method output. To investigate the differences, a descriptive comparison was done of HSMM states, acceleration values, angle values, cut-points categories, and time use diary records. To ease interpretation, only days of measurement with full 24 hours of valid data (no accelerometer non-wear time) were considered for the descriptive comparison. We will focus mostly on the HSMM models using *acceleration+angles* since we expect the addition of angle variables to give extra insights, *acceleration* only results will be reported in the supplement. We choose a sample size of 500 participants for two reasons. First, we would like to demonstrate that the HSMM also works in relatively small samples, such that it can be applied in wide range of study conditions not limited to larger cohorts. Cut-points are typically derived from laboratory studies with less than 100 participants by which 500 participants is still a large sample size. Second, adding more data would increase the computing time. To evaluate that a subsample can generalize to a larger population we tested the reproducibility. A HSMM model was trained on a random subset of 250 participants out of the total set of 500, Nmax = 10, using only the *acceleration* variable since this makes it easier to compare state distributions. The states of the original model trained on 500 participants and the model trained on the subset were sorted on acceleration mean so that a match between corresponding states of both models could be made. The model parameters (mean and sigma for the observation distributions, lambda for the duration distributions) of the corresponding states of the two models were compared. Evaluation were based on the closeness of distributions for each state using the Kullback-Leibler divergence, which is an information-theoretic measure to calculate the distance between two statistical distributions. The Kullback-Leibler divergence is denoted with KL(P\|Q), where P is the distribution of the observed values as learned by the original model on 500 participants, and Q the corresponding distribution for the subset model. # Results A total of 9122 participants accepted to wear the accelerometer, 4970 participants returned the accelerometer and time use diary, out of which a random subsample of data from 500 participants was used for the present study. The demographics of this sample are shown in. Note that the data of all 500 participants, regardless of whether the data was complete, could be used for training the HSMM model. The demographics of the participants who wore the accelerometer for 24 hours on both days are also shown in. **In the calculation of the acceleration values on average** 31% (standard deviation 5%) of the of the epochs were replaced by zero. ## Correlation analyses and PCA Ten different states were found by the HSMM training for both the models using *acceleration* only and using *acceleration+angles*. The percentage of explained variance for the number of principal components of the derived time use variables is plotted in. The lower the curve is, the more dimensions of information the dataset has. Four, five, and eight principle components were needed to explain 95% of the variance of the variables for the cut-points categories (10 variables were input), *acceleration* HSMM model (10 variables were input), and *acceleration+angle* HSMM model (10 variables were input), respectively. The absolute correlation between time spent in the 10 different cut-points defined physical activity categories was 0.27±0.24 (range: 0.01–0.85), for the states of the *acceleration* model it was 0.25±0.18 (range: 0.00–0.77) and for the *acceleration+angles* model it was 0.20±0.13 (range: 0.01–0.52), as illustrated in. For the cut-points categories, there were 8 combinations of variables (out of 45 combinations) that had an absolute correlation of more than 0.5, for the *acceleration* model and the *acceleration+angles* model there were 6 and 1 combinations respectively. ## Comparison of acceleration and angles to cut-points states Distributions in acceleration values, angle values, and durations varied by state and threshold categories, see. We labeled the states with letters in increasing order of average acceleration. In we see how much time the participants spent on average in each cut-points category and each state. A detailed description of the states in the *acceleration+angles* model is given in the supplement. We also compare the values in for parallels between states and cut-points categories. The same can be done for the *acceleration* model (available in of the supplement). The model states and cut-points categories can both be grouped in combinations of respectively states and categories that have similar acceleration levels, corresponding to sustained inactivity, inactivity, LPA and MVPA, see. States were grouped as follows: the group {A, B} corresponds to *sustained inactivity*, states {C, D, E, F} to *inactivity*, states {G, H, I} to *LPA* and state {J} to *MVPA*. The Pearson’s correlation coefficient between the time spent per grouped cut-points category and the time spent per grouped states were 0.69, 0.56, 0.72 and 0.88, respectively for each level of activity defined. The cut-points categories differed in duration by design (standard deviation of duration means = 15.9 minutes), while less difference was observed in durations between states as there were all shorter than 17 minutes and standard deviation of the duration means was 4.3 minutes for the *acceleration* model and 2.9 minutes for the *acceleration+angles* model. In contrast, the distribution of angle values is different for the states with similar acceleration levels. The average values in the cut-points categories for the x, y and z angles have a standard deviation of respectively 8.1, 9.3 and 6.6 degrees over all cut-points categories, whereas over the states this is 31.7, 21.6 and 31.7 degrees. ## Comparison against time use diary To assess whether there is a relation between the states and activities reported in the time use diary, we focused on the 10 most common activities reported in the time use diary, see for a comparison with the *acceleration+angles* model (the same table for the *acceleration* model is available as in the supplement). The sustained inactivity states {A, B} are mostly present during sleep. Strongly sedentary activities, such as “Watching tv‥” and “Playing electronic games‥” have most time steps in the inactivate states {C, D, E, F}, which is consistent with our comparisons against the cut-points approach. States {G, H, I} present more time in activities such as speaking or eating a meal. State J appears as a mix of short time in different activities. ## Reproducibility The states of the *acceleration* model trained on a subset of 250 participants were sorted on acceleration mean and then matched with the states from the model trained on the full data set of 500 participants. The model parameters (acceleration mean and variance, duration mean and variance) are plotted in Figs and. The Kullback-Leibler (KL) divergence for the acceleration distributions is below 1.0 for all state combinations except the two states with small durations. The duration distributions are less consistent over the two models, with 4 out of 10 state combinations having a KL divergence of larger than 1. The calculated KL divergences are listed in the supplement in. # Discussion This study proposes an unsupervised Hidden semi-Markov Model (HSMM) to segment and cluster data from wearable accelerometers. The HSMM is a possible alternative for the widely-used cut-points approach as it does not require resource expensive calibration studies. The increased use of raw data accelerometry in recent years has resulted also in the need for a new calibration study for every population and every new acceleration processing metric to be used when relying on the cut-points approach. Furthermore, these calibration studies are typically limited to small numbers of participants and activity types, thus generalization to different populations and conditions is a known problem. The HSMM approach, examined in the present study, learns ‘states’ from the data, which are described by the mean and variance of the observations (accelerometer derived time series) and by the lambda of their Poisson distributed duration. An important strength of our implementation of the HSMM is that it relies on a small set of input metrics that are relatively easy to interpret in the domain of movement intensity. The HSMM has this advantage of interpretability over other approaches, such as Deep Neural Networks. We intentionally do not use other signal features, because that would introduce the risk that the HSMM model detects activity types rather than intensity. Activity type is a different dimension of physical activity and mistakenly classifying types would undermine interpretation. Therefore, our comparison with the time use diary should be interpreted with care as time use categories do not only reflect variations in intensity. Our findings show that the HSMM derived states were related to cut-points categories. For example, the HSMM found short lasting states with high acceleration and long lasting states with low acceleration, which is consistent with data derived from cut-points approaches. Further, when states and cut- points categories are grouped by acceleration level, correlations of 0.56 and higher are observed between th e two approaches. The mean acceleration for some of the states was close to the threshold value used in the cut-points approach, which indicates disagreement on the range or distribution of acceleration for typical behaviors. The distributions of durations in the HSMM states is also different from the cut-point categories. However, these differences may well be explained by the fact that the HSMM is driven by the distributions in observations (accelerometer recording) from data collected outside a laboratory in the daily life of British teenagers, while the cut-points approach in our case was driven by cut-points derived from Norwegian children and adults performing a very specific set of activity types in a laboratory setting. By showing most states to last less than 17 minutes, our results contribute to the debate about current practice to quantify physical activity and inactivity in ten minute bouts and thirty minute bouts respectively. Consequently, it may not be surprising that no perfect agreement is found between the cut-points approach and the HSMM approach. More importantly, the HSMM covers a plausible range of acceleration levels (low, medium, and high), durations (from less than a minute to more than 30 minutes), and to some extent, although difficult to interpret, angle ranges. Further, it was reassuring to observe that the principal component analyses applied to time spent in states has a less steep scree plot compared to time use variables based on the cut-points approach. This indicates that research on interactions between behaviors will be less challenged by collinearity. An important strength of the HSMM is that it can account for multi-variate input, even if no prior theory exists on why or how the additional input variables could contribute. In contrast, the cut-points approach needs such a theory. Our PCA results indicates that adding the angles offers a description of physical activity with less collinearity and possibly higher dimensions compared to using acceleration only. The HSMM approach allows us to move towards a description of physical behavior based on, and driven by, the accelerometer data that can feasibly be collected in both small and large scale populations. The HSMM is not biased by the subjective nature of self-report methods, avoids the complexities of accounting for inter-individual variation in body composition in energy expenditure estimation and the variation in the relationship between body composition and energy expenditure between activity types, and avoids the difficulties with generalizability of supervised learning techniques that rely on training data composed of small numbers of participants and/or activity types. The HSMM approach will not directly fit into the research framework aimed at providing public recommendation on layman constructs like steps or minutes of moderate to vigorous physical activity per day. However, HSMM may speed up and facilitate a data driven approach that could help to understand how variations in activity characteristics, as measured by acceleration and arm angle, relate to health and disease. ## Strengths and limitations The agreements between time spent in the HSMM states and time use diary categories were poor. Reasons for this are likely to rely on the fact that the time use diary collects broad information on activity type (10 items) and activity context using low time resolution (10-minute slot) in comparison to the physical activity intensity construct as measured by the cut-points approach. Nevertheless, the time use diary allowed us to evaluate how these two constructs relate to one another. A challenge in the design of the present study was that there exists no gold standard for intensity profiling of physical activity in a real life setting in a representative population. Therefore, we combined a variety of analyses to assess the comparability of HSMM and cut-points approach from different perspectives. Future studies might complement the evaluation of models trained on real life data, with data from lab studies, where activities and energy expenditure can be directly measured with indirect calorimetry. The states found by the HSMM are based on patterns in the data and are not specific for the research question of a user. Therefore, when using the HSMM states as physical behavior descriptors in further research, it might be good practice to undertake post-processing of the data. This includes, for example, grouping states that have similar characteristics regarding the research question (e.g. similar acceleration level), or using the majority state for a larger window to adjust for the desired time granularity. In practice, it seems not to be feasible to let the model converge to very consistent state assignment (e.g. \< 1% Hamming distance). It is not clear from theory how much of this is attributed to variance in the data, and how much could be gained with more training iterations. Therefore, future research (both empirical and theoretical) is needed to investigate the relationship between data size, population characteristics, and convergence. Future research is needed to better understand the application of HSMMs on physical activity data. For example, the number of states can be varied to optimize face validity, while retaining interpretability and feasibility in terms of training time. The same holds for the size of the input time frame: instead of 5-second time frames, smaller or larger time frames can be used as input. It is also possible to include more input metrics in the model, although that may also complicate the interpretation of the states. In the present study we limited the number of metrics to facilitate a standardized comparison with the cut-points approach and to facilitate interpretation. The use of the z-angle for sustained inactivity detection in the cut-points approach does not undermine the standardized comparison, because the HSMM model also uses this information: When calculating the magnitude of acceleration that is used as input for the HSMM model, values are replaced by zero when the z-angle is constant for a five minutes. The use of different distributions to represent the data in the HSMM model could be investigated, such as a log-normal distribution for the acceleration metric. The use of metrics that describe the orientation of the device imposes challenges. Firstly, the interpretation of the distributions of the angle values is difficult, asking for visualization and comparison to specific activities to distinguish between states with different angle distributions. Secondly, the correct wear position of the device becomes crucial. In this work, we took a heuristic approach at correcting for improper worn devices. Future research is needed to build a reliable classifier that determines the wear position if signal metrics are used that are body side dependent. Separation of gravitational and calibration of acceleration sensor data is a known challenge and the estimates of acceleration are not entirely free from calibration error. By replacing the magnitude of acceleration by zero for the time segments where the accelerometer does not change orientation we ensure that: calibration error as a function of accelerometer orientation does not influence the segmentation of the acceleration data; the contribution of white signal noise to the magnitude of acceleration is minimized, and; bias caused by calibration error is as close to zero across the recording. We chose a random subset of 500 participants in this study, because we wanted to demonstrate that the HSMM method works in relatively small data sets. Suitability for small datasets is important for uncommon study populations, including the very old, rare diseases, and populations in hard to reach rural areas. The reproducibility experiment suggests that the model for a smaller subset of 250 approaches the model trained on the data of all 500 participants, except for the rarest states. This provides us with confidence that the model generalizes well to more data from the same target group and is not overfitted to the data it is trained on. However, the question remains how much the model generalizes to other populations, e.g. age groups or countries. If enough data are available, it is also possible to train the model for a specific person; this would however make it more difficult to relate the resulting states among the participants. This is a problem of finding the right balance between a model optimized in specific population, and a model that representative for the general population. The cut-points approach has known limitations, but despite these limitations it has been of tremendous value to the physical activity research community for decades. Therefore, we felt it important to aim for comparability with the traditional approach, while at the same time trying to address one of its limitations. Future analyses to compare the associations of time spent in categories assessed by the cut-point approach and states assessed by the HSMM with health outcomes in different population setting will allow to assess the value of the HSMM approach for research. We conclude that applying Hidden Semi- Markov Models results in informative states, based on the data from a real-world setting. It is possible to relate the states to conventional cut-points categories, to interpret the meaning of the states. The unsupervised model can easily incorporate multiple input metrics, so that the states provide a higher dimensional description of physical behavior. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction In recent years, new drug development has become increasingly difficult and more expensive worldwide. It has been reported that the average success rate for new drug development is no greater than about 2%. Long development times, a significant decrease in the approval of new drugs and the expiration of patents have led to a decline in profits for pharmaceutical companies, threatening the sustainability of the industry. In the past, new drug candidates have been sought by high-throughput screening of synthetic libraries and *in silico* research to identify possible drugs from the structures of target molecules. These computer technologies have since evolved into artificial intelligence (AI) systems, based on huge amounts of data, which are now used for drug repositioning. Yet the number of approved new drugs is decreasing, despite an increase in development pipelines worldwide. One reason is that, in the drug development process described above, many target- optimized drug candidates tend to be poorly water-soluble, which results in poor bioavailability when taken orally. Of the various ways of administering a drug, from the point of view of patient quality of life (QOL), oral administration is most desirable. Yet almost 40% of marketed drugs and 70% of candidate compounds were classified as poorly water-soluble at the beginning of the 2000s. Drugs are classified into four types by the U.S. Biopharmaceutics Drug Classification System (BCS) according to their solubility in water and permeability across biological membranes. Drugs in BCS Classes II (low solubility, high permeability), III (high solubility, low permeability) and IV (low solubility, low permeability) show low bioavailability when orally administered. Therefore, in the process of formulation, efforts have been made to improve bioavailability by various techniques such as particle size reduction, lipid-based formulations, solid dispersions, and absorption enhancers. We previously reported successfully improving the solubility of a number of poorly water-soluble drugs by mechanically coating the drug surface with nano- hydroxyapatite (nano-HAP), which is the main component of mammalian teeth and bone and a highly biocompatible substance. A feature of this method is that no physical or chemical pretreatment of the drug is necessary. In the case of bezafibrate (BCS Class II), we found that the area under the blood concentration-time curve (AUC) after oral administration in rats increased when using the nano-HAP coated formulation compared with bezafibrate alone. Moreover, it was found that the side effects observed on continuous administration of bezafibrate were reduced when using the nano-HAP coating. In the case of cisplatin, a highly polar broad spectrum anti-tumor drug classified under BCS Classes III and IV and administered intravenously, lowering patient QOL, we found that coating with nano-HAP improved solubility and intestinal permeability, raising the possibility of using cisplatin as an oral drug. Moreover, cisplatin orally administered in rats after coating with nano-HAP showed equivalent anti-tumor effect to the intravenously administered drug, while reducing the level of renal and hepatic toxicity normally associated with cisplatin. However, while demonstrating the ability of our nano-HAP coating to improve the solubility and intestinal permeability of drugs classified under BCS as poorly soluble and/or poorly absorbed, we had not yet elucidated the functional mechanism. Therefore, in the present study, we investigated the effect of nano- HAP coating on the BCS Class IV drug acetazolamide (AZ), studying in particular its potential routes of intestinal absorption, using *in vitro*, *ex vivo* and *in vivo* methodology. # Materials and methods ## Preparation of formulation and SEM observation AZ was purchased from FUJIFILM Wako Pure Chemical Corporation (Osaka, Japan). The nano-HAP used for coating the surface of the AZ crystals in the AZ/HAP formulation was prepared as previously reported, including a final enteric coating needed for peroral use. Briefly, chemically synthesized micron-level HAP (SKM-1, Sangi Co., Ltd) was ground with a wet grinding mill (DYNOMILL KDLA, Shinmaru Enterprises Co., Osaka, Japan) using zirconia beads 0.3mm in diameter and then dried. The primary particle size of the resulting powder was confirmed to be nano-scale by scanning electron microscope (SEM, Hitachi S-4500, Hitachi High-Tech Science Corporation, Tokyo, Japan), and the average particle size was approximately 130 nm as measured by laser scattering particle size distribution analyzer (LA-950; HORIBA, Ltd., Kyoto, Japan). Preparation of the AZ/HAP formulation was done by mechanical fusion (Mechanofusion AMS-Mini, Hosokawa Micron Group, Osaka, Japan) at 2,000 rpm for 15 min under an argon gas atmosphere. The weight ratio of AZ to nano-HAP in the AZ/HAP formulation was 1:2. The morphology of the AZ/HAP formulation was observed by SEM at an acceleration voltage of 15.0 kV. For animal testing, the AZ/HAP formulation was prepared with an enteric coating, using a mixture of acetone, cellulose acetate phthalate and diethyl phthalate at a weight ratio of 255:9:36. ## Solubility testing Solubility testing was carried out using distilled water and disintegration test solution (DTS) 2 (phosphate buffer solution; pH 6.8) specified by the Japanese Pharmacopoeia. We did not use DTS 1 (phosphate buffer solution; pH 1.2) because the formulation in our study was an enteric-coated preparation. Fifty mL of each test solution was placed in a glass tube and stirred with a 15 mm teflon bar at a constant rotation speed of 200 rpm. Respectively, 300 mg of AZ and 900 mg of AZ/HAP formulation (containing 300mg of AZ) formulation were added to each of the two test solutions. One mL samples were collected into a microtube at 3, 10, 30, 120 and 360 min after the start of the test, which was performed in an incubator at 37 ± 0.5°C. Each sample was centrifuged for 3 min at 8,060 × g, and the supernatant freeze-dried. The dry sample was then dissolved in an appropriate amount of ethyl alcohol. The concentration of AZ in each sample was determined by spectrophotometer (UV-1200, Shimazu Corporation, Kyoto, Japan) at 264 nm wavelength. Solubility testing was carried out on AZ 3 times and on the AZ/HAP formulation 6 times. ## Pharmacokinetics study Kwl:SD male rats (7 weeks; Tokyo Laboratory Animals Science Co., Ltd., Tokyo, Japan) were used in a pharmacokinetics study after single-dose oral administration of either the untreated AZ or AZ/HAP formulation. The dose of AZ in each case was 10 mg/kg, meaning that in the case of the AZ/HAP formulation 30 mg/kg of the physical mixture was administered. Rats were fasted for 16 h before oral administration. Each formulation was suspended in 5 mL water and administered by oral gavage using a rat feeding needle. Blood samples were collected in heparinized tubes from the tail vein at 0.5, 1.0, 3.0, 6.0 and 24.0 h after administration. Each sample was centrifuged for 15 min at 2,190 × g, and the supernatant stored at −40°C until measured. The AZ group and the AZ/HAP group consisted of 4 and 7 rats, respectively. All animal experiments in the study were performed according to the guidelines for animal studies specified by the Science Council of Japan and the internal regulations regarding animal experiments at Sangi Co., Ltd. Approval numbers provided by the Committee for Experimental Animal Care and Use at Sangi Co., Ltd. are as follows: 2018D3 (pharmacokinetics study), 2018D1(Ussing chamber study), and 2018D2 (localization in intestinal tissue using fluorescence-labelled HAP nanoparticles). AZ in each sample was measured with reference to the papers of Ichikawa et al. and Alberts et al.. In brief, 100 μL of chlorothiazide (Tokyo Chemical Industry Co., Ltd., Tokyo, Japan) dissolved in water at a concentration of 50 μg/mL, 0.5 mL of 0.05 M sodium acetate (Special Grade, FUJIFILM Wako Pure Chemical Corporation) and 5 mL of ethyl acetate (HPLC Grade, FUJIFILM Wako Pure Chemical Corporation) were added to each sample. Each tube was stirred with a vortex mixer for 5 min and centrifuged at 1,000 × g for 10 min. The organic layer in the tube was transferred into a separate test tube and dried at room temperature under 100 mmHg atmosphere. 500 μL of the mobile phase solution used in HPLC was added to each dried sample. Finally, samples for quantitative analysis were prepared by passing them through a 0.45 μm filter (Minisart RC4; Sartorius AG, Göttingen, Germany). The amount of AZ in plasma was determined by HPLC (PU-980, UV-970, UV/VIS detector, JASCO Corporation, Tokyo, Japan). YMC-Pack Pro C18 (150 × 4.6 mm.I.D. S-5μm, 12nm, YMC Co., Ltd., Kyoto, Japan) was used as the separation column. The measurement was performed at room temperature using a flow rate of 1.0 mL/min, a measurement wavelength of 266 nm, and an injection amount of 100 μL. The mobile phase solution was a mixture of 0.05 M sodium acetate and acetonitrile (HPLC Grade, FUJIFILM Wako Pure Chemical Corporation) at 90:10 (V/V), and finally the pH was adjusted to 4.1 with acetic acid (Special Grade, FUJIFILM Wako Pure Chemical Corporation).　The standard curve was calculated from the data of each concentration of AZ (0.5, 1.0, 2.5, 5.0, 10.0, 20.0 μg/mL). Maximum concentration (Cmax) and AUC (0–6) values for both test groups after oral administration were calculated from the AZ blood concentration data obtained. ## Gel filtration chromatography analysis of AZ/HAP formulation The components of the AZ/HAP formulation, namely AZ crystals and nano-HAP particles, each diffuse immediately upon contact with water. But it was unknown whether they exist independently in water, or are bonded. Therefore, we conducted an analysis using gel filtration chromatography. PD-10 (Cytiva, Tokyo, Japan) filled with Sephadex G-25 (Medium) was used as the column for gel filtration chromatography, and the analysis was performed using DTS 2 and rat intestinal fluid. The intestinal fluid was collected from two 8-week-old male Kwl:SD rats (Tokyo Laboratory Animals Science Co., Ltd.) according to the method described by Alsulays et al.. Two mL of the test solution was added to 10 mg of AZ, 20 mg of nano-HAP and 30 mg of the AZ/HAP formulation respectively, and each mixture was stirred at 37°C for 30 min. Each sample solution was then centrifuged at 8,060 × g for 5 min, and 1.0 mL of the supernatant used for analysis. The PD-10 column was equilibrated with DTS 2 according to the manufacturer’s instructions. One mL of each test sample was applied to the column, and elution performed by free fall. One mL of the eluate was sampled in the glass tube of a fraction collector (SF-160, ADVANTEC Co., Ltd., Tokyo, Japan), and the total elution volume was 20 mL. 500 μL of each sample was diluted 10-fold (in the case of DTS 2-treated samples) or 6-fold (in the case of intestinal fluid-treated samples) with ethanol solution (ethanol 3.5mL + water 1.0 mL) (ethanol, Special Grade, FUJIFILM Wako Pure Chemical Corporation), and the amount of AZ determined using a spectrophotometer (UV-1200, Shimadzu Corporation) at 364 nm wavelength. In addition, each sample was diluted 10-fold with 0.5 M nitric acid (Special Grade; FUJIFILM Wako Pure Chemical Corporation), and the amount of calcium determined using an inductively coupled plasma atomic emission spectrophotometer (SPS-1700R, Hitachi High-Tech Science Corporation). The amount of hydroxyapatite in each sample was calculated from the amount of calcium recorded. ## AZ and AZ/HAP formulation permeability test using a Caco-2 cell monolayer A Caco-2 cell monolayer assay kit (POCA^® Small Intestinal Absorption (Caco-2), KAC Co., Ltd., Kyoto, Japan) was used to test for the possible permeability of AZ and AZ/HAP formulation via tight junctions, an application for which this apparatus is known. The kit comprises an apical plate, perforated with holes that accommodate 24 wells, each well containing a Caco-2 cell monolayer at its base, to simulate the intestinal lumen, and a basolateral tray underneath, containing 24 receiving compartments into which the apical wells fit, to simulate the underlying tissues. The permeability test was carried out according to the instructions of the assay kit distributor, including the medium used. Transepithelial electrical resistance (TEER) was measured by epithelial voltmeter (EVOM2, World Precision Instruments Ltd., Hertfordshire, UK), and a CO₂ incubator (Direct Heat CO₂ Incubator model 320, Thermo Fisher Scientific, MA, USA) was used for culturing. Test solutions were prepared by adding AZ or AZ/HAP formulation respectively to a Hank’s balanced salt solution (HBSS) buffered with HEPES (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, final concentration 25 mM) and NaHCO₃ (final concentration 0.35 g/L), at a concentration of 0.2 g AZ/mL. To ensure complete dissolution of the AZ crystals, dimethyl sulfoxide (DMSO) was added up to achieve a concentration of 0.5%. Each test solution was stirred for 3 min and pre-warmed to 37°C before being added to the apical wells. First, 1,000 μL of HBSS was added to each receiving compartment in the basolateral tray. The medium in each of the apical wells was then decanted off, and 200 μL of HBSS was added, after which the apical plate containing the wells was fitted into the basolateral tray, and equilibration carried out for 15 min at 37°C. Initial TEER was then assessed by measuring the electrical resistance (Ω) of the current flowing from the medium on the apical side to the medium on the basolateral side of the Caco-2 monolayers. The HBSS in each apical well was then decanted off and 200 μL of either test solution respectively was added. The kit was then incubated for culturing at 37°C for 1 hour, and TEER was again assessed, after which the entire amount of culture medium from each basolateral receiving compartment was collected. The concentration of AZ in the respective samples collected was determined using HPLC according to the method described above. The apparent permeability coefficient (Papp) was calculated according to the following equation: $$Papp = \left( {dQ/dt} \right)\left( {1/\left( {AC_{0}} \right)} \right)$$ where *dQ/dt* is the steady-state flux (μg/s), *A* is the diffusion area of each monolayer (cm²) and *C*₀ is the initial concentration of the test drug (AZ) in each apical well (μg/mL). The percentage change in TEER value 60 min after addition of the respective test solutions, compared with the initial value, was also calculated. The entire experiment was carried out three times, and the average of the results was calculated. ## Permeability testing using rat small intestine in an Ussing chamber An *ex vivo* study using the Ussing chamber was carried out to investigate permeation via an intracellular route. An Ussing chamber is an apparatus for measuring epithelial membrane permeability for ions, nutrients, drugs etc., as shown in. It consists of two baths separated by a sheet of mucosa or a monolayer of epithelial cells　. Since the properties of substances absorbed and the mechanisms of absorption differ between the region forming the intestinal villi and the lymphatic Peyer’s patches in the intestinal wall, specimens of each tissue respectively were prepared for use in the Ussing chamber and the permeability of the two drug solutions tested using each type. The specimens were prepared from Kwl:SD male rats (8 weeks; Tokyo Laboratory Animals Science Co. Ltd.). Intestinal tissue was dissected from each rat after sacrifice by inhalation with isoflurane (MSD Animal Health K.K., Tokyo, Japan). Segments of intestinal mucosa for use in the experiment (approximately 5 cm each) were taken from an area extending to about 50 cm below the duodenojejunal flexure. Tissue from the mucosal cell region not including Peyer’s patches (NPP tissue) was prepared with reference to Clarke, keeping the epithelium and lamina propria intact, but removing the muscular layers. Tissue including a Peyer’s patch (PP tissue) was prepared with reference to the method of Brayden and Baird. All tissue preparation was designed to be completed within a 5-minute time-frame after euthanasia. Krebs-Ringer solution (117 mM NaCl, 4.7 mM KCl, 1.2 mM MgCl₂, 1.2 mM NaH₂PO₄, 25 mM NaHCO₃, 2.5 mM CaCl₂, 11 mM glucose, pH 7.3) was used for all experimental steps. Each tissue specimen was mounted in the Ussing chamber using a slider (P-2300 and P2305, window area = 0.49 cm² respectively, Physiologic Instruments, Inc., CA, USA). Both the mucosal and serosal baths were filled with 7 mL Krebs-Ringer solution kept at 37°C and stirred by continuous supply of 95% O₂-5% CO₂ gas. A total of 24 rats were used in the Ussing chamber experiments. Ag-Ag/AgCl electrodes (P2020-S, Physiologic Instruments Inc.) were connected to the chamber, and TEER was measured by epithelial voltmeter. TEER values were used as evaluation criteria for both NPP and PP tissue to determine whether each excised specimen was functioning normally. If the TEER value was below 70 Ωcm² in NPP tissue and below 65 Ωcm² in PP tissue, the tissue was not used in the experiment. In the PP tissue, whether or not the tissue was functioning was also evaluated by a macromolecular permeability test using horseradish peroxidase (HRP type VI; Sigma-Aldrich Co. LLC, STL, USA) with reference to Keita et al.. HRP was determined using the Pierce™ TMB ELISA Substrate Kit (Thermo Fisher Scientific). Each specimen, after mounting in the Ussing chamber, was left idle for 15 min. Since the ratio of AZ to HAP was 1: 2 w/w for the AZ/HAP formulation, 1/3 of the HAP formulation was calculated as AZ. The AZ and AZ/HAP test solutions were prepared as follows. Whether using untreated AZ or AZ/HAP formulation, the amount added to 500 μL of Krebs- Ringer solution for use in the mucosal bath was adjusted to a final concentration of either 540 ppm or 2,240 ppm of AZ, and the suspension was mixed for 1 min by vortex mixer. 500 μL of each test solution respectively was then added to the mucosal bath after removing 500 μL of Krebs-Ringer solution from the bath in order to equalize the water pressure between the two baths. Similarly, 500 μL of solution was taken from the serosal bath and the same volume of fresh Krebs-Ringer solution added at 30, 60, 90, and 120 min, and four samples were collected at each time point. Each sample solution was immediately transferred to a plastic tube and stored at −80°C until the AZ was measured. Each frozen solution was thawed at room temperature, passed through a 0.45 μm filter, and 10 μL used for HPLC measurement of AZ under the conditions previously shown. ## Intestinal localization of nano-HAP fluorescence-labelled particles HAP nanoparticles are known to be promising carriers for drugs and nucleic acids in drug delivery systems, but their routes of entry into the tissues remain unclear. Since information on the localization of the nano-HAP particles in the intestinal tract after oral administration could provide useful clues for identifying the absorption pathway of AZ in the AZ/HAP formulation, the location of particles labeled with two types of fluorescent dye was investigated. ### Fluorescence labeling of nano-HAP particles Two types of fluorescent dye were used to label nano-HAP particles for observation in the intestinal tissue. One was Calcein (Research Grade; FUJIFILM Wako Pure Chemical Corporation), a calciphilic fluorescent dye, and the other OsteoSense 680EX (Perkin　Elmer Inc., MD, USA), a hydroxyapatite-affinity fluorescent dye used by the authors in a previous paper. First, nano-HAP and Calcein were mixed, at a ratio of 1:25 w/w, with an appropriate amount of distilled water and stirred at room temperature for 2 hours. The HAP-Calcein complex was then washed by a successive process of centrifugation-decantation-washing to remove Calcein not bound to nano-HAP particles. The mixture was transferred to a plastic tube and centrifuged at 8,060 × g for 10 min. After removal of the supernatant, an appropriate amount of distilled water was added to the tube containing the precipitate, and the mixture was treated with an ultrasonic generator (USC-100Z38S-22; Ultrasonic Engineering Co., Ltd., Tokyo, Japan) in 120 W for 5 min. Fluorescence was observed in the supernatant using an epifluorescence microscope with a mercury lamp (BX60 and U-ULH respectively; Olympus Corporation., Tokyo, Japan). The further washing process was repeated until no fluorescent dye was observed in the supernatant and the precipitate was then suspended in an appropriate amount of distilled water. Subsequently, OsteoSense 680EX was added to this suspension at a ratio of 0.08 nmol to nano-HAP 30 mg, and the mixture stirred at room temperature for 3 hours. The centrifuge-decant-wash process was repeated three times to remove OsteoSense 680EX not bound to nano-HAP particles. ### Localization of fluorescence-labeled nano-HAP particles in the rat small intestine Kwl:SD male rats (8 weeks; Tokyo Laboratory Animals Science Co., Ltd.) were used for localization of nano-HAP particles labeled with Calcein and OsteoSense 680EX. Rats were fasted and given water ad libitum 16 hours before oral administration of fluorescence-dyed nano-HAP particles. The nano-HAP particles were suspended in water and orally administered at 300 mg/10 mL/kg. Only water was administered to the untreated controls. Each group consisted of 3 rats. Rats were euthanized by over-anesthesia with isoflurane 30 min after oral administration. After confirming death, the region from the jejunum to the upper ileum, the small intestine including Peyer’s patches and the liver were removed and cut to an appropriate size. Each tissue sample was fixed by immersing in a 4% paraformaldehyde/PBS(−) solution statically at 4°C overnight and then immersed in PBS(−) solution statically at 4°C for 1 hour. Each sample was cut into appropriate blocks and then embedded in Tissue Tech^® O.C.T. compound (Sakura Finetek Japan Co., Ltd., Tokyo, Japan) using liquid nitrogen. Each block was sliced to a thickness of 5 μm at −20°C using a cryostat (Leica CM3050S; Leica Microsystems Inc., Wetzlar, Germany) and attached to a glass slide (Platinum Pro; Matsunami Glass Ind., Ltd., Osaka, Japan). Each frozen section was enclosed with SlowFade™ Diamond Antifade Mountant using DAPI fluorescent dye for nuclear staining (Thermo Fisher Scientific K.K., Tokyo, Japan). Each section was observed using a confocal laser scanning microscope (TCS-SP5 II; Leica Microsystems Inc.) at a magnification of × 630. The excitation wavelengths of OsteoSense 680EX, Calcein and DAPI were 633, 490 and 405 nm respectively, and the fluorescence wavelengths were 690 ± 10, 510 ± 15 and 460 ± 15 nm respectively. ## Statistical analysis Data were expressed as the mean ± S.D. Means were compared by Student’s t-test to identify differences between groups. A value of p \< 0.05 was accepted as significant. # Results ## SEM observation shows SEM micrographs of AZ and the AZ/HAP formulation. The crystal surface of AZ was smooth, and the crystal size measured from micrographs was approximately 20 μm. In contrast, the AZ/HAP formulation particles were slightly smaller, and comprised AZ almost entirely covered by HAP nanoparticles. ## Solubility testing shows the results of solubility testing of AZ and the AZ/HAP formulation, in water and in DTS 2. In the test with water, the amount of AZ dissolved increased slightly at 3 min, then plateaued, and no further change was observed for up to 360 min. Error bars are not visible in the case of AZ alone as the standard deviation was extremely small. In the case of the AZ/HAP formulation, the amount of AZ dissolved at 3 min was about twice that for AZ alone, and it gradually increased up to 120 min, then continued to decline thereafter up to 360 min. In the test with DTS 2, the amount of AZ dissolved in the case of both test solutions increased significantly for up to 10 min, but showed no further increase thereafter for up to 360 min. No significant difference in AZ solubility between the two groups at any sampling time was observed. ## Pharmacokinetics study The blood concentrations of AZ measured for the AZ group and the AZ/HAP formulation group at each sampling time are shown in. Blood concentration in the AZ group increased relatively slowly, peaking at 1 hour after administration, in contrast with the AZ/HAP formulation group, in which blood concentration was significantly higher, increasing rapidly within the first 30 min after administration then continuing to decline slowly thereafter. It decreased to a level that was not significantly different from that of the AZ group at 6 hours. shows Cmax and AUC (0–6) calculated from the blood concentration values of AZ in each drug administration group. The Cmax of the AZ/HAP formulation group was about 5 times higher than that of the AZ group. The AUC (0–6) value was about 4 times higher in the AZ/HAP formulation group than in the AZ group. ## Gel filtration chromatography analysis of AZ/HAP formulation The results of gel filtration chromatography carried out on nano-HAP and untreated AZ, and on AZ/HAP formulation treated with either DTS 2 or with intestinal fluid are shown in. The recovery rate was almost 100% in all elution experiments. In the case of the untreated AZ solution, AZ was first observed from 4 mL, and the maximum amount was observed at 7 mL, while in the case of the nano-HAP suspension, nano-HAP was observed from 1 mL, and already showed its maximum elution rate at 2 mL. In the case of the AZ/HAP formulation treated with DTS 2, the elution pattern of AZ was almost the same as that of AZ without DTS 2 treatment, and the elution pattern of nano-HAP was also almost identical with that of nano-HAP without DTS 2 treatment. In the case of the AZ/HAP formulation treated with intestinal fluid, a small peak was observed in which AZ was eluted slightly earlier than in, while for nano-HAP, the initial peak appeared to be divided into two, showing a broad overall elution pattern. ## AZ and AZ/HAP formulation permeability test using a Caco-2 cell monolayer shows the result of permeability testing of AZ and the AZ/HAP formulation using a Caco-2 cell monolayer assay kit. The vertical axis shows the Papp value. Although the average value of Papp was somewhat higher in the AZ/HAP formulation group no statistical difference between the two groups was observed. shows the percentage change in TEER value 60 min after addition of each respective test solution, compared with the initial value. This level of change is not considered to indicate damage to cells resulting from any external cause or substance, and it was not significantly different between the two test groups. ## Permeability testing using rat small intestine in an Ussing chamber ### Permeability in NPP tissue shows the results of testing for permeability of AZ across NPP tissue when respectively using AZ and the AZ/HAP formulation. When the final concentration of AZ in the mucosal bath after addition of each test solution was 560 ppm, the degree of AZ permeation was shown to increase over time in both groups, but was small and not significantly different between the groups. When the final concentration of AZ in the mucosal bath was raised to 2,240 ppm, the degree of AZ permeation was about 5 times higher in both groups, and it also increased over time, but level of permeation was significantly higher at all sample times for the AZ/HAP formulation group. Values for TEER did not rise during the experiment and no significant difference in them was observed between the two groups, regardless of the initial concentration of AZ. ### Permeability in PP tissue shows the results of testing for permeability of AZ across PP tissue when respectively using AZ and the AZ/HAP formulation. In the case of PP tissue, whether the final concentration of AZ in the mucosal bath was 560 ppm or 2,240 ppm, the degree of AZ permeation observed for both groups once again increased over time and was about 5 times higher at the higher concentration level, but there was no statistically significant difference in permeation between the AZ group and the AZ/HAP formulation group. Values for TEER did not rise during the experiment, and no significant difference in TEER between the two groups was observed. In macromolecular permeability testing prior to the trial, using horseradish peroxidase to determine whether the PP tissue mounted in the Ussing chamber retained the structure and function of follicle-associated epithelium, our results showed that the extracted tissue was functioning normally. We therefore considered our technique sufficient to produce specimens that could be viably used in this study. ## Intestinal localization of nano-HAP-fluorescent particles Nano-HAP particles labeled with two types of fluorescent dyes were used to investigate the location of intestinal absorption after oral administration of the particles. Tissue photographs taken 30 min after administration are shown in. No fluorescence was observed at the fluorescence wavelengths in control rats treated only with water, indicating that the fluorescence of OsteoSense and Calcein observed in this experiment was not affected by any factor of autofluorescence. In addition, as shown in smaller photographs in the right column, the respective fluorescences of OsteoSense 680EX and Calcein were located in the same position, therefore we concluded that these fluorescences were from nano-HAP particles. Fluorescence observation showed that nano-HAP particles were already located in the epithelial cells of the villi and in the submucosal tissue 30 min after oral administration. The presence of many nano-HAP particles was observed in Peyer’s patches, and scattered fluorescence was also seen within the sinusoids of the liver. # Discussion We previously reported that for some poorly water-soluble drugs, coating with nano-HAP particles improved bioavailability and reduced toxicity. In the present study, we succeeded in improving the bioavailability of acetazolamide (AZ), a BCS Class IV low-solubility, low-permeability drug, by mechanically coating its crystal surface with nano-HAP particles. The target of this HAP formulation is poorly water-soluble drugs, namely BCS Class II and Class IV drugs. In the case of BCS Class II drugs, which often show acidity, adjusting the pH of the drug in solution to near neutral is known to increase the solubility of the drug and therefore its absorption from the intestinal tract \[10,11\]. HAP is known to function as a buffer in an acidic pH range. In an earlier study, we succeeded in increasing the AUC of the BCS Class II drug bezafibrate by using a HAP formulation, and we postulated that this buffer function of HAP may play a part in improving the drug’s solubility. In the present study, however, although the BCS Class IV drug AZ was uniformly coated with nano-HAP particles, its solubility in DTS 2 was not significantly improved. AZ in solution is reported to be only weakly acidic, with a pKa of 7.2, and this weak acidity might conceivably have precluded any positive buffering effect of HAP. Since AZ is a BCS Class IV drug, an improvement in bioavailability could not have been expected from the results of our solubility testing, which showed that both AZ and the AZ/HAP formulation　were poorly soluble. Yet after single-dose oral administration, the AZ/HAP formulation group showed a Cmax about 5 times higher and an AUC (0–6) about 4 times higher than for the AZ group, strongly suggesting that the nano-HAP particles play a role in the absorption mechanism. Although various methods for improving the bioavailability of BCS Class IV drugs have been reported, the use of mechanical coating of nano-HAP particles on the drug surface has not been among them. Nano-HAP particles have been widely studied as potential drug carriers, especially for their benefits of slow release and lesion targeting, and are expected before long to be put into practical use. We initially expected that the nano-HAP particles in the HAP formulation would remain combined with AZ in the intestinal tract, and that this nano-HAP-AZ complex would be absorbed by a route different from that of AZ, with the result that the blood concentration of AZ would increase. However, gel filtration analysis of the AZ/HAP formulation, whether treated with DTS 2 or intestinal fluid, showed that AZ was not bound to the nano-HAP particles. When treated with intestinal fluid, the nano-HAP particles showed biphasic separation peaks, unlike those seen when the particles were tested with DTS 2, and the overall elution pattern moved posteriorly. In gel filtration chromatography analysis, larger molecules elute first and smaller molecules later. Results in the present study showed that the nano-HAP particles aggregated in the DTS 2, but tended to disperse in the intestinal fluid. A protein corona is reported to form on polystyrene nanoparticles of 50–100 nm when they come into contact with digestive enzymes such as α-amylase and trypsin, and proteins contained in intestinal fluid are also known to cause aggregation of nanoparticles. Our results provided no direct information regarding protein corona formation or particle aggregation. However, it is possible that any aggregates of nano-HAP particles that did form may have tended to disperse under the influence of bile in the intestinal fluid, which is known to act as a dispersant, and that nano-HAP in the AZ/HAP formulation may affect the absorption mechanism of AZ in the form of dispersed particles rather than as aggregates. The main absorption pathways for nutrients and drugs in the intestinal tract are the intercellular and intracellular pathways of the epithelium, while many large molecules and/or insoluble substances are absorbed via the lymphatic system. The tight junction forms a barrier between cells, blocking the paracellular pathway, through which many substances are nevertheless absorbed by passive transport. The intracellular pathway involves transport using various channels and absorption systems that are intricately intertwined. These include many types of membrane transporters, as well as ATP-binding cassette transporters, P-glycoprotein and other entities involved in the absorption and excretion of drugs and other substances. Membrane transport may be due to morphological changes in cell membranes, including systems such as endocytosis and exocytosis, or phagocytosis and pinocytosis, or it may result from damage to the mucosal epithelium caused by certain drugs or pathologies. There is also a mechanism known as direct permeation in which substances move directly into the cell as a result of a particular stimulus. While most substances entering the body from the small intestine are taken up via the intestinal villi, as an alternative route, especially for the entry of large molecules, substances not absorbed via the villi may enter via the Peyer’s patch tissues. These contain cells called microfold cells (M cells) that are known to absorb large substances by endocytosis on the mucosal side of the cell and expel them from the basal membrane side in a transcellular process described as transcytosis. It is also known that calcium phosphate nanoparticles formed endogenously in the small intestine can trap soluble macromolecules such as protein antigens in the lumen and transport them for removal via the M cells in the follicle-associated membrane covering the Peyer’s patches, through which the macromolecule-containing nanoparticles are absorbed, initiating an antigen- specific mucosal immune response. In order to explore and elucidate the mechanism of increased absorption of AZ seen with the AZ/HAP formulation, experiments involving a large number of detailed pathways are required. As a first step in these studies, we examined which of three pathways—intercellular, intracellular, and lymphatic—may be involved. To begin with, we looked at the intercellular pathway, in which tight junctions play a barrier role. For this research we selected the Caco-2 cell monolayer assay kit, which has long been used for tight junction research, focusing especially on the electrical resistance between cells (TEER), changes in which can be a sign of cell damage and/or of junction opening. As an electrical phenomenon, this resistance can be measured when electrodes are placed on both sides of a cell sheet. Tight junction opening, for example, can be determined by a large, sharp drop in TEER value. TEER values are also known to correlate highly with cell viability, so cell damage can be assumed if there is a greater fluctuation in TEER value than when tight junctions are affected. Testing in the present study using the Caco-2 cell monolayer assay kit showed no significant difference in permeability between the AZ/HAP formulation group and the AZ group. There was almost no change in TEER value in either case and therefore no evidence of either opening of the tight junctions or of cell damage during the experiment. We therefore postulated that the nano-HAP particles in the HAP formulation do not enter the tissues by causing opening of tight junctions, as chitosan reportedly does, or by damaging cells as titanium oxide nanoparticles are reported to do. Our results suggested that nano-HAP particles enter the tissues via intracellular pathways rather than the intercellular route. Moreover, although in recent years, concern has arisen over possible nanoparticle cytotoxicity, in the case of nano-HAP in the present study, no evidence of cytotoxicity was observed. We therefore next conducted an experiment using the Ussing chamber as a test for intracellular transport. In contrast to the *in vitro* single cell-culture Caco-2 model, the *ex vivo* Ussing method uses fresh intestinal epithelium consisting of various functional cells, offering more complex physiological interaction. Another advantage of this method is that in addition to a sample of intestinal epithelium containing principally villi, a sample containing principally Peyer’s patch tissue can also be used. In the case of poorly water- soluble drugs, there is some debate about the possibility that drug solids not completely dissolved in the intestinal tract after oral ingestion may be absorbed as they are, with transport by intracellular pathways or the lymphatic system considered to be the most probable routes. In tests with the Ussing chamber, using intestinal epithelium both with and without Peyer’s patch tissue, both low and high concentrations of AZ or AZ/HAP formulation were suspended in Krebs-Ringer solution and then dropped into the mucosal bath, and as DMSO was not used, this may have caused crystals that were not yet completely dissolved to come into contact with the intestinal epithelium. In both types epithelium, at low concentration, no significant difference in AZ permeability between the test groups was observed, but at a high concentration of AZ, the AZ/HAP formulation group showed significantly greater AZ permeability than the AZ group. Since the pH of the Krebs-Ringer solution in the Ussing chamber is neutral, and our prior testing showed almost no difference in solubility between　AZ and the AZ/HAP formulation, the reason for this difference in permeability remains unclear. However, as in the Caco-2 cell monolayer test, no significant changes in TEER were observed for either group at any time point, indicating that no cellular damage or destruction by AZ or the nano-HAP particles had occurred. It is known that HAP is highly biocompatible and has a strong propensity for cellular adhesion. Therefore, it is possible that adhesion of nano-HAP to the cell surface could interfere with the systems involved in mass transport, in which case the amount of drug absorbed could be expected to decrease. Conversely, inhibition of the ATP-binding cassette transporters and their family of P-glycoproteins, which are known to be involved in drug excretion, could result in increased absorption of their substrate drug. It has been shown that AZ is a substrate for P-glycoprotein and that the rate of AZ efflux from Caco-2 monolayers was three times higher than the uptake rate, but that P-glycoprotein inhibitors significantly reduced this rate of efflux. Because of its high affinity for protein, HAP is often used for chromatographic purification of proteins, not only secreted from cells but also expressed on their surface, and ceramic HAP chromatography has been successfully used for the purification of P-glycoprotein, indicating that HAP has the potential to bind to this protein. In our experiment using NPP tissue in the Ussing chamber, at a high level of initial AZ concentration, AZ permeability was significantly higher for the AZ/HAP formulation than for AZ alone, while at a lower level of initial concentration, no significant difference in AZ permeation between the two groups was observed. This raise~~s~~ the interesting possibility that nano-HAP may have an inhibitory effect on P-glycoprotein functioning, and might possibly contribute to AZ absorbance by interfering with drug efflux and/or protein synthesis in intestinal epithelial cells. More work to confirm this hypothesis however will be necessary. In cell membrane transport, endocytosis is primarily involved in the transport of solid particles, and nano-HAP particles are known to be taken up into cells by endocytosis. Indomethacin nanoparticles have also been reported to be incorporated into the intestinal epithelium by energy-dependent endocytosis. If an increase in endocytosis, due to the presence of nano-HAP particles, also results in increased endocytosis of other substances, it is possible that any AZ crystals not yet dissolved might be absorbed into cells. In any case, it was not possible to clearly show the cause of the increased absorption of AZ from the intestinal tract observed in the AZ/HAP formulation group in this study. It is possible that there may be more than one mechanism operating to increase AZ permeability. However, our results suggest strongly that nano-HAP particles affect not tight junctions but rather aspects of the intracellular pathway. Continuing studies to clarify the mechanism of absorption of drugs mechanically coated with nano-HAP particles may lead to the discovery of new biological functions of HAP. # Conclusion BCS Class IV drugs face the problem of low bioavailability because of their low solubility and low permeability. We succeeded in improving the bioavailability of acetazolamide (AZ), a BCS Class IV drug, when orally administered, by mechanically coating its surface with nano-HAP particles, even though there was no difference in solubility between AZ and the AZ/HAP formulation. In the AZ/HAP formulation, AZ and nano-HAP particles were shown by gel filtration chromatography analysis not to be bound, either in phosphate buffer solution (DTS 2) or intestinal fluid. Therefore, it was considered that nano-HAP particles did not act as carriers of AZ but were nevertheless effective in increasing the drug’s absorption from the intestinal tract. In the present study, testing with a Caco-2 cell monolayer assay kit indicated that nano-HAP was not causing any opening of tight junctions or any damage to cells. Experiments using the Ussing chamber strongly suggested that nanoparticles’ effect on drug permeability instead involves intracellular pathways of the intestinal epithelium, and to some extent the lymphatic system, via the Payer’s patches. At a high level of drug concentration in testing with intestinal epithelium not containing Peyer’s patch tissue, drug permeability was significantly higher in the case of the AZ/HAP formulation that for AZ alone. The method of mechanically coating a drug with nano-HAP particles presented in this paper is simple and does not cause any chemical reaction or alteration to the drug in the manufacturing process. We believe it to be a unique way of improving the bioavailability of poorly water-soluble, poorly absorbable drugs. In the present study, however, we could not elucidate the mechanism of nano- HAP’s effect on the absorption of the BCS Class IV drug from the intestinal tract, and more detailed research on this topic will be required. # Supporting information We thank Tokyo Metropolitan Industrial Technology Research Institute for their excellent technical support in laser microscopic observation, and former colleague Dr. Keiichiro Kikukawa for his expert advice concerning the manuscript. 10.1371/journal.pone.0268067.r001 Decision Letter 0 Gupta Vivek Academic Editor 2022 Vivek Gupta This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 2 Feb 2022 PONE-D-21-40400Nano-hydroxyapatite improves intestinal absorption of acetazolamide (BCS Class IV drug) – but how?PLOS ONE Dear Dr. Kaneko, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Mar 19 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Vivek Gupta Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. 3\. Thank you for stating the following financial disclosure:  "This study was funded by Sangi Co., Ltd. The funder provided support in the form of salaries for authors (Kenichi Kaneko and Ryosuke Miyasaka) as well as research funds. The funder had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript." We note that one or more of the authors is affiliated with the funding organization, indicating the funder may have had some role in the design, data collection, analysis or preparation of your manuscript for publication; in other words, the funder played an indirect role through the participation of the co- authors. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please do the following: a\. Review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. These amendments should be made in the online form. b\. Confirm in your cover letter that you agree with the following statement, and we will change the online submission form on your behalf:  “The funder provided support in the form of salaries for authors \[insert relevant initials\], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section. 4\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. 5\. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: <https://www.youtube.com/watch?v=_xcclfuvtxQ> 6\. We note that you have included the phrase “data not shown” in your manuscript. Unfortunately, this does not meet our data sharing requirements. PLOS does not permit references to inaccessible data. We require that authors provide all relevant data within the paper, Supporting Information files, or in an acceptable, public repository. Please add a citation to support this phrase or upload the data that corresponds with these findings to a stable repository (such as Figshare or Dryad) and provide and URLs, DOIs, or accession numbers that may be used to access these data. Or, if the data are not a core part of the research being presented in your study, we ask that you remove the phrase that refers to these data. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: In the present study the authors have attempted to improve bioavailability of Acetazolamide, a BCS class 4 drug by surface coating using nano- HPA as well as elucidating the possible mechanism for the same. The exact mechanism although still not confirmed by author’s and required further studies which is understandable since absorption being complex and dynamic process with multiple mechanism involved. However, the authors’ have presented a good argument throughout the manuscript with well-structured methodology as well as results and discussions. Thus, I recommend the publication of this manuscript with some minor revisions. Page 5 line 100- Although coating and preparation method has been cited from earlier studies, it will be good if authors can describe the preparation method of formulation (AZ/HAP) briefly for better understanding for readers. Page 11 line 248: Reference for figure 1 is missing Page 20 line 460: As there is no significant difference (p\<0.05) between the group, please remove the asterisk mark Page 25 line 553-554: Discussion regarding no significant difference of solubility between AZ and AZ/HAP formulation in PBS 6.8 is still unclear specially since in case of AZ/HAP formulation being enteric coated. Please elaborate on this for better understanding Reviewer \#2: Article references need to check and update for example Line \#139 reference need to add/correct the reference style. Analytical method sensitivity for estimation of drug concentration need to mentioned which used in CaCo2 study. Reviewer \#3: The authors have presented very good findings with experimental data in the manuscript. The authors are appreciated for the way of explanation given in this manuscript. This manuscript can be publishable after minor revisions. 1\) Relevance of line number 55-60 with context of present study 2\) In line number 72, mention name techniques. Include the following references for the solubility enhancement techniques in line number 72: a\) Diwan, Rimpy, Punna Rao Ravi, Nikita Shantaram Pathare, and Vidushi Aggarwal. "Pharmacodynamic, pharmacokinetic and physical characterization of cilnidipine loaded solid lipid nanoparticles for oral delivery optimized using the principles of design of experiments." Colloids and Surfaces B: Biointerfaces 193 (2020): 111073. b\) Diwan, Rimpy, Punna Rao Ravi, Shubham Ishwar Agarwal, and Vidushi Aggarwal. "Cilnidipine loaded poly (ε-caprolactone) nanoparticles for enhanced oral delivery: optimization using DoE, physical characterization, pharmacokinetic, and pharmacodynamic evaluation." Pharmaceutical Development and Technology 26, no. 3 (2021): 278-290. c\) Diwan, Rimpy, Shareef Khan, and Punna Rao Ravi. "Comparative study of cilnidipine loaded PLGA nanoparticles: process optimization by DoE, physico- chemical characterization and in vivo evaluation." Drug delivery and translational research 10, no. 5 (2020): 1442-1458. 3\) Author can mention briefly the preparation of nano-HAP and the source of micron level HAP used to synthesize nano-HAP (line no. 103) ? Did author extract the micron level HAP from teeth or bone in the lab? 4\) The author can include the brief procedure of coating of nano-HAP over the AZ particles. 5\) Mention the test solution in line 115. Is it phosphate buffer pH 6.8? 6\) What is the AZ weight in the 900 mg of AZ/HAP formulation taken for solubility (line 117)? 7\) What is the anticipated mechanism of solubility improvement with nano-HAP ? 8\) What is the reason of adjusting the AZ concentration to a final concentration of either 540 ppm or 2,240 ppm on mucosal side in Ussing chamber? 9\) Did author check the solubility of physical mixture of AZ/HAP? 10\) The author should show the error bars in Figure 3A – solubility of AZ alone. 11\) What could be the anticipated reason for huge standard deviation in blood concentration of AZ in AZ/HAP formulation at 30 min.? 12\) Line 547, author mentioned HAP is known to function as a buffer in a wide pH range. Why AZ solubility is low in DTS 2 although HAP is good buffering agent? What is pKa value of HAP? 13\) The authors could have quantified the AZ concentration in the intestine after the oral administration. 14\) In authors’ experiment using NPP tissue in the Ussing chamber, at a high level of initial AZ concentration, AZ permeability was significantly higher for the AZ/HAP formulation than for AZ alone, while at a lower level of initial concentration, no significant difference in AZ permeation between the two groups was observed. Why at a high level, AZ permeability is more? \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0268067.r002 Author response to Decision Letter 0 31 Mar 2022 Dr. Vivek Gupta, Academic Editor PLOS ONE March 31, 2022 Dear Dr. Gupta, We are grateful for the valuable comments of the reviewers and the opportunity to revise our manuscript. Our responses to each of the reviewer’s comments are below in red. Please note that page and line references are those in the revised manuscript. Review Comments Reviewer \#1: In the present study the authors have attempted to improve bioavailability of Acetazolamide, a BCS class 4 drug by surface coating using nano- HAP as well as elucidating the possible mechanism for the same. The exact mechanism although still not confirmed by author’s and required further studies which is understandable since absorption being complex and dynamic process with multiple mechanism involved. However, the authors’ have presented a good argument throughout the manuscript with well-structured methodology as well as results and discussions. Thus, I recommend the publication of this manuscript with some minor revisions. 1\) Page 5 line 100- Although coating and preparation method has been cited from earlier studies, it will be good if authors can describe the preparation method of formulation (AZ/HAP) briefly for better understanding for readers. Author’s response: We agree to this comment. According to your suggestion, we have added a brief description of the method of formulation to the main body. Please refer to the inserted sentence immediately after "... for peroral use \[19\]" on page 5 lines 105-118. 2\) Page 11 line 248: Reference for figure 1 is missing. Author’s response: Figure 1 was drawn by us on the basis of our Ussing chamber system used in this experiment, and thus there is no reference to be cited. We have added the comment “Authors’ diagram.” (page 12 line 257). 3\) Page 20 line 460: As there is no significant difference (p\<0.05) between the group, please remove the asterisk mark. Author’s response: We corrected Table 2 and the asterisk mark was deleted. 4\) Page 25 line 553-554: Discussion regarding no significant difference of solubility between AZ and AZ/HAP formulation in PBS 6.8 is still unclear specially since in case of AZ/HAP formulation being enteric coated. Please elaborate on this for better understanding. Author’s response: The AZ/HAP formulation used in the solubility testing was not enteric coated. In the solubility testing, the solubility of the AZ/HAP formulation in water was higher than that of AZ, but solubility in DTS2 was not significantly different between the two formulations. AZ has been reported to show weak acidity when dissolved in water, while HAP has been reported to have a buffering effect on acidic solutions \[(Reference No. 44 in the revised manuscript)\]. We therefore considered that this buffering effect contributed to the improvement in solubility of the AZ/HAP formulation in water. On the other hand, in the solubility test using DTS2, since DTS2 itself has a buffering action, we postulated that this buffering action improved the solubility of AZ to the same level as that of the AZ/HAP formulation, and that this is why there was no significant difference in solubility between AZ and the AZ/HAP formulation in the solubility test using DTS2. Reviewer \#2: Article references need to check and update for example Line \#139 reference need to add/correct the reference style. Analytical method sensitivity for estimation of drug concentration need to mentioned which used in CaCo2 study. Author’s response 1: In accordance with your comment, all reference style in our manuscript has been checked and corrected where needed. Author’s response 2: In the Caco-2 study, HPLC analysis was used to determine drug concentration in the sample from the basolateral tray. Regarding the sensitivity of HPLC analysis, we have provided details of each concentration of the standard drug used for calculating the standard curve, in the section “Pharmacokinetics study” in materials and methods (page 8 lines 172-173). Reviewer \#3: The authors have presented very good findings with experimental data in the manuscript. The authors are appreciated for the way of explanation given in this manuscript. This manuscript can be publishable after minor revisions. 1\) Relevance of line number 55-60 with context of present study. Author’s response: We intended to show that low solubility development pipelines resulted in decreasing the number of approved new drugs and thus improvement in drug solubility could promote new drug development. To clarify relevance with our present study, the text which Reviewer indicated has been combined with that of the following paragraph and we modified several phrases (See page 3 lines 60-61). 2\) In line number 72, mention name techniques. Include the following references for the solubility enhancement techniques in line number 72: Author’s response: We agree to this comment. We added the proposed references to our reference list and the techniques have been mentioned in the main body (See page 4 lines 72-73). a\) Diwan, Rimpy, Punna Rao Ravi, Nikita Shantaram Pathare, and Vidushi Aggarwal. "Pharmacodynamic, pharmacokinetic and physical characterization of cilnidipine loaded solid lipid nanoparticles for oral delivery optimized using the principles of design of experiments." Colloids and Surfaces B: Biointerfaces 193 (2020): 111073. b\) Diwan, Rimpy, Punna Rao Ravi, Shubham Ishwar Agarwal, and Vidushi Aggarwal. "Cilnidipine loaded poly (ε-caprolactone) nanoparticles for enhanced oral delivery: optimization using DoE, physical characterization, pharmacokinetic, and pharmacodynamic evaluation." Pharmaceutical Development and Technology 26, no. 3 (2021): 278-290. c\) Diwan, Rimpy, Shareef Khan, and Punna Rao Ravi. Comparative study of cilnidipine loaded PLGA nanoparticles: process optimization by DoE, physico- chemical characterization and in vivo evaluation. Drug delivery and translational research 10, no. 5 (2020): 1442-1458. 3\) Author can mention briefly the preparation of nano-HAP and the source of micron level HAP used to synthesize nano-HAP (line no. 103) ? Did author extract the micron level HAP from teeth or bone in the lab? Author’s response: As mentioned above, a brief description of the preparation of nano-HAP and the procedure of coating AZ with nano-HAP has been added to the main body. Please check the sentence beginning with “Briefly” on page 5 lines 105-118. 4\) The author can include the brief procedure of coating of nano-HAP over the AZ particles. Author’s response: As above, a brief description has been added to the in main body, in the text beginning with “Briefly” on page 5 lines 105-118. 5\) Mention the test solution in line 115. Is it phosphate buffer pH 6.8? Author’s response: We used either water or DTS 2 as the test solution for solubility testing. To clarify this, the phrase “the test solution” on lines 124-125 was corrected to “each test solution”. 6\) What is the AZ weight in the 900 mg of AZ/HAP formulation taken for solubility (line 117)? Author’s response: 900 mg of AZ/HAP formulation contains 300mg of AZ. The phrase “(containing 300mg of AZ)” has been inserted immediately after “AZ/HAP” on line 127. 7\) What is the anticipated mechanism of solubility improvement with nano-HAP ? Author’s response: We assume that the mechanism by which nano-HAP improves the solubility of the drug is due to the buffering action of HAP. For more information, please refer to the author's response to Reviewer's comment No.12. 8\) What is the reason of adjusting the AZ concentration to a final concentration of either 540 ppm or 2,240 ppm on mucosal side in Ussing chamber? Author’s response: In the pharmacokinetic study, 3 mL of a solution of AZ or AZ/HAP formulation adjusted to 10 mg/kg in AZ equivalent was orally administered to rats. Since rats with an average body weight of 300 g were used in the experiment, the concentration of 3 mL of the administered solution was about 3,000 ppm in terms of AZ. Based on the test conditions of the pharmacokinetic study, we investigated whether a pharmaceutical solution with an AZ concentration of 3,000 ppm could be used in the Ussing chamber study. We found that when an AZ / HAP formulation of 2,500 ppm or more was used, the formulation blocked the flow path in the chamber and data collection failed. On the other hand, the results of the solubility testing revealed that the solubility of the AZ/HAP formulation was about 1,900 ppm at the maximum. Based on these findings, the drug concentrations used in the Ussing chamber study were set at 2,240 ppm, which exceeds the maximum solubility of both the AZ and AZ/HAP formulations, and 560 ppm, which falls within the solubility range of both formulations. 9\) Did author check the solubility of physical mixture of AZ/HAP? Author’s response: In the present study, we didn’t check the solubility of the physical mixture of AZ and HAP or conduct serial dilution, because in a previous study (described in reference No. 16 (Reference No. 19 in the revised manuscript)) we found that a physical mixture of HAP and the insoluble drug cisplatin showed low solubility compared to that of a mechanically HAP-coated cisplatin formulation. 10\) The author should show the error bars in Figure 3A – solubility of AZ alone. Author’s response: The error bars are actually present but are hidden by plotted data (black circles) because the standard deviation in the AZ group was extremely small. We have added the comment “Error bars not visible in the case of AZ alone as standard deviation was extremely small.” (See page 17 lines 388-389). 11\) What could be the anticipated reason for huge standard deviation in blood concentration of AZ in AZ/HAP formulation at 30 min.? Author’s response: We used enteric-coated AZ/HAP formulation in our pharmacokinetics study, and we believe that there could have been surface variance, leading to a variation in its dissolution in the rat intestine. As described in the author's response to Reviewer \#1 comment (1), the enteric- coating procedure was simple, and the quality of the coating such as uniformity and thickness, was not strictly controlled. This may have caused non-uniform enteric coating resulting a variation in dissolution of the coating and of AZ absorption in the rat intestine. We therefore consider that the large SD value for the AZ/HAP group in the pharmacokinetics study may have resulted from non- uniformity in the enteric-coating applied to the AZ/HAP formulation. 12\) Line 547, author mentioned HAP is known to function as a buffer in a wide pH range. Why AZ solubility is low in DTS 2 although HAP is good buffering agent? What is pKa value of HAP? Author’s response: In the solubility testing, the solubility of the AZ/HAP formulation in water was higher than that of AZ, but solubility in DTS2 was not significantly different between the two formulations. AZ has been reported to show weak acidity when dissolved in water, while HAP has been reported to have a buffering effect on acidic solutions. We considered that this buffering effect contributed to the improvement in solubility of the AZ/HAP formulation in water. On the other hand, in the solubility testing using DTS2, since DTS2 itself has a buffering action, we postulated that this buffering action improved the solubility of AZ to the same level as that of the AZ/HAP formulation, and this is why there was no significant difference in solubility between AZ and the AZ/HAP formulation in the solubility testing using DTS2. As a result of reexamining Reference No. 40, initially presented by us, we came to the conclusion that it was inappropriate as a reference for explaining the buffering action of HAP as a general theory. Therefore, in the revised manuscript, we have replaced it with the following paper \[\*\] and corrected the phrase "in a wide pH range" in line 559 to "in an acidic pH range". We apologize for our inadequate confirmation of the contents of the reference. \*Reference No. 44 in the revised manuscript: Okazaki M. Chemistry of apatite in teeth and bones, first ed.. Tokai University Press, Tokyo. 1992. 116-119. Japanese. 13\) The authors could have quantified the AZ concentration in the intestine after the oral administration. Author’s response: During our experiments, AZ concentration in the intestine was not recorded. However, we believe its solubility in DTS2 clearly reflects its dissolution in the intestine. 14\) In authors’ experiment using NPP tissue in the Ussing chamber, at a high level of initial AZ concentration, AZ permeability was significantly higher for the AZ/HAP formulation than for AZ alone, while at a lower level of initial concentration, no significant difference in AZ permeation between the two groups was observed. Why at a high level, AZ permeability is more? Author’s response: As postulated in the passage from lines 672-692 in the revised manuscript, nano-HAP particles may interfere with the drug efflux system in intestinal epithelial cells because of the particles’ high affinity for proteins, particularly P-glycoproteins. In addition, it has been reported that nano-HAP particles are internalized in various types of cancer cells via the endocytic pathway, resulting in an anti-cancer effect, in a dose dependent manner, based on inhibition of protein synthesis \[\*\]. Although nano-HAP particles are also taken into normal cells, the amount of nano-HAP internalized is very low compared with that in cancer cells, and it exhibits low cytotoxicity \[\*\]. Considering these findings, we hypothesize that, in the group treated with AZ/HAP formulation at an AZ concentration of 2,240 ppm, the high amount of nano-HAP particles present may have interfered with the function of P-glycoprotein by binding to it, or by inhibiting cellular protein synthesis in normal intestinal epithelial cells. \*Han Y, Li S, Cao X, et al. Different inhibitory effect and mechanism of hydroxyapatite nanoparticles on normal cells and cancer cells in vitro and in vivo. Sci Rep. 2014; 4: 7134. doi: 10.1038/srep07134. Journal Requirements When submitting your revision, we need you to address these additional requirements. 1\) Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <h ttps://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_b ody.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formattin g_sample_title_authors_affiliations.pdf> Author’s response: We have reviewed and adjusted all style in our manuscript and confirmed that it meets the Journal’s requirements. 2\) Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. Author’s response: We have reviewed and corrected the reference list. 3\) Thank you for stating the following financial disclosure: "This study was funded by Sangi Co., Ltd. The funder provided support in the form of salaries for authors (Kenichi Kaneko and Ryosuke Miyasaka) as well as research funds. The funder had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript." We note that one or more of the authors is affiliated with the funding organization, indicating the funder may have had some role in the design, data collection, analysis or preparation of your manuscript for publication; in other words, the funder played an indirect role through the participation of the co- authors. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please do the following: a\. Review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. These amendments should be made in the online form. b\. Confirm in your cover letter that you agree with the following statement, and we will change the online submission form on your behalf: “The funder provided support in the form of salaries for authors \[insert relevant initials\], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section. Author’s response to a: We have reviewed the author contributions and confirmed the roles of all authors, with only minor adjustment. Author’s response to b: We added the sentence which you proposed to the cover letter. 4\) In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. Author’s response: We have created a supporting information file which contains data underlying our results. Please find attached an excel file entitled “S1 File Minimal data set”. 5\) PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: <https://www.youtube.com/watch?v=_xcclfuvtxQ> Author’s response: At your suggestion, I have already obtained ORCID iD and linked it to my account. 6\) We note that you have included the phrase “data not shown” in your manuscript. Unfortunately, this does not meet our data sharing requirements. PLOS does not permit references to inaccessible data. We require that authors provide all relevant data within the paper, Supporting Information files, or in an acceptable, public repository. Please add a citation to support this phrase or upload the data that corresponds with these findings to a stable repository (such as Figshare or Dryad) and provide and URLs, DOIs, or accession numbers that may be used to access these data. Or, if the data are not a core part of the research being presented in your study, we ask that you remove the phrase that refers to these data. Author’s response: We have created a supporting information file entitled “S1 Fig_Localization of OsteoSense 680EX and Calcein double-labeled nano-HAP particles in rat liver 30 min after oral administration”. 10.1371/journal.pone.0268067.r003 Decision Letter 1 Gupta Vivek Academic Editor 2022 Vivek Gupta This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Apr 2022 Nano-hydroxyapatite improves intestinal absorption of acetazolamide (BCS Class IV drug) – but how? PONE-D-21-40400R1 Dear Dr. Kaneko, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, Vivek Gupta Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed Reviewer \#3: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors have addressed all my comments and made all the changes accordingly. So, I recommend acceptance of the manuscript. Reviewer \#2: (No Response) Reviewer \#3: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: **Yes: **Jayshil Bhatt Reviewer \#3: **Yes: **Himanshu Narendrakumar Bhatt 10.1371/journal.pone.0268067.r004 Acceptance letter Gupta Vivek Academic Editor 2022 Vivek Gupta This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 May 2022 PONE-D-21-40400R1 Nano-hydroxyapatite improves intestinal absorption of acetazolamide (BCS Class IV drug) – but how? Dear Dr. Kaneko: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Vivek Gupta Academic Editor PLOS ONE [^1]: I have read the journal’s policy and the authors of this manuscript have the following competing interests: Kenichi Kaneko and Ryosuke Miyasaka are employees of Sangi Co., Ltd. Roslyn Hayman is the president of Sangi Co., Ltd. This does not alter the author’s adherence to all the PLOS ONE policies on sharing data and materials, in this study.

# Introduction DNA methylation is an epigenetic regulatory mechanism essential for cellular differentiation processes and maintenance of cell type-specific gene expression patterns. Thus, each cell type possesses a stable and characteristic DNA methylation landscape that defines them. Nevertheless, environmental and physiopathological pressures can provoke DNA methylation changes. DNA hypermethylation in the promoter region of tumor suppressor genes is a common hallmark of cancer and its causal relationship with tumor progression has been clearly established. More recently, alterations in DNA methylation have been observed in numerous diseases such as Lupus, Alzheimer’s disease, diabetes, and rheumatoid arthritis. Modifications in DNA methylation also occur in response to environmental factors such as diet. Hence, a large number of publications have revealed the fundamental link between epigenetic deregulation and human disease and consequently, DNA methylation is gaining increasing importance as a source of biomarkers and promising therapeutic targets. Despite the clinical relevance of epigenetic alterations in human disease, the epigenetic approach to the study of solid organ and bone marrow transplantation has been largely overlooked. DNA methylation plays a critical role during hematopoietic differentiation and allows the generation of cell diversity in the immune system. Moreover, it is essential for the establishment of the specific T helper cells subpopulations, regulates the expression of cytolityc genes such as perforins and granzymes in NK and activated T cells, and contributes to the functionality of memory T cells. Taken as a whole, DNA methylation dynamics in blood are essential for the development of the immune function and this implies that immunological aspects of organ transplantation are partially linked to epigenetic regulation. Obviously, this is especially relevant in the case of bone marrow and hematopoietic cell transplantation since the donor’s immune system is transferred to the recipient and thus, immunotherapy with epigenetic drugs, such as hypomethylating agents, has been recently studied in hematopoietic cell transplant. In this study, we sought to explore how global epigenetic dynamics in blood are affected after transplantation of hematopoietic cells and whether or not DNA methylation patterns change in response to pathophysiological events. For this purpose, we examined global DNA methylation levels and promoter specific methylation in a cohort of 47 patients before and after allogenic hematopoietic cell transplant (HCT). The results indicated that changes in DNA methylation occurred not only during the normal evolution of the transplanted cells, but also in response to pathology. These findings exposed the clinical significance of epigenetic analysis as a diagnostic tool in the field of transplantation. # Results ## Analysis of Global DNA Methylation Levels Post-HCT In order to study whole genome methylation status, we used pyrosequencing based methylation assay of CpG sites in repetitive DNA elements. Because of the broad distribution of these sequences across the genome, their levels of methylation are considered to be equivalent to the average whole genome methylation. In this study, we analyzed DNA methylation of the LINE1 element which is an interspersed repetitive DNA element that comprises a substantial portion of the genome, and the pericentromeric tandem repeat NBL2 in 47 patients after hematopoietic cell transplant by pyrosequencing. First, we evaluated the levels of methylation in the donors and in the recipients before and after the transplant and calculated a differential of methylation (ΔMet) that simultaneously compared the methylation values of all CpG sites analyzed in the amplicon between two samples (see details in methods section). The ΔMet mean value for the NBL2 between donors and pre-HCT recipients (ΔMet = 12.110) and between pre-HCT recipients and 1 month post-HCT recipients (ΔMet = 12.610) were significantly higher than between donors and 1 month post-HCT recipients (ΔMet = 5.8450) with a *p-value* lower than 0.001. These results indicated that the patients retained the NBL2 methylation values found in the donor’s blood at 1 month post-transplant. In addition, long term analysis of NBL2 up to 12 months post-HCT demonstrated that methylation remained consistently similar to the donor. In case of LINE1, methylation levels were very similar in all samples and no significant differences were observed. Therefore, LINE1 analysis did not allow a good discrimination of the methylation status between donors and recipients 1 month post-HCT. Nonetheless, LINE1 methylation data from all samples over time (12 months) showed that the average levels of methylation remained stable and suggested that there was not a significant epigenetic divergence from the donor and recipient methylation values during the evolution of the transplant in the LINE1 element. Pyrosequencing analysis of LINE1 and NBL2 only provides limited coverage of the genome so it is possible that other specific genomic regions where these DNA repeats are underrepresented or absent do not follow this trend. To further investigate this issue, we performed a limited methylation study of the subtelomeric repeat D4Z4 in nine patients 1 month after transplant and observed that the ΔMet value between donors and post-HCT recipients (ΔMet = 5.620) was significantly lower than between pre-HCT recipients and post-HCT recipients (ΔMet = 17.930) (Wilcoxon signed-rank test, *p* = 0.003906). Thus, subtelomeric methylation levels from the donors are dominant over the recipient’s levels when blood samples are analyzed after transplantation, similar to NBL2 methylation analysis. It is not clear which parameters are affecting NBL2, LINE1 or D4Z4 methylation. It has been observed previously that age, gender and race/ethnicity may affect DNA methylation in repetitive elements. Hence, we performed a limited study with blood samples from a healthy population (n = 90). We could not observe a clear correlation of NBL2 and LINE1 methylation with age (Pearson correlation coefficient, p\>0.05) or gender (Wilcoxon, p\>0.05) (data not show), although when an arbitrary cutoff point was set at 20 years, older individual showed higher levels of methylation in NBL2 and lower in LINE1. In any case, there was not a linear correlation between DNA methylation in DNA repeats and aging and thus, the moderate effect due to aging does not appear to be sufficient to explain the large inter-individual variability observed in NBL2 methylation. In summary, these results showed that the recipients stably retained the global methylation status of the donors after transplantation. ## Changes in Global Methylation Levels are Associated to HCT Outcomes Because donor’s NBL2 methylation levels apparently remained stable in the host after transplantation, we wanted to analyze whether these values were affected by the transplant outcome. To this end, we focused on two clinical parameters, mixed chimerism and acute GVHD, during the first month post-transplant. The control group comprised patients with complete chimerism that did not develop GVHD symptoms during the follow up of the transplant (n = 9). The ΔMet mean value between donors and the recipients 1 month after transplant was higher for patients with mixed chimerism (n = 8, ΔMet = 9.167) than in the control group with complete chimerism (n = 9, ΔMet = 3.400) (Wilcoxon signed-rank test, *p* = 0.0014). This result was expected because patients with mixed chimerism had cells from both the donor and the recipient whereas patients with complete chimerism had lost the recipient’s cells. In fact, the inherent sensitivity of NBL2 methylation is notably high, with an area under the ROC (AUC) of 0.911. On the other hand, the mean ΔMet value between donors and post-HCT recipients was higher in patients with severe aGVHD (n = 18, ΔMet = 5.870) than in the control group (n = 9, ΔMet = 3.400) or in patients with moderate GVHD symptoms (n = 10, ΔMet = 3.341) although it showed a poor discriminative ability (AUC = 0.678) and lacked statistical significance (Wilcoxon signed-rank test, *p* = 0.065). ## Analysis of Promoter Specific DNA Methylation Post-HCT Since global DNA methylation patterns in post-transplant patients were equivalent to donor values and those values also appeared to change in response to pathophysiological events, we wanted to evaluate DNA methylation in gene promoters. For this purpose, we performed microarray-based DNA methylation profiling in blood samples from a recipient without acute or chronic GVHD (case 1) and from another recipient with grade III aGVHD that evolved into chronic GVHD (case 2). In this analysis, DNA methylation in each probe was considered to be altered when the value differed more than 20% compared to donor. Using this criterion, we found 227 CpG sites (216 genes) with altered methylation values in the case 1 and 238 CpG sites (226 genes) in the case 2, with only 3 genes in common between both samples (DEGS1, TRAF1 and FLJ20273). These differences were also reflected in the gene ontology analysis on the altered genes, showing a differential distribution of biological processes between case 1 and case 2 samples, with a preferential enrichment of GO terms related to immune activation in case 2. On the other hand, 6 months post-HCT, the methylation profile between donor and post-transplant recipient was notably similar, with only 74 altered CpG sites (72 genes) in the case 1 and 37 (37 genes) in the case 2. These results suggested that methylation profiles in blood samples tend to normalize during the evolution of the transplant. To further explore DNA methylation in gene promoters we selected four genes whose function has been previously associated with the immune response to HCT (IFN-γ, FASL, IL-10, and PRF1), and analyzed their methylation status in a cohort of 47 patients up to 12 months post-HCT. Because DNA methylation in gene promoters is usually associated to gene expression we used in this analysis the real percentage of methylation instead of the ΔMet value. Before transplant, we did not observed differences between donors and recipients in IFN-γ, FASL and IL-10, and only methylation at PRF1 promoter was significantly different (p = 0.0001). Nonetheless, we observed a large divergence from the donor values starting at 1 month post-HCT and, in case of IL-10 and PRF1, methylation tended to normalize during the evolution of the transplant. Although the average methylation values did not greatly differ overtime, we observed large differences between patients, particularly during the first few months post-HCT, that could be mirroring relevant clinical parameters. We focused in aGVHD 1 month post-HCT because is strongly associated to immune response and typically appears early after transplant. One month post-HCT, methylation levels of IFN-γ and FASL were statistically lower among patients with severe aGVHD (Wilcoxon signed-rank test, *p*\<0.05) but were not significant different between the control group and the recipients with moderate aGVHD symptoms. The discriminative ability in the case of severe aGVHD is relatively high, with AUC values of 0.782 for IFN-γ and 0.769 for FASL. In contrast, IL-10 methylation was higher in patients with aGVHD (Wilcoxon signed-rank test, *p* = 0.0203) with an AUC value of 0.764, but it was not able to discriminate between patients with moderate and severe aGVHD (Wilcoxon signed-rank test, *p* = 0.1541). On the other hand, PRF1 methylation values were not significant different between groups 1 month post-HCT. Finally, the correlation between percentage of methylation and the cellular composition in whole blood was analyzed in a small number of samples for which flow cytometry data was available one month post-HCT (n = 17). Although the sample size was small, we observed statistically significant correlation between the lymphoid fractions and the DNA methylation in some of the analyzed genes. These results indicate that DNA methylation is at least partially dependent of the cellular composition in whole blood. # Discussion In this study, our aims were to develop a comprehensive epigenetic approach to study hematopoietic cell transplantation. We tackled two parallel parameters: 1) global methylation levels by using pyrosequencing based analysis at repetitive DNA elements of LINE1, NBL2, and D4Z4; and 2) promoter DNA methylation profiling by bead array technology and pyrosequencing at specific gene promoters. Changes in NBL2, LINE1 and D4Z4 methylation have been reported in cancer. Nonetheless, these global methylation levels can also change in response to environmental factors such as diet, tobacco smoke, or inflammation, and during the natural aging process. Thus, the initial question that we wanted to address is whether or not the interaction between donor cells and the recipient was able to induce global DNA methylation changes in the grafted cells. First, the pericentromeric tandem repeat NBL2 and the subtelomeric element D4Z4 retained the methylation levels of the donor after HCT and, in case of NBL2, analysis up to 12 months post-HCT showed that donor’s global methylation levels remained stable. Therefore, global methylation levels of NBL2 in blood were not significantly affected by the interaction between donor and recipient. LINE1 methylation status was similar between donors and recipients, and importantly, ΔMet values did not change over time. These results further supports the observation that there is not a significant epigenetic drift after transplant in repetitive DNA elements and thus, the epigenetic traits acquired by the donor due to environmental factors or aging are likely to be maintained by the grafted cells in the recipient. The second question was whether global methylation can change in response to physiological events during HCT. Comparison of NBL2 methylation between patients with complete and mixed chimerism showed that patients with mixed chimerism had a much greater deviation from the donor’s methylation status. Since NBL2 methylation values post-HCT remained stable, this result is easily explained by the mixed contribution of donor and recipient to the blood samples. Global methylation analysis can accurately segregate patients according to chimerism status and therefore transplanted individuals also can be considered chimeric, not only at genetic level, but also at an epigenetic level. Therefore, intrinsic differences found in the NBL2 methylation values between individuals can be used to track down the source of the cells in transplant. On the other hand, we also observed elevated NBL2 ΔMet values in patients with severe aGVHD which could be due to a real methylation drift as a consequence of the disease. These results suggested that NBL2 methylation could be used as a surrogate marker in HCT. Nonetheless, the data did not reach statistical significance. In addition to the global DNA methylation analysis during HCT, we wanted to analyze DNA methylation in gene promoters which is often associated with modulation of gene expression. We performed microarray-based DNA methylation profiling in PB samples from two patients with allogenic cell transplant at two different times post-HCT. Methylation profiles post-HCT revealed that the methylation signatures in blood at 1 month post-transplant are markedly altered relative to their donors. Conversely, 6 months post-HCT, the methylation signatures were very similar to the donor suggesting that methylation profiles normalize overtime. Nonetheless, the difference observed between samples at 1 month post-HCT are unlikely to be due to a methylation drift post-HCT since this was not observed at global level by pyrosequencing analysis of repetitive DNA elements. Because we profiled DNA methylation in whole blood, the difference observed at promoter levels probably reflect changes in the cellular composition of the samples and thus normalization of the methylation profile is mirroring the immune reconstitution in the recipient. Each hematopoietic cell type possesses a specific methylation signature, and consequently the percent contribution of each cell type to the blood sample results in changes in the methylation profiles. Therefore, methylation analysis could be used to detect alterations of the cellular composition in blood samples if the methylation signature of each cell type was previously defined and consistent in different individuals, although additional studies will be needed to validate this approach. Additionally, we wanted to explore DNA methylation in gene promoters during transplant evolution in a larger sample set. For this purpose, we examined by pyrosequencing genes typically associated with the immune response to HCT. IFN-γ levels are elevated early during development of GVHD symptoms in both animal models and patients, and therefore it constituted a good candidate for validation by pyrosequencing analysis. IFN-γ is an immune activating cytokine whose expression is tightly regulated by DNA methylation. Inverse correlation between DNA methylation and IFN-γ expression has been observed during Th1 differentiation and CD8⁺ lymphocyte activation, and it is constitutively hypomethylated in NK cells. Consequently, the hypomethylation at the IFN-γ promoter observed in patients with severe aGVHD suggested enrichment of cytotoxic CD8⁺, NK, and Th1 cells in the blood samples in agreement with an expected expansion of alloreative cells in GVHD. Second, FASL is highly expressed on activated T cells, plays a major role in T-cell mediated cytotoxicity, and thus, it constitutes a key element of the cytolytic response during GVHD in HCT. FASL levels in serum and mRNA levels in CD4⁺ and CD8⁺ T cells are higher among patients with aGVHD., We found that FASL methylation is notably lower in severe aGVHD which is in agreement with the role of FASL in GVHD. Reduced FASL methylation suggests an enrichment of cytolytic cells in the blood samples from these patients, but further studies are necessary since the methylation status of FASL in blood cells has not been clearly established. Thirdly, IL-10 is a cytokine partially regulated by DNA methylation, and is mainly expressed by monocytes, Th2 and regulatory T cells which have an inhibitory effect over the expression of Th1 cytokines. Since IL-10 antagonized Th1 response, it was proposed initially that higher levels of IL-10 may be associated with lower risk of GVHD. We found higher methylation levels of IL-10 in blood samples of patients with aGVHD, suggesting that higher methylation levels yield lower levels of IL-10 in aGVHD patients. Nonetheless, IL-10 mRNA in whole blood and protein levels in plasma increased early after HCT in patients with GVHD., This result is only contradictory with our data if we assume that DNA methylation in whole blood is directly related to gene expression. Differential expression profiles in whole blood between two samples could be caused either by changes in the relative cell composition of the sample or by changes of the expression profile of a specific cell type without changes in the relative cellular composition. However, changes of the methylation profiles are more likely due to changes of the cellular composition since DNA methylation patterns are usually stable in terminally differentiated cells. Therefore, it is possible that DNA methylation in whole blood increase because cells with hypermethylation at IL-10 promoter are overrepresented while simultaneously IL-10 levels in plasma increased due to overexpression in response to stimuli in a specific cell type. Taken together, if a particular methylation signature in whole blood is going to be used as a biomarker of disease, it is important to note that DNA methylation levels do not necessarily mirror mRNA levels. Finally, perforin (PRF1) is expressed primarily in NK and CD8⁺ T cells, and is partially regulated by DNA methylation., PRF1 gene is involved in immune mediated cell lysis and mRNA levels are elevated in peripheral blood in patients with aGVHD. We found that PRF1 methylation levels decrease in almost all samples 1 month post-HCT, although we did not observe differences between aGVHD and non-aGVHD patients. In summary, our results demonstrated that DNA methylation analysis in HCT provides useful information that, with further studies, it may be useful as a diagnostic tool of relevant clinical parameters. The development of robust biomarkers for aGVHD using this strategy will require a much larger discovery set and validation set; nonetheless, our data support the viability of this approach as a source of biomarkers in HCT and paves the way for larger studies. Pyrosequencing-based DNA methylation analysis is relatively inexpensive, is very sensitive, and provides an interesting alternative to protein and mRNA detection in blood samples, allowing the development of novel biomarkers in a large array of human diseases. Further studies will be necessary to fully explore the potential of DNA methylation analysis in the field of hematopoietic cell and solid organ transplantation. # Methods ## Patients and Samples DNA samples were obtained from the immunology department of Hospital Virgen de Arraixa in Murcia (Spain) and the Hospital Universitario in Salamanca (Spain) according to approved institutional guidelines. The DNA was obtained from peripheral blood (PB) samples collected monthly from the time of transplant up to 1 year when possible. DNA samples from donors and recipients before transplantation were also obtained. All patients received allogenic hematopoietic cell transplantation and prophylactic pharmacological treatment for graft versus host disease (GVHD). The clinical characteristic of the patients are shown in. Chimerism was diagnosed by DNA fingerprinting analysis of microsatellites markers (D5S818, Vwa, D13S317, D7S820, D8S1179, D21S11, D3S1358, D18S51, FGA and AMELX/Y) according to institutional guidelines. Diagnosis and grading of acute GVHD (aGVHD) were established as previously described. Additionally, DNA samples from healthy individuals (n = 90) were obtained from the Hospital Universitario Central de Asturias (HUCA). All patients gave their written informed consent in accordance with the declaration of Helsinki and the study obtained the approval by the local ethics committee (Comité Ético de Investigación Clínica Regional Del Principado de Asturias, project number 17/2011). ## Methylation Analysis by Pyrosequencing Sodium bisulfite modification of 200 ng DNA was carried out with the EZ DNA methylation kit (D5002, Zymo Research, CA, USA) following the manufacturer’s protocol. Pyrosequencing was performed by using the PyroMark kit (Qiagen, Germany). Primers are described in. The PCR condition was 50 cycles at 95°C for 60 s, 58°C for 30 s, and 72°C for 30 s, followed by 72°C for 5 min. The biotinylated PCR product was purified, made single-stranded, and acted as a template in a pyrosequencing reaction by using the Pyrosequencing Vacuum Prep Tool (Qiagen, Germany). Methylation levels were quantified by using the PyroMark Q24 system (Biotage, Sweden). Six CpG sites were analyzed in NBL2 and D4Z4 sequences and four in LINE1. Primers for IFN-γ, FASL, IL-10 and PRF1 were designed to analyze one CpG site within the promoter region of each gene. Primers are described in. ## Differential Methylation Analysis by ΔMet Method The differential of methylation (ΔMet) was developed to allow accurate evaluation of the global methylation drift between two samples by using pyrosequencing based methylation analysis in repetitive DNA elements. The ΔMet was calculated as the Euclidian squared distance in n-space: d(p,q) = ((p₁ − q₁)²+ (p₂ − q₂)²+ …+(p_n − q_n)²)^0.5. By this method, each CpG site within the amplicon was treated as a spacial dimension and the methylation values in each CpG site defined a coordinate in an n-space. Therefore, n equaled the number of CpG sites in the amplicon. The distance between the coordinates obtained from two samples was the differential of methylation (ΔMet). This method allowed a very precise assessment of the methylation drift given that it was not affected by non-informative CpG sites or by opposite methylation drift between CpGs within the same amplicon. In addition, the ΔMet value comprised the methylation drift between two samples in a single value, facilitating statistical analysis and data plotting. ## Promoter DNA Methylation Profiling using Bead Arrays and Differential Methylation Analysis Microarray-based DNA methylation profiling was performed on blood samples from a patient without acute or chronic GVHD and those from a patient with grade III aGVHD that evolved into chronic GVHD. Samples analyzed for each case included donor pre-HCT and recipient samples at 1 month and 6 months post-HCT. Bisulfite conversion of DNA was performed using the EZ DNA Methylation Kit (Zymo Research, CA, USA) according to manufacturer’s procedures. Processed DNA samples were hybridized to the HumanMethylation27 DNA Analysis BeadChip (Illumina, San Diego, CA) and the arrays were scanned on the Illumina iScan system. Raw data were imported and analyzed with the BeadStudio software (version 3.1.3.0 Illumina, Inc) and methylation values were obtained as described. Probes with detection *p*-values of greater than 0.01 were excluded from the analysis. In this study, DNA methylation after HCT in each specific probe was considered to be altered when the value differed more than 20% relative to the donor value. The raw microarray data have been submitted to the NCBI Gene Expression Omnibus (GEO) (<http://www.ncbi.nlm.nih.gov/geo/>) under de accession number GSE36832. Gene ontology (GO) analysis was performed with the DAVID GO Web-based tool. Redundant GO terms with identical gene hits were excluded in Supporting Information Table 2. ## Statistical Analysis Analysis was performed with R.2.10 statistical software ([www.r-project.org](http://www.r-project.org)). Due to the asymmetry of the variables, the median was considered the measure of central tendency for statistical calculations. The Wilcoxon test was used to compare the difference between variables. The discriminative ability of the variables was described by receiver operating characteristic (ROC) curves and summarized by the area under the curve (AUC) values. P-values lower than 0.05 were considered significant. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: RMR CLL MFF BSA. Performed the experiments: RMR RS BSA MM MAS MGD. Analyzed the data: RMR PMC AFF MFF BSA EC. Contributed reagents/materials/analysis tools: MM MAS MGD. Wrote the paper: RMR.

# Introduction In patients with chronic kidney disease (CKD), the mortality risk increases when glomerular filtration rate (GFR) falls and becomes very high when GFR is below 15 ml/min/1.73m². But, it is not clear to what extent the high mortality risk in CKD stage 5 (CKD5) can be reduced by earlier dialysis initiation (DI), and the precise cause(s) which in clinical practice motivates chronic DI in patients with CKD5 at a particular time are not well studied. Thus, the optimal timing and motivation of DI remains controversial. In the late 1990s, in the hope of preventing malnutrition and avoiding increased mortality and complications linked to too late start of dialysis therapy, estimated GFR (eGFR) at DI increased markedly. According to the United States Renal Data System (USRDS), the proportion of patients with eGFR\>10 ml/min at DI increased from 20% in 1996 to 52% in 2008. However, the benefits of “early dialysis start” was challenged as the results of several large observational studies comparing outcomes in patients starting dialysis at various levels of eGFR showed that starting dialysis at lower levels of eGFR (= “late start”) associated with lower mortality. In 2010, the IDEAL study showed that in 828 patients randomized to early (eGFR 10–14 ml/min) or late (eGFR 5–7 ml/min) dialysis start, there was no difference in survival or complications between the two groups. On the other hand, 76% of the patients allotted to late dialysis had to start dialysis before the target of \<7 ml/min due to uremic symptoms. This led to recommendations that dialysis primarily be initiated when patients become symptomatic, or if the patient has severe renal insufficiency, \<5 ml/min (reference 5) or \<6 ml/min. Thus, the concept of terminal uremia, defined as advanced renal failure requiring permanent active treatment in order to prevent death, disease or invalidity, is now limited to the continued presence of uremic symptoms, life- threatening uncontrollable electrolyte disturbances, and/or severe renal failure. These recommendations have stopped the accelerating trend towards earlier and earlier initiation of dialysis based on eGFR. A further problem with the IDEAL study is that patients were referred early to specialist nephrological care, were closely monitored, and, as participants in a randomized controlled trial, may have been atypical of the general terminal uraemia population. Probably as a result of these factors, the incidence of urgent DI, defined as need for a temporary dialysis catheter placement, was very low (3.7% for the early start and 8.3% for the late start group respectively) compared with the incidence of about 40% in most other clinical studies. Urgent/unplanned DI is undesirable since it is associated with a substantially increased risk of septicemia and mortality. While one cause for urgent/unplanned DI is delayed referral of patients with CKD stage 4–5 to specialist nephrological care, the physicians´ motivation for starting dialysis has received little scientific attention. An international questionnaire survey in 2000, showed that 60% of physicians believed that clinical problems were the most important factors for starting dialysis (nutritional status 14%, overhydration 8%, uremic symptoms 38%), while 32% believed residual renal function (RRF) to be most important. Only 4% did not believe that early start had any clinical benefit. These figures would be expected to be radically different in the post-IDEAL era. However, an international questionnaire study in 2012 showed that a third of nephrologists considered RRF to be most important, rising to 54% for uncomplicated patients. The median eGFR requiring dialysis for uncomplicated patients was 10 ml/min, varying between 5–20 ml/min, suggesting that many physicians remain unconvinced of the implications of the IDEAL study. While the European Renal Best Practice (ERBP) guidelines recommend the mean of urea and creatinine clearance as the optimum method of evaluating RRF, most physicians—likely for practical reasons—continue to prefer eGFR, albeit often in combination with another method. There is a continuing uncertainty about the optimum timing of DI, the role of GFR and on what other grounds physicians (should) initiate dialysis. The Peridialysis project is an ongoing, multi-center, multinational prospective epidemiological study investigating clinical practices up to DI and their immediate consequences (the “peridialysis” period). In the present study, which is the first report from the project, we report data concerning specific causes and timing of DI, focusing on physicians´ stated motivations for prescribing dialysis, and correlations of causes and timing of DI with physician details. We hypothesized that physician motivation for DI is a psychological/sociological phenomenon, independent of other clinical and biochemical variables that may influence DI practices. In particular, we hypothesized that some physicians will prescribe dialysis on mainly clinical grounds, while others will prescribe on mainly biochemical grounds. This is an important difference, since it is possible that differences in physician motivation will affect the patient’s prognosis and choice of dialysis modality. # Materials and methods Eleven nephrology departments took part in this observational prospective questionnaire study of causes and timing of DI. All delivered both peritoneal dialysis and hemodialysis. All centers were publicly financed, with no dialysis costs to the patient, but with varying financial support for medicine costs. The commonest method of assessing RRF and guiding clinical treatment was eGFR as measured by the CKD-EPI formula. ## Patients All consecutive patients starting chronic dialysis therapy for ESRD between 1.1.2015 and 1.1.2016 at the participating centers were included in the current study. Some centers provided additional data up to 1.7.2016. The patient was considered as having end-stage renal disease (ESRD) at first dialysis if 1. The treating physician “believed” that the patient had ESRD at first dialysis 2. The patient received \>90 days dialysis treatment 3. If the doctor was in doubt whether the patient had acute or chronic renal failure, the patient was included retrospectively as soon as there was no doubt that the patient had ESRD The current study comprises 446 patients (median age of 67 years; 38% females) with CKD5 who were investigated in conjunction with DI. Patient data are shown in. The underlying renal diagnoses were: glomerulonephritis 19.3%, chronic interstitial nephropathy/obstructive 9.4%, polycystic disease 7.4%, type 1 diabetes mellitus (DM) 6.3%, type 2 DM 17.3%, hypertensive nephropathy 17.9%, other 10.8% and unknown 11.7%. The study protocol was approved by the ethical review boards in centers located in countries where according to the country´s regulations such perusal was required. The study was approved by the Swedish Ethical Committee (Ref 2017/7). However, in Denmark, due to the observational non-interventional design of the study using anonymized patient data, the study protocol was not considered to be eligible for ethical review. Informed consent—either written or verbal depending on the regulations in the different countries—was obtained from participants in all centers including those in Denmark The study is registered with Clinical Trials.gov, identifier NCT02488200 The datafile for this project is available on Open Science Framework, identifiers: DOI [10.17605/OSF.IO/Z63JP](https://doi.org/10.17605/OSF.IO/Z63JP), ARK c7605/osf.io/z63jp ## Methods ### Patient clinical data The following patient data at DI were registered: age, sex, height, weight, body mass index (BMI) and renal diagnosis. The presence of the following comorbidities was registered: previous myocardial infarction, heart failure, cardiac atherosclerosis, cerebrovascular disease, diabetes, peripheral atherosclerosis, previous cancer (except basocellular), chronic pulmonary disease, chronic liver disease, psychiatric disease, previous renal transplantation, and other specified chronic conditions. ### Patient biochemical data The following biochemical data prior to or in conjunction with first dialysis were registered: blood hemoglobin, plasma values of urea, creatinine, potassium, hydrogen carbonate (bicarbonate), albumin, C-reactive protein (CRP), total or ionized calcium, and phosphate. Most centers measured ionized calcium, for other centers, ionized calcium was assumed to be 50% of total calcium. ### Physician motivation questionnaire Physicians gave details of their reasons for prescribing chronic dialysis in an English language questionnaire. They could choose between clinical and/or biochemical reasons. Since the terms used in the questionnaire were deemed to be clearly defined, no questionnaire validation for local translations were performed. *Clinical reasons*: pulmonary stasis, dyspnea, hypertension, pericarditis, oedema, cardiac symptoms, fatigue, anorexia, nausea/vomiting, cachexia/weight loss, itching, insomnia, depression, diarrhea, taste disturbances, social problems, practical problems, and other specified reasons. *Biochemical reasons (based on plasma values)*: high creatinine, high urea, low GFR, high potassium, acidosis, low calcium, high calcium, high phosphate, falling GFR. A primary reason for DI had to be specified, and a secondary was voluntary. In addition, other reasons for DI could be stated by the physician who also was encouraged to register the presence of uremic symptoms at dialysis prescription. ### Physician data Data on participating physicians (n = 84) were requested and for those who agreed to have their data registered (n = 52, contributing to 74% of DIs)) data were anonymized. They were asked to provide the following personal data: age, sex, specialist qualification, duration of physician experience, and duration of specialist nephrology experience. # Statistics Data are presented as mean ± standard deviation (SD) or median (interquartile, IQ, range) or as numbers (percentage). Parametric variables were compared using the Students t-test and MANOVA, and non-parametric using the Chi square and Kruskal-Wallis tests. A probability level of \<0.05 was considered significant. Significance values were expressed as p\<0.05, p\<0.01, p\<0.001. For comparisons, variables were divided according to median value or as tertiles. # Results Dialysis initiation (DI) in 446 patients was prescribed by 84 doctors at 11 hospitals. Patient data are shown in. The reasons (19 clinical and 11 biochemical) for prescribing dialysis are shown in and and clinical symptoms in. The primary reason for DI was clinical in 62.7% of cases, 16.4% had no biochemical indication, and 16.9% no clinical indication. 48.0% had 1–2 symptoms, 21.5% 3–4 and 13.6% \>4 symptoms. Life threating conditions (pulmonary stasis, dyspnea, pericarditis, cardiac symptoms, tetanus, hyperkalemia and acidosis) were the primary reason in 102 (22.9%) of cases. All 19 pre-stated clinical symptoms that a priori were assumed to be main causes for DI were present in one or more of patients and “other” clinical causes including ascites, freezing symptoms, muscle cramps, tetanus and vertigo were also noted in some of the patients. Low renal function (high creatinine or low eGFR) was the primary cause in 19.1% but low RRF was noted as a factor prompting DI prescription in altogether 69.4% of the cases. Only 14 patients (3.1%) had an eGFR \>15 ml/min at DI. Among the 84 prescribing physicians, 52 physicians, responsible for 332 (74.4%) of prescriptions, supplied their personal details. Their age was 49 ±10 years, median 49 (interquartile, IQ, range 41–58) years; and 27 (54%) were female. They had been qualified physicians for 22 +/- 10 years, median 22 (13–45) years, and 41 (79%) were specialist nephrologists, and had been so for on average 13 ±9 years, median 13 (6–19) years. Females physicians were generally younger (45.8 ±9.1 vs. 52.0 ±9.8 years, p\<0.05). Fewer females were specialists (67% vs. 89%), and fewer had long specialist experience (\>13 years, 21% vs. 54%) as compared to the males. There were considerable differences in the behavior of physicians, as shown in. Older physicians were more likely to prescribe dialysis on primarily clinical grounds, but were also more likely to include biochemical reasons in their prescription. Females were more likely to initiate dialysis for primarily life- threatening reasons. Non-specialists were more likely to include clinical grounds in their reasoning, and to start dialysis for life-threatening reasons. Physicians with more specialist experience were more likely to give clinical reasons, and less likely to give life-threatening reasons. There were no significant differences between the prescribers in eGFR at first dialysis, though there was an insignificant trend for more experienced doctors to start at a lower renal function, and total physician experience showed a similar pattern to specialist experience. When excluding patients with primary life-threatening reasons to take into account the possibility that less experienced doctors were more likely to meet acutely ill patients, the results were generally unchanged. However, female physicians started dialysis at a higher eGFR than male doctors (8.0 ±3.6 ml/min vs. 7.2 ±3.5, p\<0.05). The biochemical relationship to the biochemical reasons given for starting dialysis are shown in. There was general agreement on the interpretation of the terms, with a considerable overlap. Patients with a “high creatinine” had a slightly higher creatinine than those without, but there was no significant difference in the corresponding eGFR. Similarly, patients with a “low GFR”, had a significantly lower eGFR, but no difference in creatinine. The choice of the terms “high creatinine” or “low eGFR” as primary reason did not relate to physician details. There was no agreement on the meaning of the term eGFR. We hypothesized that the following variables would correlate to clinical uraemia: eGFR, BMI, albumin, bicarbonate, and CRP. The data are shown in. There were considerable differences in the eGFR for the various primary problems. Patients with cachexia, anorexia and pulmonary stasis had an eGFR of 8.2–9.8 ml/min, while patients with acidosis and edema were characterized by a low eGFR (4.6–6.1 ml/min). BMI was broadly similar, but cachectic patients had as expected a lower level, on average 21.3 kg/m². Conditions characterized by hypoalbuminemia (≤33 g/l) included pulmonary stasis, hyperkalemia, dyspnea, edema and acidosis. A low bicarbonate (≤20 mM) was seen in patients with hyperkalemia, and in those with a rapidly falling eGFR, or cardiac symptoms. Pulmonary stasis, dyspnea, acidosis and hyperkalemia were generally associated with inflammation (CRP\>24 mg/l). Patients with primarily clinical grounds for DI and/or no biochemical reasons had a significantly higher eGFR. Patients with life-threatening conditions were generally biochemically abnormal, with hypoalbuminemia, acidosis and a raised CRP. Correlations to renal diagnosis, patient sex and age are shown in. eGFR was not related to diagnosis, but in a post hoc analysis, eGFR was higher in patients with diabetic nephropathy (p\<0.05). Diabetics were more likely to start dialysis on clinical indications. Patient sex was not related to eGFR or physician motivation. Younger patients (\<60 years) were more likely to start because of life-threatening conditions. The presence of comorbidity had a major influence on biochemistry and DI. Patients with more than three comorbidities started dialysis at an eGFR that was 2.3 ml/min higher than patients with none, 8.7 ±4.7 vs 6.4 ±2.6 ml/min. CRP rose continuously, and albumin fell with increasing comorbidity. A post hoc Spearman correlation analysis revealed albumin and CRP to be highly inversely correlated (R = -0.40, p\<0.001). Patients with pulmonary stasis had a significantly higher prevalence of heart failure (34% vs. 14%, p\<0.001). # Discussion There are few studies of clinicians’ motivation for prescribing dialysis, and this is the first comprehensive prospective study in the post-IDEAL era. There were 21 primary causes stated by the 84 prescribing physicians as reason for starting dialysis among the 446 patients, illustrating that terminal uraemia is a kaleidoscopic condition. While most patients (63%) started dialysis on primarily clinical grounds, low renal function was the primary reason in only 19% of cases: however, residual renal function was stated as playing a role in 69% of cases. The physicians seem generally to be following the conclusions of the IDEAL study and recommendations from ERBP, the Canadian Society of Nephrology and KDOQI that the decision to initiate maintenance dialysis should be based primarily on assessment of signs and/or symptoms of uraemia and related conditions such as fluid overload and secondly in the presence of severe renal failure (GFR \<5–6 ml/min). There has been some criticism of the conclusions of the IDEAL study. Clinical uremia was not clearly defined in the study. suggests that the primary uremic symptom is fatigue, and this symptom may not necessarily be alleviated by dialysis, particularly if symptoms are caused by comorbid disease or hemodialysis side effects. If the cause of dialysis initiation is non-life-threatening uremic symptoms, the final decision should whenever possible be made by the patient, who must balance symptomatology against the inconvenience of dialysis. There are no financial incentives for early or late start of dialysis in the countries involved in this study; it is possible that the presence of such incentives could lead to suboptimal patient treatment. Previous studies have reported uremic symptoms at DI.\[–\] The methodology of these differed from the present study in a number of ways: two were based upon patient notes, one on objective assessment rather than symptoms. In one study patients with acute uremia were excluded and another was limited to nursing home patients. DI in these studies occurred at a higher eGFR, varying from 8.9 ml/min to 11.0 ml/min. O’Hare found a DI eGFR \>15 ml/min in 16% of patients, and Kurella 18%. Significant causes of early DI were edema and dyspnea. Since only 3% of our patients started DI at an eGFR \>15 ml/min, this suggests that these symptoms can usually be controlled by conservative measures, e.g. diuretics, fluid restriction and sodium bicarbonate supplements. Seventeen percent of our patients had no uremic symptoms at DI, generally in accordance with other studies: 12% and 18%. However, there is some disagreement concerning the number of symptoms. 22% of our patients had \>2 symptoms, compared to 20% in the Kurella study and 61% in the Curtis study. There is general agreement that fatigue, nausea, anorexia, volume overload and itching are the commonest uremic symptoms. The symptoms of uraemia are often non-specific, and many patients probably started dialysis for reasons other than uraemia, in that patients with multiple comorbidity, including diabetes and with probable symptoms related to these, started dialysis earlier than non-comorbid patients. Similarly, patients with pulmonary stasis started dialysis early. This may have been due to unrelated cardiac failure (in 34% of the cases), but suggests that increased attention to volume control in these patients could have postponed dialysis requirement. Patients starting due to pulmonary stasis and dyspnea had very high values of CRP, suggesting that complicating pneumonia may have been a partial cause of their distress. One surprising fact is that cachexia and/or weight loss was the primary indication in only 2% of patients. A previous survey showed that nutritional problems were considered to be an important reason for starting dialysis. Plasma albumin has often been used as a marker of malnutrition. This is probably erroneous: in this study, albumin was a marker of increasing morbidity and inflammation as measured by CRP. The interplay between protein malnutrition and inflammation is a complex issue, since albumin is both a marker of nutrition and of inflammation. Patients who are sarcopenic often have increased inflammation (the “malnutrition-inflammation-atherosclerosis syndrome”). This will further have complicated the decision algorithm for physicians prescribing dialysis on mainly biochemical grounds. Furthermore, patients suffering from pulmonary stasis, dyspnea, or edema were characterized by hypoalbuminemia, suggesting a dilutional effect. Patients who started dialysis on life-threating clinical grounds generally had a lower p-albumin and higher C-reactive protein. There were considerable differences between doctors according to clinical experience. The higher proportion of prescription of dialysis for life- threatening reasons among non-specialists could be related to their probable closer contact with acutely ill patients. In contrast, older doctors (\>49 years) and specialists with \>13 years of specialist experience were most likely to state primarily clinical reasons for DI. It is possible that these physicians are those most likely to accept the implications of the IDEAL study. Correspondingly, more experienced physicians tended to start dialysis at a lower eGFR. Differences between male and female doctors were relatively minor, but females started dialysis at a higher eGFR than males, probably due to their higher contact with acutely ill patients. There appeared to be general agreement on the meaning of the biochemical indications for dialysis, but with considerable overlap. This overlap could either be due to real disagreement about the meaning of the terms, or disagreement about what level indicates DI, but implies that there is a considerable variation among doctors as to the optimal timing of DI. In general, doctors started dialysis purely on the basis of renal function only if the eGFR was about 6 ml/min. All in all, this paper demonstrates considerable confusion among physicians concerning definitions and their reasons to start dialysis. A priori, one would expect the terms “high creatinine” and “low eGFR” to be equivalent; this was however not the case, in that patients with “high creatinine” at DI had similar eGFR as other patients, and vice versa. The choice of terms was not related to physician experience. They thus seem to be perceived as slightly different concepts, the clinical meaning of this difference being unclear. It may be that “high creatinine” is used by physicians who are skeptical about the meaning of eGFR (vide infra). There was no agreement as to what constituted a rate of loss of eGFR that indicated dialysis start. However, rate of loss of eGFR during the last 1–2 months before dialysis was not registered, and might have revealed significant differences. There is substantial evidence that eGFR is not a reliable measure of renal function in patients with CKD5. Paradoxically, a high eGFR at DI was reported to associate with increased mortality. This may be due to the fact that many of these patients are malnourished, with reduced muscle mass and low creatinine production, and as a consequence a “falsely” raised eGFR. Our results support this hypothesis in that patients with cachexia and anorexia had the highest values of GFR, and support the current recommendations of the ERBP guidelines that the average of creatinine and urea clearances, measured from a 24-hour urine collection, should be the preferred method of assessing renal function in CKD5. However, in the present study, there was no relationship between eGFR and patient age suggesting that age related alterations in body composition such as sarcopenia did not play a role as modifier of eGFR. Another possible explanation for the paradoxical relationship of higher eGFR with increased mortality is that patients with a high eGFR may have more comorbidity, and therefore start dialysis earlier, as previously described. It is possible that dialysis may not be strictly necessary for these patients, but it is in practice often impossible for the physician to distinguish symptoms of uraemia from other diseases, and, in addition, as mentioned above, comorbid patients are usually malnourished and may have a “falsely” raised eGFR. There are several limitations, which should be taken into account when interpreting the results of the present study. One important problem that is not addressed by the study is the converse question “Why do physicians not prescribe dialysis?” This question does not involve the question as to whether to withhold active treatment either due to severe comorbidity or due to patient choice, but the fact that all contacts to patients with CKD stage 5 must address this question. The answer might be somewhat different from just the absence of reasons for starting dialysis. The study does not reliably supply insights into the true "why" of the psychology of dialysis start today and not tomorrow. For that aim, qualitative research, or vignette studies, should be better suited. In the current setting, respondents can only provide answers already selected for them. The study was limited to the “peridialysis” period and did not include analysis of patient referral to nephrology care, or GFR trajectories and appearance of symptoms and signs and how they were treated during the months or years preceding DI; such information was most likely considered and should have influenced the decisions by the prescribing physicians. So far, the prospective data on the consequences of timing and motivation of DI are not yet available, and therefore no conclusions can be made as regards the implications of the observed findings. The relatively low number of patients, and the low number of prescribing physicians, reduces the statistical power while the generalizability of the study is confined mainly to the Nordic area where the study was performed. On the other hand, a major strength of the study is that it used a more detailed questionnaire about the causes for DI and that it collected information about the prescribing physicians. Furthermore, the study addresses an area where few recent studies have been conducted. # Conclusions In conclusion, this first report from the prospective questionnaire Peridialysis study on timing and causes prompting physicians to prescribe maintenance dialysis shows that dialysis initiation was mainly motivated on clinical grounds whereas eGFR and other indices of renal function loss appear to have had more of a supportive role for their decisions. There were considerable differences in motivations stated by physicians for prescribing dialysis, which are related to their age and clinical experience. These differences may be an independent factor in the clinical treatment of patients with consequences for the risk of unplanned dialysis initiation. # Supporting information We thank all physicians and other staff members who participated in this study. The project was supported by an unrestricted grant from Baxter Healthcare, Deerfield, Illinois, USA. Baxter Novum is the result of a grant from Baxter Healthcare to Karolinska Institutet. [^1]: This study was supported by an unrestricted grant 14CECPDEU1001 from Baxter Healthcare International, [www.baxter.com](http://www.baxter.com). Baxter Novum is a result of a grant from Baxter Healthcare Corporation to support research activities at Karolinska Institutet to promote the understanding and treatment of renal disease. Bengt Lindholm is employed by Baxter Healthcare. None of the other authors declare any conflict of interest. This does not alter our adherence to PLOS ONE policies on sharing data and materials. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

# Introduction Persistently elevated T-cell activation is now recognized as one of the key pathogenic features of HIV infection. T cell activation increases cell turnover, and the high level of T cell activation is thought to be one of the factors leading to depletion of CD4+ T cells. Although both CD4+ and CD8+ T cell activation are associated with disease progression, the association, somewhat paradoxically, is strongest with CD8+ T cells. One possible explanation for this finding is that activated CD4+ T cells undergo apoptosis more rapidly, whereas activated CD8+ T cells tend persist much longer, making CD8+ T cells a better marker of the overall state of T cell activation. After controlling for viral load, the level of CD8⁺ T cell activation in untreated individuals is independently associated with risk of disease progression, as defined by rate of CD4⁺ T cell loss and risk of developing an AIDS defining complication or dying. The factors influencing T cell activation in HIV infection have not been fully identified. While some factors that may contribute to the level of activation have been identified, such as infection with co-pathogens and microbial translocation through damaged gut mucosa, they do not fully explain the observed variability in T cell activation between HIV-infected individuals. The hypothalamic-pituitary-adrenal (HPA) axis regulates secretion of glucocorticoids, endogenous hormones with potent anti-inflammatory properties. While increased secretion of glucocorticoids in periods of acute stress may prevent excessive immune activation chronic stress, may lead to decreased glucocorticoid sensitivity and impairment in the ability of the HPA axis to regulate the immune system. We hypothesized that relative dysregulation of the HPA axis, as measured by altered patterns of diurnal glucocorticoid secretion, results in greater levels of T cell activation in persons with HIV. Normal HPA physiology results in a diurnal rhythm of cortisol secretion characterized by a peak shortly after waking, a gradual decline over the course of the day, and a nadir one hour after the usual time of sleep. This rhythm is often altered in persons exposed to chronically stressful situations, resulting in elevated bedtime cortisol and flatter diurnal slopes. Flatter diurnal cortisol slopes have also been reported to be associated with depressive symptoms in population based studies. Additionally, several studies show associations between severe chronic psychological stress and low morning cortisol. A potential role of the HPA axis in HIV pathogenesis is supported by emerging observations linking stress and mood with disease outcome in lentiviral infections. For example, severe stressful life events, lack of social support, and chronic depressive symptoms are associated with accelerated progression of HIV. Of note, in one of these studies, elevated morning cortisol was associated with more rapid disease progression (18). Also, in studies of rhesus monkeys, those inoculated with the simian immunodeficiency virus (SIV) exhibited shorter survival when exposed to stressful circumstances. The association between psychological stress and mood and HIV progression is thought to be mediated by molecular messengers of the HPA axis and autonomic nervous system (ANS), and prior research has identified pathways linking the sympathetic nervous system to HIV pathogenesis. Norepinephrine upregulates expression of CCR5 and CXCR4 receptors on lymphocytes, co-receptors for HIV cell entry, which may enhance HIV replication. Other studies in an SIV model suggest that chronic stress can increase sympathetic nervous system innervation of lymph nodes, which in turn causes increased SIV replication. While there is evidence linking the sympathetic nervous system to accelerated progression of HIV, a clear mechanism linking the HPA axis to disease progression has not been established. We hypothesized that the HPA axis influences HIV disease progression by altering the level of T cell activation. To test this hypothesis, we investigated the relationship between the HPA axis and T cell activation in untreated HIV infected persons with early stage disease. Here, we describe the associations between these factors, and present a model in which impaired function of the HPA axis, as often seen in chronic psychological stress, contributes to abnormal T cell activation and accelerated disease progression. # Methods ## Ethics Statement The study was reviewed and approved by the UCSF Committee on Human Research. We obtained written informed consent from all of the participants. ## Study design and participants We performed a cross-sectional study of the relationship between glucocorticoids and T cell activation in HIV by analyzing baseline data from HIV-infected individuals in the University of California San Francisco Staying Well study. While participants went on to an intervention stage, the data presented here were all obtained prior to any study intervention. Inclusion criteria included (1) HIV-1 seropositivity, (2) CD4⁺ T lymphocyte count \>250 cells/µl, (3) plasma HIV-1 RNA\>100 copies/ml, (4) antiretroviral therapy naive or untreated for at least four months, and (5) 18 or more years of age. Exclusion criteria included a plan to initiate antiretroviral therapy within 12 months from the time of enrollment and current use or use in the past 6 months of immunomodulator drugs. Subjects were recruited in the San Francisco area through an extensive recruitment network that included referrals from physicians, community based organizations, other research studies, and self referral generated by word of mouth, advertising, and community outreach. The study received approval from the UCSF Committee on Human Research, and all subjects gave written informed consent. ## Measures We measured salivary cortisol by providing participants with a home saliva collection kit and instructions to collect three saliva samples each day over three days. In addition to providing a practical means of measuring cortisol while subjects perform normal daily activities, measurements of salivary cortisol more accurately reflect serum free cortisol concentrations than do measurements of serum total cortisol. The salivary cortisol concentration is independent of salivary flow. Participants were instructed to collect the first sample immediately after waking, the second sample at 30 minutes post-waking, and the third sample immediately before bed. To ensure that the 30 minute post- waking sample was collected at the accurate time interval from waking, participants were given a timer to set for 30 minutes upon waking in the morning to serve as a reminder. Participants were also provided with a saliva logbook and instructed to record their waking time, bedtime, and collection times. Detailed instruction in sample collection was provided to participants by trained staff. Saliva samples were collected by having participants drool into collection tubes, which were frozen and stored at −20 C until analysis. Before biochemical analysis, samples were centrifuged at 3000 rpm for 5 minutes, resulting in a clear supernatant of low viscosity. Free salivary cortisol was analyzed with a commercially available chemi-luminescence-assay (CLIA) with a high sensitivity of 0.16 ng/ml (IBM; Hamburg, Germany). Blood was drawn between 8 am and noon, in acid citrate dextrose venous vacuum collection tubes. Peripheral blood mononuclear cells (PBMCs) were isolated from whole blood by Ficoll-Hypaque density gradient centrifugation within 6 hours of blood drawing, cryopreserved, and stored at the UCSF Biological Specimen Bank. Several immunophenotypic markers can be used to quantify T cell activation in vivo, but CD38 expression on T cells is the marker that has shown the best characterized ability to predict disease progression in HIV. CD38 is a multifactorial transmembrane glycoprotein that is upregulated during T cell activation and is associated with increased cell-to-cell adhesion, increased levels of cytokine production, and more rapid CD4⁺ T cell proliferation. The prognostic significance of CD38 expression in the setting of HIV appears to be greatest when measured as the mean density of molecules per cell, rather than the percentage of cells expressing CD38, but the proportion of T cells expressing both CD38 and HLA-DR also provides useful prognostic information. We measured expression of CD38 and HLA-DR on T cells at the UCSF Core Immunology Laboratory using flow cytometry and previously described methods that have been optimized and validated for frozen PBMCs. Frozen PBMCs were rapidly thawed in warm media and counted on a Guava PCA (Millipore) with the Viacount assay (Millipore). Cells were transferred to plates and then washed and labeled with Aqua Amine Reactive Dye (AARD, Invitrogen) to discriminate dead cells. Without additional washing, cells were stained with the following fluorescently- conjugated monoclonal antibodies: CD3-Pacific Blue, CD38-PE, HLA-DR-FITC, CD4-PE CY7, (all from BD Bioscience), and CD8-QDot 605 (Invitrogen). In each experiment a fluorescent-minus one control (FMO) was included for CD38 and HLA-DR to determine the cut off for positive staining. Stained cells were washed, fixed in 0.5% formaldehyde (Polyscience), and held at 4 C until analysis (within 18 hours). Stained cells were run on a customized BD LSR II flow cytometer and 100,000 to 200,000 lymphocytes were collected for immunophenotyping. Data were compensated and analyzed using Flow Jo (Tree Star) to determine the proportion of CD4⁺ and CD8⁺ T cells expressing each of the T cell markers. Combinations of markers were calculated in FlowJo using the Boolean gate function (see for a typical flow cytometry dot plot that illustrates gating strategy). The geometric mean of relative fluorescence intensity, referred to as the mean fluorescence intensity (MFI), was determined for CD38 on CD4⁺ and CD8⁺ T cells. This approach is similar to the Quantibrite approach, but the relative fluorescence units were not converted to the number of molecules per cell as multiple parameters were analyzed simultaneously. Based on previous work showing that CD38 antigen expression on CD8⁺ T cells is the strongest marker of HIV disease progression to AIDS and death, we used CD38 MFI on CD8⁺ as our primary outcome measure. Plasma specimens were used to measure HIV-1 RNA copies per milliliter by polymerase chain reaction using the Amplicor HIV-1 Monitor Test, v1.5 (Roche Diagnostics). # Analysis Raw cortisol data were screened for outliers thought to be due to contamination (e.g. blood) or poor participant compliance; samples with values ≥100 nmol/L were deleted. Sample 1 (waking sample) was excluded if collected \>10 minutes after waking, sample 2 (30 min post-waking sample) if collected \<20 minutes or \>45 minutes after waking, and sample 3 (bedtime sample) if collected later than 3 am the following day. In addition to the cortisol variables representing specific time points across the day, two derived cortisol variables were calculated: the cortisol awakening response (CAR) and cortisol slope. The CAR, or increase in cortisol from waking to 30 min post-waking, was calculated as the difference score between samples 1 and 2 (sample 2 – sample 1). The CAR typically represents a 50% increase in cortisol levels and is larger in persons experiencing chronic worry and stress. The cortisol slope, or degree of change (typically decline) in cortisol levels from early morning to late evening, was computed as the difference between samples 1 and 3 (sample 3 – sample 1). The slope is a global measure of diurnal cortisol rhythmicity; flatter slopes are associated with chronic stress, chronic fatigue, low socioeconomic status, and low occupational status. Averages were computed for each cortisol variable across the three days of sample collection. Data for cortisol, viral load, and CD4⁺ counts were positively skewed and normalized using a logarithmic (log₁₀) transformation. In computing derived variables, raw variables were logarithmically transformed prior to other calculations. Scatterplots were created and pairwise Pearson correlation coefficients calculated to quantify the association between cortisol predictor variables and immunologic outcome variables. Multiple linear regression was used to model T cell activation as a function of diurnal cortisol adjusted for age, viral load, and CD4⁺ count. All analyses were performed using Stata 10. # Results Of 177 participants enrolled in the overall study, 128 provided adequate baseline specimens for both cortisol and immunologic assays. Because Pearson correlation coefficients were calculated in a pairwise fashion, and the number of missing datapoints differs across variables, the number of participants used to calculate each coefficient varies within a range of 101–128. The median CD4⁺ T cell count at baseline was 469 cells/mm³ (IQR 372–585) and the median plasma HIV RNA level was 4.3 log10 copies/mL (IQR 3.7–4.7). As expected, CD4⁺ count was negatively correlated with activation, as measured by CD38 MFI, on both CD4⁺ (r = −0.18, p = 0.047) and CD8⁺ (r = −0.20, p = 0.029) cells. Also consistent with previous work, there was a positive correlation between plasma HIV RNA levels and both CD4⁺ activation (r = 0.33, p\<0.001) and CD8⁺ activation (r = 0.48, p\<0.001). We next assessed T cell activation, measured by CD38 MFI, in relationship to cortisol. We found that lower waking cortisol correlated with greater CD4⁺ T cell activation (r = −0.26, p = 0.006) and a trend toward greater CD8⁺ T cell activation (r = −0.17, p = 0.08). In the multivariate regression analysis, waking cortisol was a statistically significant predictor of both CD4⁺ (β = −413 U/nmol/L, p = 0.001) and CD8⁺ activation (β = −542 U/nmol/L, p = 0.011) after adjustment for age, CD4⁺ count, and viral load. The 30 minute post-waking cortisol is typically the highest cortisol value over the course of the day. In this study, lower 30 minute post-waking cortisol correlated with greater CD4⁺ activation (r = −0.20, p = 0.049). This relationship was statistically significant following adjustment for age, CD4⁺ count, and viral load (β = −370 U/nmol/L, p = 0.015). We also examined the relationship between diurnal cortisol slope and T cell activation. Abnormal slope values (less negative or positive) were correlated with greater CD38 MFI on both CD4⁺ (r = 0.24, p = 0.018) and CD8⁺ (r = 0.24, p = 0.017) cells. In addition, cortisol slope was a statistically significant predictor of CD4⁺ and CD8⁺ activation after adjusting for covariates. Finally, we examined T cell activation as measured by the percentage of T cells that were CD38⁺HLA-DR⁺. summarizes the association between cortisol variables and the percentage of CD4⁺/CD8⁺ cells with the CD38⁺HLA-DR⁺ phenotype. Greater bedtime cortisol was correlated with greater HLA-DR⁺CD38⁺CD8⁺%, but this relationship was not statistically significant in the multivariate regression analysis. There was also a non-statistically significant trend toward flatter cortisol slope predicting greater T cell co-expression of HLA-DR/CD38. Overall, we found few associations between diurnal cortisol and percentage of cells co-expressing activation antigens. # Discussion Psychosocial stress and mood affect disease outcomes in immune-related disorders such as HIV infection. Although some potential pathways linking these psychological states to disease progression have been identified (23), the mechanisms responsible for this association require further elucidation. Given the strong link between stress, mood, and the neuroendocrine system, we examined whether altered cortisol patterns are associated with greater T cell activation, a known risk factor for accelerated progression of HIV. We found that lower morning cortisol and flatter diurnal rhythms are associated with greater activation of both CD4⁺ and CD8⁺ T cell subsets after adjustment for covariates. To our knowledge, this is the first report of the relationship between diurnal cortisol patterns and immune activation in HIV+ subjects, and the results point to the HPA axis as an important regulator of immune activation in persons with HIV. These findings are particularly significant given the importance of immune activation in the pathogenesis of HIV infection, and imply that improvements in some of the stress-related cortisol patterns could decrease immune activation during HIV infection. Our findings add to an expanding literature supporting a link between stress, the neuroendocrine system, the immune system, and the overall risk of HIV disease progression. The link between altered cortisol rhythmicity and greater immune activation may be explained by a state in which immune cells demonstrate impaired sensitivity to glucocorticoid signals, or relative glucocorticoid resistance. Studies have shown that chronic psychological stress correlates with impaired immune cell responses to anti-inflammatory signals, including down-regulation of glucocorticoid response elements in leukocytes. These results suggest a model in which chronic stress leads to relative dysregulation of the HPA axis, which in turn causes reduced sensitivity of immune cells to cortisol. Such a loss of sensitivity would impair regulatory responses to dampen inflammation, resulting in higher T cell activation and more rapid progression to advanced disease. Of note, one of the HIV-1 accessory proteins, vpr, has been reported to interact with the glucorticoid receptor and enhance tissue sensitivity to glucocorticoids. This effect might be expected to enhance the effect of normal diurnal variation of cortisol on T-cell activation. Our ability to draw conclusions from this study is limited by its cross- sectional design. While our data show an association between diurnal cortisol and T cell activation, we cannot presently assert a causal relationship between predictor and outcome variables. We suspect that the observed associations are driven by the influence of glucocorticoids on immune function, but it is also conceivable that HPA function is altered in the setting of excessive immune activation. Furthermore, a bidirectional relationship may exist. Data from prospective studies are needed to elucidate the direction of this relationship. In addition, HIV-1 infection itself has also been linked to alterations in the HPA axis, which suggests that HIV-1 may be influencing both T-cell activation and the HPA-axis. Prior studies, however, have often reported elevated basal cortisol in advanced infection, or adrenal insufficiency in a subset of patients; our study focused on persons with earlier stage infection and most patients had normal cortisol levels. Though the statistically significant correlations between diurnal cortisol and immune activation are moderate, they are internally consistent and may be biologically significant. From previous work we know that immune activation in HIV is influenced by multiple factors, including plasma HIV RNA levels, bacterial translocation across altered gut mucosa, age, gender, and infection with co-pathogens such as CMV. We therefore do not expect glucocorticoid levels to account for the majority of variance in activation. We chose to examine CD38 expression due to the evidence supporting its importance in predicting HIV disease progression. While referred to as an indicator of T-cell activation, it is also important to note that this is a molecule with multiple functions, and not a pure activation marker. In this regard, the relationship that we have shown between cortisol measures and CD38 expression may not represent an association with immune activation alone. As with most studies of basal cortisol, this study relies on self-report of times when measurements were taken and the relationship between measurement and waking time. To the extent that cortisol patterns truly influence T cell activation, this limitation is likely to cause noise in our predictor variable and underestimation of the true relationship. Some relationships in which we observed trends towards associations, such as increased cortisol awakening response and CD38 MFI on CD4+ and CD8+ T-cells, might not be statistically significant in part due to variability in the collection time of the 30 minute post-waking specimen. We invested considerable effort, however, to minimize errors in the data by screening all cortisol data for participant non-compliance and contacting participants to clarify sampling times when necessary. Given the widths of our confidence intervals and the care with which sampling was performed, we believe the lack of statistical significance in some of the hypothesized relationships does not rule out meaningful associations between particular measures of diurnal cortisol and T cell activation, but suggests that any actual associations are likely to be variable and/or of modest strength. The most unexpected finding in our study was the apparent difference in the associations between cortisol and our two measures of T cell activation (the density of CD38 on T cells versus the frequency of T cells co-expressing CD38/HLA-DR). There was a consistent association between the density of CD38 on T cells and diurnal cortisol rhythms, but similar associations appeared to be absent (though with some overlap in confidence intervals) between the percentage of cells expressing CD38/HLA-DR and cortisol. This contrast underscores the difference between relative mean fluorescence intensity and percentage measurements for quantifying levels of activation. Glucocorticoids may have more influence on the degree of immune activation than on the percentage of cells in an activated state. In conclusion, we have shown that cortisol patterns reflective of chronic stress (less diurnal decline, low waking cortisol) were correlated with greater T cell activation in the setting of HIV-1 infection. If supported by further data, these findings indicate a pathway through which psychological states and the HPA axis influence immune function in HIV and other immune-mediated diseases. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FMH SP PM EE ES MEK SGD. Performed the experiments: FMH PM ES LE CK. Analyzed the data: SP PB MA FMH. Contributed reagents/materials/analysis tools: CK. Wrote the paper: SP FMH.

# Introduction Depression is a multi-faceted disorder encompassing emotional, cognitive, behavioral, and somatic symptoms. Treatment for major depression may include various forms of psychotherapy, antidepressant medication (ADM), such as the selective serotonin reuptake inhibitors (SSRIs), or a combination of both. Placebo-controlled clinical trials typically show that SSRIs and cognitive- behavioral therapies outperform placebo. One striking aspect of these clinical trials is the large symptom improvement in the placebo group. Meta-analyses of placebo-controlled trials of most SSRIs estimate that placebo accounts for about 75% of the effects of ADM during the acute phase treatment. That is, these data suggest that no more than 25% of the observable change may be attributed to the pharmacological effects of SSRIs, whereas the majority of change is due to nonspecific placebo effects and natural course of the illness (spontaneous remission). In this light, the psychopharmacological effects of SSRIs appear rather unimpressive. This conclusion, however, is based exclusively on reported changes in total scores on depression outcome measures and treatment effects may differ by symptom clusters. The effectiveness of SSRIs for a wide range of mental disorders indicates that they provide relief on diverse sets of psychological symptoms, or, alternatively, that they may alter broader dispositions, such as maladaptive personality traits. Secondly, patients in depression studies rarely present exclusively with a “pure” set of depression symptoms, but nearly always have clinical or subclinical manifestations of other disorders, particularly anxiety, which may also be altered by SSRI treatment. Finally, depression itself is a psychometrically multidimensional construct and improvement in one dimensional symptom set (such as mood) will not automatically accompany change in another (such as insomnia). To understand the scope and the limits of SSRI effects, researchers must examine outcomes in greater detail and depth. In one such example, Tang et al. examined both the depression severity and the personality trait of neuroticism in a placebo-controlled trial of paroxetine for moderate to severely depressed patients. Neuroticism refers to one’s tendency to experience exaggerated negative emotions of sadness, anger, and anxiety under conditions of stress. While 75% of the improvement observed with paroxetine on the traditional depression measure, the Hamilton Rating Scale for Depression (HRSD) was accounted for by placebo effect, only 23% of the observed decrease in neuroticism was duplicated in the placebo condition. In addition, the specific advantage for paroxetine over placebo with respect to depression was no longer significant after controlling for change in neuroticism, whereas its specific advantage over placebo in reducing neuroticism remained significant after controlling for change in depression. It is possible that ADM substantially changes some depression symptoms and has virtually no effect (or a negative effect) on others. Consistent with this notion, meta-analysis of placebo-controlled SSRI trials show a wide range of effect sizes for the individual depression symptoms. For example, in two separate meta-analyses of ADM treatment studies of depression—one with tricyclics and the other with fluoxetine—Faries et al. found that five symptoms (depressed mood, guilt, suicidality, disinterest / reduction in work and activities, and psychic anxiety) were more sensitive to differences between placebo and SSRIs compared to the other symptoms on the HRSD. Given that the HRSD is a commonly used measure in clinical trials, several researchers have promoted the use of different subscales of HRSD items on the basis of greater responsiveness to ADM, improved psychometric properties of these scales, and the association of individual items with overall depression severity. While these studies generally demonstrate superior ADM effects for certain HRSD subscales, a clearly stated theory for the observed differential symptom effect sizes is still absent. For example, in a post-hoc analysis, Fairies et al. described the five symptoms common to the most responsive HRSD subscales as “core symptoms” of depression. This is equivalent to defining depression as that which SSRIs reduce, and implies that non-core symptoms are unimportant. A clear conceptual distinction is needed between those symptoms on which SSRIs have a “true” pharmacological effect versus those symptoms on which their effects are largely nonspecific. We hypothesize that the specific pharmacological advantage of SSRIs over placebo will be largely concentrated on the *psychological* symptoms of depression and anxiety and not on the *somatic* symptoms. This counterintuitive hypothesis is consistent with an earlier finding that paroxetine has a considerably larger specific effect on neuroticism than on depression. While four out of the nine depression symptoms articulated in the *DSM-IV* may be characterized as somatic, and somatic symptoms make up as many as 11 out of 17 symptoms assessed by the HRSD, they are entirely absent in neuroticism measurement. Somatic complaints have been found to correlate weakly with neuroticism, whereas psychological symptoms of both depression and anxiety correlate moderately to strongly. Finally, psychometric analyses indicate that somatic symptoms—fatigue, appetite loss/gain, insomnia, and anxious arousal—show relatively distinct patterns of association relative to the more general affective symptoms common to both depression and anxiety. We test this hypothesis on data generated in a placebo-controlled randomized trial of 180 moderately to severely depressed patients. In addition, we will also explore how changes in the psychological subscales of depression and anxiety relate to changes in neuroticism, given their conceptual overlap. # Method ## Participants The Institutional Review Board at the University of Pennsylvania and the Human Research Protection Program (Institutional Review Board) at Vanderbilt University approved the study protocol and written informed consent was obtained from all participants. Subjects were moderate-to-severely depressed adult outpatients; patient characteristics, treatment procedure, and depression outcome findings have been detailed elsewhere. All patients met criteria for MDD and scored 20 or higher at both screening and intake evaluations on the 17-item version of the Hamilton Rating Scale of Depression, modified to incorporate atypical symptoms. Inclusion criteria were: (1) *DSM-IV* MDD diagnosis; (2) aged 18 to 70; (3) English speaking; and (4) willingness and ability to give informed consent. Exclusion criteria were: (1) history of bipolar I disorder; (2) substance abuse or dependence judged to require treatment; (3) current or past psychosis; (4) another *DSM-IV* Axis I disorder judged to require priority treatment; (5) antisocial, borderline, and/or schizotypal disorders (all other Axis II disorders were permitted); (6) suicide risk requiring immediate hospitalization; (7) a medical condition that contraindicated study medications; or (8) nonresponse to an adequate trial of paroxetine in the preceding year. ## Clinical Trial The trial randomized 120 patients to paroxetine and 60 patients to pill-placebo. (Sixty patients were also randomized to a cognitive therapy group, but they are not included in this study of paroxetine mechanism). Thirteen paroxetine patients (11%) and eight placebo patients (13%) dropped out before week 8. The patients who dropped out did not differ significantly on depression or anxiety severity. Patients, psychiatrists, and evaluators were all blind as to whether the patients’ pills contained paroxetine. After week 8, the blind was broken and placebo patients were offered free medication treatment. ## Measurements The following three symptom measures were administered both at intake and week 8: 17-item modified Hamilton Rating Scale for Depression (HRSD), modified to incorporate atypical symptoms, the 14-item Hamilton Rating Scale for Anxiety (HRSA) and the 21-item Beck Anxiety Inventory (BAI). Both the HRSD and HRSA are clinician-administered measurements, while the BAI is self-reported. To maximize objectivity, clinicians who administered the HRSD and HRSA provided neither psychotherapy nor ADM treatment to participants in this study. Neuroticism was assessed at intake and week 8 using a 12-item scale from the NEO-Five-Factor Inventory (NEO-FFI), a widely-used self-report measure based on the Five-Factor Model of Personality. Among patients who continued with the study, some data were nonetheless missing, mostly anxiety scores at week 8. Because we intended to compare the magnitude of change across multiple measures and treatment conditions, we limited our analyses to patients who completed the HRSD, HRSA, and BAI at both intake and week 8. We excluded 7 placebo and 21 paroxetine patients from analysis (12% and 17% of intake, respectively), because these patients did not complete one or more questionnaires at either time point. Final sample sizes for analysis were 45 participants in placebo and 86 in paroxetine. The patients with missing questionnaire data did not differ significantly on intake depression or anxiety severity from the completer sample we analyzed. For each of these three measures, we divided symptoms into somatic and psychological groups following the classification of Simon et al. of the basic nine *DSM-IV* criteria for major depression. We excluded the HRSD hypochondriasis (#15) and insight (#17) items from our analyses, as they appeared unrelated to current *DSM-IV* criteria for major depression. Following the logic of Simon et al.’s symptom division, we classified symptoms that primarily describe thoughts, moods, anxiety/fears, and interest/behaviour as psychological; symptoms that describe bodily manifestations were classified as somatic (e.g., fatigue, hypersomnia, changes in weight, heart racing). Both libido (HRSD \#14) and "unable to relax" (BAI \#4) were classified as somatic; however, we recognize that these classifications might be challenged. Although we classified the HRSD libido item as somatic (as did Enns et al.), the HRSD interview emphasizes interest in and thoughts about sex, not sexual performance. The BAI item "unable to relax" could possibly refer to cognitive manifestations rather than bodily tension. Nevertheless, placing these two items in the psychological subscales does not change the results; indeed, these items show changes that lie somewhere between the average changes we report for their respective somatic and psychological subscales. Our modified HRSD included the assessment of three atypical symptoms of depression: hypersomnia, weight gain, and appetite increase, along with their typical counterparts (insomnia, weight loss, and appetite decrease). Following Reimherr et al. and DeRubeis et al., only the maximum of each patient's typical/atypical pair (e.g., weight gain or loss) was added into the total of both the HRSD full scale and the somatic subscale. We considered other scoring options, but this scoring method produced the result that was *least* favorable to our hypothesis. For example, if only the typical or the atypical symptoms are included in the subscale, the non-significant medication advantage over placebo is further cut in half. ## Descriptive Analysis Typically, effect sizes in treatment studies are calculated by dividing the difference of two group means by the pooled standard deviation (SD). Unfortunately, standardized effects also pose several problems, especially when the purpose is to compare effects across different measures, samples, or conditions. Effect sizes may be sensitive to particular sample characteristics (such as restricted range) and differences in the reliability of the measures. In addition, standardizing essentially erases scale anchors (e.g., "not at all", "moderately", "severely"), which are still meaningful in average item scores. Reporting change in average symptom scores, however, is complicated by the fact that the rating scales of each symptom differ across the three measures and, in the case of the HRSD, also within the measure. Although HRSD total scores have established ranges corresponding to depression severity, these ranges have not been established for somatic and psychological subscales. To place scores and changes therein on comparable units, we converted scores on the subscales of the HRSD, HRSA, and BAI to Percent of Maximum Possible (POMP) scores. POMP scores are calculated by linearly transforming each participant’s raw score into a percentage of the maximum possible total score of the measure. Descriptive statistics of POMP scores are informative with respect to the range of all possible scores (and the implied severity), and readily comparable to POMP scores of other samples and similar measures. POMP scores do not alter inferential statistics like t- or F-tests and are increasingly applied in clinical research. To facilitate any possible direct comparison with other studies, we reported POMP scores separately for subscales of each measure, rather than combining similar subscales or items across measures. ## Inferential Analysis To test for significant differences among treatment conditions, we completed standard ANCOVA, with treatment condition as the independent variable, the symptom measures at week 8 as dependent variables, and intake symptom measures as a covariate. We completed these tests separately for the HRSD, HRSA, and BAI total scores, and the respective psychological and somatic subscales. Expecting significant differences for the psychological measures as dependent variables, we added neuroticism intake and week 8 scores as covariates. We also completed the reverse analysis to see if treatment assignment would still predict neuroticism reduction (relative to placebo) after controlling for psychological symptom improvement. Finally, we calculated Cohen’s d for placebo vs. active treatments by dividing the difference in the respective least squares means at week 8 by the pooled SD of the respective means. Such effects can be considered large when they are 0.8 and above, medium between 0.5 and 0.8, and small between 0.2 and 0.5. # Results ## Main Effects illustrates the discrepancy in the unique effect of paroxetine on psychological compared to somatic symptoms during the acute phase of therapy. ( also shows intake and week 8 means and standard deviations for the total scale measures and their somatic and psychological subscales). For depression, the difference in score reductions between paroxetine and placebo was more than twice as great for the psychological symptoms as for the somatic symptoms: 10.4% vs. 4.6% (of maximum possible scores). The contrast between the advantage for paroxetine in psychological symptoms vs. somatic symptoms was even stronger for the anxiety scales: 12.2% vs. 1.7% on the BAI and 5.6% vs. 0.9% on the HRSA. shows the main effect statistics for the treatments on the symptom scales. F-tests on all symptom measures indicated that paroxetine significantly outperformed placebo on total HRSD scores (*p* =.004), BAI scores (*p* =.03), and marginally on total HRSA scores (*p* =.054). Paroxetine also significantly outperformed placebo on all psychological subscales, (*p* \<.001 for HRSD, *p* =.01 for the HRSA, *p* \<.001 for the BAI), but it did not show significant advantage over placebo on any of the somatic subscales (*p*s ≥.20). In short, paroxetine outperformed placebo substantially on the psychological subscales, but it did not outperform placebo on the somatic subscales. ## Change in Psychological Symptoms vs. Neuroticism Consistent with previous research, we found that neuroticism correlated more closely with our psychological subscales than with somatic subscales, both at intake (Mean *r* =.32 vs. Mean *r* =.07) and at week 8 (Mean *r* =.45 vs. Mean *r* =.29), with all differences between corresponding psychological and somatic correlations being significant, all *p*s \<.05. In an effort to understand the level of overlap between neuroticism and the psychological symptom scales, we repeated our ANCOVA of treatment assignment (paroxetine vs. placebo), but with additional covariates. First, we found that treatment assignment (paroxetine vs. placebo) had a unique effect on neuroticism reduction, even after controlling for the psychological symptoms (*p*s ≤.02). (See for details). Even when all three psychological symptom measures are entered simultaneously as covariates, paroxetine still significantly outperformed placebo in reducing neuroticism (*p* =.01) suggesting that paroxetine’s effect on neuroticism is not a mere byproduct of psychological symptom change. On the other hand, for the psychological symptoms of HRSD and BAI, the differences between reduction on paroxetine and placebo could not be entirely explained by neuroticism reduction either (*p* =.03 and *p* =.02, respectively). However, when controlling for neuroticism, treatment assignment no longer predicted reduction on the HRSA (*p* =.65). # Discussion In this study, we analyzed symptom improvements in a moderately-to-severely depressed sample during placebo and paroxetine treatments. To our best knowledge, this study is the first to characterize the SSRI advantage over placebo as primarily psychological rather than somatic. In fact, differences in treatment assignment were not significant for any of the somatic subscales. In addition, differences between patients in the paroxetine condition and the placebo condition were much greater on psychological symptoms than on somatic symptoms, particularly for anxiety. Our results are consistent with past research that empirically searched for HRSD items showing the largest treatment effects, typically without considering the nature of the HRSD items. For example, the top five items identified by Faries et al. and Entsuah et al., as well as five of the top six items identified by Santor et al. belong to our psychological subscales of the HRSD. Our results extend these results beyond the HRSD to the HRSA and BAI. More importantly, it now gives a possible theoretical rationale to these past findings. Fournier et al. also analyzed the HRSD items of this clinical trial. Unlike this study, however, Fournier et al. divided the 24-item version of the HRSD symptoms into five clusters using factor analysis. These empirically derived clusters show a somewhat complex relationship with the system of psychological vs. physiological items in this project. For example, their 3-item mood cluster includes depressed mood, anhedonia, and loss of energy. In this project, depressed mood and anhedonia are classified as psychological, but loss of energy is classified as a somatic symptom, consistent with the symptom divisions of Simon et al., Shafer, and Enns et al.. Fournier et al.’s five-item cognitive/suicide cluster included only items from our psychological symptom subscale: suicide and guilt, along with hopelessness, helplessness, and worthlessness from the 24-item HRSD. Consistent with our results and theory, this cluster showed the largest effect size in favor of paroxetine over placebo among the five clusters at week 8; and changes in the other clusters (all of which included one or more somatic symptoms) did not differ significantly from placebo. ## Separate Neurobiological Correlates Our findings indicate that SSRI treatment differentially impacts psychological and somatic symptoms of depression and anxiety, showing much greater specific effects (relative to placebo) on psychological symptoms. One of the potential mechanisms for this finding is that separate neurobiological structures and pathways may be implicated in the expression of psychological versus somatic symptoms. Thus, while depressed mood may be marked by abnormal activation of the medial prefrontal cortex and difficulty concentrating is strongly associated with hypoactivity in the dorsolateral prefrontal cortex, motor retardation may be embodied by dysregulation in the striatum and physical tiredness may be associated with dopamine depletion in nucleus accumbens. Antidepressant medications have been thought to act predominantly on neurovegetative symptoms of depression. However, these effects are primarily associated with the older tricyclic antidepressants. Tricyclics block the reuptake of norepinephrine and serotonin, but more predominantly act on norepinephrine. Further, tricyclics are potent antagonists of histamine-1 receptors, conferring strong sedating properties. By contrast, SSRIs like paroxetine have little if any actions on either norepinephrine reuptake or histamine receptors. Paroxetine is the most potent inhibitor of the norepinephrine transporter of all SSRIs but the actions on serotonin are 10-fold greater on than norepinephrine. Why should SSRIs act preferentially on psychological symptoms of depression? In 1986 Depue and Spoont proposed that serotonin has an effect of constraining both behavioral inhibition and behavioral facilitation systems. This concept was supported subsequently by Knutson et al., who showed a general reduction in negative affect with paroxetine, and by Sheline et al. who showed that the SSRI sertraline inhibited the excess left amygdala response to all faces, particularly fearful faces using fMRI. These effects are also consistent with the observations of Tang et al. of the effects of paroxetine on neuroticism noted above. This inhibiting effect of serotonin on amygdala reactivity and general distress symptoms is attributable to the actions on specific serotonin receptors on inhibitory, GABAergic interneurons. Serotonin has a complex role in CNS function given the large number of serotonin receptors in brain and their sometimes opposing roles. However, the current work continues to support the original Depue and Spoont notion of an overall constraining effect on distress- inducing brain regional activity. ## Treatment Implications Our results suggest that the effects of SSRIs on somatic symptoms are not stronger than that of placebo. Researchers and clinicians may need to look towards additional medications to reduce these symptoms further. While neurological evidence for distinct systems is still limited, it stands to reason that improvement on somatic symptoms (such as fatigue) may require separate treatments from those that address psychological symptoms. One approach may be to target multiple neurotransmitter pathways; duel-acting medications (such as duloxetine and venlafaxine) may be more effective than SSRIs in treating some somatic symptoms of depression, though the advantage over SSRIs in total depression scores is rather modest. Research on ADM treatment of low energy levels specifically in the context of depression has been limited, despite its apparent centrality to major depression. Not surprisingly, low energy (or fatigue) is among the most common residual symptom after acute SSRI treatment. Buproprion, a norepinephrine and dopamine reuptake inhibitor that targets the frontal cortex may be more effective in improving energy levels than standard SSRIs. A meta-analysis of duloxetine trials shows a moderate improvement on the somatic HRSD symptoms of energy and retardation, though this holds primarily for moderate to severely depressed patients. In addition, first-line ADM treatment may be augmented with modafinil or central nervous system stimulants, which promote wakefulness. ## The Role of Neuroticism Our results are consistent with earlier findings that SSRIs may directly target neuroticism, a broad disposition to experience negative emotions that includes no somatic content. Our psychological subscales, therefore, show more efficient empirical and conceptual overlap with neuroticism compared to the full-length symptom scales. Although measures of neuroticism, depression, and anxiety exhibit considerable construct overlap, neuroticism (as a personality measure) is nevertheless crucially different from symptom measures because of the absence of any time-frame context. A future area of investigation would be to further understand the extent to which treatment effects may be attributed to such unique aspects of personality assessment or to the intersection of personality and depression/anxiety symptoms. ## Limitations Our study needs to be replicated using both paroxetine and other SSRIs to determine if the effects are reliable and if they are limited to a single medication or medication class. The use of multiple and more comprehensive measures of depression and anxiety would also have increased confidence in the finding. In addition, we did not directly measure the neurobiological systems underlying psychological and somatic symptoms. We also did not have plasma levels of paroxetine available to verify compliance. It is arguable that splitting each instrument into two subscales increased the type-1 error rate. However, setting alpha at 0.025 (instead of 0.05) for significance testing alters the conclusion of only one test, namely, that paroxetine no longer predicts change in psychological symptoms of the HRSD when controlling for Neuroticism change. Finally, there were a number of dropouts in the trial (11–13%) and our samples were modest and limited to those who completed assessments at both time points. ## Significance Although SSRIs are the most widely prescribed treatment for major depression, the field still needs a comprehensive description and clearly stated formulation of the symptom changes caused by SSRIs. Our results contribute to this effort by suggesting that SSRIs enact improvement beyond placebo mostly for psychological, but not somatic symptoms of depression and anxiety. To further improve somatic symptoms of depression, researchers may need to further explore how to improve treatments of somatic symptoms. # Supporting Information [^1]: RS has received speaking honoraria and grant support from GlaxoSmithKline. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. Because of patient privacy, the data shared is de-identified, and the authors adhere to the definition of the minimal data set, as described in PLOS ONE policies here: <http://journals.plos.org/plosone/s/data-availability> [^2]: Conceived and designed the experiments: RD SH JA RS. Performed the experiments: RD SH JA RS. Analyzed the data: BS TT. Wrote the paper: BS TT RD SH JA RS.

# Introduction Chemical communication is widespread in mammals and can relay information about sex, reproductive state, territory, individual identity, and dominance status. Many species have scent glands specifically for the purpose of transmitting chemical information. Further, successful communication can contribute to an animal’s fitness, and past research has found fitness benefits to marking behavior. Communicating dominance and competitive abilities via olfactory signals can increase mating opportunities, which in turn can increase reproductive success. For example, female laboratory mice prefer males who scent mark more frequently. In addition, Rothman and Mech found that scent-marking was important to courtship for newly paired wolves (*Canis lupus*) as well as reproductive synchrony in established pairs, both of which are important for maximizing reproductive success. For brown bears (*Ursus arctos*), rubbing is a common behaviour and is widely believed to represent some form of communication. Brown bears use a variety of marking postures but standing upright on their hind legs and rubbing their back against a surface is most common. Bears rub on a variety of objects including trees, power poles, and fence posts. As a result of rubbing, rub objects typically develop distinguishing characteristics such as a smooth or discolored rub surface, presence of pedal marks (i.e., path worn by bears to rub object), or presence of bear hair. These characteristics make them easily identifiable in the field. While several studies have described rubbing behaviour \[e.g.,, \], comparatively little research focuses on the reasons behind bear rubbing behavior. There are currently three primary hypotheses regarding why brown bears rub, though they are not mutually exclusive. The first, and perhaps most simplistic, is that rubbing has nothing to do with communication and could simply be a way to remove hair–particularly during the spring-summer shedding period. The second hypothesis is that bears rub to communicate superior competitive ability (i.e., dominance). Third, rubbing might function to signal for mates during the breeding season. Regardless of the mechanism (i.e., mate signaling or dominance communication), if scent marking confers fitness benefits, we might expect a relationship between bear rubbing behavior and reproductive success. We expand on these hypotheses in the following paragraphs. Brown bears might rub to remove hair, particularly during molting seasons. The timing of the molt depends on the bear’s nutritional status because the energy and protein demands of hair growth compete with other physiological processes \[, C. T. Robbins, personal communication\]. Molt can begin in May or can be delayed into late summer or fall depending on the bear’s nutritional intake relative to all other demands. For example, young bears prioritize growth while females with young prioritize lactation over hair growth \[, C. T. Robbins, personal communication\]. Lactating females also might reduce movement to protect their cubs and this can limit their access to the highest quality foods. Thus, molting in adult, lactating females may be delayed relative to when it occurs in adult males \[C. T. Robbins, personal communication\]. In the context of rubbing behavior, we would expect females, particularly those with young, to start rubbing later in the year if the hair removal hypothesis was supported. Alternatively, the dominance or mate signaling hypotheses imply that rubbing has a communication component. Brown bears are wide-ranging, solitary, and have overlapping home ranges. Thus, it seems reasonable that they might use some form of chemical signaling to communicate with conspecifics. Brown bears possess both anal and pedal scent glands, and their secretions are thought to communicate information related to the sex of the animal. These scent glands may be an important component to bear rubbing behavior. For example, sitting, stomping and sniffing behavior as well as urination are common at bear rub objects \[, K. Kendall, personal communication\], allowing for the deposition of chemical compounds. Dominance hierarchies exist in many mammals, including brown bears, and rubbing might be one way by which bears can communicate their dominance. Older, larger, more-aggressive male bears typically outcompete less-dominant males in intrasexual competition for access to females during the mating season. Similarly, more-dominant individuals often outcompete less-dominant individuals for access to food and habitat resources, which can in turn affect fitness. Communicating dominance through olfactory signals can increase mating opportunities through the defence of territories, deterrence of competitors, and advertisement of competitive abilities, which might be attractive to females. In addition to removing hair or communicating dominance, rubbing also might allow brown bears to signal for mates during the breeding season. Although relatively little is known about mate choice for brown bears, both male-male competition and female choice in brown bears have been documented. Like many other mammalian species, female brown bears are the choosier sex because they typically invest more into reproduction, and females rely on visual, acoustic, and/or olfactory signals to inform their mate selection. Scent marking, for example, can convey information about an individual’s condition and genetic quality. There is increasing evidence that odours communicate genetic information that might increase fitness, including information about relatedness and nepotism. Further, females appear to prefer genetically dissimilar males. Thus, there is an interaction between good-genes indicator traits and traits signaling genetic compatibility, although this relationship is not yet well understood. If the most odoriferous males are more successful in securing females and siring offspring that inherit their scent glands, then olfactory cues and the development of scent glands are understandable in the context of sexual selection. Thus, regardless of the mechanism (i.e., mate signaling or dominance communication), if rubbing is sexually selected, rubbing and reproductive success should be positively related for male bears. However, both male and female brown bears rub. Although scent marking is often more commonly associated with males, mammalian females also scent mark, possibly to indicate their receptiveness during the mating season, to solicit male scent marks to test potential mate quality, or to mark their home range. Rubbing also might confer fitness benefits to rubbing female bears. Bears are polygamous breeders, and females often mate with more than one male. Multiple mating might be a female strategy to confuse paternity to reduce the potential of sexually selected infanticide (SSI), whereby males kill non-offspring cubs to bring the female intro estrous for mating. Thus, if multiple mating reduces SSI by confusing paternity and females rub to attract multiple mates, we might expect a relationship between female reproductive success and rubbing. Fitness is dependent on an individual’s ability to successfully reproduce, and reproductive success is initially dependent on the ability to secure mating opportunities. Our objective was to evaluate the relationship between brown bear reproductive success and rubbing behavior. Specifically, we evaluated the prediction that bears that rub more frequently will have a greater number of mates and more offspring. If rubbing is primarily for hair removal, we did not expect to see a relationship between rubbing behaviour and the number of offspring. However, if rubbing is related to communication, either by relaying dominance information or mate signaling, we predicted that we would detect a positive relationship between an individual’s reproductive success and the number of rub objects at which and occasions during which they were detected. Further, we expected to see this positive relationship between reproductive success and rubbing behavior for both male and female brown bears. # Study area Our study area was in southwestern Alberta, Canada, where the northern boundary was trans-Canada Highway 3, the western boundary was the British Columbia border, the southern boundary was Montana, USA, and the eastern boundary encompassed most of the eastern extent of brown bear range. In our study area, the mountains transition abruptly to prairie and agricultural areas. Strong winds (\>100 km/hr) were common, and the climate was characterized by cold winters and warm, dry summers. The study area was a mix of mountainous, forested public lands (48%) under the jurisdiction of the provincial and federal (Waterton Lakes National Park) governments. Oil and gas development as well as forestry and timber harvest were present on public lands. The remaining 52% of the study area was privately owned land, where the predominant land use was agriculture and included both livestock and crop production. All four native large carnivores were present; brown bears, black bears (*U*. *americanus*), cougars (*Puma concolor*), and wolves. Approximately 67.4 (95% CI 50.0–91.1) resident brown bears had home range centers in our study area, and 172 brown bears used the study area at some time each year. # Materials and methods From 2011 through 2014, we identified 899 unique rub objects throughout the study area. For this paper, we included only rub objects, defined as trees, power poles, and fenceposts (*n* = 822); we excluded stretches of barbed wire fence that bears passed through. Each rub object was uniquely numbered, and we attached 4-pronged barbed wire to the rub object to facilitate hair collection and provide discrete sampling units. We determined individual identity from DNA extracted from hair follicles. We included data from 4 years of sampling (2011–2014). The first two years of the project were primarily set-up years and rub objects were visited less frequently than the last two years (2011: 2 sampling occasions on Crown and Park lands only, 2012: 7 sampling occasions on Crown and Park lands and 2 sampling occasions on private lands). During 2013 and 2014 all rub objects were visited 8 times from late May through early November, resulting in 7 sampling occasions. Each sampling occasion for all years was 3 weeks. We also collected hair samples opportunistically from private agricultural lands in the eastern portion of our study area. After each hair collection, we passed a flame over each barb to prevent contamination in the next sampling cycle. Further details of the sampling methods can be found in and. We genotyped 213 individual brown bears (118 males, 95 females) at 24 microsatellite loci, plus the amologenin sex marker. All field methods were completed in accordance with the Canadian Council on Animal Care guidelines and approved by the University of Alberta Bio Sciences Animal Care and Use Committee (Protocol \# AUP00000008). Field permits were granted by Alberta Environment and Parks, and Parks Canada. In a parentage analysis, one of the primary causes of incorrect parent assignment is incomplete sampling of candidate parents. Brown bears in southwestern Alberta are a small part of a larger Rocky Mountain sub-population of brown bears that extends into British Columbia and Montana, USA. Thus, we obtained data from previous non-invasive genetic sampling projects in Alberta, British Columbia, and Montana and included these data in our parentage analysis to increase our likelihood of identifying complete triads (mother, father, offspring). We used 2,043 individual genotypes (977 males, 1072 females) in our parentage analysis. There were 6 cases where sex was unknown, and we analyzed those bears as both potential mothers and potential fathers. We used program COLONY to assign parentage. COLONY uses a full pedigree approach to simultaneously assign parentage and sibship. For the parentage analysis, we specified the following parameters: polygamous males and females, long run length (\~1.9 billion iterations), full-likelihood analysis, medium-likelihood precision, initial proportion of parents in the dataset at 0.4 for each sex, and genotypic error of 0.001. We used known ages (age determined by cementum annuli from extracted teeth of handled bears) to exclude potential parents if they were not at least 2 years older than a potential offspring. Further details of our parentage analysis methods can be found in. While we used genetic data from the entire Rocky Mountain sub-population to ensure our parentage analysis was robust, we assessed the influence of rub behavior on reproductive success only with brown bears that were detected at rub objects in our southwestern Alberta study area (*n* = 55 for females, *n* = 92 for males) because that was our focal area for intensive rub object sampling. For each Alberta bear, we determined the number of offspring, the number of mates, the number of unique rub objects where an individual was detected, and the number of sampling occasions (i.e., 3-week sampling period) during which an individual was detected. We considered all data cumulatively across all 4 years. Using Poisson regression, we first evaluated how the number of mates varied as a function of the number of rub objects at which and occasions during which a bear was detected. Second, we assessed how the number of offspring varied as a function of the number of rub objects at which and occasions during which a bear was detected. We standardized covariates (mean = 0, SD = 1) and examined each explanatory variable independently (i.e., two models for each response variable) because they had high correlations (Pearson’s correlation coefficient r = 0.83 for females, r = 0.93 for males) with each other. We considered the relationship significant if the confidence intervals of the estimate did not overlap zero at the α = 0.05 level. Next, because younger bears will have fewer offspring and mates than older bears and this could influence the relationships we were testing, we calculated a relative age covariate for bears known to have successfully reproduced to help control for the potential that age alone is driving the patterns in our results (*n* = 19 for females, *n* = 27 for males). We determined if a bear was a “parent” (all bears in reduced dataset were parents), “grandbear,” great “grandbear,” or great great “grandbear” for individuals with known offspring and assigned each category a numerical value. We used these generations as indices for relative age because no age data were available for most bears. Again, we analyzed data using Poisson regression using only individuals known to have reproduced and included relative age as an additional scaled covariate in each model. We refer to this dataset as the ‘reduced data.’ Analyses were completed for each sex separately in the statistical software R (R Version 3.6.2, [https://cran.rproject.org](https://cran.rproject.org/)). We calculated exponential effect sizes for each covariate to compare consistency of the reduced data analysis with the full dataset. # Results Males were detected at a greater number of rub objects than females and during a more variable number of sampling occasions. The number of detected offspring ranged from 1–5 for females and 1–10 for males. For males in the full data set, the number of mates for a bear increased with the number of rub objects at which an individual was detected. For each one unit increase in the number of rub objects at which a male bear was detected, the predicted number of mates increased by 1.38 times. Similarly, male brown bears with more mates were detected in more occasions. Likewise, males that had sired a greater number of offspring were detected at a greater number of rub objects and in more occasions. For each additional occasion during which a male bear was detected, the predicted number of offspring is multiplied by 1.37. We observed the same relationships for female brown bears. Females with more mates were detected at more rub objects and in more occasions than females with fewer mates. Likewise, there was a positive relationship between the number of offspring a female brown bear had and the number of rub objects and occasions that bear was detected at and in. For each additional rub object at which and occasion during which a female was detected, the predicted number of offspring increased by 1.42 and 1.55 times respectively. Analysis of the reduced data set including only bears that were known to have successfully reproduced (*n* = 19 for females, *n* = 27 for males) led to reduced statistical significance in our hypothesized relationships. The confidence intervals of the relative age covariate overlapped zero in several models, and the relative age covariate was not significant in any model. The positive relationship between the number of mates or offspring a bear had and the number of rub objects and number of occasions was consistent with the full data models (with the exception of model f1), but the only significant relationship was observed in model m1.O. Similarly, effect size was lower for these models in all models for both males and females. # Discussion Bears with a greater number of mates and a greater number of offspring were detected at more rub objects and during more occasions. Thus, our data supported our prediction of a positive relationship between bear rubbing behavior and reproductive success. Our results allow us to rule out hair removal as the sole motivation for rubbing because if this were the case, we would not expect a relationship between bear rubbing and reproductive success. Nevertheless, hair removal still could be a component of rubbing behavior. Although we cannot differentiate between rubbing for mate signaling and rubbing for dominance, both likely play a role in bear rub behavior. Detections of male brown bears at rub objects are typically highest during the breeding season, suggesting that male bears rub to signal for mates. However, as Lamb et al. further hypothesize, bears might rub throughout the year to establish and maintain dominance hierarchies. Because bears rub throughout the active season, we can rule out mate signaling as the sole reason for rubbing behavior. Rubbing for mate signaling can result in increased mating opportunities, higher- quality mates, and ultimately increased reproductive success and fitness. And for females, securing multiple mates might reduce the potential of SSI. Female promiscuity in mammals is a counterstrategy to SSI because by mating with multiple males, the female can confuse paternity of her offspring and potentially reduce predation by infanticidal males. Multiple-male mating has been observed in over 130 mammalian species, and females of some species will actively solicit copulations from multiple males. Our results indicate that females that had successfully reproduced were detected at more rub objects than females without offspring. Thus, the mechanism behind rubbing behavior in females might go beyond mate advertisement; actively soliciting multiple male matings might confer fitness benefits to the rubbing female. Further, females that rub beyond the mating season might be relaying information on their quality as a mate that might inform future mating possibilities (e.g., whether reproduction was successful). Contrary to Clapham et al.’s conclusion that females do not gain fitness benefits from rubbing, our results indicate that females with offspring were detected at a greater number of rub objects and during more occasions than females without offspring, suggesting there might be an individual fitness benefit to rubbing by females. Female bears are induced ovulators (i.e., ovulation occurs after hormonal, physical, or behavioral stimulation), and multiple paternity of offspring in the same litter is possible. For example, Shimozuru et al. found that 14.6–17.1% of all brown bear litters evaluated were sired by multiple males. Thus, after mating, females might have the opportunity to choose among sperm of different males (cryptic female choice). Male-male competition, however, also can occur during this post-copulatory time via sperm competition. Thus, if multiple mating reduces SSI by confusing paternity and females rub to attract multiple mates, this might explain the positive relationship between female reproductive success and rubbing. Indeed, polygamous females can be problematic for males because copulation does not assure paternity. Thus, male brown bears are faced with a choice–guard the mated female for the duration of her oestrus, thereby assuring paternity but losing other mating opportunities, or solicit more copulations and potentially sire more offspring, but leave paternity to cryptic female choice and/or sperm competition. This decision might depend on the availability of breeding female bears. When breeding females are scarce, it is likely in the males’ best interest to mate guard, but when breeding females are common, seeking additional copulations could be advantageous. We acknowledge that a limitation of our non-invasive hair sampling data is that we do not know the age of the bears in our analysis. Thus, we were not able to determine if a bear with no mates or offspring is because that bear was not successful in securing mates or because the bear was not of reproductive age. To address this as best as possible, we calculated a relative age covariate and included this in our secondary analysis. Datasets with relative age had less power because of lower sample sizes and thus, fewer degrees of freedom. The relative age covariate was not significant in any model and, for males, the same relationships between rubbing, mates, and offspring persisted. Further, previous research has found that subadults mark less frequently than adult brown bears, and the function of scent marking by young bears remains unclear. Because young bears rub less frequently than adults, it is unlikely that very many zeros in our data set are from bears of non-reproductive age. Spatial clustering of rub objects also might play a role in our observed results. The ideal free distribution predicts that higher quality bear habitat should have a higher density of bears. In turn, if bears rubbed more in higher quality areas, this spatial clustering of rub objects in high-quality habitats could be the ultimate cause of the relationship we found between rubbing and reproductive success. This line of reasoning implies a positive relationship between rubbing and bear density. However, recent work in the U.S. portion of this population did not find a consistent relationship between annual rub tree catch per unit effort and increasing density. Indeed, Lamb et al. hypothesized the opposite–rubbing might be inversely related to population density. While local habitat conditions could influence use of rub trees \[e.g., \], the lack of a consistent pattern with density suggests that spatial clustering of rub objects is not the primary driver of our observed patterns. In summary, our data suggest a fitness component to rubbing behavior. We conclude by proposing a new alternative hypothesis for consideration: female brown bears use the information obtained from olfactory cues of rubbing males throughout the season to choose offspring paternity. Data to examine this hypothesis are beyond the scope of our study but, if supported, this hypothesis would help explain the relationship between reproductive success and brown bear rubbing behavior. Because brown bears have delayed implantation and multiple paternity of offspring in the same litter is possible, females might be able to choose among sperm of different males. If so, female brown bears must rely on cues to determine which of the males that she has mated with will sire her offspring. Females might obtain this information from the olfactory and chemical signals deposited by rubbing males throughout the active season. These results indicate that rubbing is an adaptive behaviour in brown bears. # Supporting information We thank Wildlife Genetics International for their genetic analysis of the hair samples used in this project. We also thank Nate Mikle for his work on the parentage analysis using COLONY. In addition to the funding organizations mentioned, additional in-kind and logistical support was provided by the Blackfeet Nation; Blood Tribe Land Management; Confederated Salish and Kootenai Tribes; Hab-Tec Environmental; Montana Fish, Wildlife, and Parks; Montana Department of Natural Resources and Conservation; National Park Service; Northwest Connections; and the U.S. Fish and Wildlife Service. Over 200 people assisted in the collection of brown bear hair samples; this project would not have been possible without their contributions. In Alberta, over 70 landowners provided land access and/or collected opportunistic hair samples; we are thankful for their support. We thank Alberta Environment and Parks, British Columbia Ministry of Forests, Lands, and Natural Resource Operations, the U.S. Geological Survey, and the Foothills Research Institute Grizzly Bear Program for the DNA data sets provided for use in our parentage analysis. Thanks to Kate Kendall, Garth Mowat, and Gordon Stenhouse for leading hair collection in their study areas. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. [^1]: The authors have read the journal’s policy, and the authors of this study have the following competing interests to declare: AM is the owner and operator of Winisk Research and Consulting. This does not alter our adherence to PLOS ONE policies on sharing data and materials. There are no patents, products in development or marketed products associated with this research to declare.

# Introduction Spinal muscular atrophy (SMA) is a genetic recessive disorder caused by mutations in the survival of motor neuron 1 (*SMN1)* gene on chromosome 5q, leading to motoneuron loss and subsequent muscular atrophy and weakness. Classically SMA is subdivided into different types according to maximum motor function achieved, with type I to III being the most frequent forms with pediatric-onset. In type I the onset is before 6 months of age and the ability to sit independently is not achieved. In type II the onset is between 6 and 18 months; type II children achieve the ability to sit but not to walk independently. Type III patients achieve the ability to walk independently and the onset is after 18 months \[–\]. The diagnostic process is thought to be relatively easy because of the combination of typical clinical signs. Generalized weakness, more severe in the legs than in the arms, with a proximal more than distal distribution, associated with no facial weakness and severe generalized hypotonia, absent reflexes and a typical respiratory pattern are strongly suggestive of type I SMA and should direct the clinician to consider performing genetic testing, without the need to perform muscle biopsy or other investigations. Similarly, in type II patients the observation of similar motor and respiratory pattern, even if milder, is sufficient to consider diagnosis of SMA and to proceed directly with genetic testing. In type III, the milder signs may be less specific and additional investigations, such as electromyography are often used to confirm the clinical suspicion. The genetic testing is simple as the gene is relatively small and approximately 95% of the mutations are represented by deletions in the exons 7 and 8. Despite the typical clinical features and the ease in performing the genetic analysis, a recent review of the diagnostic process in SMA reports that this is not always so straightforward and that there is often a delay between the onset of clinical signs and confirmed diagnosis in all types of SMA. Achieving a diagnosis is always important, not only for genetic counseling but also to implement disease-specific standards of care. The recent advent of new therapeutic options, already commercially available, has further increased the need to confirm diagnosis as early as possible as early treatment has been associated with better outcome. The aim of the present study was to define the age of diagnosis in the three main types of SMA with pediatric-onset and the timing between recognition of clinical signs and a confirmed genetic diagnosis over the last 2 decades. We were also interested in identifying the most frequent signs that raised the suspicion for SMA and to assess the investigations performed as part of the diagnostic pathway. # Materials and methods The study included 5 tertiary Italian Neuromuscular Center involved in the diagnosis and follow- up of SMA patients (2 located in the northern part of Italy, 2 in the center and 1 in the south). The study was approved by the Ethics Committee of each centers (Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Bambino Gesù Hospital, Gaslini, Nemo Milano, Messina). Parents of participants and patients were informed that the data collected as part of our routine clinical assessment were going to be used anonymously for an observational study defining the natural history of the diseases and they all gave written consent. Data from the clinical charts were collected. As regular genetic testing for SMA has been available since 1995 it was decided to include patients born after 1996 in order to have more consistent results. All patients with a confirmed genetic diagnosis of SMA (type I, II, III) with mutations in the *SMN1* gene, including deletions, duplications and point mutations, born between January 1996 and July 2019 in whom anamnestic and clinical reports were available were included in this study. The following information was collected: date of birth, family history of SMA, age at symptoms onset, the person who suspected the diagnosis, presenting sign or symptom, age at diagnosis, the interval between SMA symptom onset and diagnosis, type of medical investigations conducted in order to obtain the diagnosis. The patients who had a positive family history of SMA were not included in statistical analysis. “First symptom” was defined as any sign or symptom reported by a physician or a parent/caregiver that was suspicious for SMA. Age at diagnosis was defined as the date of the genetic test. Descriptive statistic was conducted to analyze the data. The Mann-Whitney U test was used to compare age at diagnosis between patients born in the decade 1996–2006 and those born later. An additional analysis was used to establish possible differences with children born in the last 7 years. A similar approach was also used to compare the interval between onset of clinical signs and diagnosis in the same subgroups. The level of statistical significance was defined for p-value \<0,005. # Results We identified 494 children diagnosed with SMA in our records who were seen in our clinic after 1996. Of these, 5 were excluded because not all the appropriate information was available. Nine children (1 type I, 4 type II, 4 type III) were not included in the analysis because diagnosis was received by a positive familial history (e.g. siblings). The final cohort included 480 patients, 191 affected by SMA type I, 210 by type II and 79 by type III. ## Symptoms onset ### Type I (n = 191) The mean age at symptom onset was 2.75 months (range 0–10 months, SD ±1.96). One hundred five patients (54.97%) had symptom onset before 3 months of age, 68 (35.60%) between 3 and 6 months and 18 (9.42%) after 6 months. In 110 children the symptoms were first recognized by parents (62.83%), and in the remaining 71 (37.17%) by a pediatrician/child neurologist. ### Type II (n = 210) The mean age at symptom onset was 10 months (range 3–24 months, SD ±3.96). Twelve patients (5.71%) had symptom onset before 6 months of age, 188 (89.52%) between 6 and 18 months and 10 (4.76%) after 18 months. In 157 children the symptoms were first recognized by parents (74.76%), 2 by teachers (0.95%), and in the remaining 51 (24.28%) by a pediatrician/child neurologist. ### Type III (n = 79) The mean age at symptom onset was 32 months (SD ±37.92, range 9 months-15 years). None (0%) had symptom onset before 6 months of age, 38 (48.10%) between 6 and 18 months and 39 (49.37%) after 18 months. Of the 38 patients who had symptom onset between 6 and 18 months, only 2 (5.26%) had onset before 12 months, 17 (44.74%) between 12 and 15 months, 19 (50.00%) between 16 and 17 months. Of the 39 patients who had symptom onset after 18 months, 8 (20.51%) had onset before 24 months, 15 (38.46%) before 36 months, 7 (17.94%) before 48 months and 9 (23.08%) after 48 months. In 67 children the symptoms were first recognized by parents (84.81%), the remaining 12 (15.18%) by a pediatrician or other health-related professionals. In 2 children the concern was raised following admission to the ER for unrelated causes (severe headache). describes the first symptoms subdivided by SMA type. ## Diagnostic pathway summarize the assessments performed in the SMA patients before performing genetic testing. ## Age at diagnosis ### Type I (n = 191) The age at diagnosis ranged between 10 days and 13.23 months, mean age: 4,70 months (SD ±2.82). In 63 patients (32.98%) the diagnosis was achieved before 3 months, in 66 (34.55%) between 3 and 6 months and in 62 (32.46%) after 6 months. The mean time between symptom onset and genetic diagnosis was 1.94 months (SD±1.84; range: 0–10.3 months). ### Type II (n = 210). The age at diagnosis ranged between 5 months and 4 years 5 months (mean age: 15.6 months (SD±5.88). In 2 patients (0.95%) the diagnosis was achieved before 6 months, in 140 (66.67%) between 6 and 18 months and in 68 (32.38%) after 18 months. The mean time between symptom onset and genetic diagnosis was 5.28 months (SD±4.68; range: 0–35 months). ### Type III (n = 79) The age at diagnosis ranged between 10 months and 18 years (mean 4.34 years; SD±4.01). In none (0%) the diagnosis was achieved before 6 months, in 4 (5.06%) between 6 and 18 months and in 75 (94.94%) after 18 months. The mean time between symptom onset and genetic diagnosis was 16.8 months (SD±18.72; range: 0–102 months). summarizes statistical analysis performed between age at diagnosis, interval between onset of clinical signs and diagnosis. # Discussion The analysis of the onset of clinical signs in our large cohort of SMA patients confirmed previous reports that in the great majority of type I patients clinical signs are generally identified before 6 months of age, and less than 10% were identified after 6 months. The patients with relatively later-onset had all achieved head control and can be classified as 1.9 according to the Dubowitz decimal classification. These findings expand recent observations obtained in a smaller cohort reporting that at the mildest end of the spectrum in type I the onset may occur after the age of 6 months²⁹. As expected, the great majority of type II patients had a clinical onset between 6 and 18 months with approximately 5% of outliers at each end of the spectrum. In contrast, the number of type III patients with onset of clinical signs after 18 months was approximately 50% of all type III. In the patients in whom the signs were identified before 18 months, this generally occurred between 13 and 18 months, after independent ambulation had been achieved. The main concern reported by the parents/caregiver/HCPs was persistent unsteady gate or frequent falls a few months after ambulation had been acquired. Although these results confirm that most patients will fall within the criteria used for the SMA classification, this is less true for the type III with earlier onset (IIIA) and the possibility to have outliers is present in all types. A recent review reported a diagnostic delay in SMA, measuring the mean interval between the age at clinical onset and the age at genetic diagnosis based on a number of studies published before 2015 in which these data were available. The review showed a progressive increase in the interval from type I to type III. The results were not easily comparable because of the different design of the studies but some differences could be noted. In our type I patients the mean interval between clinical onset and diagnosis was 1.94 months. This value is lower than the mean value reported in the review. We also found that, despite having a similar age at clinical onset, in our type II patients the mean interval was also much shorter than in the type II patients in the published review. The results in the type III cohort cannot be easily compared as this is a more heterogeneous group and the results may be dependent on the percentage of patients with earlier or later onset and this information was not available in the previous review paper. Our type III cohort was subdivided into IIIa and IIIb according to whether the diagnosis was made before or after the age of 3 years. This was helpful to highlight that the diagnostic interval between IIIa and IIIb is much shorter. In the IIIa patients, parents were often concerned about persistently unstable gait or ‘clumsiness’ a few months after ambulation had been achieved and this prompted further investigations. In older patients in whom the clinical signs became obvious after the age of 3 years, the signs were often milder and less specific, and parents reported a ‘wait and see’ attitude, a longer time before referral to a specialist, that justifies the longer interval before diagnosis. Interestingly although type III had more delay than the other groups, this was the only clinical group in which a significant improvement was noted in recent years compared to the previous decade. In the other SMA types, the interval between clinical onset and diagnosis in the overall cohort was not significantly different between the two decades explored or between the first decade and the last 7 years. The choice of selecting the last 7 years, from 2012, was related to a possible increased awareness and education in the last few years, following the implementation of the standard of care recommendations published in 2007 and the advent of active clinical trials. While the time to perform the genetic test, from blood taken to results is likely to have improved in the last few years, reflecting a wider availability and improvement of the diagnostic services, the overall time between onset of clinical signs and genetic diagnosis did not change. This was particularly true for the patients in whom the diagnostic workout was initially performed in local hospitals. Although our study includes only tertiary care centers, a proportion of children were referred to us only after diagnosis. In these patients, a diagnosis of SMA was often considered only after an extensive number of tests ruling out other diseases had been performed and the delay may also reflect the time it takes to get a referral to and appointment with a specialist. As already reported in the previous review the interval increased from type I to type III. A possible explanation is that with decreasing severity, the clinical signs are milder and can be less specific, as also proven by the highest number of investigations such as brain MRI and EMG, performed more often in type III, and progressively less in type II and type I. Similarly, muscle biopsy, that used to be one of the key diagnostic elements before the gene was identified and is not part of the current recommendations for diagnosis and care, was more often performed in type III than in the other 2 types. The only tests that were performed more regularly in type I than in the milder forms were metabolic tests and other genetic tests, mainly tests for Prader Willi, Pompe disease and Myotonic Dystrophy, that in some centers are routinely performed in weak floppy infants. # Conclusions In conclusion, our results showed that age of onset of clinical signs can identify the great majority of type I and type II cases, but this is not always true for type III, as clinical signs were often identified after ambulation was achieved but before the age of 18 months. Our results also showed overall shorter diagnostic intervals compared to previous studies. The interval was shorter in type I, this probably reflects the wider availability of genetic tests even in peripheral hospitals but also partly as the result of increased awareness of the disease in the last few years following the advent of new therapies. These findings should be interpreted with caution as they reflect the experience in our country, and they may not be necessarily extrapolated to countries where healthcare systems are different. The advent of new therapies and the promising results obtained in presymptomatic patients are paving the way to the possibility of newborn screening that is already in place in a few countries and is likely to have improved time to diagnosis in the future. [^1]: The authors have declared that no competing interests exist. [^2]: ‡ MCP and GC contributed equally as co-first authors.

# Introduction S100B is a Ca²⁺ binding protein which is abundantly and constitutively expressed in the brain by astrocytes where it has both autocrine and paracrine effects on neurons and glia. To a lesser extent it is also produced by other cell types such as monocytes, macrophages, microglia and T cells. It has both intracellular and extracellular functions. Intracellular S100B is involved in cytoskeletal interactions, Ca²⁺ homeostasis and regulation of enzyme activity. S100B can also be secreted and extracellular activities are less clear cut and may depend on concentration. At nanomolar concentrations S100B is reported to be beneficial, supporting neuronal survival, growth and function. However at higher (micromolar) concentrations there is evidence that S100B can cause apoptosis in neurons and has effects similar to a pro-inflammatory cytokine on astrocytes and microglia. S100B mediates this response through interaction with the receptor for advanced glycation end products (RAGE). RAGE is a multiligand cell receptor which upon ligand binding activates NF-κB via different signalling pathways. S100B has been shown to be involved in neurodegeneration and brain injury with elevated levels seen in Alzheimer’s Disease, Parkinson's Disease, Down syndrome and stroke patients and it may act as a damage associated molecular pattern (DAMP) protein. S100B has also been associated with chronic inflammation for example in rheumatoid arthritis, diabetes and cystic fibrosis. However, it is not clear whether S100B as a DAMP has a fundamental role as a pro-inflammatory mediator, inducing or exacerbating inflammation in these situations or whether it may play a role in dampening inflammation. There is evidence to suggest that inflammation may be enhanced in these conditions by the action of S100B on macrophages/microglia. *In vitro* studies on microglia cultured from murine BV-2 microglial cell lines have suggested that excessive production of S100B by astrocytes might lead to production of TNF-α, IL-1β, NO and COX-2 by microglia and subsequent enhanced inflammation. S100B has also been shown to have a pro-inflammatory effect on the J774 macrophage cell line, for example, stimulating nitric oxide production, inducible nitric oxide synthase (iNOS) protein transcription and TNF-α production in a concentration- dependent manner. However, the influence of S100B on production of other pro- inflammatory cytokines by macrophages, and in primary macrophages has not been studied. The aim of the study was to clarify the involvement of S100B in inflammation and determine whether its influence is likely to be via macrophages. We have therefore examined its effects *in vitro*, on both a macrophage cell line (RAW 264.7) and on primary macrophages, and also *in vivo*, in a model of retinal inflammatory disease in which pathogenesis is highly dependent on macrophage infiltration (Experimental Autoimmune Uveoretinitis, EAU). S100B has long been known to be present within the eye particularly associated with Mϋller cells and is involved in signal transduction in the photoreceptor-bipolar cell region. RAGE has also been shown to be present in the retina and expression is increased in EAU. *In vitro* studies showed that S100B had direct effects on macrophages, enhancing CCL22 and IL-1β expression in particular. EAU disease severity was shown to be reduced in mice in which S100B was deleted and this was related to a reduction in IL-1β and CCL22 expression in the retina. This suggests that S100B may play an active role in enhancing inflammation via its action on macrophages. # Materials and Methods ## Animals Animal studies were conducted under a project licence granted by the UK Home Office according to the Animals Scientific Procedures Act 1986. The project was also subject to the University of Aberdeen’s Ethical Review Process and was approved by its Animal Welfare and Ethical Review Body, in accordance with the University Code of Practice for Research Involving the Use of Animals. C57BL/6 wild type (WT) control mice and *S100B* knockout (S100B KO) mice on a C57BL/6 background were supplied by the Medical Research Facility, University of Aberdeen. S100B KO mice were originally established and obtained from the RIKEN Brain Science Institute, Japan. A breeding colony was established in the Medical Research Facility, University of Aberdeen (UK). To confirm the absence of S100B, genotyping was routinely done using PCR as previously described by Nishiyama *et al*., 2002. Mice were gender and aged matched and used between 8 and 12 weeks old. ## Macrophage cell culture and S100B treatment To determine the effect of S100B on macrophages, a murine macrophage cell line RAW 264.7 (ATCC, Manassas, USA) was cultured in DMEM plus 10% FCS and 1% penicillin and streptomycin at 37°C 5% CO₂. Primary macrophages were also cultured from bone marrow (BMDM) of C57BL/6 mice, aged between 8 and 12 weeks. Tibias and femurs were taken and excess tissue removed before sterilizing in 70% ethanol and rinsing in PBS. Bone marrow was flushed out using DMEM/F12 and cells were grown in 6 well plates in DMEM/F12, plus 10% FCS and 1% penicillin and streptomycin, containing 15% L929 conditioned DMEM. After 6 days of culture the cells were characterised as bone marrow derived macrophages by flow cytometry showing high expression of F4/80 (87.4±1.82%; n = 10, ±SEM), CD11b (94±0.35%; n = 10, ±SEM) and low expression of GR1 LY6C (6.63±1.04%; n = 10, ±SEM) and FLT3 (0.626±0.20%; n = 10, ±SEM). When the RAW 264.7 cells reached 80% confluence or on day 6 of primary bone marrow culture, where cells were also approximately 80% confluent, the media was replaced with serum-free media overnight before addition of bovine S100B protein (Sigma Aldrich, Dorset, UK). The S100B preparation was filtered (0.22 μm) and underwent endotoxin removal treatment using endotoxin removal beads according to the manufacturer’s protocol (Miltenyi Biotec, Bisley, Surrey, UK). LAL endotoxin test (ToxinSensor,GenScript. NJ, USA) confirmed that the endotoxin levels in the S100B protein preparation were minimal at \< 6 pg/ml following this treatment. S100B was heated to 95°C for 15 min and used as a control to confirm no endotoxin contamination. ## Cell viability To determine whether 2 μM S100B was influencing cell viability, morphology was checked and an acid phosphatase cell viability assay done. RAW 264.7 or BMDM were treated with S100B for 24 h. Cells were imaged using a Widefield Ziess Observer microscope (Carl Zeiss, Oberkochen, Germany) and an A1R confocal microscope (Nikon Instruments BV Europe, Amsterdam, Netherlands). For acid phosphatase assay, cells were washed in PBS and incubated with 100 μl of substrate, sodium acetate buffer containing 5 mM p-nitrophenyl phosphate substrate (Sigma-Aldrich, UK). One substrate tablet was dissolved in 2.4 ml sodium acetate buffer (0.1 M sodium acetate and 0.1% Triton). Cells were incubated for 2 h at 37°C. After the incubation period the reaction was stopped by adding 25 μl of 0.1 M sodium hydroxide. The plate was read at 405 nm in a microplate reader (Dynatech MR5000, Dynex Technologies, UK). ## RAGE expression RNA was extracted from RAW 264.7 and BMDM cells using TRIzol reagent (Invitrogen, Paisley, UK) following manufacturer’s instructions. RNA was quantified (NanoDrop 1000, ThermoScientific, Wilmington, DE, USA) and cDNA generated using superscript II (Invitrogen Paisley, UK) according to manufacturer’s instructions with 2.0 μg total RNA in each reaction and oligo-(dT) priming. The reverse transcription was performed at 42°C for 50 min and 95°C for 10 min. PCR was in a 25 μl total volume reaction mixture containing 12.5 μl Go Taq DNA polymerase reaction buffer (Promega, Southampton UK), 9.0 μl nuclease free water, 0.4 μM forward primer, 0.4 μM reverse primer and 1.5 μl cDNA template. RAGE primers were forward primer, 5’-CAGCATCAGGGTCACAGAAA-3’, and reverse primer, 5’-CTGGTTGGAGAAGGAAGTGC-3’. β-Actin was used as a reference gene, using forward primer, 5’-TGTGATGGTGGGAATGGGTCA-3’, and reverse primer, 5’-TTTGATGTCACGCACGATTTCC-3’. Primers were designed to be intron spanning using the NCBI and primer 3 software. The PCR program consisted of, 1 cycle of denaturation at 95°C for 3 min followed by 30 cycles of 95°C for 30 s, 63°C for 1 min and 72°C for 5 min before a final extension at 72°C for 5 min. The amplified products were analysed by gel electrophoresis on a 1.8% agarose gel containing ethidium bromide. PCR products were visualised under UV light. The resulting cDNA product was visualised at approximately 350 bp for RAGE and 500 bp for β-Actin. ## Western blot To confirm RAGE protein expression, western blot was carried out on BMDM. BMDM were collected, washed with PBS, centrifuged and re-suspended with 100 μl of 1% NP-40 PBS lysis buffer with the addition of protease inhibitor (cOmplete Protease Inhibitor Cocktail tablets, Roche Products Ltd, Welwyn Garden City, UK) and vortexed and placed on ice for 10 min before a further vortex for 20 s. The samples were then centrifuged at 162 g for 15 min and supernatant removed and stored at -80°C. Protein concentration for the supernatant was determined using BCA protein assay (Pierce BCA Protein assay kit, ThermoScientific). Approximately 60 μg in LDS sample buffer without reducing agents (NuPAGE LDS sample buffer, Life Technologies, Thermo Fisher Scientific) was applied to a non-reduced 12.5% PAGE gel. Mouse lung tissue was homogenised by passing the tissue through a 25G needle and syringe in 1ml NP-40 PBS lysis buffer with protease inhibitor and then treated as for the BMDM and used as a positive control. To identify protein band size, 8 μl MagicMark XP Western Protein Standard (Life Technologies) was also loaded onto the gel. Proteins were transferred to a PVDF membrane and blocked in 5% non-fat dry milk (AppliChem, Darmstadt, Germany) in 0.1% Tween in PBS for 1 h followed by incubation with 2 μg/ml of Rat-Anti-Mouse RAGE monoclonal antibody (R&D Systems, Abingdon, UK) for 1 h at room temperature. The membrane was then washed 3 x 5 min with 0.1% Tween PBS and incubated with a HRP-conjugated goat anti-rat secondary antibody (1 in 10000, Santa Cruz Biotechnology, Inc, Heidelberg, Germany) for 1 h at room temperature. The membrane was washed again in PBS Tween as previously with an additional 15 min wash. Protein bands were visualised by enhanced chemiluminesence detection system (Pierce ECL Western Blotting Substrate, ThermoScientific) by mixing 1 ml ECL reagents 1 and 2 for 1 min and reading on a luminescent image analyser (ImageQuant LAS-4000 Mini, GE Healthcare Europe, Freiburg, Germany). ## Pro-inflammatory cytokine and chemokine PCR Array To compare pro-inflammatory cytokine and chemokine gene expression in RAW 264.7 cells cultured for 6 h with or without 2 μM S100B, RNA was extracted using an RNeasy Micro kit (Qiagen, Manchester, UK Ltd) according to manufacturer’s instructions. RNA quality was checked using an Agilent RNA 6000 Nano kit (Agilent Technologies Ltd, Wokingham, Berkshire, UK) and only RNA samples showing clear 18S and 28S peaks, with the 28S:18S ratio higher than 2 were accepted for analysis. RNA was also extracted from retinas from S100B KO or WT mice with EAU. Retinas were removed from mice which had been carefully matched for disease level using TEFI grading and RNA was similarly prepared and checked as described for the RAW 264.7 cells. From the total RNA extracted, 1 μg was used in cDNA synthesis, using RT² first strand kit (Qiagen, Manchester, UK Ltd). The cDNA was then used in a 96 well RT² Profiler PCR Array for inflammatory cytokines and chemokines (SABiosciences, Qiagen, Manchester UK Ltd) all following manufacturer instructions. A Light Cycler 480 (Roche) was used for the PCR analysis. Cycle threshold values (C_T) were converted into fold change using Qiagen data analysis software. The RT² Profiler PCR Array for inflammatory cytokines and chemokines has been shown to be a robust method for this type of comparison and to give a good indication of the cytokines and chemokines which warrant further study at the protein level. ## Real-time PCR to confirm CCL22 and IL-1β gene expression To confirm the PCR array results from RAW 264.7 cells, which identified *CCL22* and *IL-1β* as responding to S100B, quantitative real-time PCR was done and the time and dose response to S100B determined. RNA was extracted and quantified and cDNA generated as described above. Real-time PCR was set up with 5 μl SYBR Green 2x concentration master mix (Roche), 1 μl forward primer (5 μM), 1 μl reverse primer (5 μM) and 3 μl cDNA and analysed using a Light Cycler 480 (Roche). Primer efficiency was determined by preparation of a standard curve (10 fold) using pooled cDNA, and primers with an efficiency of 1.85–2 were accepted for use. The primer sequences used were designed by Roche universal probe library to be intron spanning and were for CCL22, forward 5’-TCTTGCTGTGGCAATTCAGA-3’ and reverse 5’-gagggtgacggatgtagtcc-3’; and for IL-1β, forward 5’-agttgacggaccccaaaag-3’ and reverse 5’-agctggatgctctcatcagg-3’. Cycle threshold values were normalised to GAPDH gene expression using GAPDH primers, forward 5’-gggttcctataaatacggactgc-3’ and reverse 5’- ccattttgtctacgggacga-3’. The PCR program was 95°C for 5 min followed by 40 cycles of 95°C for 10 s, 60°C for 10 s and 72°C for 10 s. Results are expressed as relative fold change as calculated by the delta delta C_T method. ## Analysis of CCL22 production by ELISA To confirm that S100B increases CCL22 production by RAW 264.7 macrophages or BMDM, supernatant was collected after 24 h incubation with or without S100B, centrifuged at 167 g for 5 min at 4°C and stored at -80°C. Supernatants were analysed for CCL22 using a DuoSet ELISA kit (R&D Systems) as recommended by the manufacturer. ## Analysis of IL-1β pro-form production by flow cytometry For studies on IL-1β pro-form production, RAW 264.7 cells or BMDM were treated with 2 μM S100B and 5 μg/ml Brefeldin A for 4 h (Sigma Aldrich, UK). Cells were harvested in 2% FCS in PBS for flow cytometric analysis of intracellular pro- IL-1β. Cells were incubated on ice for 20 min with Fc block (4 μg of CD16/CD36, BD Biosciences, Oxford, UK) before surface staining with PerCP-CY5.5 rat monoclonal anti-mouse CD11b (BD BioSciences, 0.5 μl of 200 μg/ml). Cells were then fixed and permeabilised in BD BioSciences fixative and permeabilisation buffer, following manufacturer guidelines. PE-conjugated rat anti-mouse IL-1β pro-form (eBioscience Hatfield, UK at 0.06 μg per test in permeabilsation buffer) was added and incubated at room temperature for 30 min in the dark. Cell suspensions were then washed twice in permeabilisation buffer, mixing before each wash, with a final wash in FACS buffer (2% v/v FCS in PBS containing 2% w/v sodium azide). Cells were re-suspended in 300 μl FACS buffer before data acquisition on the LSR II flow cytometer (BD BioSciences). Analysis of FACS data was carried out using FlowJo Software, (Treestar, OR, USA). ## EAU induction EAU was induced using a standard method. Both S100B KO and C57BL/6 mice were immunised subcutaneously in 2 thighs (50 μl/thigh) with a total concentration of 500 μg per mouse retinol binding protein-3 peptide 1–20 (RBP-3₁₋₂₀; GPTHLFQPSLVLDMAKVLLD; New England Peptide LLC, MA, USA) emulsified with Complete Freunds Adjuvant (CFA-H37Ra, BD Biosciences,) containing additional 2.5 mg/ml *Mycobacterium tuberculosis* (1:1 v/v, BD Biosciences). An additional 100 μl, 10 μg/ml *Bordetella pertussis* toxin was administered by intraperitoneal injection. Mice were sacrificed by CO₂ asphyxiation and cardiac puncture and eyes collected. ## EAU clinical grading and histological scoring Disease progression was followed using Topical Endoscope Fundal Imaging (TEFI) at day 15, day 21 and day 24 post peptide immunization (pi) to compare S100B KO and C57BL/6 WT mice. Mice were anaesthetised, pupils dilated using 1% (w/v) tropicamide and 2.5% (w/v) phenylephrine hydrochloride (Bausch & Lomb Minims, Chauvin Pharmaceuticals Ltd, London, UK), and Viscotear liquid gel (Novartis Pharmaceuticals, Frimley, UK) was applied to each eye to provide good endoscope contact to cornea and avoid eyes drying. Fundus images obtained were clinically graded on a scale 0–4 by scoring changes in retinal infiltrate and lesions, retinal vessels and clarity of the optic disc according to Xu *et al*.,2008. At approximately peak disease, day 24 pi, mice were culled and eyes were snap frozen in OCT compound (Tissue-Tek, Agar Scientific, Essex, UK). Eyes were serially sectioned at 6 μm and sections fixed in acetone, haematoxylin stained and dehydrated in ethanol. Sections were used to grade disease based on cell infiltration and retinal structure, as described in Copland *et al*., 2008. Sections were masked and graded using 3 sections at each of 3 spaced intervals through the eye. ## Immunohistochemistry for S100B in retina To examine the expression of S100B in the retina, immunohistochemistry was carried out using a Nova Red peroxidase substrate kit prepared as per manufacturer instructions (Vector Laboratories, CA). Sections were fixed in 100% ethanol for 5 min before being rehydrated in PBS and blocked with 10% normal rabbit serum for 30 min. F(ab’)₂ Fab polyclonal anti-mouse IgG (Serotec) was applied for 30 min before incubation with anti-mouse S100B primary antibody (Sigma, UK) at a 1 in 500 dilution in 1% normal swine serum for 1 h at room temperature followed by washing in PBS. Sections were then incubated with the secondary antibody, 1 in 200 dilution, for 1 h before washing in PBS. ABC elite peroxidase (Vector PK6100, Vector Labs, Ca, USA) was applied for 30 min at room temperature, slides washed in PBS as previously, then the Nova Red substrate (SK4800) was applied. The substrate reaction was stopped after 30 s by washing in water. Sections were counterstained with haematoxylin before being dehydrated and mounted in DPX (Sigma). ## Immunohistochemistry for macrophages Sections were stained for macrophages with MOMA-2 antibody (purified rat anti- mouse polyclonal antibody; AbDSerotec, Oxfordshire, UK) using a 1 in 25 dilution. The secondary antibody was used at a 1 in 100 dilution (biotinylated anti-rat IgG; Vector Labs). In control staining, primary antibody was replaced with tris-buffered saline. Sections were masked and total stained and unstained infiltrating cells present across the retinal layers, or within the rod outer segments were counted using a graticule to count cells in an exact area (1.5 mm²) in 3 sections at 3 spaced intervals through the eye. The average count for the sections was calculated for each eye. ## Statistics Statistical analysis was carried out using Mann-Whitney non-parametric test, Student’s unpaired T Test and ANOVA. # Results ## S100B up-regulates pro-inflammatory gene transcription in macrophages To investigate the direct effect of S100 on macrophages we first used the mouse macrophage cell line, RAW 264.7. RAGE mRNA expression by RAW 264.7 and BMDM cells was confirmed by routine RT-PCR. RAGE protein in RAW cells (previously shown) and BMDM was confirmed by western blot. The effect of S100B on RAW 264.7 macrophages in terms of inflammatory cytokine and chemokine response was examined at a transcriptional level using real-time PCR array analysis. Unlike previous studies which have only focused on a limited selection of pro- inflammatory cytokines this allows a wide range of pro-inflammatory cytokines, chemokines and their receptors to be examined. RNA extracted from cells was confirmed to be high quality before use in the array. Real-time PCR cycle threshold (C_T) value, (the value where there is a significant detectable increase in fluorescence above the baseline value), was calculated using the second derivative max, and allowed fold change calculations to be made. Each array provided quality control checks for PCR reproducibility, reverse transcription and genomic DNA contamination. Five reference genes were available to select the optimum gene normalisation. All samples were normalised to GAPDH and C_T values greater than 35 were not reported. RNA samples from RAW 264.7 cells treated with or without S100B and meeting PCR array quality control standards were reverse transcribed, analysed using the PCR array and normalised to GAPDH. The PCR array showed an overall up- regulation of pro-inflammatory mediators in S100B treated compared to untreated cells. In particular, expression of *CCL22* and *IL-1β* was increased over seven-fold in response to S100B treatment. Gene expression for CCL22 and IL-1β was investigated further using real-time PCR and a range of S100B doses and incubation times. S100B was confirmed to up- regulate CCL22 and IL-1β gene expression. CCL22 showed its highest fold increase in mRNA expression (4.2 fold, *P* = 0.0131) in response to incubation of RAW 264.7 cells with 2 μM S100B for 9 h. There was also a significant increase (*P* = 0.0170) after 6 h incubation at this concentration. *IL-1β* showed a significant increase in RAW 264.7 cells compared to untreated cells with 1 μM and 2 μM S100B. ## S100B increases CCL22 production in macrophages To determine whether S100B increased CCL22 production as well as mRNA expression, CCL22 in cell supernatant was analysed by ELISA. S100B increased CCL22 production by RAW 264.7 cells in a concentration dependent manner with a significant increase in CCL22 in the presence of 2 μM (*P* = 0.0032) and 1 μM S100B (*P* = 0.0078) compared to untreated control cultures. This was also confirmed in primary macrophages, BMDM, where a significant increase in CCL22 production was observed at 2 μM S100B treatment. No increase in CCL22 production occurred with cells incubated with S100B that had been boiled. To check that these responses to S100B were not related to the effects of S100B on cell viability, cell morphology was monitored and viability was measured by acid phosphatase assay following treatment of cells with 2 μM or 5 μM S100B for 24 h. There were no morphological differences and there was no significant difference in cell viability following treatment of either RAW 264.7 cells or BMDM with 2 μM or 5 μM S100B for 24 h. ## S100B increases pro-IL-1β production by macrophages To confirm that S100B up-regulated transcription of IL-1β results in protein translation, flow cytometry was used to detect pro-IL-1β after treatment of RAW 264.7 cells and BMDM with 2 μM S100B for 4 h. Cells were gated using CD11b expression. Compared to untreated control cultures, S100B treated RAW 264.7 cells and BMDM showed a significant increase in the percentage of cells expressing pro-IL-1β (*P* = 0.0338). ## S100B is increased in EAU diseased retinal sections To investigate the relevance of the pro-inflammatory effect of S100B on macrophages *in vivo* we used a murine model Experimental Autoimmune Uveoretinitis (EAU), in which retinal pathogenesis is highly dependent on macrophage infiltration and activation. To determine whether S100B protein levels are up-regulated in the retina in response to EAU induction, immunohistochemistry was carried out on retinal sections taken from mice in which EAU had been induced or untreated mice. Positive staining was observed in the naïve retina, specifically in the retinal ganglion layer and outer plexiform layer which confirms previous studies which have identified S100B present within the retinal layers. An increase in S100B positive staining was observed in EAU diseased sections, specifically in the rod outer segments, where positive staining was located around infiltrating cells, and in the retinal ganglion layer. ## Removal of S100B dampens the inflammatory response in EAU To determine whether the absence of S100B affects the inflammatory response in EAU, S100B KO and C57BL/6 WT controls were immunised with RBP-3_1−20. Progression of disease was followed using TEFI at day 15 and day 21 pi. Fundus images were graded for retinal inflammation. By day 15 pi both groups showed signs of inflammation; however, the S100B KO mice showed a significant decrease in EAU grade compared to WT (\**P*\<0.05). Inflammation had increased in both strains by day 21 pi but S100B KO mice still had significantly less inflammation. Eyes were collected at day 24 pi for histological evaluation. Histology sections were graded according to cell infiltration and confirmed TEFI observations showing significantly reduced inflammatory infiltrate in the S100B KO mice compared to WT (\**P* = \<0.05). To determine the effect of *S100B* deletion on macrophages in the inflammatory infiltrate the infiltrate was stained using MOMA-2. Eyes were serially sectioned and 3 sections at different points through the eye were examined. A graticule was used to count total stained cells within an equivalent area in the retina from S100B KO mice or C57BL/6 WT mice. A significant decrease in macrophage cells was observed in retinal sections taken from the S100B KO mice compared to C57BL/6 WT mice. A significant reduction in macrophage infiltration specifically in the rod outer segments was also observed. No positive staining was observed in healthy control retina sections from either C57BL/6 mice or S100B KO. ## CCL22 and IL-1β gene transcription were down-regulated in the diseased retina in S100B KO mice Real-time PCR was used to compare retinas from WT and S100B KO mice in which the disease was of the same severity, grade 3 as determined by TEFI, and confirmed that CCL22 and IL-1β were reduced specifically in S100B KO mice compared to WT with a fold reduction normalised to GAPDH of 3.58 for CCL22 and 4.63 for IL-1β. # Discussion Our aim was to determine whether S100B could influence the production of cytokines and chemokines by macrophages and whether this was likely to be relevant in inflammatory disease in which macrophages are key effector cells such as EAU. Although S100B has previously been shown to have a pro-inflammatory effect on macrophage cell lines, production of a full range of pro-inflammatory cytokines and chemokines by macrophages in response to S100B has not been investigated and production has not been confirmed in primary macrophages. The macrophage cell line, RAW 264.7, was used for initial studies and we confirmed the expression of RAGE in these cells and in BMDM. RAGE has also been shown to be expressed by infiltrating macrophages in EAU. PCR array analysis, using an established, commercially prepared PCR microarray, on RAW 264.7 cells treated with 2 μM S100B for 6 h showed that TNFα was up-regulated, as shown previously, as well as other cytokines. However the largest increases were seen in IL1β and CCL22. Rapid up-regulation of IL-1β transcription in macrophages in response to S100B was confirmed by real-time PCR and corresponding up-regulation of intracellular pro-IL-1β production was shown by flow cytometry in RAW 264.7 and for also the first time, to our knowledge, in primary BMDM. Up-regulation of pro-IL-1β does not confirm the presence of the mature protein and its secretion. Complex regulation of the inflammasome is required for IL-1β secretion. It is not possible to show mature IL-1β in RAW 264.7 cells due to absence in these cells of an essential adaptor protein, ASC, needed for caspase 1-mediated IL-1β processing in inflammasomes. Thus, although we are not able to confirm this *in vivo* in this study, we propose that under fully functioning inflammasome conditions *in vivo*, mature IL-1β production by macrophages will be up- regulated by S100B. S100B has previously been seen to promote IL-1β production in primary rat astrocyte cultures, primary rat microglia and a murine microglial cell line BV-2. IL-1β mRNA expression in this microglial cell line was shown to be elevated in the presence of S100B via RAGE. S100B treatment of a human monocytic cell line, THP-1, has also been reported to result in increased IL-β gene transcription and secretion. IL-1β is a highly pro-inflammatory cytokine and a major source of it in the inflammatory response is the macrophage. Locally it acts to enhance the recruitment of further inflammatory leukocytes. It is a key pro-inflammatory cytokine involved in EAU. Its proposed increased production in response to S100B *in vivo* would therefore be likely to exacerbate inflammatory disease. S100B up-regulated the expression of CCL22 by RAW 264.7 cells up to 7 fold and up-regulation of protein production was also shown by ELISA with CCL22 secretion increased in response to S100B in both RAW 264.7 and primary BMDM. CCL22 is an important chemoattractant produced by macrophages and DC under inflammatory conditions. Although its production is associated with M2 macrophages and it can recruit Th2 cells and T regulatory cells via its receptor CCR4, indicating a more regulatory role, it may also increase pathogenesis. It has been shown to promote tissue injury in bleomycin-induced pulmonary fibrosis in mice by inducing the M1 macrophage phenotype. In addition, CCL22 is elevated in the CSF of patients with MS and both CCL22 and CCR4 have been shown to contribute towards the recruitment and function of inflammatory macrophages and disease development in experimental autoimmune encephalitis (EAE). Enhanced levels of S100B have also been reported in EAE. Increased CCL22 has been reported in EAU and it is possible that as in EAE it is acting to exacerbate disease. If S100B acts predominantly as a direct pro-inflammatory mediator it should exacerbate inflammation. However, for these *in vitro* experiments on macrophages a relatively high dose 2 μM was shown to be effective. This is consistent with other studies where higher doses of S100B were shown to be most pro-inflammatory. S100B has been shown to be present at these levels in tissue for example in brain damage where it is reported to accumulate in the extracellular space but it is not known whether in other situations it is present at levels which would be likely to influence inflammation. To test the relevance of S100B in inflammatory disease *in vivo* we have investigated its role in retinal inflammation in EAU in which macrophages play a crucial role in defining the severity of the disease and the subsequent damage incurred. Although S100B has been reported to be present within the eye it is unclear whether it is present in sufficient concentration to influence inflammation so initially we examined S100B expression in the retina. We have shown for the first time, to our knowledge, an increase of S100B in the retina of mice with EAU compared to healthy retina. In the eye S100B will be produced by glial cells, astrocytes and Mϋller cells and will accumulate in the extracellular matrix and there may also be S100B release by damaged cells. Our immunohistochemical staining showed evidence of this. In the brain, where enhanced synthesis of S100B has been shown by reactive astrocytes accumulating around the infarct area after cerebral artery occlusion, it was concluded that in areas of intense S100B staining S100B tissue concentration reaches micromolar concentration albeit temporarily. In our tissue sections there was also evidence of S100B production associated with infiltrating cells. It has been suggested that T cell production of S100B, although less than that of astrocytes, may, when polarized in the immunological synapse, be sufficient to trigger macrophage activation. Recently S100B has been shown to be up-regulated in the eye in response to laser photocoagulation which generated choroidal neovascularisation. Thus it is likely that S100B is present in the retina in EAU at the concentrations we have shown *in vitro* are able to influence macrophage response. To examine the role of S100B in inflammation we investigated the effect of deletion of S100B on induction of retinal inflammation in EAU. Clinical grading showed that EAU severity was significantly reduced in S100B KO compared to WT mice from day 15 pi. This was confirmed by histological examination at day 24 pi which showed a significant reduction in cell infiltration in S100B KO mice compared to WT. Immunohistochemistry indicated a reduction of macrophages within the retina of S100B KO mice with EAU compared to C57BL/6 WT mice with EAU. Overall reduction in the infiltrate in S100B KO mice is likely to be due to a general reduction in adhesion molecule and chemokine levels for inflammatory cells. IL-1 in particular has a major influence on these factors. Initial T cell and macrophage recruitment to the retina in S100B mice with EAU is unaffected as onset of disease is not delayed. However, macrophages in the retina of S100B KO mice then produce less IL-1 and CCL22 and further inflammatory cell recruitment is reduced. In EAE it has been reported that CCL22 enhances myeloid cell recruitment. Production of S100B during uveitis either by resident cells in the eye or by infiltrating cells is therefore likely to have up-regulated CCL22 and IL-1β in particular resulting in further recruitment of inflammatory cells and exacerbation of disease. Although average disease severity was reduced in S100B KO mice compared to WT and consequently the overall average levels of CCL22 and IL-1β would be decreased we were interested in comparing levels of CCL22 and IL-1β in mice selected for the same level of disease. Matching disease in WT and in S100B KO mice enabled us to highlight changes that might be more directly related to S100B and its underlying role rather than those which would generally follow from reduced disease severity. PCR analysis of retinas matched for disease severity (Grade 3 TEFI) confirmed reduction of CCL22 and IL-1β in retinas where S100B had been deleted. Thus we have shown for the first time that S100B is an important component in the development of retinal inflammation. It is likely to be present at levels that may have a direct action on macrophages, the key effector cell in uveoretinitis, up-regulating chemokine/chemokine receptor and cytokine expression, CCL22 and IL-1β in particular. Recently it has been shown that S100B can increase migration of BMDMs and this, in conjunction with any up-regulation of CCL22 and IL-1β, would lead to the recruitment of further cells and an exacerbated inflammatory response. The effect of S100B on infiltrating macrophages will be combined with effects on resident microglia cells in the retina further enhancing this response. The reduction in disease severity in S100B KO mice may also be due to effects on other cell types expressing RAGE such as DC and further experiments will be required to determine the extent to which the reduction in severity in EAU as a result of S100B deletion is directly due to its effects on macrophages, and in particular the role of IL-1β and CCL22. However, it is likely that S100B is able to exacerbate inflammatory disease at least in part through its action on macrophages. We would like to thank University of Aberdeen Medical Research Facility staff. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: IC JN. Performed the experiments: JN JH DM GD. Analyzed the data: JN JH DM. Contributed reagents/materials/analysis tools: SI PT MG. Wrote the paper: IC JN PT SI.

# Background In order to guarantee a safe blood supply blood donors must meet certain eligibility requirements. Eligibility is determined through the donor history questionnaire and interview, assessing risk factors for transfusion transmissible infections (TTI) such as tattooing, injecting drugs, and sexual behaviour such as “men who have sex with other men” (MSM). MSM have the highest prevalence of human immunodeficiency virus (HIV) in the Western world, with rising incidence in many countries. Other sexually transmitted diseases (STD) which are also transmissible via blood or transfusion, are hepatitis B and hepatitis C, with a high prevalence of hepatitis B virus (HBV) infection in MSM and a rising incidence of hepatitis C virus (HCV) infection in HIV-infected MSM. In addition, recent increases of syphilis in MSM have been documented, and MSM are also more often diagnosed with other STDs which are transfusion transmissible, such as chlamydia. Whether that in itself justifies excluding MSM from donating blood has been the subject of public dispute, the question being whether the risk of infection is high enough to support exclusion, and if so for how long? One side brands the measures as discriminatory against gay men. The other side argues that exclusion is based on MSM behaviour rather than sexual orientation, that the epidemiology of TTI is clear, and that the right of the patient to receive safe blood takes precedence over the wish of a particular donor group to be allowed to donate blood. Measures differ widely between countries, ranging from lifelong to a five or one year exclusion, or even no exclusion at all. This article is the first systematic review on this topic, and uses the principles of Evidence-Based Medicine at its most rigorous level, by applying the methodology of the Cochrane Collaboration. The aim was to find studies that describe the relationship between MSM and safe blood donation, not the prevalence of TTI in MSM in general for which systematic reviews already exist. The following ‘PICO’ question was developed: For male blood donors (Population) is having sex with other men (Intervention) a risk factor for TTI (Outcome) compared to not having sex with other men (Comparison) in Western countries? # Methods We followed the PRISMA statement for the reporting of this systematic review. No protocol for this systematic review existed or was published beforehand. ## Selection criteria We used the following inclusion and exclusion criteria for the selection of articles: Population: Inclusion: blood donors (or people eligible to give blood), living in areas most relevant for our Blood Service: Northern, Western, and Southern Europe (Albania, Andorra, Austria, Belgium, Bosnia and Herzegovina, Croatia, Denmark, Estonia, Finland, France, Germany, Gibraltar, Greece, Iceland, Italy, Ireland, Kosovo, Latvia, Liechtenstein, Lithuania, Luxembourg, Macedonia, Malta, Monaco, Montenegro, the Netherlands, Norway, Portugal, San Marino, Serbia, Slovenia, Spain, Sweden, Switzerland, Vatican City), USA, Canada, Australia, New-Zealand; Exclusion: a population containing blood donors, but not exclusively consisting of blood donors. Intervention/Risk factor: Inclusion: men having sex with other men after 1977. Comparison: Inclusion: men not having sex with other men. Outcome: Inclusion: markers of transfusion-transmissible infections from the following pathogenic micro-organisms in the donor blood (which are sexually transmitted and also transfusion transmissible): HIV, HBV, HCV, *Chlamydia*, and *Treponema pallidum* (causing syphilis). Study design: Inclusion: Intervention studies: randomized controlled trials, controlled clinical trials, before- and after studies; Observational studies: cohort studies, case-control studies, cross-sectional studies (surveys); these study types are included independently of the potential risk of bias (however, risk of bias will transparently be reported); Exclusion: non-controlled studies, case reports, case series, letters, comments, opinion pieces, narrative reviews, modelling studies. No language criteria were used. ## Search strategy and study selection There are three types of studies that can potentially answer our question: (1) “Type 1 studies” comparing the incidence of TTI in blood products of MSM and non-MSM donors, (2) “Type 2 studies” comparing two types of deferral strategies for MSM donors, (3) “Type 3 studies” comparing infected blood donors (cases) versus non-infected blood donors (controls), identifying risk factors (e.g. MSM) of both groups (case-control studies). Search strategies were composed to retrieve the three study types. The following databases were searched from their date of inception to 26 March 2014: MEDLINE (using the PubMed interface), Embase (using the [Embase.com](http://Embase.com) interface), The Cochrane Central Register of Controlled Trials, Cinahl, and Web of Science. Full details of the search strategies are given in. Study selection was performed in parallel by two independent reviewers (EDB, TD). Titles and abstracts of the studies identified by the search were scanned. When a relevant article was found, full text articles were retrieved. Studies that did not meet the selection criteria were excluded. The citation and reference lists of included studies were searched, and the first 20 related items in PubMed were scanned for other potentially relevant studies. Any discrepancies among the two reviewers were resolved by consensus. ## Data collection Data concerning study design, study population, outcome measures (markers of transfusion-transmissible infections expressed as risk ratio, odds ratio or incidence rate ratio), and study quality were extracted independently by two reviewers (EDB, TD). In the event that data were lacking, authors were contacted to obtain more detailed information. No statistical methods were used to pool the data because of heterogeneity of the studies. ## Quality of evidence The GRADE approach was used to assess the overall quality of evidence included in this review and the online available Guideline Development Tool (GDT), developed by the GRADE Working Group, was used to develop the GRADE table, including quality assessment and a summary of findings (<http://www.guidelinedevelopment.org/>). This table was adapted manually in case our data were not compatible with the template. A level of evidence was assigned for each outcome, which were all rated as critical outcomes. GRADE downgrades the level of evidence because of risk of bias (limitations in study design) of the included studies, inconsistency between results of different studies (due to differences in populations, interventions or outcomes), indirectness (of population, intervention or outcome), imprecision and publication bias. All studies included in this systematic review were observational studies, which results in an initial “low level of evidence” according to the GRADE approach. For none of the observational studies included there were reasons to upgrade the level of evidence. Risk of bias in the individual studies was analysed at the study level by evaluating the presence of eligibility criteria, adequate control of confounding, follow-up and correct measurement of exposure and outcome, as detailed in the Help section of the GDT. # Results ## Study selection and study characteristics provides a flowchart of the identification and selection of studies. The searches yielded 18,987 references. After removing duplicates and screening relevant titles and abstracts, 317 references were selected by reviewer 1 and 418 references by reviewer 2 for which the full text was evaluated. In an overview is given of all studies excluded by at least one of the reviewers, with the reason for exclusion. The major reason for exclusion was study design (56%) as many references were opinion pieces, letters, comments or uncontrolled studies. After resolving disagreements among the two reviewers, we retained 14 observational studies (no references of a higher study design were found), which are described below and in detail in. ### Type 1 studies: comparison of TTI in MSM vs non-MSM donors In two studies comparing MSM and non-MSM donors, TTI were searched for in respectively 25,168 male blood donors with different time intervals since the last MSM contact (self-reported in an anonymous mail survey) versus control donors, and in 52 MSM-donors who were eligible as blood donors versus 209 non- MSM blood donors. ### Type 2 studies: comparison of TTI between different deferral strategies for MSM donors Two studies compared different deferral strategies. In the first study blood samples from donors, including MSM, were examined for HIV prevalence during the five years preceding and the five years following a change from a permanent/five-year deferral versus a one-year deferral for MSM. The second study compared permanent deferral for MSM donors, with individual risk assessment of sexual behaviours. ### Type 3 studies: comparison of infected vs non-infected blood donors for risk factors 10 case-control studies tried to identify risk factors for TTI in blood donors. The cases included blood donors with a viral infection with HIV-1, HBV, or HCV, and the controls were donors who did not have this particular infection. Studies only looking at blood donors with infections were not included. Two of the 10 studies, where “homosexuality” was measured as a risk factor, did not explicitly mention if this concerned only men, but we assumed this was the case. The majority of studies did not explicitly mention which donor selection strategy was in place with regards to MSM at the time of the study, however from other references we know that all studies had a policy of lifelong exclusion at the time of the study. ## Synthesis of findings An overview of the synthesis of findings of all included studies can be found in. Since we were not able to pool any studies, a narrative overview of the results of the individual studies is given below ### Type 1 studies: comparison of TTI in MSM vs non-MSM donors There is limited evidence (in number and quality) from two observational studies from 1985 and 2005, that male blood donors who had sex with other men after 1977 have a statistically significant higher risk of HIV-1, compared to male blood donors who did not have sex with other men. The study of Alpaugh et al. found a risk ratio of 9.19, 95% CI 3.99–21.17 for the presence of antibody to hepatitis B core antigen (HBc). In this study the presence of antibodies to HBc was used to measure HBV infection. However this is an older test with poor specificity, which means that a high percentage of HBV negatives is incorrectly identified as HBV positives. Because results of this study are influenced by the test that was used to measure the outcome, we did not take the HBV data—in contrast to its HIV data—into account into our final conclusion. For the presence of antibody to human T Cell Lymphotropic virus type III antigen (= HIV-1), the risk ratio was 6.90, 95% CI 2.10–22.68 for the initial reactive test, and 27.60, 95% CI 3.47–219.37 for the repeatable reactive test. In the study by Sanchez et al., including 25,168 male donors, 6% of the donors who had sex with other men after 1977 had reactive screening test results, versus 1.7% of non-MSM donors, which was statistically significant. Analysis of the subpopulation between 1977 and 2010 indicated that the statistically significant difference was due to the MSM population with recent (\< 12 months) sexual contact. A statistically significant increased risk of infections could not be demonstrated for male blood donors who had sex with other men before 1977 (2.8% of MSM donors versus 1.7% of non-MSM donors) or for MSM more than one year ago since 1977, due to a large variability of the results (wide confidence intervals). ### Type 2 studies: comparison of TTI between different deferral strategies for MSM donors In two studies, from 2010 and 2013, no statistically significant differences between types of deferral policy were measured. In the first study, no statistically significant change in the number of HIV-positive male donors when applying a five versus a one year deferral period could be demonstrated. In the same study, the number of HIV-positive donors with MSM as a risk factor was also measured, however the number of male donors who disclosed risk factors is not known, and thus no conclusions can be made of these data. In the second study, no statistically significant difference in the number of HIV-positive MSM donors could be demonstrated when a permanent deferral was changed to an individual risk assessment of sexual behaviours. The results of these studies cannot be considered precise as the number of HIV-infected donors was too low. ### Type 3 studies: comparison of infected vs non-infected blood donors for risk factors Ten case-control studies investigated whether MSM is a risk factor for TTI such as HIV-1, HBV, or HCV in blood donors, comparing the risk profile of donors with TTI and matched healthy donors. All studies were performed when a permanent deferral strategy for MSM was in place, and thus any cases would have occurred despite this deferral strategy. There is limited evidence from one observational study, published in 1994, about the correlation between MSM and the risk of HIV-1 infection in blood donors. In this study it was shown that MSM is a statistically significant risk factor for HIV-1 infection in blood donors, reporting an odds ratio of 45, 95% CI 10.66–189.84. However, the results cannot be considered precise because of a low number of events (low number of donors with MSM as a risk factor among the cases and controls). For the risk of HBV infection, we found limited evidence from 2 observational studies from 1997 and 2001, but a statistically significant correlation could not be demonstrated between MSM and HBV infection in blood donors. The reason for this is a low number of events and a large variability of the results, reflected in wide confidence intervals. The majority of the studies were searching for MSM as a risk factor for HCV in blood donors. A statistically significant increased risk of HCV-infection in blood donors could not be demonstrated for MSM in six of these studies (all performed in the early ‘90s and one in 2000), because of a low number of events (all studies) and/or a wide confidence interval. In one more recent (2012) study it was shown that MSM blood donors have a statistically significant increased risk of HCV, reporting a risk ratio of 8.79, 95% CI 1.18–65.25, compared to non- MSM blood donors. However this risk ratio is a crude risk ratio, which represents a correlation between HCV data and MSM, not taking into account that the MSM donors could have other risk factors such as intravenous drug use or history of transfusions. Since the latter risk factors are confounding variables and no measures were taken to control for confounding in this study, this crude risk ratio, which we calculated ourselves based on the raw data provided in the study, was not used as part of the evidence base. ## Quality of the evidence All studies included in this systematic review were observational studies, which results in an initial “low level of evidence” according to the GRADE approach, and there were no reasons to upgrade the level of evidence. In order to determine the final level of evidence, first of all the risk of bias was assessed at the individual study level. In all studies, risk of bias was found because of limitations in study design or execution. For several studies there was no description of possible confounders nor of the method that was used to adequately control for confounding, or it was unclear if measures undertaken to adequately control for confounding were sufficient. For 5 of the case-control studies inappropriate eligibility criteria were used for the selection of cases and controls. In 5 studies questionnaires were used and in 2 studies interviews were performed to obtain information about risk factors, which is prone to recall bias. The studies by Seed et al. and Suligoi et al. are historically controlled, but groups where no change in deferral policy was implemented were not included. In addition in the study by Seed et al., different states/territories had different policies (≥ 5 year deferral) before implementation of the new policy, and implementation of the new policy did not occur at the same moment in each state/territory. In the study by Sanchez et al. and in the 11 case-control studies, donors who reported MSM since 1977 were donating despite being excluded from this practice. It cannot be excluded that these donors may have different characteristics from men with a history of MSM since 1977 who did not donate because of the current policy, but would if it were changed. In addition to the risk of bias, the level of evidence was further downgraded because of imprecision (low number of events, large confidence intervals and lack of data). There was no reason to downgrade for inconsistency or indirectness. Overall, the strength of the body of evidence, as defined by the GRADE approach, is “very low” for all outcomes, which means that the estimates of effect are uncertain and further research is very likely to have an important impact on our confidence in the estimate of effect. An overview of how we obtained the level of evidence for all outcomes is shown in. # Discussion In this systematic review we searched for evidence on the association between MSM blood donors and TTI. We identified 14 studies: two studies directly comparing MSM with non-MSM donors, two studies comparing different types of deferral strategy for MSM and 10 studies comparing infected versus non-infected donors during a permanent deferral policy for MSM. Two studies however provided (statistically significant) data that were not taken into account to formulate our conclusions: the HBV data in the study of Alpaugh et al. were measured with an older diagnostic test with low specificity, and the HCV data in the study of Allison et al. were not taking into account any confounding variables. We identified 3 studies looking at the risk of HIV-1 infection in MSM versus non-MSM donors. These studies showed a significant correlation between MSM and the risk of HIV-1 infection in blood donors. In addition, we found 3 studies looking at the risk of HBV infection and 7 studies looking at the risk of HCV infection, but none of these studies could demonstrate an increased risk of infection. The available evidence is too limited to be able to unambiguously/clearly recommend a certain deferral policy, however one study suggests exclusion of MSM donors for at least 1 year after the last MSM contact. The latter however is based on very low level evidence from only 1 study and should therefore be interpreted with caution. This study showed that 8.3% of the donors who had sex with other men the last 12 months had reactive screening test results, versus 1.7% of non-MSM donors. We found no evidence that MSM blood donors are at a higher risk of HBV and HCV infection. In the majority of the studies there was no information available about the period between the last MSM contact and the blood donation. The main limitations of our analysis are: the quality of the available evidence is defined according to the GRADE approach as “very low”, because of the study design (observational studies), a low number of infected individuals, a low number of donors with MSM as a risk factor, and/or large confidence intervals. Furthermore, the majority of the studies are older studies, including one study from 1985 and 7 studies from the early 1990s, which could result in bias since test specificity and sensitivity has improved significantly since then. our analysis does not account for (non-)compliance in filling out the donor history questionnaire, as it is impossible to deduce from the studies what percentage of donors were honest about MSM behaviour, and of those who were not, what percentage were later forthright about this. This in itself could compromise the quality of further studies. A recent study from Australia confirmed high compliance to a 12-month deferral for MSM, however compliance was calculated against the total population of male donors and not against the population of MSM donors, and no comparison between different deferral strategies was made. High quality studies about the impact of different deferral strategies on non- compliance are currently lacking. our systematic review did not capture unpublished surveillance data that are probably being collected by many blood services. our analysis is limited to a certain geographic area, namely Western countries (Northern, Western, and Southern Europe, USA, Canada, Australia, New- Zealand) as defined in the selection criteria, since the populations in these countries are most relevant for our Blood Service. Because the epidemiology of STDs (which are also transfusion transmissible), sexual risk behaviour and STD prevention is different in developing countries, the results of this systematic review cannot be generalized. Taking into account the limited evidence available, further higher quality research is necessary. There is a clear lack of studies directly comparing MSM and non-MSM donors, and studies comparing different deferral strategies. When comparing risk factors of infected and non-infected donors, it is important to describe the studied risk factors in detail and to mention the donor selection policy used. The classic triad of evidence-based work consists of the best available evidence, complemented by expert opinion and by preference of the target population. In the absence of strong evidence it is not surprising that expert opinion and preference (in this case both of patients and donors) play a greater role in determining policy than they would if the quality of the evidence were stronger. Donor preference (MSM group) is clear: many demand to be allowed to donate blood because they feel discriminated against. Some do not contest exclusion on the basis of MSM, but rather the lifelong character of the exclusion, noting that it is applied to no other group on the basis of risk behaviour, except for, for example, intravenous drug users or commercial sex workers. The preference of the patient population is also clear: they expect to receive the safest blood possible, but since most people are only future patients, they tend to be less informed and vocal, with the exception of haemophilia patient groups, who are both knowledgeable and very concerned about receiving safe blood. Expert opinion mostly favours exclusion of MSM as the experts also take into account (1) evidence that is excluded in this systematic review because of its study type (e.g. uncontrolled studies); (2) evidence from outside the field of transfusion medicine such as higher prevalence of HIV and other TTI in MSM; (3) that the right of the patient to the safest blood possible has predominance over the wish of a particular donor group to be allowed to donate blood; (4) the tradition of the sector to make blood ever safer, following the precautionary principle (stating that, in the interest of public health, risk management action should be taken in the absence of certainty about risk) rather than the principles of health economics (not reimbursing measures that are not cost- effective) or “risk-based decision making” (implying that risk management actions should be proportionate to the level of demonstrated risk). # Conclusions In summary, high-quality studies investigating the link between MSM blood donors and TTI are scarce. The available evidence suggests a link between MSM blood donors and HIV-1 infection. In one study it was shown that the significant correlation between MSM and HIV-1 infection was related to recent (\< 12 months) MSM contact. This is however very low level evidence and more high quality studies are needed to be able to unambiguously/clearly recommend a certain deferral period for MSM. In absence of strong evidence, the length of the exclusion period mainly depends on regulators’ choices about whether the precautionary principle should continue to be applied in the blood banking sector or not, and on whether the estimated impact of exclusion/length of exclusion policy on non-compliance with the donor history questionnaire is greater than the gains made by applying the precautionary principle. The scarcity of high-quality evidence within the field of transfusion medicine together with these choices may explain existing policy differences between countries. # Supporting Information [^1]: The authors have read the journal's policy and have the following competing interests: All authors are employees of Belgian Red Cross- Flanders, which is responsible for supplying adequate quantities of safe blood products to hospitals in Flanders and Brussels on a continuous basis. [^2]: Conceived and designed the experiments: EDB TD VC PV. Analyzed the data: EDB TD. Wrote the paper: EDB PV. Formulated research question and selection criteria: EDB TD VC PV. Performed the literature search and study selection: EDB TD.

# Introduction By the end of 2010, the Chinese government estimated that 780,000 people were infected with human immunodeficiency virus/acquired immune deficiency syndrome (HIV/AIDS) in the People's Republic of China. In Si-chuan province, which is located in the southwest of China, 48,357 cases of HIV infection were reported by the end of 2011, which is among the five Chinese provinces with the highest HIV prevalences. More than half of the HIV/AIDS patients (53%) in Si-chuan province live in the Liangshan Yi autonomous prefecture, which shares borders with Yunnan and is in close proximity to the Golden Triangle region where large amounts of heroin are manufactured and trafficked. Almost half of Liangshan's population of 4.73 million are members of the Yi minority and live in rural areas. Rugged mountainous terrain and the sparsely scattered rural populations have hindered the economic development of this region, and Liangshan remains one of the poorest areas in the People's Republic of China. Sociological research has linked the high HIV prevalence in this area to a range of historical and cultural factors, including poverty, low education, and indigenous customs. For more than 10 years since the first HIV infection was documented in 1995, the Liangshan AIDS epidemic has been concentrated among intravenous drug users (IDUs). In recent years, however, the proportion of cases among IDUs has decreased; on the other hand, heterosexual transmission has begun to rise. Studies performed in other provinces showed that co-infection with the hepatitis C virus (HCV) is common among HIV positive individuals and IDUs because they share the same transmission routes. Heterosexual transmission of HIV among non- IDU sexual partners was also reported and is currently attracting increasing attention. Co-infection of HIV and HCV can result in liver cirrhosis and can increase HIV morbidity and mortality. Understanding the transmission routes of both HIV and HCV has significant indications for implementing anti-retroviral therapy. In 2008, a large scale HIV screening was conducted in Butuo county.The target population was local residents aged from 14 to 60 years old, and the size was 84 357, but the study only got 30 111 samples, the response rate was 35.7 percentage. The result showed that HIV epidemic was extremely severe, and HIV prevalence was highest among China. In order to fully understand HIV and hepatitis C virus (HCV) prevalence and related risk factors in this region, a national HIV strategy has been implemented in two towns of Butuo since 2009, conducting population-based surveillance and testing campaigns and implementing the Four Frees and One Care policy, which includes free HIV testing and antiretroviral medication for HIV-affected individuals. The population surveillance program provides a unique opportunity to understand the scale and determinants of HIV and HCV co-infection epidemic as well as risk factors for the spread of infectious disease in this remote and isolated region. This study uses baseline data from the population survey and testing data obtained in 2010 from two counties with 99% of its population consisting of the Yi minority group. In this paper, we principally focused on the impact of social factors, including educational status, occupation, drug abuse, knowledge of transmission types and multiple sexual partners. We will assess the associations between HIV and HCV prevalence and will examine patients' socio-demographic characteristics, knowledge of HIV, and drug use status by gender and marital status. # Methods ## Study setting and sample The population survey and testing study were conducted in two towns (A and B) of the Butuo district between April 2010 and December 2010. These towns were selected out of 30 towns, which had similar social demographic characters (such as, age structure, education level, constituent ratio of Yi people, economic level, etc.), meanwhile this two towns had high prevalence rate of HIV. To be eligible for the study, participants had to: (1) be a resident of the town or have lived there for more than 3 months at baseline, and (2) give consent to participate in the study. To improve participant enrollment, the local government and Si-chuan Center for Disease Control (CDC) completed a series of preparation steps with the 11 village chiefs (six in town A and five in town B). The chiefs were first contacted and invited to health education sessions provided by the CDC. Next, the village chiefs were also asked to collect basic demographic information to register all villagers, including name, age, gender, marital status, migrant status, drug use status, and so on. Based on this information, a schedule was made for each village leader to indicate when the survey and physical examination would take place and who should attend each component. The village chiefs then invited towns people to a meeting where they described the study, including all tests and procedures. If villagers were interested in participating, they were instructed to schedule an appointment for their physical exam and survey. The surveys were administered at the township hospital when the villagers came for their scheduled appointment. A free physical exam was provided to each subject as an incentive for participating in the survey and included measurement of height and weight, a general physical exam, an electrocardiogram, a chest x-ray, an abdominal B ultrasound, and blood tests including routine HIV, HCV, HBV, and syphilis tests. The physical examination took place first and was followed by an interview in a private room by a trained Yi interviewer who used a questionnaire to guide the interview. During the interview, the interviewer transcribed participant responses onto the questionnaire, and all questionnaires were collected by a quality control checker to ensure that all questions had been answered and that no logistical mistakes had been made. The questionnaires were then handed to a researcher responsible for storing the data in a special, locked file cabinet. Four students were hired and trained to input the data into the electronic database. We obtained oral consent from participants prior to participation because most participants were illiterate and unable to provide written consent. On this basis, we requested that all 11 village chiefs would organize a meeting attended by all adults before this survey was conducted. In this meeting, the village chiefs explained the survey to the participants, and the aims of the Chinese government in conducting this project. All participants were informed that this survey included a questionnaire and free physical examination, and the questionnaire included details of health-related practices and high risk behaviors, such as drug use and multiple sexual partners. The physical examination included measurements of height and weight, a general physical examination, an electrocardiogram, a chest x-ray, an abdominal B ultrasound, and blood tests including routine blood tests, HIV, HCV, HBV, and syphilis tests. Some items were not provided to individuals outside of the designated age bracket or if they were pregnant. Regarding children, if their parents agreed, they also underwent this survey at the township hospital, including the questionnaire and physical examination. At the end of the survey, each participant received 20 RMB and their fingerprint was recorded on a list of individuals in the region. All of these procedures and the survey content were approved by the institutional review board (IRB) of the Si-chuan Center for Disease Control and Prevention. ## Measures Socio-demographic data were collected with a 29-item face-to-face interview that was designed to assess HIV related risks and pilot tested to be sensitive to the local economic and cultural situation. The questions used in this study involved three sub-sections in the questionnaire: demographic characteristics (such as name, age, gender, marital status, education and occupation, and so on), high risk behaviors (such as drug use and multiple sexual partners), and knowledge of HIV/AIDS. All residents were provided the physical examination and blood tests. Because of the possibility that children under age 14 were infected through mother-to-child transmission and because local customs permit young people (over 14 years old and unmarried) to have multiple concurrent sexual partners, we only analyzed data from subjects over 14 years old. A sub-group analysis was conducted among single participants. Age was divided into the following groups: 14–24, 25–34, 35–44, and 45+. Because most people were illiterate, and only a few went to junior high school, we combined those that had attended primary school or higher education into one group (primary school or above). Marriage status was divided into three groups: never married, married, and divorced/widowed. Most people reported their occupations as student, farmer, or migrant worker (worked outside the town); very few people had other careers, so they were combined into the “student and other” group. The local Yi culture allows multiple sexual partners for both young men and women before marriage, but extramarital sex is not acceptable after marriage. Due to these customs, only participants who were 14 years of age or younger and single were asked about their sexual partners in the interview. If the response to the question, “Do you have a sexual partner?” was “Yes”, the subjects would be asked follow-up questions such as “how many sexual partners do you have?” Drug use status referred specifically to heroin use and was determined from two sources of information: 1) survey questions, such as “Is there any heroin use among your friends?”, “Have you ever used heroin?”, and 2) drug use status reported by village chiefs. Ultimately, we decided to rely on the village chiefs' report of drug use to determine individual drug use status. We found that survey questions were too sensitive due to government drug control enforcement; thus, severe reporting bias existed and very few participants reported drug use. Village chiefs were determined to be a reliable source of information because of their status and the intimacy of the local population. People in a village communicate frequently with each other and drug use status is often known by the entire community. Furthermore, the village chiefs' prestige facilitated the acquisition of additional information about individuals, and they often help the government in enforcing drug control efforts. To determine HIV knowledge, participants were first asked “Do you know of AIDS?”, if they answered “No”, they would be coded into the “no knowledge group”, if the answer was “Yes”, they would be asked four questions about their knowledge of HIV blood transmission, including injectable drugs, sexual transmission, and pregnant mother-to-child transmission. If the participant answered “Yes,” that he or she knew of the respective type of HIV transmission to all four questions, the participant would be coded into an “all knowledge” group. If the participant did not answer all four questions correctly, he or she would be coded into a “some knowledge” group. Our survey did not include questions about HCV, so our analysis of the relationship between knowledge and HIV infection does not take into account HCV knowledge. ## Blood testing All blood samples were transported from township hospitals to the Butuo county hospital for serum separation by laboratory technicians within 24 hours, and then transported to the Liangshan Prefecture CDC for HIV and HCV testing. HIV antibodies were screened using enzyme-linked immunosorbent assay (ELISA; Anti- HIV Antibodies ELISA Diagnostic Kit, Livzon Diagnostics Inc, China; Anthos 2010, Anthos Fluido 2, Biochrom Ltd.); if the result was positive, two additional assays (the original assay plus a different, confirmatory assay) were conducted in parallel. If both were positive or the results were discordant, a confirmatory test was done using a bolt assay (Gebelabs Diagnostics Pte Ltd., Caverdish, Singapore). An ELISA for anti-HCV IgG antibodies was conducted to determine HCV infection status (ELISA, HCV Antibodies Kit, King Hawk Pharmaceutical, Ltd.; Anthos 2010, Anthos Fluido 2, Biochrom Ltd.), according to the manufacturer's protocol. All tests were performed strictly under national operation standards. The results were filed in the participants' physical examination forms by the researchers. ## Data analysis All data collected by paper-and-pencil surveys were inputted manually into a custom designed database and analyzed using SPSS for Windows Version 17.0 (IBM Corporation, Armonk, NY, USA). Descriptive statistics were generated for each general characteristic variable. We used Chi-square tests to compare differences between different demographic and risk groups. Adjusted and unadjusted logistic regression analysis was performed to test risk factors associated with HIV, HCV, and co-infection. All statistical tests were two-sided with a significance level of *P*\<0.05. # Results ## Demographic characteristics Of the 10,939 residents registered with the village chiefs in the two towns, 10,104 (92.4%) participated the survey and 9,179 provided blood samples for HIV and HCV testing between April 2010 to December 2010. Since local Yi indigenous culture considers age 14 as the start of adulthood, when sexual debut and marriage are acceptable, we included only the 6,072 participants who were at least 14 years old and who completed both the survey and HIV/HCV testing. The demographic characteristics of the study participants are shown in. Participants were predominantly of Yi nationality (99.6%), nearly half (48.6%) were males, and most participants were married (80.7%), illiterate (84.4%), and worked as farmers (74.5%). Only 3.7% of participants reported seeking job opportunities outside the district. ## HIV/HCV prevalence by demographic, drug use status, and HIV knowledge The rates of HIV, HCV, and HIV/HCV co-infection for all residents over 14 years old were 11.4%, 14.0%, and 7.7%, respectively. Compared to other age groups, the 25–34-yr age group had the highest prevalence of HIV, HCV, and HIV/HCV co- infections, reaching rates of 24.4%, 26.2% and 16.0%, respectively. Males had a much higher prevalence of all infections compared to females (HIV: 16.3% vs. 6.8%, HCV: 24.6% vs. 3.9%, HIV/HCV co-infected: 14.7% vs. 1.1%, respectively; *P* = 0.000 for all), and married participants had higher rates of HIV, HCV, and co-infection, followed by single participants and divorced or widowed participants. Using the drug use information provided by village chiefs, we found that about half of drug users tested positive for HIV (48.7%) and more than half (68.4%) tested positive for HCV. In addition, migrant workers were more likely to test positive for HIV and HCV and co-infection than other work groups (*P*\<0.01). There was no significant difference in rates of HIV, HCV, and co-infection by education level. Participants who had the least knowledge about HIV had the lowest rate of HIV infection (*P* = 0.00). Bivariate and multivariate logistic regression analysis were undertaken to identify risk factors associated with HIV and HCV infection. presents the adjusted odds ratio (AOR) and unadjusted odds ratios (UOR) for HIV, HCV, and co- infection. Significant risk factors indicated by the AOR were age group (especially 25–34 years old, AOR: 13.6, 95% \[confidence interval\] CI 9.4–19.7 for HIV, and AOR: 13.8, 95%CI 9.7–19.5 for HCV) and drug abuse (AOR: 6.9 for HIV and AOR: 11.8 for HCV). Other risk factors, including being male, illiterate, working as a farmer, migrant worker, divorce/widowed, were marginally significant in multivariate logistic regression models. There was no significant relationship between knowledge and HIV infection. Males were more frequently infected by HCV than by HIV; the AOR rose from 2.4 for HIV (95%CI: 2.0–2.9) to 7.8 for HCV (95%CI: 6.3–9.7). ## Subgroup analysis: HIV/HCV prevalence among singles shows that HIV and HCV infection rates in the singles sub-population were 9.3% and 12.4%, respectively, both lower than the overall study population, especially among women (HIV: 6.8% overall vs. 2.6% among single women; HCV: 3.9% overall vs. 2.9% among single women). Males had a higher rate of HIV and HCV infection than females (HIV: 13.3% vs. 2.6%, HCV: 18.0% vs. 2.9%, respectively); these differences were significant (*P*\<0.01). Participants who reported having multiple sexual partners had higher HIV and HCV infection rates (HIV: 13.9%, HCV: 18.9%, HIV and HCV co-infection:11.8%) than those reporting none or only one sex partner. There was no significant difference in infection rates between groups stratified by HIV knowledge. The difference in HIV and HCV infection rates by different education levels was significant (*P*\<0.05; OR = 2.29). Drug use was the most important risk factor for HIV prevalence. HIV prevalence in IDUs reached 54.8%, and the rate of HCV infection was even greater, at 71.0%. We also carried out univariate logistic regression analysis, which showed that five factors were significantly associated with HIV and HCV infection: gender (OR = 5.8), education (OR = 2.29), occupation (student as reference; farmer: OR = 5.02, migrant worker: OR = 6.12), drug abuse (OR = 18.0), and multiple sexual partners (OR = 2.92;); knowledge was not significant. # Discussion In this study, the prevalence of HIV was 11.4% in the \>14 years age group. This is the highest ever reported rate by annual surveillance performed in Liangshan Prefecture or other areas in the People's Republic of China, and far greater than the total population prevalence in the People's Republic of China (0.057%) ; these numbers are even higher than mean infection levels in Sub-Saharan Africa (5.0%). These numbers imply that HIV prevalence in this region is very serious, and surrounding areas should increase HIV testing efforts (we have already started extending the scope of this survey). The results also showed that the rate of HIV infection among males (16.3%) was much higher than among females (6.8%). This finding is consistent with surveillance results in Liangshan and suggests that there are greater risk factors for HIV in males than in females. Results of risk behavior analyses confirmed that men did indeed have higher rates of intravenous drug use (4.5% vs. 0.8%) and multiple sexual partners (6.9% vs. 2.1%) than women. Our study also found high levels of HCV prevalence (14.0%) compared to worldwide estimates of HCV infection (3.3%), and the total population prevalence of HCV in the People's Republic of China. The rate of HCV infection among men was also higher than among women. This finding is consistent with higher HIV rates among men than women, providing evidence that HCV shares the same transmission routes as HIV. However, the proportion of residents with HIV that were co-infected with HCV (62.5%) was lower than what has previously been reported with respect to HCV transmission through drug use. This implies that blood transmission is not the only primary transmission route for HIV in this region, in which the HIV epidemic was historically driven by injection drug use, and that sexual transmission is now accelerating its spread. This was comparable to annual surveillance results in this region, which show that the proportion of residents infected with HCV by sexual transmission is rising. Furthermore, men had a higher risk of HIV and HCV co-infection than women. The results in also showed that the AOR was 2.4 for HIV and 7.8 for HCV, which suggested that the greater risk of co-infection in men compared to women could be attributable to a greater risk of HCV infection. This may be due to men having high rates of needle sharing, and is consistent with results that HCV can transmit through intravenous drug use 10 times more efficiently than HIV. The results of the logistic analysis showed that drug use (OR = 6.9 in the over 14 age group analysis; OR = 18.0 in unmarried participants over 14 years of age), multiple sexual partners (OR = 2.9 in unmarried participants over 14 years of age), being a migrant worker (working outside of the town, OR = 3.36), and being married (OR = 1.33) were risk factors for HIV infection. However, illiteracy and lack of knowledge did not increase the risk of HIV infection. There may three reasons for this finding. First, the education level in this region was very low: 84.3% of the total population was illiterate and few people went to school. Even those that did go to school achieved very low literacy scores. Second, many subjects were already infected when they learned about HIV. It is possible, then, that HIV risk behaviors occurred prior to acquiring knowledge of HIV. This result, however, should not undermine the importance of HIV education. Instead, our results stress the need for the government to enhance overall education in these areas. In this study, we also found that the prevalence of HIV in the 14–24-yr and 25–34-yr age groups were higher than in the other groups, at 24.4% (OR = 16.3) and 11.9% (OR = 6.9), respectively. This finding implied that individuals in these two age groups had more risk factors than those in the other groups, especially in the 25–34-yr age group. Individuals in this age bracket were found to have higher rates of drug abuse, were more sexually active and had a longer period of risk for exposure of HIV, as the first HIV infection was reported in 1995 in this region. Among all participants, men had a higher HIV infection rate than women (OR = 2.67), suggesting that men have greater risk factors for infection. A higher rate of drug use (male: 4.5%, female: 0.8%) among men confirmed this point. Due to the local culture in this ethnic minority, we did not ask about multiple sexual partners for participants who reported that they were married; however, we have collected sufficient data from the groups that were aged over 14 years and those that were unmarried, and hope that this data will make up for this deficit. Butuo is a minority area and local customs dictate that before marriage, young people are allowed to have multiple sexual partners (for example, on special dates each month young people are allowed to freely look for sexual partners at public market gatherings), but after marriage, multiple sexual partners are strictly prohibited. As we did not ask married individuals about multiple sexual partners, our analysis here focuses only on unmarried participants. We found that HIV and HCV prevalence in the married versus unmarried groups were 9.3% and 12.4%, respectively. These are both higher than the divorced/widowed group (HIV: 5.6%, HCV: 6.1%). Combined with the results of multiple sexual partners (male: 29.9%, female: 17.1%), which were higher than in other, low HIV prevalence regions, we think that having multiple sexual partners is a very important risk factor for HIV infection. In contrast to the total study population, we found that education was related to HIV infection, but knowledge was not. This suggests that education, which improves comprehension, is necessary to control HIV. Only focusing on several knowledge points is not enough. Compared to the total population, we also found that the OR in the male group increased from 2.7 to 5.8 in the over 14 years old group and also in the unmarried group. This implies that this group is the most important risk population and, aside from drug use, multiple sexual partners may be more common and important in this group than other groups, and more attention should be paid to these groups when developing intervention strategies. Our study has several limitations. Firstly, we did not collect data on multiple sexual partners in the married population because the local culture prohibits extramarital behavior after marriage. Because of this custom, very few participants were likely to acknowledge that they had had multiple sexual partners after marriage. Secondly, drug abuse data was not self-reported by participants. At the time we conducted the survey, the local government was carrying out an anti-drug project in which all drug users would be subjected to mandatory drug treatment. Therefore, many drug users did not dare say they were drug users for fear of having to undergo the mandatory drug treatment. This may have resulted in lower rates of reported drug use and multiple sexual partners in our data. Thirdly, due to the local culture, we did not ask about multiple sexual partners for participants who reported that they were married. Most study participants were unwilling to discuss sexual behaviors, especially because the study was conducted over such a short time. In order to make up for this deficiency, we will conduct separate in-depth interviews about sexual behaviors during 2014, including extramarital sex and sex within marriage. Fourthly, publication has been delayed by the length of time needed to analyze and check the raw data. Fifth, Liangshan includes 17 counties while the vast majority of Yi minority live in 8 counties, which are Zhaojue, Butuo, Puge, Xide, Meigu, Yuexi, Jinyang and Ganluo. These eight counties are adjacent to each other and share a great amount of similar cultural characteristics and social traditions. Therefore, in this study we only choosed two towns for a population-based survey, we know this can not represent the whole minority region of Liangshan. and hope we can expand the scope of the study in the future. In summary, we believe that drug use, multiple sexual partners, and low education levels were the three main risk factors leading to HIV prevalence in this region. The government should pay more attention to improving education and personal health consciousness while enhancing drug control programs and changing bad habits, as well as unhealthy minority customs. We are grateful to the Liangshan Government for supporting the project. We thank doctors from the People's Hospital of Liangshan Prefecture and Butuo County Hospital, and lab technicians from the Butuo County Hospital and Lianshan Center for Disease Control and Prevention for their assistance. We thank all the patients and their families who participated in this study. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: CD ZJH WY. Performed the experiments: HL GQ DW. Analyzed the data: CD ZJH. Contributed reagents/materials/analysis tools: JH BD WL LL YY YH LZ ZS LN QW HD JZ HL GQ DW. Wrote the paper: CD ZJH MCM FYW.

# Introduction Western equine encephalitis virus (WEEV) is an alphavirus endemic to and enzootic in North America. The virus, first isolated in 1930 by inoculation of brain material from a sick horse into a healthy one, is maintained in the zoonotic cycle which involves transmission from (primarily) *Culex* (*Cx.*) *tarsalis* mosquitoes to passerine birds. The range of the virus principally corresponds to the geographic distribution of the mosquito vector. Periodically, humans or equines exposed to the virus via *Cx. tarsalis* or other bridge vectors have resulted in large epidemics or epizootics of encephalitis. The first equine outbreaks recorded in the 1930's affected several hundred thousand equines with mortality rates of up to 50%. At least 10 states in the US and 5 Canadian provinces had human and/or equine outbreaks during the 1930's–1950's. While human epidemics were not as severe as equine epizootics, the case fatality was still high ranging from 8 – 15%. Outbreaks continued to occur in the western parts of Canada and the US through the 1970's but the magnitude of these episodes was decreasing over time. The last documented human case of WEE in the US occurred in 1999 when a single patient in Minnesota was diagnosed with the disease. Canada has seen a similar decline in WEEV presence with little or no activity detected since the 1990's. The disappearance of WEEV disease in humans and equids suggested that WEEV had become extinct. However, the periodic isolation of the virus from mosquitoes during non-epidemic periods combined with the detection of seropositive animals in surveillance programs indicated that the virus is indeed still present in zoonotic cycles. Several studies using multiple North American WEEV isolates failed identify a general temporal decline in mammalian, avian, host vector, or *in vitro* virulence since the mid-1900's peak in mammalian cases. Our previous work examined the virulence of several distinct strains of WEEV and demonstrated significant differences in both viremia and mortality in an outbred mouse model. In contrast to earlier studies, among the specific isolates tested, virulence did appear to decline with time. In agreement with previous outbred mouse studies, the earliest isolate we examined, McMillan (McM), originally isolated from a human case in 1941, was lethal to CD-1 mice by 4 days post-infection (dpi) in a dose independent manner. This is in contrast to the most recent isolate in that study, IMP-181 (IMP; isolated in 2005 from mosquito pools), which induced no mortality in the same animal model system. Based on those results, we hypothesized that there exists a viral genetic explanation for the differing levels of disease activity in that model. To understand the role of specific viral determinants that decrease virulence and to identify viral genetic regions or mutations responsible for this phenotypic variation, we developed a series of McM/IMP chimeric viruses and point mutants that would precisely locate which, if any, viral genetic elements regulated virulence in an animal model. Additionally, we used this same panel of viruses in *Cx. tarsalis* to evaluate the ability of various viral genetic elements to modulate infection or dissemination within the vector. Our results identify specific amino acids which impart significant changes in virulence within mice and infection within the vector mosquito. # Methods ## Cell culture BHK-21 (BHK; baby hamster kidney), C6/36 (*Aedes albopictus*), DF-1 (chick), and Vero E6 cells were obtained from American Type Culture Collection (Manassas, VA) or from the CDC cell culture reference section. The Neuro-2a (N2a) murine neuroblastoma cell line was a gift from Dr. Mark Zabel (Colorado State University) and was originally obtained from ATCC (catalog \#CCL-131). All cells were propagated in minimal essential medium with 10% heat-inactivated fetal calf serum at 37°C (28 °C for C6/36) in a 7% CO₂ incubator. ## Virus strains WEEV McM, originally isolated in 1941 from the brain of an infected human, isolate was obtained from the Arbovirus Reference Collection at the Centers for Disease Control and Prevention, Fort Collins, Colorado, USA. The WEEV McM infectious clone was constructed by Dr. Thomas Welte, CSU. WEEV IMP was isolated from a *Cx. tarsalis* in 2005 in Imperial County, CA. The WEEV IMP infectious clone was constructed by Dr. Michael Anishchenko and obtained from Dr. Aaron Brault (CDC, Ft. Collins, CO). Passage history prior to cloning was as follows: WEEV McM-MP2, SMB1, V2; WEEV IMP-V2 (MP, mouse; SMB, suckling mouse brain; V, Vero cells) (Logue 2009). Seed stocks for these experiments were made by electroporation of viral RNA from infectious clones into BHK cells grown in minimal essential medium with 10% fetal calf serum. Cell culture supernatant was collected when 90% of cells showed cytopathic effect (48–72 hrs) and stored in aliquots supplemented with 20% FBS at −80°C. Virus titer was determined by plaque assay on Vero cells as previously described. ## Chimeric virus construction Where possible, chimeric viruses were constructed using the existing conserved, unique restrictions sites, KpnI and AvrII, present at WEEV McM nucleotides 7576 and 9682, respectively. Construction of the chimeras receiving material not bordered by these sites and the point chimeras was based on a strategy using type IIs restriction enzymes adapted from Blakqori and Weber and is outlined in detail by Saxton-Shaw et al for the generation chikungunya/o′nyong nyong chimeric viruses. This method allowed the generation of chimeras and point mutants without the introduction of nonnative nucleotides to the virus. In general, cloning was completed in a three step process. First, an amplicon was generated by PCR from the parental backbone virus clone extending from the AvrII or KpnI site toward the other (inward). The PCR primers would add a site for convenient cloning into the pUC19 polycloning site. To the other end of the amplicon, a SapI recognition site oriented inward was added so that the restriction site was the exact sequence at which the change was to occur, the chimera junction. A second restriction site was added outside the SapI site for cloning into pUC19. This amplicon was cloned into a modified pUC19 plasmid that previously had the BsmBI sites and SapI site removed. A second PCR amplicon was generated from the clone donating the fragment from the AvrII or KpnI site to the chimera junction. Outside the AvrII or KpnI site, this amplicon contained the same convenient recognition site as was added outside the SapI site in the previous amplicon. On the opposite terminus was added a SapI site oriented inward such that restriction would occur at the same site as on the first amplicon, leaving complementary overhangs. Digestion of both the previous plasmid and the second amplicon with SapI and the terminal convenient restriction site and subsequent ligation generated a plasmid cassette containing a chimeric AvrII-KpnI region. Digestion of this plasmid and the parental virus clone with AvrII and KpnI and subsequent ligation resulted in a full-length chimeric WEEV clone free of artifactual base changes introduced by the cloning process. Generation of pair 5, which exchanges a region not bordered by AvrII or KpnI, required an additional step. Generation of the first cloning plasmid was performed with an amplicon that contained a linker enzyme site (MfeI) between the added SapI site and selected pUC19 cloning site. The second amplicon was generated from the same parent and also contained a pUC19 cloning enzyme site outside AvrII or KpnI and an MfeI site outside the SapI site. This amplicon was cloned into the first plasmid using a second pUC19 cloning enzyme site and MfeI, generating a -SapI-MfeI-SapI- motif at the point of foreign DNA insertion. The DNA to be inserted was amplified from the other parental clone, adding SapI sites to both termini, oriented inward such that the restriction sites were situated over the desired cloning junction. Digestion of this amplicon and plasmid with SapI and ligation created a cloning cassette containing the chimeric region from the AvrII to KpnI sites that were then inserted back into the parental clone as above. Finally, the point mutants were generated by a three-step process. The region of interest, including the site to be mutated and flanking sequence to include convenient, ideally unique, cloning sites, was broken into two PCR fragments. Each PCR fragment was comprised of the sequence from the mutation site through the cloning site. The PCR primers contained SapI recognition sites oriented inward toward the mutation site and cut sites modified to leave complementary overhangs containing the desired base change. Digestion of the PCR amplicons with SapI and the relevant enzymes to restrict the opposite ends followed by a three-part ligation with a modified pUC19 plasmid yielded a plasmid construct containing the region of interest with the desired base change. Once the change was confirmed by sequence analysis, the region was digested and removed from the provisionary plasmid construct and inserted into the full-length WEEV plasmid. Full-length plasmid constructs were confirmed to have incorporated the desired change by sequence analysis. ## Virus growth curves T-25 flasks or 24-well plates (Corning, Corning, NY) were seeded with BHK, N2a, C6/36, or DF-1 cells. When approximately 90% confluent, cell monolayers were infected with virus at low multiplicity (MOI = 0.01–0.05). At specified times, supernatant was removed from three wells or two flasks for each virus on each cell type and placed in a screw-cap cryovial at −70°C until titration by plaque assay. Titration results for each virus were compared at all time points by the two tailed t-test. ## Chick infection studies The use of chickens was reviewed and approved by the Animal Care and Use Committee at Colorado State University (protocol \#12-3418A). Care and handling of chicks was in accordance with the PHS Policy and Guide for the Care and Use of Laboratory Animals. Three day old chicks (n = 6/virus) were infected with 10⁴ plaque-forming units (pfu) of McM and IMP viruses by subcutaneous injection and monitored daily by a veterinarian for 7 days. Chicks were bled at days 1, 2, 3, and 4 dpi to determine viremia. The humane endpoint for these experiments was at the termination of the experiment (7 dpi). ## Mouse virulence studies The use of all mice was reviewed and approved by the Animal Care and Use Committee at Colorado State University (Protocol \#11-2605A). Five week-old female CD-1 mice (The Jackson Laboratory, Bar Harbor, ME or Charles River Labs, Wilmington, MA) were allowed to acclimate for 7–10 days. Viral stocks were diluted in serum-free MEM to a concentration of 2×10⁴ pfu per mL. Mice were infected in groups of 5–10, at least two replications per virus (final n = 10–15), by subcutaneous (s.c.) injection inside the left thigh with 10³ pfu in 50 µL. Inocula were titered by plaque assay on Vero cells to confirm dosage. All infected and uninfected control mice were visually monitored by laboratory personnel at a minimum of two times daily. Additionally, mice were monitored once daily by Laboratory Animal Resources and results of these observations reported to the University Veterinarian. Our protocols allow continuation until clinical signs are evident. Mice exhibiting lethargy, unresponsiveness, or neurological signs (postural instability, seizures, piloerection, and/or paralysis) were euthanized by CO₂ inhalation. The day a mouse was euthanized was considered the day of death for calculation of mean time to death (MTD). Survivorship was followed for a period of 14 days and compared among virus constructs by Fisher's exact test. ## Mosquito infections Three to four day-old adult *Cx. tarsalis* mosquitoes were fed on a blood meal containing WEEV. The blood meal contained equal parts of virus, FBS with 10% sucrose, and goose blood (Colorado Serum CO., Boulder, CO) washed with phosphate-buffered saline and packed by centrifugation. A Hemotek feeding system (Discovery Workshops, Accrington, Lancashire, UK) was used to deliver the blood meal to the mosquitoes at 37°C. The mosquitoes were allowed to feed for 1 hour on the apparatus. The fully engorged females were separated and placed into a humidified environmental chamber (Thermo Scientific 3960, Houston, TX) and held at 28°C for 8 days until processing. Blood meal titers were determined by plaque assay and ranged from 2.0×10⁵–2.0×10⁶ pfu/mL. After the 8 day holding period, mosquitoes were cold anesthetized and decapitated, placing the bodies into 1.7 ml tubes (Eppendorf, Hauppauge, NY). A 400 µl aliquot of Dulbecco's minimal essential medium (DMEM) (Gibco, Carlsbad, CA) supplemented with 10% fetal bovine serum (FBS), 100 U/ml of penicillin and streptomycin, 1 U/ml of fungizone and gentamycin was placed into each tube and the sample was homogenized using a pestle (Kontes, Vineland, NJ). The supernatant was clarified by filtration through a 0.2 µM syringe filter (Pall, Ann Arbor, MI) into a clean tube and placed at −70°C until use. Virus presence was determined by pipetting 100 µL of clarified supernatant onto Vero cells in a 96-well plate. The plate was incubated at 37°C under 5% CO₂ and monitored daily for cytopathic effect (CPE). Samples were considered virus positive if CPE was observed. # Results ## WEEV McM E2 amino acid 214 is necessary for mouse virulence A panel of McM/IMP chimeras was constructed to identify the determinants of WEEV McM virulence in CD-1 mice. Rescued virus was administered by subcutaneous inoculation of 6 week-old CD-1 mice. Chimera pair 1 containing exchanges in the E1 gene region retained phenotypes similar to the parental viruses with 86% mortality and 0% mortality for the McM-based and IMP-based chimeras, respectively. Exchange of the region from the unique AvrII restriction site at McM nt 7576 to the unique KpnI site at McM nt 9552 (pair 2;) did transfer the virulent phenotype with the McM-based chimera losing virulence and the IMP-based chimera becoming virulent (80%) but with a relatively delayed mean time to death of 7.9 days. These results suggested that virulence is determined by capsid, E3, and/or E2 genes. Chimera pair 3 and 4 exchanged the capsid gene from the AvrII site to the 3′ end and the E2 gene from the 5′ terminus to the KpnI site. Exchanging the capsid gene did not confer or remove the virulent phenotype. The IMP and McM-based chimeras exhibited 0% and 80% mortality in CD-1 mice, respectively. However, exchanging the E2 gene did have an effect on virulence. The McM-based chimera containing the IMP E2 gene completely lost its virulence while the IMP-based chimera with the McM E2 gene exhibited 40% mortality (MTD = 7.7 days). The E2 gene was further manipulated by exchanging E2 nts 1–518 or E2 nts 519–1226. Pair 5, exchanging E2 nts 1–518 retained the phenotype of the parental viruses, though mortality was further decreased for the McM-based chimera (30%). Placing the IMP-derived E2 nts 519–1226 in the McM backbone reduced mortality to 0% (pair 6). The reciprocal chimera also exhibited reduced mortality (10%). Therefore the minimal necessary determinant essential for mouse virulence was narrowed to a region of McM E2 from nt 519 to the KpnI site at 1226. Alignment of the IMP and McM E2 sequences (accession numbers GQ287641 and GQ287640, respectively) revealed 19 nucleotide differences between the strains in this region. Of these, 11 were silent and eight led to seven amino acid differences (two nucleotide changes are involved in a single amino acid change; ). Fourteen point mutants were constructed representing the reciprocal chimeras for each of the seven loci. Exchanges at six of the seven loci (aa 181, 217, 224, 231, 277, and 390) retained the parental neurovirulence phenotype in the mouse. Mortality rates associated with the McM-based point mutants ranged from 70–100%. Despite retaining a generally virulent phenotype, placement of the IMP-like amino acid at positions 181 and 224 yielded significantly less mortality than the McM parent (p\<0.05). Mortality among IMP-based mutants ranged from 0–10%, none significantly different than the IMP parent. Only the exchange of amino acid 214 (Q to R) affected the pathogenic phenotype. Mortality following infection with the IMP-based virus containing the McM E2 Q214 (IMP E2 R214Q) residue remained at the previously observed 0%. The reciprocal chimera \[McM-based mutant containing IMP-like E2 R214 (McM E2 Q214R)\] showed significant loss of virulence relative to the McM parent virus as only a single animal succumbed to disease (n = 10; p\<0.001) indicating that while not sufficient to confer virulence in an otherwise attenuated backbone, the McM E2 amino acid 214 is necessary for virulence in our mouse model of WEEV infection. The single animal that succumbed following infection with the McM E2 Q214R appeared to follow an atypical disease course (hind limb paralysis only) that will be examined in future studies. That animal and the single mouse that succumbed to infection with IMP E2 S390A are the only examples in these studies of infection with a virus containing arginine at E2 aa 214 resulting in a lethal outcome. ## Baby chicken infection results with structural chimeras and E2 point mutations As avians are the natural reservoirs for WEEV, we further sought to examine the characteristics of these viruses in a readily available avian model. 10⁴ pfu of McM, IMP, McM E2 Q214R and IMP E2 R214Q viruses were subcutaneously injected into 3 day old chicks to observe any differences among the viruses in their resulting viremia within an avian model. Viremia in chicks was detected for McM at 1 dpi but no viremia was detected at days 2–3 dpi. Viremia was detected at 1–3 dpi in chicks infected with IMP or IMP E2 R214Q virus. Though none of the viremia differences reached statistical significance, the highest measurable viremias occurred on day two and were associated with IMP or IMP E2 R214Q viruses. In the present study, none of the infected chicks showed signs of disease during the 7 dpi monitoring period. Hardy et al showed 1 day old chicks were highly susceptible to WEEV-induced disease. It is not clear if the different results are due to an age-dependent susceptibility to WEEV disease or some other detail of our model. However, that viremia was more often associated with IMP infection suggests that the observed McM virulence in mice is not due to an extraordinary general virulence in vertebrates. Additional studies in avian hosts, including passerine birds, are proposed for the future. ## Mosquito infection results with structural chimeras and E2 point mutations In addition to the mouse virulence phenotype, we used McM/IMP chimeras to investigate the determinants of the observed difference in mosquito infectivity. *Cx. tarsalis* mosquitoes were allowed to feed on blood meals containing each of the 12 structural chimeras and 14 E2 point mutation exchanges. The infectivity of the parent strains was 10.0% and 78.9% for McM and IMP, respectively. Clone pairs 1 and 3 yielded no significant change from the original parent infection phenotype; however pair 2 restored the infection phenotype and suggested a determinant in the capsid, E3, or E2 regions. 87.9% of *Cx. tarsalis* given McM backbone virus were infected and 30% of mosquitoes given IMP backbone virus were infected. Clone pairs 4, 5, and 6 further identified the determinant of infection as residing in the same region of E2 (nts 519–1226) that determined mouse virulence. Pair 6 had a demonstrated infectivity of 23.8% for both rescued viruses, representing a deviation from the parental infection phenotype. Infectivity for the point mutations at E2 amino acids 181 and 214 were 5.1% and 11.7%, respectively, for the IMP strain mutants and 13.3% and 0% for the McM mutants, representing a significant reduction of the IMP parental mosquito infection phenotype (p\<0.001 for both IMP-based viruses), but not restoration. The remaining pairs resulted in a slight reduction and an approximately 3-fold increase in infectivity for the IMP and McM-based chimeras, respectively (the pairs exchanging amino acid 224 and 231 will all significantly different than the parent viruses, p\<0.05). These results suggest that while there may be multiple determinants, the IMP arginine at E2 position 214 is critical but not sufficient for mosquito infectivity. ## Multi-step growth curves of McM, IMP, and E2 aa 214 point chimeras To determine whether the observed phenotypes were due to general replicative differences, multi-step growth curves of McM, IMP, or the McM or IMP-based E2 aa 214 point chimeras were performed in both mammalian and insect cells. In BHK cells, all four viruses grew similarly well, rapidly increasing in titer from 10–100 pfu/mL to 10⁶–10⁷ pfu/mL. No consistent statistical differences were observed among the four viruses, though the titer of McM E2 Q214R was significantly greater than McM, IMP, and IMP E2 R214Q 72 hpi (p\<0.05). N2a murine neuroblastoma cells were used to determine whether growth differences among viruses might account for the observed differences in mouse mortality due to neuron infection. Again, no overt growth difference was observed among, McM, IMP, McM E2 Q214R, and IMP E2 R214Q viruses through 72 hpi. Though at 72 hpi, IMP trended toward having lower virus titers than McM, at both 48 and 72 hpi, no statistically significant difference in virus titer was observed between any of the four viruses. Cultured DF-1 chicken embryo fibroblast cells were infected with McM, IMP, McM E2 Q214R, and IMP E2 R214Q viruses to observe any growth differences that would be regulated by the virus-vertebrate reservoir dynamics. In this cell line, significant growth differences were observed, especially through 24 hpi. All 4 viruses were different at 18 hpi, with the McM and McM E2 Q214R McM viruses 1–2.5 logs higher than IMP and IMP E2 R214Q viruses at 24 hpi. Peak viral titers were significantly different between McM (24 hpi) and IMP (36 hpi), but no other viruses. The titers for the point mutants were significantly higher than the parent viruses at 48 and 72 hpi. In C6/36 mosquito cells, all four viruses showed similar replication kinetics with no statistically significant differences observed. Mean titer measurements at 72 hpi ranged from 6.8 Log10 pfu/mL for IMP (E2-R214Q) to 7.5 Log10 pfu/mL for McM. # Discussion Extensive pathogenic studies have been performed on several alphaviruses strains and the availability of infectious clones for many alphaviruses has allowed reverse genetics studies to identify the molecular basis of various phenotypes, including vertebrate virulence and vector adaptation. In different models nsP2, nsP3, capsid, E1, and E2 were all identified as pathogenic determinants,. In this study, we identified minimal infection and virulence determinants of two distinct WEEV phenotypes. A single amino acid, E2 Q214, was identified as necessary for WEEV McM-mediated pathogenesis in a murine model of infection. Importantly, the same amino acid position, E2 R214, was identified as necessary for efficient WEEV IMP-mediated infectivity of the primary mosquito vector, *Cx. tarsalis*. That these two phenotypes are determined by different amino acids at the same position suggests that lethality and transmission are mutually exclusive in this model, suggesting an explanation as to why there are so few mammalian cases in the presence of active transmission. Although IMP E2 R214Q virus had viremic profiles similar to IMP virus infected chicks, the R to Q change significantly affected mosquito infectivity. Curiously, in neither case did the reciprocal point chimera confer the examined phenotype to the previously incompetent parental virus. Such a transfer of increased virulence phenotype likely requires multiple loci, a notion supported by the variable mouse mortality among the chimeric viruses in this study. Our *in vitro* studies conclude that the recombinant viruses used in this report do not have defects in replication as indicated by the growth curve analysis in multiple cell lines. In general, McM showed the most rapid replication kinetics compared to all other viruses. However, *in vitro* results did not correlate well with *in vivo* findings. Mouse virulence or mosquito infectivity of each virus, including the single point mutants, did not correlate with any differences in growth kinetics *in vitro*. These findings underscore the limitations of *in vitro* assays in predicting virulence in a mouse model of infection or infectivity of the primary vector mosquito species. *In vitro* studies remain necessary, however, to demonstrate that recombinant viruses are not overtly compromised with respect to replication kinetics. Among alphaviruses, relatively little is known about WEEV structure and pathogenesis. Much must be extrapolated from studies of other alphaviruses, including Sindbis virus (SINV). The region of the SINV E2 glycoprotein from approximately amino acid 172 through 220 has been implicated in cell recognition, and roughly corresponds to E2 domain B, a disordered domain from amino acid 167 through 254. Cryo-electron microscopic (cryo-EM) reconstruction of SINV E1-E2 complexes suggests that E2 amino acid 214 is located on the extreme tip of the E2 spike and within domain B, supporting a direct role for E2 amino acid 214 in cell recognition or receptor binding. The exchange of a polar neutral glutamine for a positively charged arginine, and vice-versa, at position 214 could be expected to significantly alter such molecular interactions. Similarly, E2 amino acid 181, also involved in vector infectivity is likely located in domain B near the tip of E2. The amino acid switch at this position is more significant, exchanging positively charged lysine and negatively charged glutamic acid. This major change could affect interactions with cellular recognition molecules directly or indirectly by altering the conformation of the E2 spike tip. Previous studies examining SINV vector adaptation have primarily implicated residues located in domain A. Pierro et al. found E2 amino acids 55, 70, 95, 96, and 116–118 to be involved in enhanced SINV infectivity of *Aedes* (*Ae.*) *aegypti*. While the mechanism is likely different based on the location within the protein, amino acids 55 and 70 present a situation analogous to the current report. SINV E2 Q55H/E70K results in increased virulence in suckling mice and increased binding to neuronal cells. Viruses containing the reciprocal mutations, E2 H55Q/K70E demonstrate increased infectivity for *Ae. aegypti*. Chikungunya E2 amino acid 211 and Venezuelan equine encephalitis virus E2 amino acids 207, 213, and 218 were implicated in adaptation to their respective vectors. X-ray crystallographic and cryo-EM structures also place these amino acids at or near the tip of the E2 spike. Nagata et al. previously examined the molecular determinants of WEEV pathogenesis in mice. That study differed from the current study in several ways: 7–20 week old BALB/c mice were used, mice were inoculated by the intranasal route, and all isolates used were 100% lethal to mice in this model. The authors showed by correlating structural protein phylogenetic analysis and virulence that the three most virulent strains clustered together and the five least virulent strains clustered together with sequence divergence apparently correlating with changing virulence. Interestingly, two of the three most virulent strains contain E2 Q214. One of these two strains is McM and the other California, a strain phylogenetically closely related to McM. The remaining strains, including Fleming, the most virulent strain used in that study, contain IMP-like E2 R214. That five different uniformly lethal strains contain E2 R214 is not surprising. Given that our data suggest the presence of multiple loci responsible for pathogenesis, it is possible that other differences between these five strains and the McM/California genotype compensate for R214 to render them pathogenic to mice that is not present in IMP or McM. Additional studies will be required to address this possibility. In summary, we have used our previously defined mouse model of WEEV pathogenesis to determine that different amino acids at the same position within the E2 glycoprotein are necessary for both WEEV McM mouse virulence and WEEV IMP vector infectivity. WEEV IMP may have an advantage in the transmission cycle by efficiently infecting the avian host and producing high vector competence in the *Cx. tarsalis* strain. Our studies indicate that increased virulence in CD-1 mice is accompanied by decreased mosquito infectivity, suggesting a plausible explanation for the current lack of epidemic WEEV. The authors thank Cori A. Mossel (CSU) for cloning assistance. The authors would also like to thank Dr. Richard Bowen and the CSU Animal Models Core for performing the chicken studies. The findings and conclusions in this report are those of the author(s) and do not necessarily represent the official position of The Centers for Disease Control and Prevention. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ECM ATP AMP KEO. Performed the experiments: ECM JPL ATP EMB. Analyzed the data: ECM JPL ATP AMP KEO. Contributed reagents/materials/analysis tools: AMP. Wrote the paper: ECM JPL ATP AMP KEO.

# Introduction Epigenetic regulation refers to the molecular events where gene expression is regulated without alterations in the DNA sequence. Epigenetic regulation has been recognized as an extra level of the genetic codes with diverse mechanisms. Previous studies have identified several epigenetic mechanisms, including chromatin remodeling, DNA methylation, histone tail modifications, and non- coding RNAs. Regulation of these epigenetic codes, or called epigenomes, is critical for multiple biological processes, including cell division, reprogramming, and differentiation, which are vital for tissue and organ development. Furthermore, dysregulation or aberrant alterations of the above epigenetic mechanisms have been shown to be associated with developmental abnormalities, biological disorders, and diseases. The accumulating evidence of how important epigenetic modifications are to human health also calls for an increasing demand of epigenetic studies. The diverse post-transcriptional modifications of RNAs, or called epitranscriptomics, have been recognized as another key player in epigenetic regulation mechanisms. The pioneer studies of RNA chemical modifications were performed more than 50 years ago. Owing to the development of advanced biotechnological techniques, including next-generation sequencing and mass spectrometry techniques, the studies of RNA modification and epitranscriptomics have gained popularity again in recent years. Up to date, more than 100 RNA modifications have been identified in multiple RNA types from almost all known living organisms, including messenger RNAs (mRNAs), housekeeping non-coding RNAs, which include transfer RNAs (tRNAs) and ribosomal RNAs (rRNAs), and regulatory non-coding RNAs, which include microRNAs (miRNAs) and long non-coding RNAs (lncRNAs). These modifications have been shown to regulate processing or metabolism of RNAs, including RNA splicing, translocation, stability, and translation efficiency. However, the studies of RNA modifications are still immature, and some recent researches are focused on mapping of RNA modifications, elucidating the biological roles of these modifications, and identifying what molecules are involved in these processes. Previously, the RNA modifications and epitranscriptomics were regarded as a relatively static status for a specific RNA structure. Whereas, recent studies have shown that the modifications of RNAs are a dynamic and reversible process. Similar to other epigenetic mechanisms, the RNA modifications are also controlled by the “writers, erasers, and readers” proteins. Writers are proteins being capable of adding chemical groups to specific sites of RNA molecules, erasers are proteins to remove the modified chemical groups added by writers, and readers are a group of proteins with specialized domains, which can recognize and bind to the modified RNA sites. These proteins work together as a complex network in the regulation of dynamic RNA modifications. Furthermore, dysregulation and mutation of these currently known RNA modification proteins have been shown to be related to human diseases, including cardiovascular diseases, metabolic diseases, neurological disorders, and cancer. These findings illustrate the importance of studying the expression and function of RNA modification proteins. Ontogenic development is a complicated process involving the buildup of genetic and epigenetic signatures. This phenomenon is applicable to specific organs, including liver, which is the key organ in the metabolism of both endogenous and exogenous compounds. The development of the liver and its functions is critical to protect infants and children from exposure to environmental toxicants. The ontogenic expression patterns of several liver-specific genes during development have been reported. There is an increasing amount of evidence showing that epigenetic mechanisms are contributing to the regulation of ontogenic expression of genes, where the expression and function of epigenetic modifying proteins are key players. However, whether RNA modifications, as relatively new epigenetic mechanisms, are also involved in the regulation of liver development and maturation is still largely unknown. In the current study, the ontogenic mRNA expression of RNA modification proteins involved in several frequent RNA modifications, including N⁶-methyladenosine (m⁶A), N¹-methyladenosine (m¹A), 5-methylcytosine (m⁵C), 5-hydroxymethylcytosine (hm⁵C), N⁷-methylguanosine (m⁷G), and pseudouridine (Ψ), was studied in mouse liver at different ages during postnatal maturation using RNA sequencing. # Materials and methods ## Animal experiments, total RNA extraction, cDNA library construction, and RNA-Seq The procedures including animal experiments, RNA extraction, cDNA library construction, and RNA sequencing were described in previous publication. ## RNA-Seq data analysis The RNA-Seq reads from the FASTQ files were mapped to the mouse reference genome (NCBI37/mm9) and the splice junctions were identified by TopHat 1.2. The output files in BAM (binary sequence alignment) format were analyzed by Cufflinks 1.0.3 to estimate the transcript abundance. ## Data visualization and presentation The selected RNA modification enzyme genes were retrieved from Cufflinks output for further analysis. Significant gene expression was determined by the drop-in- deviance F test of the fitted FPKM data to a generalized linear model with a Poisson link function, a statistic designed to measure the significance of a gene’s measured FPKM value relative to a zero FPKM value. Data were presented as the mean ± S.D. Ontogenic expression patterns were presented using the GraphPad Prism 7 software program (GraphPad Software, Inc., La Jolla, CA). # Results ## Ontogeny of mRNA expression of writers, erasers, and readers involved in *N*⁶-methyladenosine (m⁶A) N⁶-methyladenosine involves the addition of a methyl group at the N-6 position of an adenosine base. High-throughput sequencing following m⁶A-specific antibody immunoprecipitation has revealed a high prevalence of this modification in multiple RNA types across different species. The m⁶A is believed to be the most abundant RNA modification in mRNAs, but it also exists in other RNA types, including miRNAs, lncRNAs, tRNAs, rRNAs, and other small RNAs. The m⁶A modification has been shown to play a role in multiple biological processes, including development, metabolism, and circadian rhythm, through regulation of RNA metabolism, transport, and translation. The addition of m⁶A is accomplished by a methyltransferase complex (writers) consisting of several components, including methyltransferase Like 3 (METTL3), METTL14, Wilms tumor 1 associated protein (WTAP), Vir like m⁶A methyltransferase associated (VIRMA or KIAA1429), RNA binding motif protein 15 (RBM15), and zinc finger CCCH domain-containing protein 13 (ZC3H13). Each component in the complex has specific roles in the installment of m⁶A modification at a precise location. As showed in, the expression of m⁶A writers followed specific trends. The expressions of METTL14 and ZC3H13 were the lowest among all tested components in the m⁶A methyltransferase complex. Compared to the prenatal expression, both METTL14 and ZC3H13 mRNA levels dropped \~50% at birth. The postnatal expressions of METTL14 and ZC3H13 showed a constant drop until day 20 and 15, respectively. The expression of METTL3, KIAA1429, and RBM15B showed higher mRNA expression levels in mouse liver compared to METTL14 and ZC3H13. However, similar to METTL14 and ZC3H13, the expression of METTL3, KIAA1429, and RBM15B at birth decreased to only half of the prenatal levels. KIAA1429 showed a constant decrease during development after birth, whereas METTL3 and RBM15B were relatively stable between day 20 and 60. The expression of WTAP was highest among all tested components in m⁶A methyltransferase complex. The expression of WTAP dropped at birth, rebounding during day 0 to 5 after birth. After day 5, WTAP showed a constant sharp drop until adulthood. The dynamic regulation of m⁶A is also controlled by another group of demethylases (erasers). So far, two enzymes have been identified as demethylases for the m⁶A modification, including fat mass obesity-associated protein (FTO) and AlkB Homolog 5 RNA Demethylase (ALKBH5). Furthermore, dysregulation of these erasers has been shown to induce many types of disorders, indicating the important role of homeostatic m⁶A status in health. As shown in, the expression of ALKBH5 had a 45% decrease at birth, remaining stable after birth. The expression FTO had three phases, where the expression level increased slightly before day 5 postnatal, and decreased from day 5 to 30, followed by another increase until day 60. Most m⁶A-mediated cellular functions depend on reader proteins, which can recognize and bind to m⁶A-modified RNAs. The YTH domain containing proteins are the best characterized m⁶A readers. Recruitment of reader proteins has different impacts on the target RNA molecules. Recruitment of YTHDF1 and YTHDF3 mainly regulates translation process. YTHDF2 mainly regulates RNA stability. Two additional members of the YTH protein family, the YTHDC1 and YTHDC2, were also detected in mice liver across different ages. According to previously reported, YTHDC1 was able to influence mRNA splicing process by affecting the function of pre-mRNA splicing factors. In terms of YTHDC2, the protein was initially found to be important in spermatogenesis in male mice, through mechanisms including translation efficiency enhancement. A more recent study also showed that YTHDC2 had the potential to regulate mRNA stability through interaction with exoribonuclease XRN1, which indicated a more general role of YTHDC2 in gene regulation. However, that are the roles of YTHDC2 in liver function is still largely unknown. The proline rich coiled-coil 2A (PRRC2A) is a recently discovered m⁶A reader protein, involved in glial development in mice. The results in showed that the expression of PRRC2A was the highest among all the tested m⁶A readers in mouse liver. The expression of PRRC2A had a constant decrease during liver development. At day 60, the PRRC2A expression was only 20% compared to prenatal expression. YTHDF1 was the most abundant YTH domain containing reader, followed by YTHDF2, YTHDC1, YTHDF3, and YTHDC2. Regardless of the differences in expression levels, all these YTH domain containing readers showed a sharp decrease in expression at birth, ranging from 70% decrease (YTHDC2) to 32% decrease (YTHDF2). ## Ontogeny of mRNA expression of writers, erasers, and readers involved in *N*¹-methyladenosine (m¹A) The N¹-methyladenosine consists of addition of a methyl group at the N1 position of adenosine. The m¹A modification was first identified in tRNAs and rRNAs decades ago. Recently, transcriptome-wide mapping also revealed the existence of m¹A modification in mRNAs. The m¹A modification in tRNAs was important for tRNA folding, stability, and tRNA-protein interaction. The m¹A modification was also observed at several sites in rRNAs, which affect the tertiary structure of the ribosome and downstream gene translation. In contrast to non-coding RNAs, m¹A modification has a relatively low abundance in mRNAs, with a potential function in regulating translation process. TRM6 and TRM61A are two identified m¹A methyltransferases responsible for tRNA modification. Nucleomethylin (NML or RRP8) was reported to catalyze m¹A modification at multiple rRNA sites in both human and mouse cells. These known tRNA/rRNA methyltransferases were also reported to be able to modify mRNAs, but specific mRNA m¹A writers remain unknown. As shown in, the expression of all three m¹A writers showed a decrease trend during development. The expression of ribosomal RAN-processing protein (RRP8) was the highest among all tested m¹A writers and it decreased constantly in mouse liver with the highest level at the prenatal age. The expression of TRM6 decreased from prenatal day 2 to postnatal day 10 to about 40%, followed by a relatively stable expression until adulthood. The expression of TRM61A dropped constantly, but the dropping rate was reduced after postnatal day 10. Two proteins in the AlkB family have been identified to be m¹A demethylases, ALKBH3 and ALKBH1. Overall, the expression of ALKBH1 decreased during mouse liver development, with some elevations at days 1, 5, and 25 after birth. The expression of ALKBH3 was much lower compared to ALKBH1 with a constant decrease at all tested ages. The YTH-domain containing proteins YTHDF1, YTHDF2, YTHDF3, and YTHDC1, but not YTHDC2, are also reported to be m¹A reader proteins, which have a role in translation regulation. The expression patterns of m¹A readers were the same to that of m⁶A readers, which were presented above in. ## Ontogeny of mRNA expression of writers, erasers, and readers involved in 5-methylcytosine (m⁵C) The 5-methylcytosine (m⁵C) modification was first identified in DNA molecules, but also found later in RNAs. Utilizing transcriptome wide mapping technique, the m⁵C modification has been identified to be widely distributed in multiple types of RNAs, both coding and non-coding. Known functions of m⁵C modification includes regulation of translation, RNA metabolism, and RNA trafficking. The RNA m⁵C methyltransferases (MTases), which contain S-adenosyl-L- methionine (SAM)-dependent MTase domains, are the major writers of m⁵C modification. Several members of the NOP2/Sun RNA methyltransferase family (NSUN) have been identified to methylate different RNA molecules. Among the tested NSUNs, NSUN2 had the highest expression. The expression of NSUN2 dropped 60% at birth, but rebounded during postnatal day 0 to 10. At day 10, NSUN2 was expressed at similar levels compared to day 0 and started to increase in liver until adulthood. NSUN2 was reported to mediate mammalian mitochondrial tRNAs in several different positions and involve in processes including cell differentiation and motility. The expression of NSUN1 (or NOP2) was the second highly expressed NSUNs following NSUN2, which showed a constant decrease from prenatal to adulthood. Unlike NSUN2, NSUN1 was reported to be a methyltransferases for rRNAs and regulate processes including cell proliferation or drug response. NSUN4 and NSUN5 were expressed in a relatively stable manner across all tested ages. Both NSUN4 and NSUN5 were able to methylate rRNAs. NSUN4 was able to methylate 12S rRNA and played a key role in ribosome biogenesis while NSUN5 was able to methylate 25S rRNA and regulate lifespan and differential stress response in yeast. Three members of the NSUNs, the NSUN3, NSUN6, and NSUN7 were barely expressed in mice liver across all tested ages. Other than NSUNs, the tRNA aspartic acid MTase 1 (TRDMT1 or DNMT2), which was previously believed to be a DNA methyltransferase, was also reported to mediate the formation of m⁵C modification in tRNAs. However, the detected TRDMT1 expression was very low in mouse liver at all ages. The study of demethylase of m⁵C is still very limited. There are currently no identified proteins, which can remove the methyl group at m⁵C modified sites. But it has been reported that m⁵C modification can be converted to other types of modifications, for example hm⁵C. This indicates that the erase or elimination of m⁵C might be accomplished by further modification of the methyl group instead of removing it. The Aly/ REF export factor (ALYREF) is the first identified m⁵C reader protein, which regulates the exportation of m⁵C modified mRNAs. As shown in, compared to prenatal expression, the expression of ALYTEF dropped sharply to 37% at birth. Its expression kept a decreasing trend until adulthood, at which age the expression level was just 14% the prenatal level. The Y-box binding protein 1 (YBX1) is another recently identified m⁵C reader protein, which was reported to promote the pathogenesis of human urothelial carcinoma of the bladder in an m⁵C-modification-dependent mechanism, where the binding of YBX1 maintained the stability of target mRNAs. From, the overall expression level of YBX1 was much higher than ALYREF. But similarly, a 53% drop in expression level at birth was also observed in YBX1, followed by a rebound between postnatal day1 to day 10. After postnatal day 10, the expression showed a relative stable expression. ## Ontogeny of mRNA expression of writers and readers involved in 5-hydroxymethylcytosine (hm⁵C) Similar to m⁵C, the hm⁵C was first identified in DNAs, which was later identified in mammalian RNAs. However, recent studies on hm⁵C are still focused on its role in DNA modifications with very limited knowledge on RNA modifications. Studies have shown that the hm⁵C in the coding sequences of mRNAs increases translation efficiency through unknown mechanisms. The ten-eleven translocation (TET) family proteins, including TET1, TET2, and TET3, were reported to mediate the oxidation of m⁵C to hm⁵C in both DNAs and RNAs. As shown in, the expression of TET3 was the highest in mouse liver followed by TET2 and TET1. The expression of TET3 dropped to 50% at birth and rebounded to 90% at postnatal day 1. This was followed by a constant decrease during liver development. The expression of TET2 increased between postnatal days 5 to 30, where the peak at postnatal day 15 was 1.6-fold higher than the expression at prenatal day 2. The expression of TET1 dropped 43% at birth and remained low during liver development. However, in a previous study, TET-null mouse embryonic stem cells also exhibited significantly high levels of hm⁵C in RNAs, which indicates the existence of hm⁵C writers other than TETs. Currently, there are no identified hm⁵C erasers in RNAs. Future studies are still needed to cover this knowledge gap. The ubiquitin-like with PHD and ring finger domains 2 (UHRF2) proteins might have the potential to bind to hm⁵C modified RNAs. The UHRF2 was reported to bind to hm⁵C modified DNAs. shows that the expression of UHRF2 dropped sharply at birth, with a 60% decrease comparing to prenatal day 2. Its expression increased slightly during postnatal days 0 to 5, followed by a constant drop until adulthood. ## Ontogeny of mRNA expression of writers involved in *N*⁷-methylguanosine (m⁷G) N⁷-methylguanosine (m⁷G) is another most prevalent modifications existing in multiple types of RNAs, including tRNA, rRNA, and mRNA. In tRNAs, catalyzed by the methyltransferase like 1 (METTL1)/WDR4 complex in human and mouse, m⁷G was reported to regulate the stability and decay pathways of tRNAs, which further involved in the regulation of physiological processes including cellular development and human disease. The m⁷G is also conserved in rRNAs in multiples species. Catalyzed by the WBSCR22/TRMT112 complex, where the TRMT112 protein acts as a co-activator of the methyltransferases WBSCR22, m⁷G modification was reported to regulated the processing, export, and maturation of several pre-rRNAs and rRNAs. In mRNAs, m⁷G cap modification was first identified, which is mediated by the RNA (guanine-7-) methyltransferase (RNMT)/RAM methyltransferase complex, and regulate the export, stability, splicing, transcription, translation, and decay of mRNA. Utilizing more advanced sequencing techniques, two more recent studies mapped the distribution of m⁷G in a transcriptome wide manner and found the existence of m⁷G in internal mRNAs and long non-coding RNAs, with a potential role in translational regulation. There are three identified writer enzymes of m⁷G modification. METTL1 was reported to catalyze the formation of m⁷G on both tRNAs and mRNAs. As showed in, the expression of METTL1 was relatively stable during the development of mice across all tested ages. WBSCR22 is another reported m⁷G methyltransferase catalyzing the formation of m⁷G on pre-rRNAs and rRNAs. The expression of WBSCR22 dropped at birth and rebounded to its prenatal level at day 5, followed by another drop. RNMT is the enzyme mediating cap m⁷G modifications. The expression of RNMT was found highest in prenatal mice liver. At birth, the level of RNMT dropped and stayed low till adulthood. The research on identifying specific reader and erasers of m⁷G are still limited and future studies are needed to address these knowledge gaps. ## Ontogeny of mRNA expression of writers involved in pseudouridine (Ψ) Pseudouridine (Ψ), a C-C glycosidic isomer of uridine (U), is derived from the incorporation of C5 into the glycosidic bond. Recent sequencing results have revealed the wide distribution of the Ψ modification in different species of RNAs. However, Ψ still predominantly exists in non-coding RNAs, tRNAs, and rRNAs, with relatively lower expression in mRNAs. In tRNAs, the Ψ modification has been shown to regulate stability of tRNAs. In rRNAs, the Ψ modification has been shown to regulate the ribosome assemble process, which is important to protein synthesis initiation. The function of the Ψ modification in mRNAs is still largely unknown. The Ψ modification was found in multiple regions of mRNAs with no specification preference. Previous studies showed that the existence of the Ψ modification increases the stability of mRNAs against heat shock stress and is associated with higher translation efficiency. In RNAs, uridine is transformed into pseudouridine by a class of enzymes known as pseudouridine synthases (PUSes). Different PUSes have shown preferences for different residues in tRNAs or rRNAs. Several PUSes were identified in mouse liver by RNA-sequencing, as shown in. The expression of PUS1 was the highest among all tested PUSes with a constant decreasing trend from prenatal day 2 to 10, followed by a relatively stable expression. At postnatal day 10, the expression level of PUS1 was only 47% compared to its peak level at prenatal day 2. The remaining PUSes, PUS3, PUS7, PUS7L, and PUS10 all showed a \~50% decrease at birth, in comparison to prenatal expression. The erasers and readers of the Ψ modification are still largely unknown. Future studies are needed to fill in these knowledge gaps. # Discussion Even though RNA modifications have been known for decades, the biological significance of these modifications has only been recognized in recent years, owning to the development of genome-wide and high resolution sequencing techniques. Recent studies have identified more than 150 types of RNA modifications covering almost all RNA types in the genomes of different species. Although the knowledge about the biological functions of these modifications is still limited, it has been shown that RNA modifications are able to regulate RNAs in multiple ways, including metabolism, transport, and translation. Development involves dynamic regulation of gene expression in cells, where RNAs play a non-negligible role in this process. Chemical modifications of RNAs have emerged as promising mechanisms in modulation of development process. Previous studies have shown the importance of RNA modifying proteins in development. Deletion of *Mettl3*, the m⁶A writer gene, has been shown to be embryonically lethal in mice with dysregulation of pluripotency in embryonic stem cells. Knockout of some m⁶A modifiers also affected neuronal development and reproductive function (fertility, spermatogenesis, and oogenesis). Aside from the m⁶A modification, some other types of RNA modifications are also involved in development regulation. Knockout of m⁵C modification writer *Nsun2* or *Dnmt2* has been reported to affect stem cell differentiation and endochondral ossification, respectively. All these findings underscore the importance of RNA modifications and RNA modifying proteins in the development process. In the present study, the mRNA expression of genes involved in several of the most common RNA modifications, including m⁶A, m¹A, m⁵C, hm⁵C, m⁷G, and Ψ, were studied in mouse liver at different ages during development. Most of these RNA modifying proteins showed ontogenic changes in mRNA levels along with liver development, where the majority of them showed dramatic drops at birth followed by downregulations during postnatal development. These data indicate that RNA modification status might also have such ontogenetic changes, which is regulated by the alterations in RNA modifying proteins. The data also showed that most RNA modifying enzymes are high expressed in prenatal compared to postnatal mouse liver, at which stage the cells are more stem-like. The studies mentioned above have suggested the important role of RNA modifications in stem cell functions. So it may be interesting to study how the RNA modifying proteins are expressed in the early stages of prenatal development. However, in the current research, only prenatal day two liver samples were collected and future experiments are needed to analyze the prenatal expression of RNA modifying proteins. Many other liver-specific genes have been reported to show specific ontogenic expression patterns during development, including phase I and phase II drug metabolizing enzymes. However, the factors contributing to this phenomenon are still largely unknown. The currently discovered RNA modifications might involve in this ontogenic regulation of gene expression. Moreover, discrepancies have been reported between mRNA and protein expression for multiple genes. This is very common for cytochrome P450 genes. As post-translational modification mechanisms, RNA modifications might be also responsible for this phenomenon. However, the study of RNA modifying proteins is not sufficient to confirm the involvement of RNA modifications in these biological events. The expression patterns of RNA modifying proteins reported here is not sufficient to explain all genes with different expression patterns. Direct detection of RNA modification types and sites on the RNA transcripts of the genes is still needed to study how RNA modifications involved in the regulation of a specific gene. In summary, the present study provides new knowledge about the ontogenic mRNA expression of multiple RNA modifying proteins during development of mouse liver. Such knowledge can serve as foundation for future studies on the impact of RNA modifications in gene regulation during liver development. Understanding the biological significance of RNA modifications in liver development or functions will also benefit the study of drug metabolism or liver diseases in the future. [^1]: The authors have declared that no competing interests exist.

# Introduction In recent years, a large amount of evidence has been gathered indicating that reach-scale river restoration projects often fail to meet their predefined goals and, in particular, that reach-scale restoration of the local hydromorphological conditions often does not lead to the re-establishment of natural communities. Based on this rapidly expanding body of published case studies, Bernhardt and Palmer noted that river restoration research should progress to identifying the drivers that determine the success or failure of restoration projects. To date, very few attempts have been made to integrate the results of multiple river restoration projects and determine relevant variables for restoration outcomes. However, a sound evaluation of different methods and measures is necessary to increase the success of future restorations. In the few comparative studies that exist, the effects of restoration on the species assemblage was primarily related to ‘abiotic’ variables, such as the restoration measures that were carried out, or the hydromorphological quality of the newly created habitats. These studies found it difficult to identify variables that were conclusively associated with restoration success. Analysis on the outcome of 13 lowland river restoration projects in which flow deflectors and artificial riffles were installed showed inconsistent effects on fish assemblages in terms of species richness, diversity and equitability. Also the effectiveness of wood placement for enhancing fish assemblages varied between individual projects. In addition to the potential effects of ‘abiotic’ variables on the outcome of river restoration projects, the ‘biotic’ components that may determine whether river restorations meet their goals have received increasing attention. In this context, some authors focused on the role of the regional species pool. Applying a filter model, local assemblages in restored sites may be forecasted from the regional species pools based on species dispersal capabilities, and local hydromorphological and ‘biotic’ constraints. In particular, dispersal should play a vital role in structuring the regional population networks within the dendritic structure of a river system. A recent study on the spatial extent of the species pool available for the colonization of restored reaches in streams and rivers in Central Europe found that 96.6% of the fish species recorded in restored reaches had nearby source populations within a range of 5 km around the restored reach. Species with their closest source populations further away than 5 km rarely colonized restored reaches. Where dispersal is not artificially limited, the probability that a restored reach will be colonized by a fish species should depend on the species population structure within the regional species pool. Thereby the colonization of a restored site is largely determined by the propagule rain it receives, which in turn is affected by the number and size of potential source populations within a critical dispersal distance. To date, the role of the population structure in the species pool on the outcome of restoration projects is largely unknown. In addition to quantitative characteristics of the regional species pool, a species’ ability to colonize a restored reach will be determined by its ecological traits. In the context of dispersal and colonization, in particular morphological traits that potentially affect the mobility of species are important. Also other traits like habitat preference and foraging type may be indirectly related to mobility and the ability of species to establish at restored reaches. Furthermore taxonomic affiliation has been shown to be a good indicator for the dispersal abilities of fish. In this study, these variables on population structure and species traits are summed up as ‘biotic’ variables to contrast them to ‘abiotic’ variables characterizing the hydromorphological quality of the restored reaches. To gain a better understanding of the key variables that are best able to explain species assemblages at restored river reaches, we applied separate statistical models to explain the fish assemblages at restored reaches using ‘biotic’ and ‘abiotic’ variables and compared the proportion of variability that they account for. We used data from 18 river restoration projects and associated species pools from Germany to address the following questions: (1) How well do ‘biotic’ variables relating to the regional species pool explain the fish assemblages that colonise restored reaches? (2) Are these ‘biotic’ variables better suited to predict fish assemblages in restored reaches than ‘abiotic’ hydromorphological habitat characteristics? This information is highly relevant for restoration managers, as it can be used for a realistic and target-oriented approach to river restoration, and thus can help to increase river restoration effectiveness. # Materials and Methods ## Ethics Statement All animal work has been conducted in accordance with relevant national and international guidelines. ## Sampling Sites We investigated 18 reach-scale river restoration projects in third- to seventh- order rivers in the low-elevation mountain ranges of the German federal states of Hesse and North Rhine-Westphalia. The goal of these projects was to restore the animal and plant communities to a more natural state by improving and diversifying the hydromorphological structure of the rivers. To achieve this overarching restoration goal, a variety of measures were applied, including the removal of bank fixations, the creation of a more braided and meandering planform and the placement of large wood. We only considered projects for which the length of the restored reach was a minimum of 200 m; this criterion ensures that the extent of the projects was sufficient to potentially enhance fish assemblages. The mean length of the 18 selected restored reaches was 1.2±1.1 km (± SD). The time between the implementation of the restoration measures and the monitoring in 2007 and 2008 was in the range of 1 to 19 years. An almost identical set of river restoration projects was analysed in Stoll et al.. Only one restoration project was excluded due to lack of quantitative fish assemblage data in the surroundings and replaced by the newly obtained restoration project No. 18. Comparing the fish assemblages at restored and unrestored control reaches, this latter study found a small, however significant positive effect of the restorations on the naturalness of the local fish assemblages. Assemblages at restored sites comprised on average 2.8±1.8 additional species that were part of the stream-type specific natural reference lists compared to the unrestored control reaches, whereas at the same time only 1.3±1.5 of such species were lost. ## Monitoring of Hydromorphological Conditions The local hydromorphological conditions in restored and unrestored control reaches were assessed according to the German hydromorphological survey method ; this method is also described by Kamp et al. and Kail and Hering. Twenty-five parameters of the six main groups (planform, longitudinal profile, bed structure, cross-section, bank structure and floodplain corridor) were assessed at each site with a scoring system ranging from 1 (natural) to 7 (completely altered). For further analyses, averaged values for each of the six main groups were used. Analyses of the hydromorphological data from both restored and unrestored control reaches showed that the restorations significantly improved the hydromorphological conditions. In all six main parameter groups, the restoration projects achieved hydromorphological conditions that can be expected from successful restorations. The hydromorphological conditions in all six main parameter groups were rated similarly, with an average rating around quality level 3 ‘moderate alteration’. Bed structure of restored reaches was rated best; however, on average bed structure also showed the least deficits in unrestored conditions. Greatest improvements as a result of the restorations were achieved in river planform, where on average restored reaches rated 2.6 quality classes better than unrestored reaches. ## Monitoring of Fish The restored reaches were electrofished in August and September of 2007 and 2008 following the EU Water Framework Directive compliant protocol for the assessment of river fish assemblages in Germany. According to this protocol, wadeable streams were sampled by electrofishing on foot over section lengths of approximately 40× stream width. In non-wadeable streams, electrofishing was executed from a boat. When fishing by boat, sections of a length of approximately 100× stream width were sampled to compensate for the decreased sampling efficiency. The sections were never shorter than 100 m and contained representative proportions of all habitat types present in a reach. Electrofishing was conducted against the current as single passes with generator-powered DC electric fishing gear. Fishing was conducted at stable, low-flow conditions, and extreme discharge events or other adverse conditions were avoided. All stunned fish were placed in trays until the end of fishing, counted and released. Electrofishing permits for this project were obtained from the Regierungspräsidien Darmstadt, Gießen and Kassel in Hesse and the Untere Landschaftsbehörden in North Rhine-Westphalia. Private land owners were kind enough to provide access to the sampled river reaches. Also protected species were sampled, however not harmed, as all specimens were released at the end of the sampling procedure. ## Data from the Regional Species Pool The spatial extent of the relevant species pool for the colonization of restored reaches in Central Europe was analysed by Stoll *et al.*. They demonstrated that virtually all fish species that colonized the restored reaches were present in source populations within a distance of 5 km up- or downstream, whereas species for which the closest nearby population was more distant than 5 km rarely colonized restored reaches. Based on these findings, the fish populations in the river network 5 km up- and downstream (including tributaries) were considered to be potential source populations constituting the relevant regional species pool for colonization of the restored reaches. The species compositions of these regional species pools were analysed based on electrofishing data gathered by governmental environmental agencies of the German federal states of Hesse and North Rhine-Westphalia from 1998 to 2008. Electrofishing data from a total of 320 reaches were available and represented 35 different species. On average, 7.2 species occurred per sampled reach. The mean length of the sampled reaches was 257 m (±290 SD). Electrofishing in these surrounding reaches was performed in the same way as in the restored reaches. ## Fish Species Traits The ecological species traits of habitat preference, feeding type, flow preference and migratory ability were assigned to all present species according to the trait database ‘[www.freshwaterecology.info](http://www.freshwaterecology.info)’. For statistical reasons, individual traits that occurred in fewer than three fish species were pooled for the analyses. Specifically, pelagic habitat preference was pooled with bentho-pelagic habitat preference, in contrast to demersal habitat preference. Furthermore, the feeding types herbivorous, piscivorous and filter-feeding were pooled as ‘specialist’ feeding types because these types all rely on only one food source, in contrast to inverti-piscivorous and omnivorous fish. Mobility of the species was estimated from the trait swimming factor (SF), which is defined as the aspect ratio of the minimum depth of the caudal peduncle and the maximum caudal fin depth. Fish with a small ratio are capable of strong, sustained swimming. In the [www.freshwaterecology.info](http://www.freshwaterecology.info) database, fish species are assigned to three swimming factor categories, small (SF1), medium (SF2) and large (SF3). Additionally, the species were taxonomically classified to the order level. The orders of perciform, salmoniform and cypriniform fish were differentiated; species from orders with fewer than three present species were clustered as ‘other’. All species trait information is presented in the Supporting Information. ## Statistical Analysis The composition of the regional species pools was characterized using two indices, species occurrence rate and species density. The occurrence rate of a species within a species pool was calculated as the fraction of reaches in which this species was present out of the total number of sampled reaches within that species pool, corresponding to ‘per cent occupancy’ in the study by Albanese et al.. Species density was calculated as the average density of all known occurrences of this species within the species pool. All statistical analyses were performed in *R* 2.9.1. In all analyses, each species at each restored reach was regarded as an independent replicate. To analyse the species presence-absence data from the restored reaches, generalized linear models (GLM) for binomial data were used. At each restored reach, present species were assigned the value ‘1’, species that were not present at restored reaches but in the respective regional species pool were assigned ‘0’. Species without proven occurrence in a regional species pool were excluded from the analysis. Two models were used to analyse fish presence-absence data. The first model used ‘biotic’ variables as independent predictors, namely occurrence rate, fish density, taxonomic affiliation, feeding type, habitat preference, flow preference and migratory ability. In the second model, the effects of the ‘abiotic’ variables on species presence at restored reaches were tested. These ‘abiotic’ variables included catchment size, stream order, length of restored reach, time since restoration, planform, longitudinal profile, bed structure, cross-section, bank structure and floodplain corridor. All significant variables from the ‘biotic’ and ‘abiotic’ model were thereafter combined into one model. Linear models (LM) were used to analyse species densities in the restored reaches. Again two models were fitted using the same independent variables as the GLM models on species presence-absence data. To obtain normal distributions, the species densities in the restored reaches and in the species pools were ln(x+1)-transformed. Again, all significant variables from the ‘biotic’ and ‘abiotic’ model were subsequently combined into one model. All analyses were initiated with a model containing all variables and second- degree interactions. The models were backward-selected until the minimal Akaike Information Criterion (AIC) was reached. Our approach to consider each species at each restoration project as an independent replicate may be considered as a potentially pseudo-replicative structure of the dataset. Nevertheless, we chose this approach as the individual colonization events are the basic entities that we were interested in. To consider potential effects involved with pseudo-replication of this approach, a restoration project identifier was added to each model as an additional independent variable; however, this identifier was excluded from the models during the backwards selection procedure. This also points to the broad applicability of the results. ## Data Deposition Data on regional species pools belonged to the environmental agencies Hessisches Landesamt für Umwelt und Geologie, Hessen-Forst, and Landesamt für Natur, Umwelt und Verbraucherschutz NRW. These data may be requested from the above-mentioned agencies directly. A summary of our own sampling data is published in an electronic appendix to Stoll et al.. # Results The ‘biotic’ variables, relating to the species composition of the regional species pools and ecological species traits, were much better suited to explain the species assemblages in restored reaches than were the ‘abiotic’ variables, characterizing the restoration projects and local hydromorphological conditions of the restored reaches. The best model using ‘biotic’ descriptors explained 34% of the variability in the species presence, while the best model using ‘abiotic’ descriptors explained only a marginal share of 2%. In terms of variability in fish densities within restored reaches, the best model using ‘biotic’ descriptors explained 38%, while the best model using ‘abiotic’ descriptors only explained about half of that share, i.e. 21%. Combining the significant ‘biotic’ and ‘abiotic’ variables into one model did not increase the share of explained variability in species presence data beyond 34%, but for species density data, the share of explained variability increased to 57%. In these combined models, all significant variables from the individual models were retained. The presence of a species within the restored reaches was particularly dependent on the occurrence rate of the species in the species pool. For example, species that occurred at 10% of the sampling sites in a regional species pool showed an average probability of 45% to colonize the restored reach of that river. Species with occurrence rates in the regional species pool greater than 75% colonized every restored reach. The density of a species within the regional species pool had a marginally significant effect on the presence of that species in restored reaches. Species with a density of about 7 ind. ha⁻¹ showed a probability of about 50% to colonize a restored reach. All species with average densities greater than 470 ind. ha⁻¹ in a regional species pool colonized the respective restored reach. Salmonid species, including brown trout (*Salmo trutta*), rainbow trout (*Oncorhynchus mykiss*), Atlantic salmon (*Salmo salar*) and grayling (*Thymallus thymallus*) exhibited the lowest probabilities of colonizing restored reaches, while fish species that belonged to the mixed group labelled ‘other’, including eel (*Anguilla anguilla*), brook lamprey (*Lampetra planeri*), pike (*Esox lucius*), three-spined stickleback (*Gasterosteus aculeatus*), ten-spined stickleback (*Pungitius pungitius*) and bullhead (*Cottus gobio*), showed the highest probabilities of emerging in restored reaches. Rheophilic species exhibited a higher probability of colonizing restored reaches than did limnophilic species; eurytopic species that are indifferent to flow conditions showed an intermediate response. Higher density of a species within a restored reach was related to a combination of high occurrence rate and the density of that species in the regional species pool. Cyprinids exhibited the highest densities in restored reaches, e.g. minnow (*Phoxinus phoxinus*) 1022±547 ind. ha⁻¹, stone loach (*Barbatula barbatula*) 676±147 ind. ha⁻¹ and gudgeon (*Gobio gobio*) 463±141 ind. ha⁻¹(all mean ± SD). Percids and ‘other’ species occurred at the lowest densities, e.g. pike 10±4 ind. ha⁻¹, eel 6±2 ind. ha⁻¹ and pikeperch (*Sander lucioperca*) 12±11 ind. ha⁻¹. Densities in restored reaches also differed between feeding types. Species that consume fish as some part of their diet occurred at low densities in restored reaches, while invertivorous species exhibited the highest densities in restored reaches. Among the set of ‘abiotic’ variables, the presence of a species in a restored reach depended only on the time between restoration and sampling. Within the first year after the restoration work was completed, the probability of a species being present in a restored reach was about 65%; this percentage further increased with the time elapsed since the restoration work was performed, to about 80% after 19 years. A positive influence of two hydromorphological quality metrics, planform and floodplain quality, on the species density in restored reaches was detected, and higher fish densities were found in reaches of lower stream order. # Discussion Our study showed that, in total, 57% of the variability in the fish density data and 34% of the variability in the fish presence-absence data at restored reaches is explained by a set of simple ‘biotic’ and ‘abiotic’ variables. Strikingly, ‘biotic’ variables characterizing the regional species pools were much better suited to explain fish presences and abundances in the restored reaches than ‘abiotic’ variables characterizing the restoration projects and hydromorphological structures in restored reaches. A particular importance of the regional species pools for river restoration outcomes has already been assumed, but not quantitatively demonstrated. The river restoration projects in this study, in line with other studies on reach-scale river restoration projects, were successful in removing hydromorphological limitations and providing natural (or at least near-natural) habitat conditions. So if species do not colonize such restored reaches, it is more likely that this is because of absence or rarity in the regional species pools and limitations in the dispersal process than because of lack of habitat suitability at a restored reach. The most important individual variable for the probability of species to colonize a restored river reach was the occurrence rate of this species within the regional species pool, while high population densities in surrounding areas were more important for the size of a population in a restored reach. Large populations provide not only a higher number of emigrants when a fixed rate of individuals is expected to emigrate but may also have an overall higher rate of emigrants to avoid intraspecific competition. Using a modelling approach, already Huxel and Hastings postulated that species establishment at restored reaches depends on habitat occupancy of these species in neighbouring reaches. In a fish removal experiment, Albanese et al. confirmed the role of species abundance in the regional species pool for the colonization of emptied reaches, and further highlighted the role of fish mobility in colonizing such reaches. Also Radinger and Wolter found a relation between fish mobility and dispersal distance of fish species, whereat they also estimated a species mobility from a species swimming factor. In the present study, fish mobility was not retained in the best models explaining the fish assemblages in the restored reaches. Nonetheless, there is a considerable overlap between the ecological traits, mobility and rheophily, as many rheophilic species are strong swimmers. However, some rheophilic species, despite living in riffle or run environments avoid fast flow by choosing appropriate micro-habitats in the interstices of the bottom substratum, e.g. bullhead. Inversely, our results suggest that stretches of riffles in river networks may pose considerable dispersal impediments, particularly for limnophilic species. Furthermore, rheophilic species may be generally more avid dispersers as they have to be able to compensate for drift and it has been shown for a number of species that peak flow triggers upstream movement. An alternative interpretation of the higher probability of rheophilic species to be present at restored sites is that the newly created habitats are more suitable for such species. However, this appears relatively unlikely because the restorations did not increase flow velocities, but as a result of river widening, elongation of river courses and reconnection of backwaters, the variability of flow velocities is significantly increased. Thus, additional habitats for both rheophilic and limnophilic species were created. This evidences that the mechanism behind the different colonization success between rheophilic and limnophilic species is rather connected with dispersal than with habitat suitability of restored reaches. In addition to the quantitative metrics of the regional species pools, taxonomic affiliation, i.e. species identities, had an effect both on the probability to colonize a restored reach and on the resulting species densities. Salmonids showed a sub-average probability to colonize restored reaches, particularly compared to species of ‘other’ orders. Among these latter species, brook lamprey, three-spined sticklebacks and bullhead, in particular, are known from other studies to be strong colonizers. The result that the heterogeneous ‘other’ group showed the highest colonization probabilities makes it difficult to reveal the relevant underlying ecological traits. Nevertheless, even though being successful colonizers, these ‘other’ species did not build up high densities at restored reaches. Instead, cyprinids showed almost as high colonization probabilities and additionally were most likely to reach high densities in restored reaches. Most cyprinid species are gregarious, and fish density in the regional species pool has been revealed as a major factor for the density of a species at a restored reach. Furthermore, as cyprinids typically feed on lower trophic levels, habitats support higher fish densities than in taxonomic groups that predominantly forage at the upper end of the food chain, such as for example percids that occur only at lower densities. Among the ‘abiotic’ predictors of species presence at restored reaches, time since the restoration was carried out was most important. It is often assumed that, on small spatial scales, the colonization process will proceed rapidly. However, after this initial colonization by nearby species, colonization by species through long-distance dispersal has been shown to be a slow and highly stochastic process. Because datasets are often small and comprise only data of one or few restoration projects, it is difficult to demonstrate rare colonization events following long-distance dispersal; as such incidents may be masked by the natural species turnover within local fish assemblages or by the limitations in the detectability of small populations by electrofishing. None of the six metrics of river hydromorphological quality affected the probability of a species to colonize a restored reach; however, fish densities at restored reaches were positively influenced by river planform and floodplain quality. Planform is a good proxy for the overall hydromorphological state because high sinuosity and braidedness is usually associated with high habitat diversity (e.g., bars, pools and undercut banks) and lateral channel dynamics. Dynamic changes in a river course due to relocating gravel bars, erosion and deposition provide essential habitat for many riverine species, especially as nursery areas , and emerging shallow water areas reduce current velocity and provide shelter from predatory fish. Also good rating results of the hydromorphological parameters related to floodplain quality, comprising information on riparian features including provision of shading and land use across the entire floodplain, is often associated with high fish densities. For instance, fish often aggregate under riparian structures providing protection from aerial predators. Also differences in land use affect fish, albeit on larger spatial scales than the effects of shading, leading to impoverished fish fauna and lower fish densities in intensely used systems. In small, low-order rivers, the fish densities in restored reaches were higher than in large high-order rivers. On a per-area basis, fish abundance is often higher in small and medium-sized rivers compared to large rivers, as the former typically provide more diverse and more complex habitat structures, which typically aggregate fish. Part of the difference in fish densities between restored reaches in low- and high-order rivers may also be explained by the decreasing effectiveness of electrofishing with increasing river size. ## Implications for Restoration Planning This study demonstrates that ‘biotic’ data on the regional species pools may be used to estimate the probability of fish species to colonize a restored reach. These ‘biotic’ data are much better suited to explain fish presence and densities at restored reaches than ‘abiotic’ data, which are often used as the base for such attempts in other studies. The results of this study highlight the paramount importance of appropriate spatial prioritization in river restoration planning. Only if the regional species pools are intact and diverse, will the removal of hydromorphological deficits succeed in enhancing the naturalness of local fish assemblages. Therefore, in addition to focusing on local hydromorphological conditions and engineering aspects, the regional species pools should receive more attention when planning river restoration projects. In the prioritization of alternative restoration sites, knowledge on the status of the regional species pool permits estimation of the likelihood of restoration projects to reach specific targets. Only species with a sufficient abundance in a regional species pool can be expected to colonize a restored reach, whereat the critical values of individual species may vary, depending on their ecological traits. Species with low abundance in the regional species pool are unlikely to colonize restored reaches. Thus, restoration projects aiming to improve general habitat structure have little chance of success in supporting endangered species, which typically have small and scattered populations. In cases where restorations are designed to support individual species (e.g. endangered species with particular habitat needs) a distinct focus on the limiting habitat features may be more promising. To support such species with scarce or only fragmented source populations, stocking, sometimes also referred to as assisted migration, may be an option; however, this practice has been heavily debated in recent years. # Supporting Information Data from the surroundings of the restoration projects were kindly provided by Hessisches Landesamt für Umwelt und Geologie (HLUG), Hessen-Forst (FENA), and Landesamt für Natur, Umwelt und Verbraucherschutz NRW (LANUV). Jonathan Tonkin provided linguistic advice. Constructive comments of Brendan Ebner and one anonymous reviewer helped to improve this manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SS JK AWL AS PH. Performed the experiments: AWL AS. Analyzed the data: SS. Contributed reagents/materials/analysis tools: SS JK PH. Wrote the paper: SS JK AWL AS PH.

# Introduction In humans, mycobacteria often colonize without causing overt disease. In India 5% (equivalent to ∼65 million) of the population shed *M. leprae* DNA in their nasal secretions, although \<400,000 have clinical leprosy. Similarly, the number of individuals who have “latent” tuberculosis (30% of humankind) far exceeds the number with clinically evident tuberculosis., *M. avium* subspecies *paratuberculosis* (MAP) causes a chronic wasting enteritis in ruminants called Johne's disease that is highly evocative of Crohn's disease (CD). Prevailing medical dogma considers that MAP is not zoonotic. Yet it is of great concern that humans worldwide are continually exposed to viable MAP. MAP has been cultured from USA chlorinated potable municipal water, pasteurized milk in the USA, and Europe, breast milk of mothers with CD and from the blood of patients with inflammatory bowel disease (IBD). Intriguingly, although Koch's postulates may already have been met for MAP and CD they have still not been met for *M. leprae* and leprosy. We posit that the pivotal reason that MAP has not been acknowledged as a human pathogen is that, unknowingly, the medical profession has been treating MAP since 1942, when sulfasalazine was introduced. Until recently, it was unrecognized that the “anti-inflammatory” 5 amino salicylic acid (5-ASA) and the “immune modulators” methotrexate, azathioprine and its metabolite 6 mercapto- purine (6-MP), are antiMAP antibiotics. We concluded that all antecedent studies evaluating the potential zoonotic character of MAP need to be reevaluated, as their control groups were not placebo., We hypothesized that MAP may asymptomatically colonize apparently healthy individuals. We further hypothesized that the unwitting use of antiMAP antibiotics may be associated with a decrease in the incidence of MAP in patients with IBD. Accordingly, we evaluated the blood of healthy individuals and IBD patients treated with antiMAP agents for the presence of MAP DNA. # Methods The Ethics Committee from each institution approved this study. Every participant signed an Informed Consent form that was in compliance with all relevant national and European Union regulations. There were no therapeutic interventions or alterations in the concurrent therapy, or initiation of therapy, as a consequence of participation. The Control group comprised 100 subjects. The majority were healthy blood donors (HBD) recruited from a regional Basque Country Blood Bank. They had met all European Union requirements for allogenic blood donation, and each denied major infections or concomitant active disease. None gave any history of gastrointestinal disease. The remainder of the Controls were healthy laboratory workers from the same region. The IBD group consisted of 246 subjects recruited from three hospitals in the Basque Country in Northern Spain; The Quirón Donostia Clinic in Gipuzkoa (65), the Hospital de Txagorritxu in Araba (81), and the Hospital de Galdakao in Bizkaia (100) The clinical diagnosis of IBD had been predetermined by each treating physician. Patients were stratified into Crohn's disease, (CD), ulcerative colitis (UC) or Indeterminate Colitis (IC) by clinical histories and routine endoscopic, histological, and radiographic criteria. Current disease status and concurrent treatment were documented on a standardized eight item, 54 choice questionnaire that was completed by each patient with the physician's help. ## Blood sampling Three 4 mL whole blood tubes were obtained from each subject (two sterile EDTA and one heparin-lithium Vacutainer® tubes (BD)). All blood samples were coded to conceal the patient's identity and diagnosis to laboratory workers. All samples were processed within 4 hours after extraction in a class II bio-safety cabinet. ## Nested PCR Genomic DNA was extracted from buffy coat cells. Briefly, one volume blood was incubated with one volume 155 mM ammonium chloride for 20 minutes to lyse the red blood cells. The tube was centrifuged (10 mins. 200×g) the cell pellet washed twice with PBS, recentrifuged (10 mins. 200×g). DNA was extracted and purified (QIAamp DNA Blood Mini Kit (QIAGEN GmbH, Hilden, Germany) and stored at −20°C until amplified. Nested PCR was used to amplify IS900 as described. In brief, the first round primers (P90 and P91) amplify a 398 bp fragment and the second set (AV1 and AV2) identify a 298 bp fragment. For the first round, 10 µl of genomic DNA were added to 40 µl of PCR buffer mixture. The PCR buffer mixture consisted of 5 mM MgCl₂, 0.2 mM dNTP, 6% DMSO, 2 µM primers and 2.5 U of Taq Polymerase (Invitrogen Ltd., Paisley, UK). In the second round, all conditions were the same except that 5 µl of the PCR product from the first round were used as DNA template. PCR cycling conditions were: 95°C for 5 min, 34 cycles of: 95°C for 1 min, 58°C for 1.5 min, 72°C for 1.5 min, with a final extension phase of 10 min at 72°C. Amplification products were separated using 2% agarose gel electrophoresis (150 volts; 50 mins). The positive MAP control was ATCC 19698 DNA and the negative controls were distilled water, identically processed with the clinical samples. A band co-migrating with the ATCC 19698 DNA at the predicted amplicon size of 298-bp was considered positive. The identity of the amplified 298 bp amplicon was confirmed from two positive healthy controls and 2 IBD patients. Bands were excised, extracted, purified (GFX PCR DNA and Gel Band purification kit. Amersham Biosciences, Buckinghamshire, UK) and commercially sequenced (Centro Superior de Investigaciones Científicas, Madrid, Spain). The sequence identity of the final 298 bp. amplicon was compared with Genebank accession X16293 sequence for MAP IS900 using BLAST (NLM) and sequence alignment analyses. ## Contamination Avoidance Procedures Stringent controls were adopted to minimize the possibility of contamination. These include using a Level II Bio-safety hood to process buffy coat for DNA extraction and using separate uniforms, rooms, pipettes and thermocyclers for the primary and the secondary rounds of PCR amplification. ## Statistical Analyses Comparisons of MAP DNA in blood frequencies for overall IBD, IBD type, lesion location and type of treatment were made using the Fisher's exact test (SAS Institute Inc., Cary, NC 27513, USA). Treatments were grouped by chemical structure or presumed mechanism of action in three categories: anti- inflammatories or salycilic acid derivatives (SAD) (mesalamine and sulfasalazine), immuno-modulators (azathioprine, 6-mercaptopurine, methotrexate and Tacrolimus®) and conventional antibiotics (metronidazol and ciprofloxacin). Statistical significance was accepted at p\<0.05. # Results Of the 100 Controls in this study, 90 were healthy human blood donors and 10 healthy laboratory workers. The majority of IBD subjects (54%; 132/246) had CD, 42% (103/246) had UC and 3% (8/246) had IC. As a group the CD were the youngest, and IC the oldest. When anatomical location of maximal pathology was reported, a minority (5%; 6/116) of CD subjects had disease confined to the colon (Data not presented). Among subjects with UC, a minority (42%; 38/92) had ulcerative proctitis (Data not presented). The PCR data shows bands that co-migrate with the positive control at the predicted amplicon size of 298 bp. The sequenced DNA of the representative sample bands from each clinical subset (Controls and IBD) showed \>99% identity in all cases with the Genebank accession X16293 sequence for MAP IS900 (Data not presented). Forty seven % (47/100) of the controls \[47% (42/90) blood donors and 50% (5/10) healthy laboratory workers\] had a band with the predicted 298 bp amplicon ( & ). In contrast, 16% (40/246) IBD patients had the 298 bp amplicon (p\<0.0001 compared to Controls) ( &). There were no significant differences when stratified by IBD diagnosis or location of maximal pathology (Data not presented). When comparing the prevalence of MAP DNA in each group of drugs to the Controls, conventional anti-inflammatories (p\<0.0001), immuno-modulators (p\<0.0001) and antibiotics (p = 0.0293) were significantly less likely to have MAP DNA in their blood. For IBD patients receiving “No medications,” MAP DNA prevalence was slightly higher (22% compared to 15% for all IBD subjects combined). However, all IBD subsets were significantly lower than the non-IBD control group. ( Right hand columns). Finally, within each chemical group or class of agents, we evaluated individual medications. As a *caveat*, although these data were collected prospectively, the analysis is *post hoc* and the numbers are small. The SAD “Anti-inflammatories” were sulfasalazine and mesalamine. 17% receiving mesalamine (103 on mesalamine/143 not on mesalamine) were MAP DNA positive. Among those receiving sulfasalazine, 6% (16 on sulfasalazine/230 not on sulfasalazine) were MAP DNA positive. The “Immuno-modulators” prescribed were azathioprine, its metabolite 6-MP, methotrexate and Tacrolimus®. With azathioprine 18% were MAP DNA positive, similar to controls (50 on azathioprine/196 not on azathioprine). In contrast, no MAP DNA was detected when either 6-MP (3 on 6-MP/ 243 not on 6-MP), methotrexate (9 on methotrexate; 237 not on methotrexate) or Tacrolimus (3 on Tacrolimus; 243 not on Tacrolimus) were used. The conventional antibiotics prescribed were metronidazole and ciprofloxacin. Twenty five per cent of patients receiving metronidazole were MAP DNA positive (12 on metronidazole; 234 not on metronidazole). In contrast, none of the patients on ciprofloxacin (5 on ciprofloxacin; 241 not on ciprofloxacin) were MAP DNA positive. Steroid therapy had no effect on the presence of MAP DNA (16%) when combined (any steroid n = 44/no steroids n = 202) or individual steroids were studied. “Disease Activity” data were available for 98% (240/246) of IBD patients. There was no difference in the presence or absence of MAP DNA when analysed according to disease activity and/or concomitant medication usage at the time of phlebotomy. # Discussion The presence of MAP DNA does not address potential MAP viability, anymore than the presence of *M. leprae* DNA in nasal secretions determines possible *M. leprae* infectivity. Nevertheless, our data shows a disquietingly high 47% of healthy individuals have MAP DNA in their blood. These data thus corroborate and extend a prior study showing 20% MAP DNA positivity in non-IBD subjects. We conclude that the possible viability of MAP in blood, that may have allogenic use, should be expeditiously clarified. “Anti-inflammatory” (5-ASA), “immuno-modulatory” agents (azathioprine its metabolite 6-MP and methotrexate) and the “immuno-suppressive” agent Tacrolimus® are used to treat IBD and multiple “inflammatory” and “autoimmune” diseases. This is despite the fact that the “mechanism of action (of 5-ASA) in the therapy of IBD is unclear” and all are used simply because of empirical efficacy. In this study, we show that 5-ASA decreases the incidence of MAP DNA from the blood of patients with IBD. We additionally show that the “immuno-modulators” 6-MP and methotrexate and the “immuno-suppressive” Tacrolimus actually clear MAP DNA. We found that all of our IBD subset groups, including those on “no active medications”, had a lower incidence of MAP DNA than the non-IBD control group. The most plausible explanation for the fact that IBD patients on no medication have a lower incidence of MAP DNA than the non-IBD controls is that our questionnaire did not request a history of medications that had been used by the IBD patients prior to the day of phlebotomy. Thus, those patients who were reported as being “On no medications” at the time of phlebotomy may well have been exposed to antiMAP agents in the past. ( &). Because of small numbers, our *post hoc* analyses should be considered as tentative. Nevertheless, our observations are compatible with the thesis that 5-ASA, azathioprine, 6-MP and methotrexate are acting as antiMAP antibiotics. They also corroborate our *in vitro* data that 5-ASA is a far less potent antiMAP antibiotic than are 6-MP and methotrexate. Tacrolimus is an “immuno-suppressive” medication most used to prevent organ transplant rejection and more recently in the therapy of IBD. It is of considerable interest that Tacrolimus is from the macrolide antibiotic family of medications, amongst the most potent anti *M. avium* antibiotic families. Ciprofloxacin clears MAP DNA. This is an antibiotic, which has shown an activity against different strains of MAP “*in vitro.*” In contrast, metronidazole an alternative antibiotic used in this study does not clear MAP DNA. Our data suggest that ciprofloxacin is a more effective antiMAP antibiotic than is metronidazole. These initial observations provide insights that should be of use when determining which agents should be evaluated when designing future, pivotal, clinical trials. The use of the TNF alpha antagonist Infliximab® is associated with reactivation of latent tuberculosis and requires concomitant use of anti-tuberculosis prophylaxis. We find only 9% (2/13) of our patients treated with infliximab are MAP DNA positive. Our observations are therefore at variance with the 80% (4/5) culture of MAP from the blood of patients who were receiving infliximab in a prior study. The most plausible explanation for these discrepant observations are the potent antiMAP agents that were concomitantly used in our patients.. Prudence suggests that, until the potential MAP zoonosis conundrum is finally resolved, any individual being treated with TNF alpha antagonists, for any disease, should continue to receive antiMAP agents such as 5-ASA, sulfasalazine, methotrexate, 6-MP, ciprofloxacin, Tacrolimus or another macrolide antibiotics. When steroids are used to treat the inflammatory reaction in tuberculosis, anti- tuberculosis medications are always co-administered. , MAP has been cultured from 60% (3/5) IBD patients receiving steroids. However, our data do not show an increase in the incidence of MAP DNA when steroids are used. Again, the most reasonable explanation is the potent antiMAP agents that were co-administered in our patients being given steroids. We conclude that caution suggests that whenever steroids are used in the therapy of IBD, concomitant antiMAP agents should always be used. If MAP zoonosis is accepted, MAP antibiotic susceptibility studies will need to be performed. However, few laboratories can successfully culture the cell wall deficient form of MAP that exists in humans, a process that may take up to 18 months. Nucleic acid based methods, to more rapidly determine the presence of MAP RNA (indicating viability), potential infectivity and antibiotic susceptibility may need to be developed. Our data suggest that until such nucleic acid based methodologies are available, identifying MAP DNA prior to, and its clearance following initiation of antiMAP therapy, may provide the most readily available practicable surrogate to *in vitro* MAP antibiotic susceptibility information. [^1]: Conceived and designed the experiments: RG RJ JG IO. Performed the experiments: NE JG AP MG IS RC AI. Analyzed the data: RJ. Contributed reagents/materials/analysis tools: NE AP MG JC AT FG RC AI. Wrote the paper: RG. [^2]: RAJ. is President of the International Paratuberculosis Association. He has had expenses paid by a company for giving a conference, attending a meeting and making an expert report for registration of a vaccine against paratuberculosis. NEIKER has received funding for a study on the efficacy of paratuberculosis vaccination in cattle. RAJ, NEL, JGA, MVG and ISE are involved with a small company that is an spin-off of NEIKER in a project for the development of an improved vaccine for paratuberculosis. RJG has submitted patents on some of the hypotheses addressed in this and prior manuscripts. There are no potential conflicts of interest for the other authors.

# Introduction Everyone experiences a variety of emotions throughout their lives. Thus, the nature of emotion has always been of interest to scholars, as understanding the nature of emotion could lead to a clearer understanding of the human experience and yield useful results for improving people’s emotional health. However, the necessary standardization of stimulus materials presents a challenge in emotion research. To address this issue, the National Institute of Mental Health (NIMH) research center for emotions and attention established a series of stimulating material systems, including pictures, sounds, and words, that have been quantified and evaluated. As the study of emotion continues to grow, there are more complete picture-, sound-, image-, and word-emotion systems used in different countries. The “word-emotion system” is mainly the affective-words system, which is used primarily in the study of unconscious emotions. At present, many researchers are exploring the influence of positive and negative emotions on individuals’ various cognitive activities by initiating explicit emotions. Language is a potent tool for communicating emotions, but previous studies have found that words are less effective than pictures for arousing emotions. The key to effectively activating emotions is ensuring that participants become psychologically engaged in the corresponding situation because the situation constitutes a segment of life and has a good activation effect for exposing emotions. Consideration of the need to psychologically engage emotions leads to the question, “What if we replace words with emotional statements?” Situational sentences may have a better priming effect on emotions than words, as browsing through situational sentences arouses more emotional responses. Most depictions of the human emotion system in prior studies contain basic emotions, such as fear, disgust, anger, sadness, and happiness; further, prior studies confirm that different emotions correspond to different qualities. Anxiety is common among college students as they experience pressure from interpersonal relationships, life changes, environment changes, and academic and employment-related demands. Moreover, college students are also easily affected by emergencies. This was confirmed by Hoyt et al. in a survey of 707 American college students. The study results found that during the COVID-19 pandemic, most students experienced anxiety and stress, while their happiness continued to decline. Additionally, anxiety among college students is likely to lead to depression. Therefore, it is crucial to understand anxiety through research. However, some studies have reported that pictures, sounds, and words cannot effectively activate an individual’s anxiety, on their own. This is possibly because anxiety is a compound emotion. For instance, in a picture, the context of the picture and the faces of individuals may not be easily identifiable in a snapshot alone. Additionally, while a scene in a movie may trigger anxiety, a single frame does not provide the full experience of the film. However, some studies indicate that words may be effective in triggering anxiety. As situational statements can present varied and complex emotions with brief descriptions, in addition to the six basic emotions, we added “anxiety” to the present study to explore the activation of different emotion types more comprehensively. The first aim of the study was to examine the effectiveness of a new ESSS in eliciting emotional responses in college students. To verify the validity of ESSS, it is crucial to compare it with other emotional materials. However, few studies have achieved direct comparisons of emotional effects between stimulus domains. Therefore, the second aim of this study was to compare the utility of the ESSS to that of emotion-inducing pictures and explore the advantages of the different materials. # Design and methods We designed two studies to explore the activation of emotions. The purpose of Study 1 was to explore a new mechanism of stimulating explicit emotions using emotional statements. To establish an ESSS relevant to college students, Study 1 drew upon standard procedures used by previous scholars to examine emotional systems. At present, the emotion picture is one of the most commonly used emotion priming materials in emotion research, and contains a wide range of emotion types, with good reliability and validity. Study 2 aimed to examine the calibration validity of the ESSS as well as ecological validity. As emotional pictures and ESSS contain similar types of emotions, they have the possibility of comparison, and the comparison with authoritative materials can better illustrate their respective advantages. The study was conducted by comparing the items with the highest arousal in the five basic emotions of the ESSS and the emotional pictures through behavioral experiments. The respective advantages and disadvantages of the ESSS and emotional pictures were analyzed in detail to provide directions for future research. # Study 1 ## Participants The participants of Study 1 were divided into two groups. The first group, 607 undergraduates and postgraduates randomly selected from Jiangsu province (mainly Suzhou and Nanjing) completed the emotion-inducing situations questionnaire. These participants included 234 men and 373 women aged 21.57 ± 4.19 (*M* ± *SD*) years. The second group graded the sorted materials. This group comprised 91 undergraduates randomly selected from a university in China, including 23 men and 68 women aged 18.19 ± 0.61 (*M* ± *SD*) years. 11 participants with incomplete answers were excluded, and the number of final participants was 80, including 17 men and 63 women, aged 18.19 ± 0.62 (*M* ± *SD*) years. According to the suggestion of the reviewer, 30 participants (10 men and 20 women) aged 20.07 ± 2.16 were randomly selected to reevaluate all the items to test the retest reliability. All participated in this study voluntarily and signed an informed consent form before we began the experiment. They were each paid 10 yuan after the experiment. All participants were healthy, without mental illness, and had normal or corrected-to-normal vision. This study was approved by the Ethics Committee of our school and conformed to ethical standards. ## The compilation process of the ESSS ### Collection of situation statements A collection of situation statements was compiled from the first group’s responses to the emotion-inducing situations questionnaire. This questionnaire consists of seven questions on seven dimensions of the situational emotion system (see for details.). An example of a questionnaire item is, “Please describe, in one sentence, a situation that makes you feel sad, such as ‘My beloved grandmother died suddenly’ (please write down at least three such sentences).” In the first round, we filtered and modified the collected situational statements, deleting sentences that had already been documented. To make the meaning of the sentence clear to the participants quickly, we ensured that each situational statement comprised approximately 10–15 words. ### Filtering situation statements The collected situational statements were screened by 15 experts, including one professor, two associate professors, six doctoral students, and six master’s students. Determining whether the items in each dimension evoked the appropriate emotions was the most important part of the selection process. If any one expert found an item to be inappropriate, all experts discussed and made a decision regarding deletion of the item. All the filtered statements involved common situations, had clear semantics, were not repetitious, and could stimulate the corresponding emotions in individuals. Finally, a total of 778 situational statements were selected, among which 121 reflected situations corresponding to fear, such as “waking up late at night looking at a dark room”; 119 corresponded to disgust, such as “walking into a room full of smelly garbage”; 111corresponded to anger, such as “someone read my diary without permission”; 94 corresponded to sadness, such as “my beloved grandparents passed away suddenly”; 116 corresponded to happiness, such as “I’m going to marry the one I love”; and 118 corresponded to anxiety, such as “I broke a valuable vase at my teacher’s house by accident”; the remaining 99 statements were neutral, such as “I saw a few people studying in the classroom.” The content mainly included scenes of campus life, family life, family relationships, classmate relationships, loving relationships, nature scenes, and other scenes closely related to contemporary college students’ lives. ### Rating situation statement According to Osgood’s theory, we used a self-report method to evaluate the material from the three components of valence, arousal, and dominance. All participants were assessed for each emotional dimension. To reduce the fatigue effect of participants, the same participant completed the assessment of seven emotional dimensions in seven sessions, each lasting approximately 15 minutes. Participants completed the assessment through an electronic questionnaire. To understand the feelings of the participants more accurately, we asked them to rate the relationship of each item to the three components using a 9-point Likert-type scale (1 = very slightly/not at all; 9 = extremely). Specifically, in the valence component, a score of 1 indicates “extremely unpleasant” (very painful, annoyed, dissatisfied, sad, disappointed), and 9 indicates “extremely happy” (extremely happy, happy, satisfied, hopeful). In the arousal component, 1 means “extremely calm” (calm, relaxed, little stimulation, little attention), and 9 means “extremely uncalm” (extremely excited, exciting, interesting, alarming, bright). In the dominance component, 1 indicates “completely controlled” (completely affected by the content of the sentence and feeling weak, directed, controlled, manipulated), and 9 indicates “completely in control” (feeling dominant, having full control, fully restricted, influential). ## Results ### The reliability analysis The second group evaluated 778 sentences from three components: valence, arousal, and dominance. The Cronbach’s α of the total average score of each of the three components were 0.987, 0.987, and 0.988, respectively, indicating a high level of credibility. We conducted reliability analysis based on previous studies. The Cronbach’s α for the results of the seven emotional situational categories ranged from 0.986–0.998, and the split-half ranged from 0.895–0.986, indicating that the ESSS has high credibility. As shown in, the retest reliability was represented by the correlation coefficient between the two assessments, and the retest reliability of the ESSS is 0.981, 0.938 and 0.726. ### Descriptive statistical analysis also reported a descriptive statistical analysis of 778 situational sentences in three components. Each sentence was rated by 80 participants. Then, the average score of 80 participants for a sentence was used as the sentence’s score. The mean (*M*) focused on the average of the sentences, and the standard deviation (*SD*) focused on the differences between the sentences. For the total, *SD* indicated the consistency of the score; moreover, dominance was the smallest, indicating that the score was more uniform than valence and arousal. The range of valence and arousal was larger than dominance, which indicated that they have a wider range of scores. In terms of different types of emotions, first, we analyzed negative emotions, among which there were four basic emotions. In terms of *SD*, the four emotions of fear, disgust, anger, and sadness showed little difference in the three components, that is, their scores were relatively consistent. For fear, the range of the three components was similar. For disgust, dominance was the most extensive and had the largest range. For anger, valence was the most extensive and had the largest range. In sadness, arousal was the narrowest and the range was the smallest. Anxiety is a complex emotion; it is rated low on valence, and can be considered a negative emotion. Among positive emotions—happiness had the highest *SD* of arousal. In terms of range, the arousal was the widest and the range was the largest. Finally, neutral sentences were analyzed, and the *SD* of valence was the largest, indicating that arousal and dominance degree were more concentrated than valence. In terms of range, the valence was the widest and the range was the largest. ### Scatter plot analysis *Scatter plot analysis of total*. Ggplot 2 package in R language was used for scatter plot analysis. First, the scatterplots of total average scores of valence, arousal, and dominance were analyzed. As shown in, the score distribution is relatively wide; however, there is a phenomenon of grouping, which may be due to different emotional dimensions (negative, positive, and neutral). Therefore, we will specifically analyze the score scatter diagrams for different types of emotion. *Scatter plot analysis of negative emotions*. In the ESSS, fear, disgust, anger, sadness, and anxiety belong to the category of “negative emotions.” As reported in, the scatter diagram of negative emotions shows a similar distribution. For fear, valence and arousal were significantly negatively correlated (*r* = -0.84, *p* \< 0.001)—the higher the valence, the lower the arousal. Arousal and dominance were significantly negatively correlated (*r* = -0.79, *p* \< 0.001)—the higher the dominance, the lower the arousal. Dominance and valence were significantly positively correlated (*r* = 0.80, *p* \< 0.001)—the higher the pleasure, the higher the dominance. For disgust, the valence and arousal were significantly negatively correlated (*r* = -0.87, *p* \< 0.001)—the higher the valence, the lower the arousal. Arousal and dominance were significantly negatively correlated (*r* = -0.90, *p* \< 0.001)—the higher the dominance, the lower the arousal. Dominance and valence were significantly positively correlated (*r* = 0.86, *p* \< 0.001)—the higher the valence, the higher the dominance. For anger, valence and arousal were significantly negatively correlated (*r* = -0.72, *p* \< 0.001)—the higher the valence, the lower the arousal. Arousal and dominance were significantly negatively correlated (*r* = -0.59, *p* \< 0.001)—the higher the dominance, the lower the arousal. Dominance and valence were significantly positively correlated (*r* = 0.52, *p* \< 0.001)—the higher the valence, the higher the dominance. For sadness, valence and arousal were significantly negatively correlated (*r* = -0.93, *p* \< 0.001)—the higher the valence, the lower the arousal. Also, the higher the dominance, the lower the arousal; thus, arousal and dominance were significantly negatively correlated (*r* = -0.90, *p* \< 0.001)—the higher the dominance, the lower the arousal. However, dominance and valence were significantly positively correlated (*r* = -0.89, *p* \< 0.001)—the higher the valence, the higher the dominance. For anxiety, valence and arousal were significantly negatively correlated (*r* = -0.89, *p* \< 0.001)—the higher the valence, the lower the arousal. Also, the higher the dominance, the lower the arousal; thus, arousal and dominance were significantly negatively correlated (*r* = -0.90, *p* \< 0.001)—the higher the dominance, the lower the arousal. Furthermore, dominance and valence were significantly positively correlated (*r* = 0.82, *p* \< 0.001)—the higher the valence, the higher is the dominance. *Scatter plot analysis of positive emotions*. As shown in, for happiness, there was a significant positive correlation between valence and arousal (*r* = 0.66, *p* \< 0.001)—the higher the valence, the higher the arousal. Arousal and dominance were significantly negatively correlated (*r* = -0.73, *p* \< 0.001)—the higher the dominance, the lower the arousal. Dominance and valence were negatively correlated (*r* = -0.67, *p* \< 0.001)—the higher the valence, the lower the dominance. *Scatter plot analysis of neutral emotions*. As shown in, for neutral, valence and arousal were significantly negatively correlated (*r* = -0.49, *p* \< 0.001)—the higher the valence, the lower the arousal. Arousal and dominance were significantly positively correlated (*r* = 0.25, *p* \< 0.05)—the higher the dominance, the higher the arousal. Dominance and valence were negatively correlated (*r* = -0.33, *p* \< 0.05), the higher the valence, the lower the dominance. # Study 2 ## Participants Thirty participants (14 men and 16 women) from Suzhou, aged 20.10 ± 1.32 (*M* ± *SD*) years were recruited through advertisements. The sample size was calculated by G\*Power (*α* = 0.05, *β* = 0.8), and the results showed that the sample size was standard and sufficient. Participants volunteered for this experiment and had not participated in Study 1. They signed informed consent forms prior to the experiment and were paid 10 yuan as compensation on completing the experiment. In addition, 39 college students (25 men and 14 women) from Jiangsu Province, aged 20.54 ± 1.83 (*M* ± *SD*) years completed the supplementary questionnaire. All participants were healthy, without mental illness, and had normal or corrected-to-normal vision. This study conformed to ethical standards and was approved by the Ethics Committee of our school. ## Experimental material ### Emotional pictures In this study, a total of 156 pictures were selected from the NimStim facial expression system. Twenty-six pictures were selected for each of the six emotions: fear, disgust, anger, sadness, happiness and neutral; another four were randomly selected for the exercise stage. The model in this system had expressions with their mouth open as well as with their mouth closed. Langeslag, Gootjesand van Strien found the former expression to be better as it could cause the individual to pay attention, and induce emotional effect. Therefore, the picture of the model with their mouth open was chosen in this study. To ensure that the selected pictures could more effectively stimulate the corresponding emotions among participants, before the final experiment, 31 participants, aged 18.35 ± 0.93 (*M* ± *SD*), were randomly selected from a university in Suzhou. They were asked to rate the pictures with the opened mouth in the NimStim facial expression system for arousal and valence. Finally, 13 pictures with the highest arousal for men and women in each dimension were selected as the formal materials of the experiment. The following details the selected picture number. Each emotional dimension has its own code. The fear dimension included 13 pictures of women (No. 1, 3, 5, 7, 8, 9, 10, 11, 13, 14, 16, 18, 19) and 13 pictures of men (No. 21, 22, 26, 29, 30, 34, 35, 37, 38, 39, 40, 42, 43). The disgust dimension 13 pictures of women (No. 2, 3, 5, 6, 7, 8, 9, 10, 13, 14, 16, 17, 19) and 13 pictures of men (No. 23, 24, 26, 29, 30, 31, 32, 33, 36, 39, 40, 41, 43). The anger dimension included 13 pictures of women (No. 1, 3, 6, 7, 8, 9, 10, 12, 13, 14, 17, 18, 19) and 13 pictures of men (No. 20, 22, 23, 25, 26, 27, 29, 31, 33, 34, 36, 38, 39).The sadness dimension included 13 pictures of women (No. 1, 3, 5, 6, 7, 9, 10, 11, 12, 13, 14, 18, 19) and 13 pictures of men (No. 20, 23, 24, 25, 27, 30, 31, 32, 33, 39, 40, 41, 42).The happiness dimension included 13 pictures of women (No. 1, 2, 3, 5, 6, 7, 8, 9, 11, 13, 14, 18, 19) and 13 pictures of men (No. 22, 23, 25, 27, 30, 31, 33, 34, 35, 36, 39, 40, 41). The neutral dimension included 13 pictures of women (No. 1, 2, 3, 5, 6, 7, 8, 9, 10, 13, 14, 16, 18) and 13 pictures of men (No. 21, 23, 26, 28, 29, 30, 32, 33, 35, 37, 38, 40, 41). The average valence and arousal of fear were 3.05 ± 0.46 and 6.69 ± 0.34 respectively. The average valence and arousal of disgust were 2.61 ± 0.46 and 7.04 ± 0.37 respectively. The average valence and arousal of sadness were 2.95 ± 0.37 and 6.28 ± 0.45 respectively. The average valence and arousal of happiness were 7.08 ± 0.28 and 6.44 ± 0.44 respectively. The average valence and arousal of neutral were 4.49 ± 0.35 and 4.31 ± 0.28 respectively. This indicates that the pictures used in the study can effectively enable the participants to experience the corresponding emotions and activate the participants’ strong corresponding emotional. ### ESSS In ESSS, 26 emotional statements of fear, disgust, anger, sadness, happiness and neutral were selected respectively, and the other 4 were randomly selected for the exercise stage. The first 26 items with the highest arousal were selected from each emotional dimension. The average valence and arousal of fear were 2.45 ± 0.14 and 7.20 ± 0.07 respectively. The average valence and arousal of disgust were 2.23 ± 0.23 and 7.10 ± 0.12 respectively. The average valence and arousal of anger were 2.44 ± 0.17 and 7.16 ± 0.10 respectively. The average valence and arousal of sadness were 2.51 ± 0.27 and 6.69 ± 0.19 respectively. The average valence and arousal of happiness were 7.78 ± 0.14 and 6.87 ± 0.10 respectively. The average valence and arousal of neutral were 5.77 ± 0.24 and 3.84 ± 0.08 respectively. Similarly, the statements used in the study can effectively make the participants experience corresponding emotions and activate their strong corresponding emotional experience. ## Procedure The experiment was produced and presented using E-prime 2.0 (Psychology Software Tools, Inc., Sharpsburg, PA, USA). As shown in, this experiment consisted of two types of tasks. In task one, a “+” lasting 500 ms was initially presented; next, the emotional pictures were presented. After 3000 ms, two emotional evaluation screens were presented. The first screen required participants to evaluate the emotional experience after looking at the picture. Ratings were given using a 9-point scale with “1” meaning extremely unpleasant (very painful, annoyed, dissatisfied, sad, disappointed), and “9” meaning extreme happiness (extremely happy, happy, satisfied, hopeful). The second screen required participants to evaluate arousal after looking at the picture. Arousal was also rated on a 9-point scale with “1” meaning extremely calm (calm, relaxed, little stimulation, least attention), and “9” meaning extremely uncalm (extremely excited, exciting, interesting, alarming, exciting, bright). Task two was similar to task one, except that the emotional statements from the ESSS were rendered in task two. Before starting the experiment, the participants participated in practice experiments, and entered the formal experiments once they fully understood the experimental process. The participants were asked to complete six parts in total, each part belonging to an emotional dimension, including 26 pictures and 26 sentences, a total of 52 items. The six parts were presented randomly, and the 52 items in each section were also presented at random. At the end of the experiment, the participants were asked to answer two questions: 1. According to you, which picture is closer to college students’ lives, the emotional picture, or the emotional situation statement?; 2. According to you, which picture is more interesting, the emotional picture or the emotional situation statement? The participants completed the experiment in approximately 30 minutes. To further compare the differences between emotional sentences and emotional pictures, 39 participants were recruited to complete the supplementary questionnaire after the experiment. In this questionnaire, the participants were shown two kinds of emotional priming materials, and then asked to evaluate how interesting the two materials were and how close they were to college students’ lives. The higher the number, the more interesting/close they were. ## Results ### Descriptive statistical analysis In this study, the items with the highest arousal of five basic emotions in the ESSS and NimStim were compared. reports the results of emotional pictures and emotional sentences analysis using paired sample *t*-test. First, the participants’ assessments of different negative emotions were analyzed. When different materials were used to stimulate the individual’s fear, disgust, anger, and sadness, the valence of emotional sentences was significantly lower than that of emotional pictures. On the other hand, the arousal was significantly higher than that of emotional pictures. This result showed that when negative emotions were aroused, emotional sentences activate more unpleasant emotions in participants compared to emotional pictures. For reaction time, when the participants’ negative emotions were activated, they took longer to judge the valence and arousal of sentences significantly. Second, when different materials were used to stimulate participants’ happiness, the valence and arousal of emotional sentences were significantly higher than that of emotional pictures. For the reaction times of valence and arousal, the reaction times for emotional sentences were significantly longer than those of emotional pictures. Finally, when different materials were used to stimulate the participants’ neutral emotion, the valence of emotional sentences was significantly higher than that of emotional pictures, but the arousal was significantly lower than that of emotional pictures. The reaction times for emotional sentences were significantly longer than those of emotional pictures, whether it was the judgment of valence or arousal. ### Analysis of essay questions Of the participants, 76.92% thought that emotional sentences were closer to college students’ lives than emotional pictures, and 66.67% thought that emotional sentences were more interesting than emotional pictures. Paired sample *t*-test was used to analyze the differences between emotional sentences and emotional pictures in terms of being interesting and college life closeness. The results also indicated that emotional sentences (6.77 ± 1.39) were more interesting than emotional pictures (6.03 ± 1.93) (*t* = 2.09, *p* \< 0.05, Cohen’s *d* = 0.44), and emotional sentences (6.62 ± 1.84) were more relevant to college students’ lives than emotional pictures (5.23 ± 1.95) (*t* = 3.69, *p* \< 0.01, Cohen’s *d* = 0.73). # Discussion ## A standardized and ecological emotional situation sentence system is established The results of the two studies indicate that the ESSS closely relates to college students’ lives, and is suitable for emotional priming. The process of emotional material standardization includes collecting and selecting the material, determining the evaluation dimensions, implementing the evaluation, and analyzing the reliability of the evaluation and content. According to all the indexes of the measurement results, the ESSS conforms to the requirements for each measurement, with good reliability and validity. It is a standardized, situational, and ecological emotional situational sentence system that is different from the existing emotional priming materials. In the process of standardization of the statements, according to Osgood’s theory, the valence, arousal, and dominance of materials can be standardized by using the method of self-reported assessments. Through the Cronbach’s α, the split-half and the retest reliability, we found that the reliability of each dimension was good and that the measurement requirements were met. Additionally, according to the scatterplot, all negative emotions showed a significant negative correlation between the valence and the arousal, a significant negative correlation between the arousal and the dominance, and a significant positive correlation between the dominance and the arousal. For positive emotions, there was a significant positive correlation between the valence and the arousal, a significant negative correlation between the arousal and the dominance, and a significant negative correlation between the dominance and the arousal. For neutral emotions, although there was significant correlation between the dimensions, the correlation coefficient obviously decreased. These results are similar to those reported by previous studies, indicating that the relationships are consistent with the rules of basic human emotions. The results also demonstrate that the seven emotions involved in this study could be further divided into three categories: negative emotions, positive emotions, and neutral emotions. This design could provide expedient conditions for scholars to explore positive and negative emotions, and neutral emotions can be used as a control group, increasing the scientific neutral of the research design. The scientific nature of the ESSS is confirmed by the statistical indicators, was ensured by the selection principles for the situational statements, and is evident in their contents. First, regarding the selection principles, which is similar to previous studies, we found that, in principle, the ESSS requires items to express a clear meaning to quickly stimulate participants’ corresponding emotions without requiring them to think carefully and consume too much of their cognitive resources. Second, the items in each emotion are drawn from the typical daily lives of the participants. Thus, they can easily arouse emotional resonance for the participants and stimulate the emotional experience. Third, we tried to enrich the content. For example, to elicit positive emotions, the ESSS included parents, friends, travel, study, money, work, games, exams, and so on. For negative emotions, the ESSS included things such as buildings, animals, movies, food, friends, and parents. For eliciting neutral emotion, the ESSS included classrooms, books, the college campus, landscapes, sounds, plants, and so on. In the compilation of the emotional system, the richer the stimulus content, the better it is for inducing the emotions needed for the experiment. At the same time, the rich scenes help maintain the interest of the participants, ensure the effect of emotional stimulation, and guarantee the effectiveness of emotional activation. ## The differences between ESSS and existing emotional activation materials The ESSS differs from existing emotional priming materials in many ways. Previous studies also used emotional systems involving words, such as the affective norms for English words (ANEW) and the Chinese emotion adjective words system. However, while these emotional systems are related to words, they are often used to study unconscious emotions, especially implicit emotions. Unlike a prior study, the present study established the ESSS using many statements with complete scenes. The purpose of the ESSS is the same as that of earlier systems using pictures, videos, sounds, and other emotional systems—to activate the explicit emotions of individuals. However, the ESSS is more likely to stimulate strong corresponding explicit emotions among participants in a short period due to the situational background. It is of great significance in examining emotion recognition of college students, the cultivation of college students’ network social ability, and the neural mechanism of how emotion affects cognitive activities. In addition, the items involved in this study are from college students’ own lives as well as prior studies; thus, they are closely related to college students’ lives, making the ESSS more ecologically relevant. Hence, it may be easier to activate the corresponding emotions of college students using the ESSS than by using other materials. It is undeniable that further revisions can be applied in the future, depending on the intended usage. # The advantages of ESSS Although the ESSS is an original emotional activation system, data analysis has demonstrated its many advantages over the emotional picture system in several ways. In valence, all negative emotions (fear, disgust, anger, and sadness) showed that the valence of emotional sentences was significantly lower than that of the pictures (the lower the valence, the less happy the participants felt); however, for happiness, the valence of emotional sentences was significantly higher than that of the pictures. From the perspective of arousal, the arousal of the six emotions compared in this study was significantly higher than that of the emotional pictures. With the emotional sentences, when the three emotions were activated, arousal was better than it was with pictures; moreover, no significant difference between the two was found in other dimensions. This finding suggests that emotional sentences are more effective, regardless of whether negative or positive emotions are needed. The results of negative emotions are different from the results of the comparison of words and pictures, which may be the advantage of emotional sentences and emotional words. The increased effectiveness of the ESSS may indicate that the included items are more closely related to the lives of college students than the items used by other systems; hence, college students may have more emotional responses to this system. During their college-going age, individuals have a stronger sense of identity, which enables them to experience stronger emotional resonance prompted by the faces of strangers or strange environments. Furthermore, some participants may find the ESSS content more interesting than emotional pictures. In the survey, some participants reported that “emotional sentences are more attractive,” “sentences make people more comfortable, and pictures are more rigid,” and so forth. When people are attracted to interesting objects, they tend to have higher levels of motivation, which may be one of the reasons why the ESSS can activate participants’ emotions better than pictures. Second, from the dimensions measured, the ESSS not only effectively activates the basic emotions described above but also can activate an individual’s anxiety emotions. In fact, there are few studies on the emotion of anxiety evoked by ecological conditions and situations, even though tension is known to be a common negative emotion among college students. At the beginning of college, the strange environment can bring out anxiety around interpersonal communication. College students must attain professional knowledge, and the diverse range of examinations they must pass may produce academic anxiety. Participating in campus activities such as speeches, debates, and music performance can create further anxiety relating to being on stage. Anxiety in college students must be properly controlled so that they can perform better in their studies, lives, and work, and they could enjoy the experiences of college lives. Previous studies on anxiety in college students have often been conducted through questionnaires and situational tests (such as Trier Social Stress Test, TSST). Due to the complexity and diversity of anxiety, there are few perfect methods for initiating anxiety in the subjects in experiments. The description of the development of the ESSS and its use may enrich the literature and make up for the gaps in previous studies, providing standardized tools for future scholars to explore anxiety, which is one of the innovations and the highlights of the ESSS. Finally, from the researchers’ point of view, it is necessary to evaluate the valence and arousal degree of the instrument, no matter which tool is used to activate the subjects’ emotions. When using previously established methods for eliciting emotion, men and women participants should be matched with gender- relevant emotional pictures. However, using ESSS to activate emotions does not involve preparing different stimuli based on gender because there is no difference between men and women in the design of the situations. Thus, the linking of gender matching could be reduced, allowing the overall workloads of researchers to be reduced. ## The advantages of emotional pictures According to the results of this study, although emotional sentences are better for activating emotions in valence and arousal, the use of emotional pictures still has some advantages. From the perspective of reaction time, when the participants evaluated valence and arousal, they responded significantly faster to the emotional pictures and the neutral pictures for the four emotions of disgust, anger, sadness, and happiness. This can be explained through the beliefs of some participants. For example, some participants believed that “pictures are more intuitive, sentences need more comprehension and understanding,” and “emotional pictures can make my experience more direct.” Prior studies on both emotions and emotional regulation, that have explored experimental materials’ priming effect on participants, and the differences between various emotional strategies on emotional regulation, usually require participants’ subjective evaluations of their emotional experiences. The results of the present study show that the participants were able to judge their current emotional experience more quickly after seeing the emotional pictures. Moreover these quick responses could reduce the overall time of an experiment, especially during ERP experiments. Another important factor is that compared with emotional situational statements, emotional pictures usually only need 1000 ms to arouse a participant’s emotions. However, it is obviously difficult to finish reading the emotional situation statements within 1000 ms. The presentation time of the emotional situation statements in this study was 4000 ms. Further verification is need to understand whether the emotion can be activated using emotional situation statements in a shorter time duration. From the perspective of applicability, in previous studies, emotional pictures (such as the NimStim set of facial expressions used in this study as an example) have been widely applied to participants of all groups and ages. For instance, emotional pictures have been used to study patients with depression, patients with schizophrenia, children, and college students. Emotional pictures have a wide range of practicality, and when different groups see pictures of emotional faces, they can activate the corresponding emotions for all the groups. However, the projects involved in the ESSS are closely related to the lives of college students, so the ESSS is more suitable for research with college students as participants. Therefore, because its use is more targeted, its scope of application is inevitably narrowed. From the perspective of cultural background, emotional pictures (also exemplified by the NimStim expression system) are widely used not only in Western studies but also in experimental research by Chinese scholars. Emotional pictures show facial expressions, they can include racial diversity, and they can equally represent men and women; therefore, their cultural background relevance is broad, which increases their applicability. However, the ESSS is more suitable for Chinese college students because only Chinese college students can recognize and understand some situations, such as those involving the CET-4 (a college English test used in China) and Taobao (an online shopping APP in China). # Conclusion The collection, screening, evaluation, and analysis of the ESSS are in line with the process of measurement. The ESSS comprises a set of standardized emotional scenario statements well-suited to the emotional activation of college students. The ESSS has significantly better arousal and potency than other stimuli and can be applied to experimental studies of “anxiety” emotions. However, using emotional pictures yields shorter response times, includes a wider application range, and may be better suited for examining cross-cultural characteristics. According to different research needs, researchers can choose the emotional initiation tool most suitable for their research. This study also has some limitations. First, it follows the traditional research method where in the data is obtained from subjective participant reports. However, we can obtain more objective data if the scoring method is explained in greater detail and the physiological indexes of the participants are observed in future studies. Second, the ESSS is closely related to the lives of Chinese college students; therefore, there are certain limitations to its application. In the future, research tools suitable for middle school students, workers, college students in other cultures or countries, and other groups should be further developed to expand the research scope and the application population. In addition, although this study included seven emotions, the types of emotional priming, such as pride, surprise, and guilt, could be further expanded in future research. Finally, the ESSS is a newly established emotional statement library, and its effectiveness in activating emotions must be verified in more experimental studies. In the future, study samples can be expanded to provide more references for emotional problems of college students. # Supporting information We wish to thank all participants for their participation and the reviewers for their valuable suggestions. 10.1371/journal.pone.0252671.r001 Decision Letter 0 Li Zezhi Academic Editor 2021 Zezhi Li This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 Mar 2021 PONE-D-20-40437 Can situations awaken emotions? The compilation and evaluation of the Emotional Situation Sentence System (ESSS) PLOS ONE Dear Dr. Zhao, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Apr 26 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Zezhi Li, Ph.D., M.D. Academic Editor PLOS ONE Journal Requirements:  When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and [https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf](about:blank) 2\. Please change "female” or "male" to "woman” or "man" as appropriate, when used as a noun (see for instance <https://apastyle.apa.org/style-grammar- guidelines/bias-free-language/gender>). 3\. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: <https://www.youtube.com/watch?v=_xcclfuvtxQ> 4.Thank you for stating the following financial disclosure: \[Yes\]. At this time, please address the following queries: Please clarify the sources of funding (financial or material support) for your study. List the grants or organizations that supported your study, including funding received from your institution.State what role the funders took in the study. If the funders had no role in your study, please state: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.”If any authors received a salary from any of your funders, please state which authors and which funders.If you did not receive any funding for this study, please state: “The authors received no specific funding for this work.” Please include your amended statements within your cover letter; we will change the online submission form on your behalf. 4\. Thank you for stating the following financial disclosure: "Yes" At this time, please address the following queries: Please clarify the sources of funding (financial or material support) for your study. List the grants or organizations that supported your study, including funding received from your institution.State what role the funders took in the study. If the funders had no role in your study, please state: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.”If any authors received a salary from any of your funders, please state which authors and which funders.If you did not receive any funding for this study, please state: “The authors received no specific funding for this work.” Please include your amended statements within your cover letter; we will change the online submission form on your behalf. 5\. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For more information on unacceptable data access restrictions, please see <http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data- access-restrictions>. In your revised cover letter, please address the following prompts: a\) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially sensitive information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent. b\) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. We will update your Data Availability statement on your behalf to reflect the information you provide. 6\. We note that Figure 5 includes an image of a participant  in the study. As per the PLOS ONE policy ([http://journals.plos.org/plosone/s/submission- guidelines#loc-human-subjects-research](about:blank#loc-human-subjects- research)) on papers that include identifying, or potentially identifying, information, the individual(s) or parent(s)/guardian(s) must be informed of the terms of the PLOS open-access (CC-BY) license and provide specific permission for publication of these details under the terms of this license. Please download the [Consent Form for Publication in a PLOS Journal](about:blank) ([http://journals.plos.org/plosone/s/file?id=8ce6/plos-consent-form- english.pdf](about:blank)). The signed consent form should not be submitted with the manuscript, but should be securely filed in the individual's case notes. Please amend the methods section and ethics statement of the manuscript to explicitly state that the patient/participant has provided consent for publication: “The individual in this manuscript has given written informed consent (as outlined in PLOS consent form) to publish these case details”. If you are unable to obtain consent from the subject of the photograph, you will need to remove the figure and any other textual identifying information or case descriptions for this individual. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This study established and evaluated a standardized emotional situation sentence system (ESSS) relevant to the lives of college students. This ESSS used many statements with complete scenes, which is innovative and necessary. The results showed that the ESSS could better stimulate the emotions of the subjects than emotional pictures. I have some confusion and suggestions for modification. (1)In study 1, there were two groups, and the first group put forward statements with complete scenes. However, I did not see any result of the first group in Results. A few examples for each emotion and expert-rating consistency reliability should present. The content in “Filtering situation statements” should be the results rather than methods. After all, the purpose of Study 1 is to build an emotional database, so the contents of this database should be the first and most important result. Then the results of reliability and validity of the database should be showed. (2)I am confused by the Table 2. As I understand it, there are two kinds of analysis to get M and SD. The first one is as follows: Each emotion had many sentences, just like a questionnaire dimension with a lot of questions. A participant rated all the sentences for this emotion, for example, fear. Then the average score of all sentences for fear is used as the participant's score for fear. Therefore, each participant had a score for fear. The M was obtained by taking the average of 80 participants, and the SD was obtained by taking the standard deviation of 80 participants. If you take this analysis approach, the M will focus on the average of the 80 participants, and the SD will focus on the individual differences of the 80 participants. The second one is as follows: Each emotion had many sentences, just like a questionnaire dimension with a lot of questions. Each fear sentence was rated by 80 participants. Then the average score of 80 participants for a fear sentence is used as the fear sentence's score. Therefore, each fear sentence had a score. The M was obtained by taking the average of 121 fear sentences, and the SD was obtained by taking the standard deviation of 121 fear sentences. If you take this analysis approach, the M will focus on the average of the 121 fear sentences, and the SD will focus on the differences of the 121 fear sentences. There are slight numerical differences but substantial qualitative differences between the two analysis approaches. In my opinion, the second approach is more suitable for the purpose of building a database, such as getting the average number of comments per sentence, as the author showed in study 2. Please indicate which method was used, and if the first method was used, I suggest using the second or both methods. In addition, only 80 participants rated, which was small sample size for building and testing a database. less. To avoid fatigue effects, each participant did not evaluate all the sentences, so how was the data analyzed? Then the number of raters for each sentence was reduced further, which had a great impact on the reliability and validity of the database. Therefore, it is suggested to increase the number of participants to at lest 300. (3)Study 2 compared the differences between emotional sentences and emotional pictures, and found that the valence of emotional sentences was closer to the emotional type, had higher arousal degree, higher authenticity and higher interesting. Why only use proportion, not 9 point scale, to evaluate authenticity and interesting? Authenticity and interesting should be two of the core values of the database and quantified data should be obtained using a 9-point scale. Study 2 actually examined the calibration validity of the ESSS (its similarity to the emotional pictures) and ecological validity (better than the emotional pictures). Study 1 only examined reliability, not validity. Therefore, Study 2 should point out that it actually examined validity. In this way, study 1 and study 2 will form a unified whole for the purpose of building ESSS. Of course, to test validity, study 2 need to increase the participants to about 100. (4)Previous studies also used emotional systems involving words, such as the affective norms for English words (ANEW) (Fairfield et al., 2017) and the Chinese emotion adjective words system (Lei & Zhang, 2013). So why this study did not compare ESSS and emotion words? Such studies should be supplemented to reflect the advancement, necessity and validity of ESSS. Since this study needs to further supplement data analysis and experiments, it is suggested that the editorial department give the author 3 months to modify. Reviewer \#2: The ESSS compiled in this study is suitable for experiments with college students as subjects. At present, quite a lot of researches on emotion are based on college students, so it is of great significance to compile an emotion system suitable for college students. Study 2 illustrated their respective advantages through the comparison of ESSS and emotional pictures, which increased the reliability of the research results. The following questions need to be considered by researchers. 1\. Study 2 mainly compared the emotional pictures with ESSS, and the reasons for the comparison should be explained in more detail. 2\. The pictures presented in the research results are not clear enough. Figure 5 shows a Chinese sentence that needs to be translated into the manuscript. It is suggested that researchers reformat the figure 1 to figure 4 to make them more beautiful and clear. 3\. Study 1 mentioned that the number of words in the situational statements consisted of approximately 10–15 words, please explain the rationale for that. 4\. The innovation of Study 1, compared to existing database, is that added “anxiety” in building ESSS. It is suggested that researchers should give adequate explanations of why it's necessary. 5\. The results of Study 2 mainly indicated that the ESSS had significantly better arousal and potency than pictures, and the emotion images had shorter response times. Their difference in practical application, however, is less discussed. And so, to supplement and perfect that is suggested. 6\. In the process of writing, it is necessary to increase the amount of reference literature in recent three years. 7\. To further improve the quality of the language, researchers can invite English major teachers to help with sentence by sentence modification, especially in the discussion section. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Jianxin Zhang Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0252671.r002 Author response to Decision Letter 0 24 Apr 2021 Reviewer \#1: This study established and evaluated a standardized emotional situation sentence system (ESSS) relevant to the lives of college students. This ESSS used many statements with complete scenes, which is innovative and necessary. The results showed that the ESSS could better stimulate the emotions of the subjects than emotional pictures. I have some confusion and suggestions for modification. (1)In study 1, there were two groups, and the first group put forward statements with complete scenes. However, I did not see any result of the first group in Results. A few examples for each emotion and expert-rating consistency reliability should present. The content in “Filtering situation statements” should be the results rather than methods. After all, the purpose of Study 1 is to build an emotional database, so the contents of this database should be the first and most important result. Then the results of reliability and validity of the database should be showed. Reply: Thanks for the expert's suggestion. According to the experts' suggestion, we give an example of each emotion in filtering situation statements. This will allow readers to have a more direct understanding of the ESSS. Similar to previous studies (e.g. Bai et al., 2005; Liu et al., 2006; Fairfield et al., 2017), in this part, 15 experts are required to select and retain suitable items according to the principles. (2)I am confused by the Table 2. As I understand it, there are two kinds of analysis to get M and SD. The first one is as follows: Each emotion had many sentences, just like a questionnaire dimension with a lot of questions. A participant rated all the sentences for this emotion, for example, fear. Then the average score of all sentences for fear is used as the participant's score for fear. Therefore, each participant had a score for fear. The M was obtained by taking the average of 80 participants, and the SD was obtained by taking the standard deviation of 80 participants. If you take this analysis approach, the M will focus on the average of the 80 participants, and the SD will focus on the individual differences of the 80 participants. The second one is as follows: Each emotion had many sentences, just like a questionnaire dimension with a lot of questions. Each fear sentence was rated by 80 participants. Then the average score of 80 participants for a fear sentence is used as the fear sentence's score. Therefore, each fear sentence had a score. The M was obtained by taking the average of 121 fear sentences, and the SD was obtained by taking the standard deviation of 121 fear sentences. If you take this analysis approach, the M will focus on the average of the 121 fear sentences, and the SD will focus on the differences of the 121 fear sentences. There are slight numerical differences but substantial qualitative differences between the two analysis approaches. In my opinion, the second approach is more suitable for the purpose of building a database, such as getting the average number of comments per sentence, as the author showed in study 2. Please indicate which method was used, and if the first method was used, I suggest using the second or both methods. In addition, only 80 participants rated, which was small sample size for building and testing a database. less. To avoid fatigue effects, each participant did not evaluate all the sentences, so how was the data analyzed? Then the number of raters for each sentence was reduced further, which had a great impact on the reliability and validity of the database. Therefore, it is suggested to increase the number of participants to at lest 300. Reply: Thanks to the expert for the question. Based on your suggestions, we made the following corrections: A. “The second approach is more suitable for the purpose of building a database”. As suggested by expert, the method for data analysis we actually used was the second approach. In order to make the readers understand more clearly, we have carried on the improvement to this part as follows: Table 2 reported a descriptive statistical analysis of 778 situational sentences in three components. Each sentence was rated by 80 participants. Then, the average score of 80 participants for a sentence was used as the sentence's score. The mean (M) focused on the average of the sentences, and the standard deviation (SD) focused on the differences between the sentences. B. Maybe we didn't express it clearly enough before, which caused some misunderstanding. To reduce the fatigue effect of participants, the same participant completed the assessment of seven emotional dimensions in seven sessions, and each session last about 15 minutes. In other words, every participant rated all seven emotional dimensions and they only evaluated one emotional dimension every time which lasting about 15 minutes, instead of “each participant did not evaluate all the sentences”. The final number of participants was determined based on previous research on the compilation of emotional system (Bai, Ma, Huang, & Luo, 2005). Initially, 91 participants were enrolled in the study, but to improve the accuracy of the results, we eliminated participants who did not complete the seven assessments. The following is the number of participants participating in the evaluation in some of the studies. It can be found that the sample of our study is larger or similar to that of previous related studies. 1\. Bai, L., Ma, H., Huang, Y. X., & Luo, Y. J. (2005). The development of native Chinese affective picture system-A pretest in 46 college students. Chinese Mental Health Journal, 19(22), 719-722. Sample: 46; Cited: 400 times; Downloads: 7178 2\. Liu, T. S., Luo, Y. J., Ma, H., & Huang, Y. X. (2006). The establishment and assessment of a native affective sound system. Psychological Science, 29(2), 406-408. Sample: 50; Cited: 102 times; Downloads: 1587 3\. Mikels, J. A., Fredrickson, B. L., Larkin, G. R., Lindberg, C. M., Maglio, S. J., & Reuter-Lorenz, P. A. (2005). Emotional category data on images from the international affective picture system. Behavior Research Methods, 37(4), 626-630. Sample: 60; Cited: 320 times 4\. Lei, Y., Sun, X. Y., Dou, H. R. (2019). Specifically inducing fear and disgust emotions by using separate stimuli: The development of fear and disgust picture systems. Journal of Psychological Science,42(03):521-528 Sample: 84; Cited: 1 times; Downloads: 645 (3)Study 2 compared the differences between emotional sentences and emotional pictures, and found that the valence of emotional sentences was closer to the emotional type, had higher arousal degree, higher authenticity and higher interesting. Why only use proportion, not 9 point scale, to evaluate authenticity and interesting? Authenticity and interesting should be two of the core values of the database and quantified data should be obtained using a 9-point scale. Study 2 actually examined the calibration validity of the ESSS (its similarity to the emotional pictures) and ecological validity (better than the emotional pictures). Study 1 only examined reliability, not validity. Therefore, Study 2 should point out that it actually examined validity. In this way, study 1 and study 2 will form a unified whole for the purpose of building ESSS. Of course, to test validity, study 2 need to increase the participants to about 100. Reply: We carefully considered the expert's suggestion and improved the manuscript according to the expert's suggestion. A: The scale description is indeed not more accurate than the 9-point scale, so we re-surveyed 39 college students and asked them to compare the interestingness and familiarity of ESSS with emotional pictures using a 9-point scale, and the results are as follows. Paired sample t-test was used to analyze the differences between emotional sentences and emotional pictures in terms of being interesting and college life closeness. The results also indicated that emotional sentences (6.77 ± 1.39) were more interesting than emotional pictures (6.03 ± 1.93) (t= 2.09, p \< 0.05, Cohen's d= 0.44), and emotional sentences (6.62 ± 1.84) were more relevant to college students' lives than emotional pictures (5.23 ± 1.95) (t= 3.69, p \< 0.01, Cohen's d = 0.73). B: Study 2 actually examined the calibration validity of the ESSS (its similarity to the emotional pictures) and ecological validity (better than the emotional pictures). This suggestion from the expert is very accurate and useful. Therefore, we have made supplements and improvements in Design and Methods. C. In terms of sample size, calculated by G\* POWER 3.1 (α=0.05, β=0.8), the sample size of this experiment has reached saturation state (30 \>27).In addition, our sample size was similar to, or even larger than the sample size of previous related studies. For example, through 17 participants (30\>17), Xie and Yang (2016) compared the differences in emotional stimulation of four commonly used emotional-inducing methods: pictures, music, movies and meetings. Bayer and Schacht (2014) compared the difference between emotional words, pictures and faces with 24 participants in the experiment (30\>24). Schacht, Adler, Chen, Guo and Sommer (2012) compared the difference between emotional pictures, faces and words with 16 participants (30\>16). 1\. Xie, Y. Z., Y, Z. (2016). A comparative study on the validity of different mood induction procedures (MIPs). Studies of Psychology and Behavior, 14(5), 591-599. (Sample: 17) 2\. Bayer, M., & Schacht, A. (2014). Event-related brain responses to emotional words, pictures, and faces - a cross-domain comparison. Frontiers in Psychology, 5, 1106. (Sample: 24) 3\. Schacht, A., Adler, N., Chen, P., Guo, T., & Sommer, W. (2012). Association with positive outcome induces early effects in event-related brain potentials. Biological Psychology, 89(1), 130-136. (Sample: 16) 4\. Bayer, M., & Schacht, A. (2014). Event-related brain responses to emotional words, pictures, and faces - a cross-domain comparison. Frontiers in Psychology, 5, 1106. (Sample: 25) 5\. Kensinger, E. A., & Schacter, D. L. (2006). Processing emotional pictures and words: Effects of valence and arousal. Cognitive, Affective, & Behavioral Neuroscience, 6(2), 110-126. (Sample: 21) (4)Previous studies also used emotional systems involving words, such as the affective norms for English words (ANEW) (Fairfield et al., 2017) and the Chinese emotion adjective words system (Lei & Zhang, 2013). So why this study did not compare ESSS and emotion words? Such studies should be supplemented to reflect the advancement, necessity and validity of ESSS. Reply: Thanks for the expert's suggestion. Previous studies used emotional systems involving words, such as the affective norms for English words (ANEW) (Fairfield et al., 2017) and the Chinese emotion adjective words system (Lei & Zhang, 2013). Although these emotional systems are related to words, they are often used to study unconscious emotions, especially implicit emotions (Lei & Zhang, 2013). ESSS plays a similar role to emotional pictures, and its main purpose is to activate the explicit emotions of the participants in the experiment. In particular, the complement of anxiety makes up for the deficiency of the previous emotional system. According to the suggestion of the expert, we made a supplement to “The differences between ESSS and existing emotional activation materials” on the basis of previous studies (Smith & Smith, 2019; Lin, Li, Cao, Lv, & Ke, 2018; Ding et al., 2020). In the future research, we will pay attention to this problem. According to the suggestion of the expert, we explain the shortcomings of the study at the end of the manuscript. Since this study needs to further supplement data analysis and experiments, it is suggested that the editorial department give the author 3 months to modify Reviewer \#2: The ESSS compiled in this study is suitable for experiments with college students as subjects. At present, quite a lot of researches on emotion are based on college students, so it is of great significance to compile an emotion system suitable for college students. Study 2 illustrated their respective advantages through the comparison of ESSS and emotional pictures, which increased the reliability of the research results. The following questions need to be considered by researchers. 1\. Study 2 mainly compared the emotional pictures with ESSS, and the reasons for the comparison should be explained in more detail. Reply: Thank you for your valuable suggestion. Based on previous studies (Cui, Song, Si, Wu, & Feng, 2021; Mowle, Edens, Ruchensky, & Penson, 2019), we clarified why the ESSS were compared with the emotional pictures, and the specific reasons are as follows. At present, emotion picture is one of the most commonly used emotion priming materials in emotion research, and contains a wide range of emotion types, with good reliability and validity. Study 2 aimed to examine the calibration validity of the ESSS as well as ecological validity. As emotional pictures and ESSS contain similar types of emotions, they have the possibility of comparison, and the comparison with authoritative materials can better illustrate their respective advantages. 2\. The pictures presented in the research results are not clear enough. Figure 5 shows a Chinese sentence that needs to be translated into the manuscript. It is suggested that researchers reformat the figure 1 to figure 4 to make them more beautiful and clear. Reply: The expert's suggestion is very useful, the previous figures are really not clear enough. To better present our results, ggplot 2 (Wickham, 2010) package in R language (R Core Team 2020) was used for scatter plot analysis. Translating the Chinese in figure 5 will help more readers to understand the experimental process. However, as the editor proposed that the picture in Figure 5 should be modified, in order to correspond with the description in Task 1, we modified Figure 5, so this sentence in Chinese was deleted. 3\. Study 1 mentioned that the number of words in the situational statements consisted of approximately 10–15 words, please explain the rationale for that. Reply: According to the expert's suggestion, we have made a supplement to this part, as follows: To make the meaning of the sentence clear to the participants quickly\[19, 20\], we ensured that each situational statement comprised approximately 10–15 words. 4\. The innovation of Study 1, compared to existing database, is that added “anxiety” in building ESSS. It is suggested that researchers should give adequate explanations of why it's necessary. Reply: Thanks for the expert's valuable suggestion. Anxiety is an important part of our study, we have made a supplement to this part, as follows: Moreover, college students are also easily affected by emergencies. This was confirmed by Hoyt et al. \[12\] in a survey of 707 American college students. The study results found that during the COVID-19 pandemic, most students experienced anxiety and stress, while their happiness continued to decline. 5\. The results of Study 2 mainly indicated that the ESSS had significantly better arousal and potency than pictures, and the emotion images had shorter response times. Their difference in practical application, however, is less discussed. And so, to supplement and perfect that is suggested. Reply: The advantages of emotional pictures in reaction time are discussed in the first paragraph of the advantages of emotional pictures. According to the expert's suggestion, we have made a supplement to this part. We further demonstrate that priming participants' emotions with emotional pictures is more suitable for the study of neural mechanisms and can shorten the duration of the experiment. 6\. In the process of writing, it is necessary to increase the amount of reference literature in recent three years. Reply: In the process of this revision, we paid more attention to the suggestion of the expert when referring to previous studies, and revised the manuscript according to the research of the recent three years (e.g. Cui, Song, Si, Wu, & Feng, 2021; Ding et al., 2020; Hoyt, Cohen, Dull, Maker Castro, & Yazdani, 2021; Mowle, Edens, Ruchensky, & Penson, 2019). 7\. To further improve the quality of the language, researchers can invite English major teachers to help with sentence by sentence modification, especially in the discussion section. Reply: Thanks for the expert's suggestion. In order to improve the quality of the article language, we first conducted self-proofreading. Important parts of the manuscript were then revised through the Edtage, a professional language organization. The supporting materials are as follows. The editor also put forward a lot of valuable suggestions, according to the editor's suggestion, we have revised the manuscript. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. Reply: We revised the manuscript at the editor's request. As this is my first submission to PLoS ONE, I am worried that there is something wrong with it. If there is any problem, I am very sorry. So, if the editor finds any problem, please feel free to contact me. I will try my best to modify it. 2\. Please change "female” or "male" to "woman” or "man" as appropriate, when used as a noun (see for instance <https://apastyle.apa.org/style-grammar- guidelines/bias-free-language/gender>). Reply: We have changed "female” or "male" to “woman” or "man" as appropriate. 3\. PLOS requires an ORCID ID for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Reply: According to the editor's requirement, I registered an account with ORCID. 4\. Thank you for stating the following financial disclosure: \[Yes\]. At this time, please address the following queries: a\. Please clarify the sources of funding (financial or material support) for your study. List the grants or organizations that supported your study, including funding received from your institution. b\. State what role the funders took in the study. If the funders had no role in your study, please state: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” c\. If any authors received a salary from any of your funders, please state which authors and which funders. d\. If you did not receive any funding for this study, please state: “The authors received no specific funding for this work.” Please include your amended statements within your cover letter; we will change the online submission form on your behalf. Reply: This study was supported by the Qinglan Project of Jiangsu Universities and the police lie detection method with micro-expression recognition (Applied Innovation Project of Ministry of Public Security, Project No. : 2018YYCXJSST029). 5\. We note that you have indicated that data from this study are available upon request. Reply: The data of this study has been uploaded in the system. As a final note, the full version of ESSS is available via the link. You can also email the corresponding author for ESSS and related data. 6\. If you are unable to obtain consent from the subject of the photograph, you will need to remove the figure and any other textual identifying information or case descriptions for this individual. Reply: We deleted the people photo in Figure 5. 10.1371/journal.pone.0252671.r003 Decision Letter 1 Li Zezhi Academic Editor 2021 Zezhi Li This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 20 May 2021 Can situations awaken emotions? The compilation and evaluation of the Emotional Situation Sentence System (ESSS) PONE-D-20-40437R1 Dear Dr. Zhao, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, Zezhi Li, Ph.D., M.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#2: No 10.1371/journal.pone.0252671.r004 Acceptance letter Li Zezhi Academic Editor 2021 Zezhi Li This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 8 Jul 2021 PONE-D-20-40437R1 Can situations awaken emotions? The compilation and evaluation of the Emotional Situation Sentence System (ESSS) Dear Dr. Zhao: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Zezhi Li Academic Editor PLOS ONE [^1]: No authors have competing interests.

# Introduction ‘Patient flow’ describes the flow or movement of patients through the different stages of required hospital care and considers whether they are subject to unnecessary delay. Poor patient flow is especially apparent when incoming emergency department (ED) patients cannot immediately be admitted into the main hospital due to the lack of beds available. However, hospital bed management is frequently reactive and so delays in discharge, and by extension the release of hospital beds, are commonplace. The effects of poor patient flow are amplified during periods of viral infection outbreaks, such as seasonal influenza, and the Coronavirus disease 2019 (COVID-19) pandemic. Delays in the release of hospital beds from all patient types lead to hospitals being unable to accept surges of patients arriving with infection. Anticipating the recovery of patients from infections, as well as other illnesses, is therefore a key step in facilitating safer and more efficient releasing of hospital beds, thereby improving overall patient flow in hospitals at times of critically high occupancy. The recent proliferation of electronic health record (EHR) systems by hospitals provides an opportunity to employ promising data-driven approaches, such as deep learning, to challenging medical problems such as patient discharge prediction. Research in this field to-date has typically focused on classifying, at a single point in time, a patient’s length of stay (LOS) into short, medium, or long stays, a task that is usually performed on admission or pre-operatively. Most studies to date have restricted themselves to making predictions for patients of a specific diagnostic category. By contrast, only a small number of studies make more operationally-focused discharge predictions, out of which four use machine learning (ML) methods and two use deep learning methods. However, these papers restrict themselves in their predictions, to LOS within the intensive care unit (ICU), or to patients who have had a surgical procedure, or those that are in certain wards. Our main contributions are as follows: - Proposal of a strategy for using machine learning models to make operationally focused, real-time discharge predictions for almost all individual patients in hospital at any given time, to improve patient flow in hospital during periods associated with spikes in hospital admissions due to, for example, infection outbreaks such as the seasonal influenza. - The use of separate models for patients discharge prediction, where independent models are trained independently based on patient admission type and number of elapsed days since admission. - Feature analysis of variables used within the models; variables learned as being of predictive value can be incorporated in future related studies. # Methods ## Data We analysed patient data collected in the EHR of the John Radcliffe Hospital, within the Oxford University Hospitals NHS Foundation Trust, between January 2013 and April 2017, a period that was studied due to the annual resurgence of influenza. This is a teaching hospital group serving a population of 600,000 and providing tertiary services to the surrounding region. De-identified patient data was obtained from the Infections in Oxfordshire Research Database (IORD). One of the largest datasets of its kind, the extracted data contains 431,458 records of unique admissions to hospital from 225,009 de-identified, adult patients. This study considers a subset of 49,832 admissions, recorded across the four years of the study period, who met the criteria of normal discharge and had full vital-signs observation sets. To select the cohort of patients for which a discharge prediction would be most clinically useful, we considered only patients who are likely to have required a hospital bed. We identified these patients by selecting only patients admitted to general hospital for longer than 6 hours. These 6 hours do not include any time spent in the ED and therefore we do not consider patients who only visited ED. In the UK healthcare system, patients remain under the care of ED for up to 4 hours and only those requiring longer hospital observation or treatment are admitted to main hospital. We also excluded patients attending only as outpatients, for example those attending regular haemodialysis sessions. Patient admissions were categorised as either planned or emergency admissions, where planned admissions were those scheduled in advance whilst emergency admissions describe patients whose entry into the main hospital was through the ED. While planned admissions are often for surgery, followed by a relatively predictable trajectory of recovery, emergency admissions, which are frequently precipitated by infection, generally present a more challenging patient type for hospital bed managers to predict discharge. The cohort of emergency patients with infection broadly reflects the patient admission type which would spike during a seasonal influenza outbreak, with this cohort having the longest average LOS and with the highest variability in their LOS. Within our dataset the median (IQR) length of stay was 2.9 (0.85–6.3) days, with detailing the LOS variability for the patient cohorts considered. The top ten most presented primary diagnostic codes in the international classification of disease (ICD-10) format, were: J181, I251, N390, I639, S7200, I214, I500, S0650, N179, A419 (lobar pneumonia, atherosclerotic heart disease, urinary tract infection, cerebral infarction, femur fracture, myocardial infarction, heart failure, subdural haemorrhage, acute kidney failure, sepsis). Our predictions were therefore made in a cohort typical of those admitted to hospital, who frequently have complex multifactorial care needs and whose recovery trajectories can be difficult to forecast. ## Ethics De-identified patient data was obtained from the Infections in Oxfordshire Research Database (IORD) which has generic Research Ethics Committee, Health Research Authority and Confidentiality Advisory Group approvals (19/SC/0403,19/CAG/0144) as a de-identified electronic research database. We describe an approach for utilising data from the electronic health records of patients admitted to hospital, to develop models to predict readiness for discharge for patient cohorts within hospital, including those with infection. ## Study design The system proposed in this work aims to provide operationally focused clinical decision support for periods of crises in hospital. We propose a strategy in which hospital bed managers run these models from within a decision-support tool during a period of high influx of patients with infectious disease. The models would identify the patients who are most likely to be ready for discharge within the next 24 hours. A medical professional would then be assigned to screen the highest ranked patients to confirm the models’ predictions. Once confirmed, hospital bed managers would be able to proactively make discharge arrangements for that patient, to release them from the hospital as quickly as possible and to save valuable time during a critical situation in hospital. Predictions can be made for all patients currently in hospital at any time and thus can incorporate new data as it becomes available. In this study, we simulated predictions being made every 24 hours, with the initial prediction being made on the day of a patient’s admission to main hospital. We constructed individual models for each patient admission group (planned and emergency admissions) and for each day elapsed since a patient’s admission to hospital. Elapsed times since admission t ∈ {0,1,…,7} were considered, with t = 0 representing the day a patient was admitted to the general hospital. For this study, patient stays were truncated at 7 days. Consequently, 16 different independent models, per model architecture, were developed. The sub-datasets used to train and evaluate the models are denoted *D*_pt and *D*_et, respectively, with the first subscript indicating the patient admission type, and the second indicating the time elapsed in days since admission. For example, as shown in, if Patient 1 is a planned patient, who arrives in hospital on 02/02/2016 and stays in hospital for 2 days, they will be included in datasets *D*_p0 and *D*_p1. If Patient 3, a different planned admission, arrives in hospital on 03/02/2016 and stays in hospital 6 days, they will also be included in datasets *D*_p0 and *D*_p1 along with Patient 1, and will additionally be included in datasets *D*_p2, *D*_p3, *D*_p4 and *D*_p5. Each of the sub-datasets were balanced by down-sampling to improve the training and to allow for unbiased testing of the models, details of the down-sampling strategy can be found in Appendix A in. The resulting size of each sub-dataset is summarised in. Diminishing quantities of data were available for increasing *t*, as the sub-datasets only include patients who have not been discharged after *t* days. In this work, a prediction by a model that a patient will be discharged within the next 24 hours is denoted a *positive* prediction, whilst a prediction that a patient will not be discharged in the next 24 hours is denoted a *negative* prediction. Based on the probability score predicted for each patient, each proposed model ranks patients based on their likelihood of discharge. ## Model development ### Model architecture In this study, four supervised ML classifiers were considered. Random forest (RF) and support vector machine (SVM) models, which have previously been shown to give good performance were compared with deep neural networks (DNN) in the form of multilayer perceptron (MLP) models. Logistic regressor (LR) models were also included to serve as a baseline, being a strong comparator from medical statistics. The different classifiers were assessed on their ability to predict whether an inpatient would be discharged within the next 24 hours and the probability scores given by the classifier were used to rank patients in order of their likelihood of discharge. Model hyperparameters were selected through a nested K-fold cross-validation scheme on the *D*_e0 dataset, where the outer- and inner-loops consisted of 5 and 3 folds respectively. The 5-fold scheme partitioned the data into training and evaluation folds, whilst the additional 3-fold partition was applied in an inner-loop on the training set folds, to create a training- validation set to assess performance of different hyperparameter choices. A grid-search approach was used to test different hyperparameter combinations, with the combination giving highest average AUROC across all validation folds eventually selected for all models. The hyperparameter values determined and used are detailed in. ### Feature engineering Domain knowledge and prior literature were used to determine which information within the dataset would be most useful for predicting patient discharge. Handcrafted features used to train the models included: age, day of the week, procedures information, ICU information and statistical representations of the National Early Warning Score (NEWS) metric, which encodes vital signs information, binned into 24-hour periods. Temporal features such as ‘time elapsed since procedure’, ‘time elapsed since ICU discharge’ and features relating to NEWS were populated in ‘real-time’, only being included into the models for which the information would be available. A maximum of 79 features were engineered, the full list of which is summarised in. For operational purposes in hospital, it is preferable for a decision support tool to be able to make predictions for all patient groups in the hospital at any given time. Patient diagnosis is typically classified using international classification of disease (ICD) or “Clinical Classifications Software” (CCS) groupings, both of which contain too many diagnostic groups to be easily included as ML features directly. As stated earlier, most prior studies restrict themselves to a handful of patient diagnostic categories or a specific patient type. In this study, to directly capture the effects of a patient’s diagnostic category on LOS, features containing the historic mean and variance of the LOS of patients within the same diagnostic category as the patient-under-test were developed. The historic mean and variance of LOS for a particular CCS category were calculated using the training dataset. These mean and variance values were then assigned to patients of the same CCS category in both the training and the test datasets. For patients in the test set with an unseen CCS category, the average of all diagnostic categories was assigned for each feature. Under the present hospital processes, diagnostic categories are assigned and recorded on a patient’s discharge. As such, the information used in this study can be thought of as a proxy for the working diagnosis assigned by clinicians during a patient’s stay. If implemented as a decision support tool, suspected CCS category could be recorded by clinicians and used within the models in real- time. ### Feature selection For the SVM models, which are particularly sensitive to the inclusion of features with low predictive value, feature selection techniques were applied. Spatially Uniform RelieF (SURF) feature selection algorithm was used to select features, as we found it to be the most robust against white noise features and to be one of the most consistent at picking similar sets of features across 3-fold cross-validation in a comparison between feature selection algorithms. This algorithm uses the proximity of samples in feature space to describe how feature interactions relate to the sample’s class. The normalised scores from running the SURF feature selection algorithm over the engineered features were generated. The detailed methodology of running this algorithm can be found in Appendix C in. For the other non-SVM ML models, all features as described in were used. # Results ## Feature importance Feature selection can provide medical practitioners with valuable insight into the importance of each feature in the predictions made. The results of the feature selection method show that for both planned and emergency admission types, the feature deemed most important by the SURF algorithm was feature no. 78, the historic mean LOS of patients in the same diagnostic category. This feature, described earlier, aims to capture the effect of a patient’s diagnosis. Age (feature no. 1), Charlson Comorbidity Index CCI (no. 2) and NEWS features (nos. 52–77) were shown to influence discharge predictions significantly, with age and CCI being of particular importance for emergency admissions. For both planned and emergency admissions, abnormal blood test results (nos. 36, 45) were informative. Whether blood tests were taken within the last 48 hour period (nos. 34–42) were seen to be informative features for planned admissions; with albumin blood tests (no. 34) found to be particularly important. This was the only blood test included as a feature which would not be carried out in the hospital by default, but rather would have been requested as an additional test for a patient by a clinician. Information about procedures and operating theatres were shown to be of high predictive value for patients with planned admissions (nos. 24–31), while ICU features (nos. 10–23) were shown not to be of importance for patients in either dataset. ## Predictive performance In this study, the models developed were evaluated to indicate the efficacy of the models’ use in an operational hospital setting during crises. We propose that, in this setting, 20% of all patients in hospital at a given moment with a positive prediction by the model would be a reasonable proportion of patients to be considered as candidates for discharge. However, this threshold could be adjusted to match the needs of the hospital at any point. Hospital bed managers would oversee the use of these models. We would expect these patients to then have a further screening by a clinician to confirm whether they are ready for discharge or not. The models were evaluated in terms of their mean and variance in positive predictive value (PPV) over a 5-fold cross-validation. A positive classification was given to any sample with a probability score of 0.5. Each dataset was randomly split into five-folds, with 80% of the data used to train the model, and the remaining unseen 20% used to evaluate the model’s performance on each iteration. PPV represents the proportion of these patients who would have been correctly deemed as ‘ready for discharge’ after having the second screening by a clinician. When computed for the top *x*% of ranked predictions, this metric can be regarded as an evaluation metric particularly well suited to assessing the efficacy of a decision-support tool in clinical practice. For example, if a model achieves a PPV of 0.8, this is equivalent to saying that, for every 10 patients that are prioritised to have a secondary screening by a clinician, 8 patients can subsequently have discharge arrangements proactively made for them, for their release within 24 hours. The performance of each model, developed for each of the datasets *D*_pt and *D*_et, *t* ∈ {0,…,7}, was evaluated. Moreover, additional analysis on the results of the models trained using emergency admissions *D*_et was carried out on the subcategory of these admissions where patients had been diagnosed with infection. This subcategory corresponds to 37 CCS categories. The results for this subcategory are hereafter denoted by *D\**_et. The mean PPV performance of the different models, calculated for the 20% of patients with the highest positive classification scores within each patient category considered (*D*_pt, *D*_et and *D\**_et) are presented across three separate subplots. The mean and standard deviation PPV results, as well as the corresponding NPVs, are presented in Tables and. # Discussion During outbreaks of disease such as seasonal influenza or the global COVID-19 pandemic, healthcare systems across the world have struggled to cope with an increased demand for hospital beds. This has resulted in situations where patients who required beds in hospital were unable to be admitted, forcing clinicians to make difficult decisions regarding which patients should receive care. Previous work has shown that introducing an ML prediction system can have statistically significant impact on improving overall patient flow. We therefore hypothesize that, with improved patient flow, hospitals would have a greater chance of coping with sudden surges in admissions during crisis periods. However, use of ML techniques to make operationally focused discharge prediction for a broad patient base is an under-researched area, particularly through the use of more advanced ML techniques. This retrospective study attempts to address these issues through the development of models which are able to reliably classify whether patients will be ready for discharge within the following 24 hours and rank them according to their probability of discharge readiness. The expectation is that these rankings would be used by hospital-bed managers to identify patients to prioritise for a secondary screening. Four different model architectures, LR, RF, SVM and DNN, were compared in their abilities to make this classification. Planned and emergency admissions within the dataset were studied separately, with custom models developed for each. The predictions were made for each day of a patient’s admission, from the first day of their arrival up to 7 days into their stay. It was found that the DNN models often outperformed the other models considered. Furthermore, we observed that models generally performed best in predicting the discharge of planned admissions, *D*_pt, rather than emergency admissions, *D*_et, and were better for predicting the discharge for emergency admissions as a whole, compared to the sub-cohort of emergency admissions with infection, *D\**_et. This is likely due to the higher variance in LOS which was present in emergency admissions, and even more so in emergency admissions with infection. A higher variance suggests that by the nature of their admission, these patient groups were less predictable and thus more difficult to classify correctly. Although there were differences in PPV between models developed for the different patient admission types, overall, the results were comparable. This indicates that, if implemented in a hospital setting, we could robustly predict 24 hour discharge readiness for all admission types, and could confidently predict discharge for patients recovering from infection using the models trained on general emergency admission data. This has clear implications during a pandemic. It is also worth reiterating that during periods of crises, the discharge of all patients across hospital is important, as prioritizing one planned-admission type patient for discharge would release a hospital bed for an incoming emergency-admission type patient with infection. It was seen that PPV is higher and more consistent in models trained and evaluated on datasets where *t* is lower, i.e., datasets for patients with shorter LOS, or earlier into the admission of patients with a longer LOS. This trend could be a combination of two factors. Firstly, a lack of training data for models with higher *t* is likely to impact performance, particularly for DNN models which generally require more training data than traditional ML models. Secondly, it is possible that it is simply harder to predict next day discharge for a patient who has already been in hospital a considerable length of time, who therefore represents a more complex case. It is also a possibility that the model hyperparameters, which were based on analysis of dataset *D*_e0 could be overfit to this dataset and not generalize as well to datasets with higher *t*. Nevertheless, if implemented in a hospital setting, it is likely that the performance of the DNN models would improve for datasets containing patients with longer stays as more data is collected. Lastly, it is worth noting that in general, higher mean NPV results were obtained, which could be interpreted as it being easier to predict when a patient was not ready to be discharged. This can also give us confidence that the models were not making suggestions for patients to be discharged too early, which would be unsafe. # Limitations Within the dataset used, there was no information indicating when a patient was medically ready for discharge, therefore the timestamp of the true discharge was used as proxy for this status. Patients who need to be relocated to a subsequent care facility at the end of their stay often have their discharge delayed due to factors out of the hospital’s control. Consequently, the time that they left the hospital is more likely to differ from the time that they were medically fit for discharge. Therefore, as stated earlier, we excluded these patients and restricted our study to only patients who were discharged under normal conditions, to their usual place of residence. A further limitation is that, although this research considered all patients from across a hospital, from different departments and wards, this research was limited to a single hospital. However, prior studies have shown electronic tools to be effective in improving patient flow in other hospital centres, suggesting that the research is generalizable. If implemented in a hospital setting, it would be advised that the hospital records when a patient is medically ready for discharge and that the models should be retrained with this information, and either with data from across multiple centres or with data from the specific hospital where it is intended to be deployed. Furthermore, if not implemented carefully, there is a potential risk that the use of these ML models in hospital could harden any bias in the discharge process. A suggested mitigation of this risk is for hospital bed managers to use the tool, rather than the clinicians directly, thus decoupling the discharge process from clinical prognoses and preventing clinicians from altering their behaviour in response to the models’ output. Finally, this study did not include data from any period associated with a pandemic. Due to the substantial changes within the healthcare system due to COVID-19, this period should be studied separately; research in this area is on- going. # Conclusion We have proposed an operationally focused ML classifier which is able to make predictions as to whether a patient will be ready to be discharged within the next 24 hours, for all patients in hospital at any given moment. This classifier is intended to be used during periods that result in a large influx of admissions to hospital, such as peaks in seasonal influenza cases. The intention is for the classifier to be implemented within a well-engineered decision- support tool and for it to be used by hospital bed managers to identify and prioritize patients for discharge. This would improve the efficiency of the safe release of hospital beds and therefore overall patient flow in the hospital. Generally high PPVs were achieved for the top 20% of patients ranked by the models, showing promise that ML systems could prove to be a valuable tool for improving patient flow in clinical settings. Furthermore, variables learned as being of predictive value can be incorporated in future studies which aim to predict real-time discharge or LOS, for individual patients. # Supporting information We thank all the people of Oxfordshire who contribute to the Infections in Oxfordshire Research Database. This work uses data provided by patients and collected by the NHS as part of their care and support. Research Database Team (Oxford): R Alstead, C Bunch, DW Crook, J Davies, J Finney, J Gearing (community), H Jones, L O’Connor, TEA Peto (PI), TP Quan, J Robinson (community), B Shine, AS Walker, D Waller, D Wyllie. Patient and Public Panel: G Blower, C Mancey, P McLoughlin, B Nichols. [^1]: DWE has received lecture fees and conference expenses from Gilead. No other author has a conflict of interest to declare. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Neonatal hypoxic-ischemic brain injury is a prominent cause of neurological disability in neonates. Since it is not possible to conduct controlled studies in children, it is necessary to perform experimental studies in suitable animal species to obtain information that is likely to be applicable to humans. In this respect, pigs have for long been used as an experimental model given that many of their anatomical and physiological characteristics closely resemble those of humans, more so than other non-primate species. The retina pertains to the central nervous system (CNS) and it is one of the most metabolically active tissues in the body. Its high-energy demand is due to the highly sensitive and efficient system that converts light energy into neuronal signals, the reason why the retina consumes oxygen more rapidly than other tissues. Thus, in times of increased energy demand, oxygen becomes one of the most limited metabolites in the retina. For this reason, the retina is susceptible to alterations in oxygen tension and specifically, the retina is sensitive to hypoxia, a condition defined as an inadequate supply of oxygen for an organism, tissue or cell. At the cellular level, functional studies suggest that retinal ganglion cells (RGCs), the neurons that relay visual signals to the brain, may be the most sensitive cells in the retina to experimental transient ischemia or systemic hypoxia. Indeed, a reduction in oxygen tension could be associated with the development of retinal pathologies, such as retinal vessel occlusion, proliferative diabetic retinopathy, retinopathy of prematurity, glaucoma, age-related macular degeneration or high altitude retinopathy. Death of RGCs is a hallmark of retinal diseases in which hypoxia and/or ischemia are assumed to play an etiological role. While the brain represents 2% of our body weight, it consumes 20% of the body’s oxygen demand. Moreover, the immature foetal and neonatal brains are particularly vulnerable to severe alterations in oxygen tension and they may develop neurovascular malformations when oxygen levels are low. However, in mammalian neonates certain physiological responses and adaptations exist to respond to a limited oxygen supply. The superior colliculus is a multilayered structure in the mammalian midbrain, and it is the structure in the brain where among inputs from retinal axons and the visual cortex converge. As hypoxia triggers apoptosis, it is often best to study this phenomenon by counting the number of recently activated apoptotic cells. Moreover, RGCs are sensitive to hypoxic conditions and the death of these cells provokes a gradual loss of vision that will ultimately lead to blindness. As such, we evaluated the apoptosis induced by hypoxia by examining the distribution of cells with active caspase-3 in the superior colliculus and retina, the activation of which is a marker of this form of cell death. Glial cells play crucial roles in regulating neuronal development and neuronal activity, and astrocytes in particular are susceptible to reductions in oxygen tension. Given their importance in linking vascular and neuronal function, astrocytes are fundamental in the induction of neuronal deficits, and hypoxia is known to induce astrocyte-dependent protection of neurons. They are the first cells exposed to the damage from hypoxic or ischemic insults. Moreover, hypoxia also affects the ability of astrocytes to sustain neuronal viability and it induces specific molecular responses in astrocytes. Indeed, the degeneration of astrocytes is associated with the functional failure of the blood retinal barrier in oxygen-related neuropathies. For these reasons, we assessed the morphological changes to the cytoskeleton of astrocytes in the retina and superior colliculus. It is important to note that within the retina, another macroglial cell type is present in addition to astrocytes that are not present in the brain, the Müller glia. These cells are specialized in maintaining the homeostatic and metabolic support of retinal neurons, as well as fulfilling other functions. In addition, there is some metabolic heterogeneity and distinct vulnerability to hypoxia within different tissues that could be responsible for producing a distinct response to hypoxia of the tissues of interest here. Given that hypoxia has a negative effect on neuronal metabolism and that it may be detrimental to the function of neurons, and since astrocytes have the capacity to sustain normal neuronal activity and to regulate the development of the vasculature, here we evaluated whether low oxygen conditions induce changes in neurons and astrocytes in the retina and superior colliculus. # Materials and methods ## Animal preparation This study was carried out in strict accordance with the recommendations for the Experimental Research Committee of the Cruces University Hospital, which is registered in the Official Register of Breeders, Suppliers and Users of animals for experimental and other scientific purposes in the Basque Country (Spain). All the experimental protocols met with the European (2010/63/UE) and Spanish (RD53/2013) guidelines for the protection of experimental animals and they were approved by the Ethics Committee for Animal Welfare of the Cruces University Hospital. On the same day as the experiment, animals were obtained from a local farm authorized by the Basque Country Regulatory Agency to supply animals for research. All animals were free of any disease and they were transported with a certificate of health. The protocol to produce hypoxia used in the present study has been described extensively elsewhere. Neonatal pigs (*Sus scrofa*, n = 8) aged 2–4 days old (1.7± 0.2 kg) were sedated with an intramuscular injection of ketamine (15 mg/kg) and diazepam (2 mg/kg). Before performing the surgical procedure, anaesthesia and analgesia were induced by intravenously administering fentanyl (5 μg/kg) and propofol (1.2 mg/kg), and this state was maintained by continuous intravenous infusion of fentanyl (titrated as necessary: 5–20 μg/kg/h), propofol (titrated as necessary: 2–3 mg/kg/h) and midazolam (titrated as necessary: 0.5–2 mg/Kg/h). In addition, the animals used as controls were paralysed by continuous intravenous infusion of vecuronium bromide (3 mg/kg/h). In all cases, an ear vein was catheterized to deliver the anaesthesia and analgesia. A tracheotomy was performed, and a tracheal tube (4.0 mm ID) was inserted and connected to a ventilator. Animals were then positive pressure ventilated and changing ventilator parameters were performed to maintain adequate blood gas values of arterial oxygen pressure (PaO₂) 80–110 mmHg, adequate arterial pressure of carbon dioxide (PaCO₂) 35–45 mmHg and a pH 7.30–7.45. An arterial catheter was inserted into the femoral artery to monitor the mean arterial blood pressure (MABP) and heart rate (HR), and to obtain arterial blood samples for blood gas analysis: PaO₂, PaCO₂, pH, Base Excess (EB), oxygen saturation and lactic acid (GEM Premier 4000, Instrumentation Laboratory). Animals were also monitored continuously by three- lead electrocardiogram during the experimental procedure. ## Hypoxia The hypoxia model is based on a swine model of neonatal asphyxia. Two experimental groups were established, control (n = 4) and hypoxic (n = 4) pigs. In the hypoxic group, the animals were stabilized for 30 minutes and hypoxia was then induced by decreasing oxygen levels to 12–14% for 120 minutes, followed by 240 minutes of normoxia. Hypoxia was induced by reducing the fraction of inspired oxygen (FiO₂, the fraction or percentage of oxygen in the space being measured) to 0.1–0.15 while increasing the concentration of inhaled nitrogen gas. Control pigs were not subjected to reduced oxygen concentrations. Following hypoxia, the FiO₂ levels were increased to 21–35% in order to maintain normoxia for 240 minutes. At the end of the experiments (4 hours after the onset of hypoxic injury), the animals were sacrificed with an intravenous overdose of potassium chloride (0.35 mg/kg). ## Tissue collection Neonates eyes were enucleated, and the cornea, crystalline lens and vitreous humour were removed. Each eyecup containing the retina was fixed overnight by immersion in 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (PB, pH 7.4) and after then washing in phosphate buffered saline (PBS, pH 7.4), the retina was carefully separated from the rest of the eye. In addition, the superior colliculus was extracted from the brain and immediately fixed overnight in 4% PFA. Control and hypoxic tissues were cryoprotected for 24 hours at 4°C in 30% sucrose diluted in 0.1 M PB, and the tissue was then embedded in OCT medium to obtain cryosections (14 μm thick) that were stored at −20°C. ## Immunohistochemistry Whole mount retinas were immunostained as described previously, with some minor modifications. The retinas were washed with PBS and non-specific antibody binding was blocked by incubating them overnight at 4°C in a solution of PBS- TX-100-BSA (0.25% Triton-X 100 and 1% bovine serum albumin in PBS), with shaking. The retinas were then incubated (for one day with shaking at 4°C) with the primary antibodies diluted 1:1,000 in PBS-TX-100-BSA: an anti-Brn-3a goat polyclonal antiserum (Santa Cruz Biotechnology, Santa Cruz, USA) to detect RGC nuclei; and an anti-GFAP mouse monoclonal antibody (Sigma, Steinheim, Germany) to detect the cytoskeleton of astrocytes. Subsequently, the retinas were washed thrice in PBS for 15 minutes and antibody binding was detected (5 hours at room temperature with shaking) with secondary antibodies diluted 1:1,000 in PBS-BSA (1%): an Alexa Fluor 568 conjugated donkey anti-goat antibody (Invitrogen, Eugene, Oregon, USA) and an Alexa Fluor 488 conjugated rabbit anti-mouse antibody (Invitrogen, Eugene, Oregon, USA). Finally, the retinas were washed 3 times for 10 minutes in PBS, flat mounted onto slides in PBS-Glycerol (1:1) and coverslipped. Cryostat sections of the retina and superior colliculus were immunostained as described previously. After washing the sections twice in PBS-TX-100 for 10 minutes, they were incubated overnight with the primary rabbit anti-active (cleavage) caspase-3 (Asp175) antibody (1:1,600, Cell Signaling Technology \#9661, Danvers, Massachusetts, USA) and an anti-GFAP mouse monoclonal antibody (1:1,000, Sigma, Steinheim, Germany). After washing twice in PBS, antibody binding was detected for 1 hour with an Alexa Fluor 555 conjugated goat anti- rabbit antibody (1:1,000, Invitrogen, Eugene, Oregon, USA) and an Alexa Fluor 488 conjugated goat anti-mouse antibody (1:1,000, Invitrogen, Eugene, Oregon, USA) diluted in PBS-BSA (1%). The sections were washed twice with PBS for 10 minutes and mounted with a coverslip in PBS-Glycerol (1:1). ## RGC quantification To analyse the RGCs and astrocytes in the retina, confocal microscopy images of whole mount retinas were obtained at a resolution of 2048 x 2048 pixel with a 20X objective (Olympus FV500, Olympus, Tokyo, Japan). A Z-stack was obtained of five images with a 4 μm separation and a total of 8 whole mount retinas were studied (4 control and 4 hypoxic). Four different areas of the retina were analysed, the peripheral and central areas of the dorsal and ventral retina. Thus, a total of 16 images from each retina were captured (four pictures of each analysed area). Since the porcine optic nerve is located ventrally, the dorsal area is larger than the ventral area, we selected the peripheral area 15 mm from the optic nerve in the dorsal domain and 6 mm from the optic nerve in the ventral domain. The central area refers to tissue 2 mm from the optic nerve in both the dorsal and ventral directions. These criteria were applied to the 8 retinas studied and thus, the total retinal area analysed was 8 mm² per retina (2 mm² per area). The method to quantify RGCs was slightly modified from that used previously, employing AxioVision 4.7.2.0.Software. The number of Brn3a labelled RGC nuclei in each retinal area of the experimental and control eyes was counted, and compared in the dorsal and ventral peripheral and central retina. ## Quantification of active caspase-3 positive cells The cells with active caspase-3 were quantified in the retina and superior colliculus using a fluorescent microscope (Zeiss, Jena, Germany) and the Zeiss Zen software, analysing 5 sections from the same central and peripheral areas of each retina, and 20 sections from the same rostro-caudal level of each superior colliculus from 4 control and 4 hypoxic animals. Five images from each section were acquired at 20X on a Zeiss Axiocam MRM (Zeiss, Jena, Germany) and the active caspase-3 cells were counted double blind by two experimented researchers, as described previously. In the retina, the active caspase-3 positive cells in the ganglion cell layer (GCL) were counted and the linear distance of each micrograph was measured (Zen, Zeiss, Jena, Germany). Thus, the number of labelled cells were expressed per linear millimetre of the RGC layer (cells/mm). In the superior colliculus, the number of active caspase-3 cells was counted in the superficial layers and in the total surface of the superior colliculus, including the superficial layers, calculating the average number of active caspase-3 positive cells per mm² of the superior colliculus (cells/mm²). ## Astrocyte morphometry To analyse the cytoskeletal morphology of astrocytes in the retina, images were taken of whole mount retinas using the same microscope and criteria as those used for RGC quantification. The astrocytes in the pig retina are organized mainly like those in the human retina, running parallel to the RGC axons. A semi-automatic method to measure astrocyte organization was used to measure the morphological changes in astrocytes. The aim was to quantify the differences in the astrocyte network between the control and hypoxic retinas based on the local orientation of the astrocytic processes. An *ad hoc* programme was developed in Matlab R2010b (MathWorks, Inc) to quantify the morphological changes to astrocytes. This programme estimates the local orientation of the GFAP-positive lines in an image, taking advantage of the SURF extraction algorithm and its capacity to define local pixel orientation based on their neighbouring relationships. Following local characterization, a histogram of this orientation is extracted that allows the frequency of each direction to be measured (1-degree bins). This involves measuring the randomness of the histogram using Shannon’s Entropy equation, a well-known function to measure the predictability of a variable. The higher the entropy the less predictable and consequently, the more disorganized the system. With this information, the degree of disorganization of control astrocytes and hypoxic astrocytes can be compared. The average entropy from the histogram of the images was calculated and a statistical analysis was performed to evaluate our hypothesis that the entropy is greater in hypoxic retinal astrocytes. To analyse the astrocytes in the superior colliculus, a Zeiss Axiocam MRM fluorescent microscope (Zeiss, Jena, Germany) and the Zeiss Zen software was used, obtaining 20 images (20X) at the same rostro-caudal level of the superior colliculus of the 4 control and 4 hypoxic animals. Since the astrocyte organization in the superior colliculus does not follow a well-established parallel pattern, the method used to analyse the astrocytes in this structure differed to that used in the retina. The morphology of the astrocyte cytoskeleton was analyzed using ImageJ (version 1.49), the image processing program developed at the National Institutes of Health (NIH). Using the “threshold color” tool, the region formed by colored pixels (labeled with the antibody against GFAP) was selected, and this area of the cytoskeleton of selected astrocytes was measured. Given the entire image area, we could calculate the proportion of the area occupied by astrocyte´s cytoskeleton in the control and hypoxic superior colliculus. ## Statistical analysis RGC density, the number of active caspase-3 positive cells and the morphological data from the astrocyte’s cytoskeleton were described as the mean and standard error of mean, and these parameters were compared between control and hypoxic tissues. Statistical analyses were carried out using IBM SPSS Statistics software v. 21.0 and the homogeneity of the variances was assayed with Levene´s test (p \< 0.05). To assess whether there were significant differences in the number of RGCs, the number of active caspase-3 positive cells and in astrocyte morphology between control and hypoxia conditions, a Student T-test was used. In addition, a Mann-Witney U test was used to verify the differences between the control and hypoxic groups. For both tests, the minimum value of significance was defined as p\<0.05. # Results ## Gas exchange At baseline, all animals displayed adequate gas exchange, whereby the fraction of inspired oxygen (FiO₂) was 30 ±2%, the PaO₂ was 108 ±16 mmHg, PaCO₂ was 42 ±6 mmHg, pH was 7.34 ±0.08, EB (base excess) was -2 ±3 mmol/l and lactic acid was 2 ±1 mmol/l. In the hypoxic animals, the decrease in oxygen levels produced a FiO₂ of 12 ±2%, and a significant impairment of gas exchange, with a PaO₂ of 31 ±8 mmHg, PaCO₂ of 45 ±8 mmHg, pH of 6.84 ±0.05, EB of -24 ±2mmol/l and lactic acid of 17 ±2mmol/l. Moreover, the MABP (mean arterial blood pressure) decreased significantly after hypoxia (32 ±5 vs. 78 ±12 mmHg), while the HR (heart rate) remained unchanged relative to the basal values (184 ±36 vs. 196 ±33 bpm). These parameters partially reverted during the re-oxygenation period to a FiO₂ of 33 ±1%, a PaO₂ of 92 ±17 mmHg, a PaCO₂ of 45 ±5 mmHg, a pH of 7.22 ±0.12, an EB of -8 ±6 mmol/l and lactic acid of 4 ±4 mmol/l. However, there was no change in either the MABP (36 ±9 mmHg) or HR (193 ±26 bpm). In the control animals that were not subjected to hypoxia, these parameters remained at the basal values throughout the experiment. Finally, the hemoglobin levels (7.0 ±0.7 g/dl) remained constant throughout the experiment and all pigs remained alive throughout. ## RGC density The number of RGCs in the control and hypoxic retinas was analysed in the four selected areas (dorsal periphery, dorsal centre, ventral periphery, ventral centre:). There were no significant differences in RGC number in the retinas from control and hypoxic animals in any of the four areas. While there was a mild tendency towards a loss of RGCs in the dorsal periphery of the hypoxic retinas, although this did not reach significance. ## Caspase-3 activation Numerous active caspase-3 positive cells were found in the inner nuclear layer (INL) and ganglion cell layer (GCL) of the retina, both in control and hypoxic animals. Indeed, hypoxia did not produce a significant change in the number of positive active caspase-3 cells in the GCL, with 34 (±3) active caspase-3 positive cells/mm in control retinas and 38 (±4) cells/mm in hypoxic retinas. By contrast there was a 37.6% increase in the number of active caspase-3 positive cells after hypoxia in the superior colliculus, rising from 279 (±13) cells/mm² in the control animals to 384 (±30) cells/mm² after hypoxia (p = 0.013). This increase was even stronger if only the superficial layers of the superior colliculus were taken into account, where the number of apoptotic cells increased by 56.93% with respect to the controls: 610 (±7) active caspase-3 positive cells/mm² in the controls as opposed to 957 (±2) cells/mm² in the hypoxic animals, (p = 0.002). ## Astrocyte morphology in the retina When we analysed the cytoskeletal morphology of astrocytes in the control and hypoxic neonatal pig retina, retinal astrocytes appeared to be more disorganised following hypoxia, with an increase in the lateral extension of processes. Using the algorithm described above, we quantified the disorganisation of the astrocyte networks in the retina, which was translated into different histograms of the orientations depending on the characteristics of the input. As such, some hypoxic retinas displayed a higher degree of randomness in the histograms, whereas control retinas seemed to have a greater difference in the frequency between the main orientation angles and other angles (: parallel axon processes in red and non-parallel ones in green). The *ad hoc* routine to quantify the differences in the retinal astrocyte networks in function of the local orientation of their processes highlighted the differences in the distribution of retinal astrocyte processes. However, no significant differences between control and hypoxic retinas were found in the area analysed. While these results indicate there was no significant difference between these populations (p \>95%), there does appear to be a difference in the degree of organisation (a difference of approximately 93%, p = 0.069). ## Astrocyte morphology in the superior colliculus Activation and hypertrophy of astrocytes was evident in the hypoxic superior colliculus, implying the emergence of gliosis. Thus, we cannot rule out that hypoxia induces changes in the number of astrocytes in the superior colliculus. Moreover, the disposition of astrocytes in the superior colliculus meant that we could not use the same method as that employed in the retina to measure the changes in the patterning of their processes. When the area occupied by the astrocyte cytoskeleton was quantified, the mean proportion of the area occupied by the astrocyte’s cytoskeleton in the control superior colliculus was 4.07% (± 0.50), while in the hypoxic superior colliculus it was 17.75% (± 2.41). Moreover, hypoxia induced an increase in astrocyte density of 13.78%, representing a 4.36-fold increment (p = 0.0037). # Discussion Experimental studies in suitable animal models commonly provide insights into human neonatal situations. In the present study we have used the pig as an animal model because beside other primates, the porcine retina is the most similar to the human retina among the large mammals. Neonatal pigs have been used previously to study brain alterations and to test different drugs, and here we have used hypoxic conditions followed by a short recovery to detect early damage, and to compare the effects of hypoxia on the retina and brain. We found that hypoxia produces significant neuronal apoptosis in the superior colliculus but not in the retina. Moreover, we found significant gliosis of the astrocytes in the superior colliculus but not in the retina. However, in the retina some changes in RGC density in the peripheral retina were found, although they were not significant. The changes described probably reflect the earliest events that result from the hypoxic insult and we cannot rule out that longer recovery times will produce more significant damage in the retina. Limiting oxygen in the retina contributes to RGC neurodegeneration, which results in a loss of vision and ultimately, blindness. Cell viability can be dramatically compromised by hypoxic stress and we found that the dorsal- periphery of the retina was the area where the RGCs are more vulnerable to hypoxia, even though significant differences were not found. This is consistent with the early events that take place in glaucoma, a neurodegenerative disease caused by elevated intraocular pressure (IOP) that possible causes mild hypoxia and where peripheral RGCs are more sensitive to damage, in accordance with others studies in pigs, rats and mice. Although several mechanisms of cell death are activated by oxygen depletion in neural tissues, changes in apoptotic RGCs were not found and the basal number of active caspase-3 positive cells in the ganglion cell layer was similar in control and hypoxic retinas. By contrast, the proportion of active caspase-3 positive cells after hypoxia increases in the superior colliculus as a whole, and this increases further in the superficial layers of the superior colliculus with respect to the control. Cell death has been described as a natural event in the superior colliculus during development and postnatally. Moreover, hypoxic insult in the developing brain triggers the same apoptotic pathway as that activated during development. The induction of apoptosis following hypoxia is evident just a few hours after insult in brain areas like the cortex and hippocampus of neonatal pigs. Furthermore, the intensity of cell death in the superficial layers of the superior colliculus could reflect the elevated cell density in these layers relative to the rest of the superior colliculus. Thus, in hypoxic conditions, the cells will have more limited access to oxygen in these superficial layers, which could induce them to more readily undergo apoptosis. Hypoxia induces astrocyte activation in many neurological disorders. Reactive astrocytes divide and become hypertrophic, producing long, thick processes, as well as overexpressing GFAP. In the superior colliculus, signs of astrocyte activation and reactive gliosis were noted, such as an increase in the surface area occupied by GFAP, although no changes in the orientation of the astrocyte’s prolongations were found in the retinas. The distinct susceptibility to hypoxia between the retina and the superior colliculus could be explained by the heterogeneity in metabolic and molecular regulation between different areas of the CNS that produce a distinct vulnerability to oxygen depletion, as well as a different degree of neuroprotection. The more severe vulnerability of brain neurons to limited oxygen may be due to the brain tissue failing to up-regulate glycolysis sufficiently in order to compensate for the loss of respiratory ATP. By contrast, the retina has the capacity to metabolize the glucose converted to lactate from glycolysis more efficiently thanks to the Müller glia, only present in the retina. In addition, the stronger resistance to hypoxia in the retina could be due also to the presence of Müller glia cells, which are not present in the brain. Although they share many characteristics with astrocytes, Müller glia have specific functions related to the homeostatic and metabolic support of retinal neurons. Müller cells resist hypoxia and low glucose conditions by activating anaerobic glycolysis and by oxidizing alternative substrates in order to obtain energy in the form of ATP, as well as providing neurons with lactate. Furthermore, surfactant protein A (SP-A) is found in Müller cells, RGCs and astrocytes in the retina, and it is up-regulated during hypoxia. Since SP-A influences neovascularisation, its expression may represent a protective response against systemic inflammation and it may participate in the maintenance of the blood retinal homeostatic barrier. Finally, in transient ischemia we described an increase in brain derived neurotrophic factor (BDNF) in RGCs after ischaemic insult, as well as changes in the neurotrophic p75 receptor in Müller cells. These changes may reflect the neuroprotection in the retina that makes it more resistant to hypoxia in neonates. The lack of morphological changes in the retina after hypoxia only reflect early events, suggesting that the brain is more vulnerable to hypoxia at these early time points. However, after longer reperfusion times, retinal cells might also be sensitive to such insult. Moreover, given the limited sample size in this study some changes could remain unnoticed, with only major differences becoming evident, such as those in the superior colliculus that appears to be more sensitive to hypoxia. Consequently, a therapeutic window appears to exists in which a neuroprotective action protects the retinal neurons even though the brain has suffered damage. The results of the present study highlight the differences in the reaction of neurons and astrocytes in different parts of the CNS. As heterogeneous responses to hypoxia were detected in different brain regions, interesting opportunities may be open to design therapeutic and preventative treatments specific to certain areas and structures in the nervous system. # Supporting information [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** NR CR VM EV. **Data curation:** NR XP. **Formal analysis:** NR XP EV. **Funding acquisition:** EV. **Investigation:** NR CR VM XP EV. **Methodology:** NR CR VM. **Project administration:** EV. **Resources:** CR XP EV. **Software:** NR. **Supervision:** EV CR. **Validation:** EV. **Visualization:** NR XP. **Writing – original draft:** NR. **Writing – review & editing:** NR CR VM XP EV.

# Introduction It is well established that the hypothalamic-pituitary-adrenal (HPA) axis or stress system is particularly sensitive to programming by early life events. Dysregulation of immune/inflammatory responses may play a central role in mediating early programming effects. Studies in the field of early programming have focused primarily on heightened stress reactivity later in life, often measured in terms of HPA responses to acute stressors. In contrast little is known about effects of programming on endogenous stress levels long after a period of early stress has terminated. Infants born very preterm (≤32 weeks gestation) are exposed to considerable procedural pain-related stress during weeks to months of life-saving procedures during hospitalization in the neonatal intensive care unit (NICU), that appears to impact the HPA axis long after NICU discharge. Importantly, the neuroendocrine system, the immune system and the central nervous system are linked in a complex regulatory network, with extensive bidirectional communication between and among these systems. Glucocorticoid hormones can regulate expression of immunologically related genes, and conversely, dysregulation of immune/inflammatory responses may play a central role in mediating early programming effects of the HPA axis. The potential role of stress hormones in mediating immune/inflammatory function and in turn, the role of immune function/inflammation in mediating changes in the HPA axis in children born very preterm have not been addressed to our knowledge. Cortisol levels are known to be relatively low while preterm infants are in the NICU. Long after NICU discharge, the trajectory of cortisol activity over time appears to be altered. In a longitudinal cohort, we found that greater cumulative neonatal pain-related stress (higher number of skin-breaking procedures from birth to term adjusted for neonatal medical confounders) was associated with altered baseline cortisol levels at 8 and 18 months corrected age compared to infants born full-term term. Importantly, endogenous cortisol levels play an important role in brain function – and in physiology and metabolism. Thus alterations in HPA activity and responsiveness at school age in children born preterm have important functional implications. Stress and adversity in early life can program a phenotype of exaggerated adrenocortical and inflammatory responses to challenge. Environmental stress can enhance or suppress aspects of the immune response by altering activity of the cellular signaling pathways that regulate inflammation, including the proinflammatory transcription factor NF-κB. Long-term alterations in HPA responsiveness that result from early life stress and adversity may also play a role in modulating immune function. Moreover, early life stress, including pain, shows sex-specific effects in rodents, however the direction varies depending on numerous factors such as age and type of exposure; therefore, we examined gender differences. Several studies have reported associations between maternal depression and altered child basal cortisol levels, therefore we examined maternal depression and anxiety as potential social confounders. In the present study we first determined whether the increased cortisol levels observed in children at 8 and 18 months persist to school age in children born very preterm. Then to address the etiology of HPA alterations in these children, we examined whether neonatal inflammation may be involved in programming of the HPA axis by early life stress. Specifically, we investigated whether common genetic variants in the promoter region of the *NFKBIA* gene modulate the long- term associations of neonatal procedural pain-related stress with HPA axis programming in children born very preterm. The *NFKBIA* gene encodes IκBα, a critical negative regulator of the transcription factor NF-κB. NF-κB regulates the expression of the majority of proinflammatory cytokines, chemokines and leukocyte adhesion molecules, as well as pro-survival and anti-apoptosis genes; moreover, dysregulation of NF-κB is a known consequence of early life stress. Specifically, we assessed functional genetic variants in the promoter of *NFKBIA.* It has been established that individuals carrying the minor allelic variants at rs3138053 and rs2233409 have lower expression of both the *NFKBIA* gene expression and IκBα protein and this decrease in negative regulation is associated with higher Toll-like receptor-mediated inflammatory responses. Importantly, in the present study, we used hair cortisol as an integrated measure of chronic stress that provides a well-validated index of endogenous HPA axis activity over time – in contrast to salivary or serum cortisol that provides an acute measure at a single point in time, varies with time of day, and reflects the more immediate stress context. Furthermore, in a subset of children born preterm that consented to provide a blood sample at age 7 years, we compared the levels of NF-κB-driven secretion of interleukin 6 (IL-6) and tumor-necrosis-factor alpha (TNFα) between subjects who varied in the promoter regions of the *NFKBIA* gene. Cumulative neonatal pain-related stress was operationalized as the sum of all skin-breaking procedures, adjusted for medical confounders. The magnitude of response to a procedure in preterm neonates does not directly reflect the degree of invasiveness. Rather, sensitization to what has occurred in the past hour, past 24 hours, and cumulatively since birth, underlies reactivity. Sensitization is mediated at the spinal cord level, and is a phenomenon of the immature central nervous system that is excitable to incoming stimuli, inducing a wind-up phenomenon that affects reactivity to subsequent stimuli. Therefore we use the number of skin-breaking procedures from birth to hospital discharge as an index of “pain-related stress” and did not adjust for extent of each procedure. Our approach reflected the theoretical framework of long-term effects of major psychological childhood stress proposed by Miller, Chen and Parker , and aimed to extend their model to long term effects of physical stress in very immature preterm neonates. We hypothesized that at age 7 years: 1) hair cortisol levels will differ in children born very preterm compared to full-term, and differ by gender; 2) among the very preterm children, procedural pain-related stress in the NICU, quantified as number of skin-breaking procedures from birth to term equivalent (adjusted for medical and social confounders), will be associated with altered hair cortisol levels at age 7 years; and 3) genetic variation that impacts the regulation of NF-κB will interact with the degree of pain-related stress in the NICU and be reflected in hair cortisol levels at age 7 years. To our knowledge, this is the first study in very preterm children to address the etiology of HPA activity in relation to a transcription factor critical in stress and immune function, and the first study in this population of a cumulative index of stress long after hospital discharge. # Materials and Methods ## Ethics Statement The study was approved by the Clinical Research Ethics Board of the University of British Columbia and the Research Ethics Board of the Children’s & Women’s Health Centre of BC, and conforms to the conventions set out in the Declaration of Helsinki. ## Participants A total of 133 school age children, 91 born preterm (PT) at ≤32 wk gestational age (GA) (M = 42, F = 49; mean gestation 29.6 wk; SD = 2.4; mean age at testing 7.7 yrs; SD = 0.3) and 42 full-term (FT) healthy controls (M = 15, F = 27; mean gestation 39.9 wk; SD = 1.0; mean age 7.8 yr; SD = 0.8) were included in the present study as part of a longitudinal study on the long-term effects of pain- related stress in children born very preterm. Excluded were children with major sensory, motor, or cognitive impairment (Wechsler Intelligence Scale for Children - Fourth Edition (WISC-IV, full scale IQ \<70). In addition, children currently on glucocorticoids or other medications that affect cortisol (e.g. to treat asthma or attention deficit hyperactivity disorder) were not included (11preterm, 4 full-term). A flow chart of recruitment is shown in. Written informed parent consent and child assent was obtained. Hair samples were collected and psychometric assessment conducted on all the children who attended the study visit at age 7 years. We invited a subset of these children to undergo magnetic resonance imaging (MRI) at the B.C. Children’s Hospital, on a subsequent occasion. In order to measure cytokine levels, from the 91 preterm children who participated in this study, 54 (M = 23, F = 31) provided blood samples for cytokines assay, in conjunction with the visit to acquire MRI. ## Neonatal Characteristics Detailed systematic nursing and medical chart review was carried out by an experienced research nurse from birth to term equivalent age as described previously. Variables included, but were not limited to gestational age, birth weight, early illness severity (SNAP-II) on day 1, number of skin-breaking procedures (e.g. heel lance, intramuscular injection, intravenous or central line insertion), days of mechanical ventilation, postnatal infection confirmed on clinical laboratory testing, number of surgeries, and daily dose of morphine adjusted for daily weight. Cumulative neonatal pain-related stress was operationalized as the sum of all skin-breaking procedures, adjusted for medical confounders. Cumulative exposure to morphine was calculated as the average daily dose (i.e. intravenous dose plus intravenous-equivalent oral dose) adjusted for daily body weight, multiplied by the number of days of morphine administration. A summary of the neonatal characteristics of the preterm children is provided in. ## Hair Cortisol Cortisol was assayed from hair as an integrated measure of HPA activity in the prior two months. Hair samples were collected from the vertex posterior of the head, as this area has been shown to have the lowest coefficient of variation in hair cortisol concentrations. A cluster of hair strands (∼2–3 mm diameter) were cut at the base of the hair shaft from five small spots. The hair samples were secured on a cardboard card with tape, labeled, and stored in individual sealed plastic bags at room temperature until analysis. The most proximal 2 cm from the scalp of each hair sample was utilized for the assay. The hair sample was weighed, and 10–15 mg from each sample was placed in scintillation vials. To remove external contaminants, hair samples were washed twice by immersing each sample in 3 mL of isopropanol and incubating it at room temperature in a centrifuge at 100 RPM for 3 minutes. After decanting the isopropanol, samples were allowed to dry for at least 12 hours. After drying, 1 mL of methanol was added to each scintillation vial. Hair was then finely minced with surgical scissors, cleaning the scissors between samples. The vials were sealed with paraffin film and incubated for 16 hours at 50°C in a centrifuge at 100 RPM. Following methanol extraction each cortisol-containing methanol solution was transferred into 5 mL test tubes and evaporated on a test tube hot plate under a steady stream of nitrogen gas. The residue was then reconstituted with 250 µL of phosphate buffered saline. These reconstituted samples were analyzed using the salivary enzyme linked immunoassay kit (Alpco Diagnostics, Salem, NH, USA). The intra-assay and inter-day coefficients of variation were 8.9% and 5.1%, respectively. The kit reported a sensitivity of 1.0 ng/mL. ## Parent Questionnaires Beck Inventory –2^nd Edition (BDI-II) : is a 21-item self-report questionnaire widely used to assess the presence and severity of symptoms of depression in adults in both research and clinical settings. The BDI-II has an alpha coefficient of 0.80. State-Trait Anxiety Inventory (STAI Form Y-2; : Trait (T-Anxiety) is a 20-statement self-report scale widely used to assess how the respondent generally feels. Trait anxiety, but not state anxiety was used, since hair cortisol is a cumulative index of longer-term stress. ## Genotyping DNA was collected from each of the 91 preterm children and extracted from saliva using Oragene OG-500 and prepIT-L2P collection kit (DNA Genotek/OraSure Technologies Inc., Bethlehem, Pennsylvania). The *NFKBIA* (rs3138053, rs2233409) single nucleotide polymorphisms (SNPs) were genotyped using commercially available TaqMan SNP Genotyping Assays (C_73866_10, C_15945891_10) on the Applied Biosystems 7300 Real Time PCR System (Applied Biosystems, Carlsbad, California). SNPs were deemed acceptable for analysis if they had call rates \>99% and frequencies did not deviate from Hardy-Weinberg Equilibrium (HWE) (p-value \>0.05). Standard TaqMan protocols were followed as recommended by the manufacturer. ## Cytokine Assay Samples were collected in 4 ml 75 USP unit sodium heparin tubes (BD Vacutainer®, Mississauga, ON), incubated at 36°C for 18–24 hrs with 0.01, 0.1, and 1 µg/mL of lipopolysaccharide (LPS) (Ultra pure E. coli LPS 0111:B4, InvivoGen, San Diego CA). The LPS-induced release levels of interleukin-6 (IL-6) and tumor necrosis factor alpha (TNFα) were then assessed on 60 µL of whole blood and 140 µL of RPMI media with LPS (total of 200 µL/well) using Enzyme-Linked Immuno-Sorbent Assay (ELISA) kits (Human IL-6 ELISA Ready-SET-go!®, and Human TNF alpha ELISA Ready-SET-go!®, eBioscience, San Diego CA). ELISA was performed according to the manufacturer’s manual. ELISA plates were read using SpectraMax® Plus384Absorbance Microplate Reader (Molecular Devices, Sunnyvale CA), and quantified by the SoftMax Pro software provided by Molecular Devices. ## Statistical Analysis In preliminary analyses, t-tests were used to compare characteristics between the preterm and full-term groups. The *NFKBIA* rs3138053 and rs2233409 polymorphisms in our sample did not deviate from Hardy-Weinberg equilibrium (p = 0.55 and 0.49 respectively). Pearson chi-square test was then used to compare the gene allelic distribution between preterm girls and boys. To address the central study questions, univariate ANOVA was carried out on the hair cortisol level, with group and gender as between-subjects factors, with concurrent self-reported mother anxiety and depressive symptoms as covariates in ANCOVA. Generalized Linear Modeling (GZLM) was used, for the preterm children only, to examine cumulative neonatal pain adjusted for multiple clinical confounders, in relation to hair cortisol levels. R was used for GZLM, using F statistics and the car package for type III tests. Confidence intervals were computed using R. To establish adequate power the confidence intervals should not include zero. SPSS17 was used for the other statistical analyses. t-tests were used to compare area under the curve (AUC) of LPS-induced release levels of IL-6 and TNFα between CC and CT/TT variants of *NFKBIA* rs2233409 and AA and AG/GG of *NFKBIA* rs3138053 polymorphisms. # Results ## Participant Characteristics Characteristics of the preterm and full-term groups are provided in. As expected, the preterm group had significantly lower gestational age and weight at birth. The proportion of boys and girls in each group and child age at the 7 year visit did not differ between the groups. Mothers of children born preterm self-reported more symptoms of depression on the Beck questionnaire and fewer years of education. However both groups had relatively high educational background (preterms mean 15.8 years, and controls 18.3 years). Neonatal characteristics and genotype distribution between preterm girls and boys are shown in and respectively. There were no statistically significant differences in neonatal characteristics between boys and girls among the preterm children. ## Hair Cortisol, Group and Gender Hair cortisol values were log transformed. Outliers \>3 SD from the mean were winsorized such that the highest value within 3 SD was assigned. On ANOVA, hair cortisol levels were lower in the preterm compared to the full-term children (p = 0.018), and lower in girls than in boys (p = 0.007). Subsequently, ANCOVA was used to control for concurrent maternal factors (self-reported depressive symptoms, trait anxiety, and years of education), and the main effects for group and gender remained the same (p = 0.006 and 0.007 respectively). There was a significance effect for maternal depressive symptoms (p = 0.041) but not maternal trait anxiety or number of years of education. The estimated marginal means for each group by gender are shown in. ## Neonatal Pain, Morphine Exposure and Hair Cortisol in Preterm Boys and Girls Since hair cortisol level differed by gender, we examined the relationship between neonatal factors and hair cortisol level separately for boys and girls. A GZLM model was constructed with number of skin-breaking procedures (cumulative pain from birth to term equivalent), number of surgeries, cumulative morphine (daily intravenous and oral doses calculated for equivalence, adjusted for daily weight), early illness severity (SNAP II on day 1), and days of mechanical ventilation as covariates, and postnatal infection as a factor, to predict hair cortisol level at age 7 years. In preterm boys, greater cumulative neonatal pain was associated with lower hair cortisol (B = −1.11, *p* = 0.037), independent of morphine exposure, early illness severity, days on mechanical ventilation, gestational age at birth, and postnatal infection, none of which were statistically significant. The GZLM parameters are shown in. There was no significant relationship between any of the neonatal predictors and hair cortisol level for girls (B = 0.085, *p* = 0.72). The association between cumulative neonatal pain and hair cortisol (adjusted for confounders) is shown for boys and girls in. ## Neonatal Pain, Hair Cortisol and *NFKBIA* Genotype in Preterm Boys and Girls The allelic distribution for *NFKBIA* promoter SNPs rs2233409 and rs3138053 is provided in. For statistical analysis, children who were heterozygous and homozygous for the minor allele (CT or TT) were grouped together, and compared with children homozygous for the major allele (CC). The GZLM model was run again adding the *NFKBIA* polymorphism as a main factor in predicting hair cortisol level, and an interaction term (*NFKBIA* polymorphism X cumulative neonatal pain), in separate analyses for rs2233409 and rs3138053. In preterm boys, there was a significant relationship between hair cortisol and the interaction term of *NFKBIA* rs2233409 genotype and cumulative pain (B = −1.15, *p* = 0.03). For boys with a copy of the minor allele (CT or TT), lower hair cortisol was associated with greater neonatal procedural pain-related stress independent of morphine exposure, early illness severity, days on mechanical ventilation, gestational age, and postnatal infection. The GZLM parameters are shown in. For girls, there was no significant relationship between hair cortisol and the interaction term of *NFKBIA* rs2233409 genotype and cumulative pain (B = −1.12, *p* = 0.66). The interactions between the genetic variants and cumulative pain in relation to hair cortisol are shown in. The GZLM was re-run for the *NFKBIA* rs3138053 genotype, with no statistically significant relationships for boys or girls. We then re-ran the models adding the concurrent stressors (maternal depressive and anxiety symptoms). Results remained the same for the interaction term *NFKBIA* rs2233409 genotype and cumulative pain for boys, *p* = 0.046; there was no statistically significant effect for the concurrent stressors for boys or girls. As before, there were no significant relationships with the *NFKBIA* rs3138053 genotype. ## *NFKBIA* Regulated Cytokine Secretion in Preterm Boys and Girls In preterm boys only, LPS-induced IL-6 and TNFα secretion from peripheral blood mononuclear cells was significantly higher in children with the CT/TT variant of *NFKBIA* rs2233409 compared to children with the CC variant. Likewise, children with the AG/GG variant of *NFKBIA* rs3138053 had significantly higher LPS- induced IL-6 and TNFα releasing levels compared to the AA variant. # Discussion This study provides the first evidence, to our knowledge, that the HPA programming effects of procedural pain-induced stress in the neonatal period in children born very preterm persist to school-age. Furthermore, we utilized normal variation in the *NFKBIA* gene to examine whether genetic modulation of the inflammatory phenotype influences the effects of early stress. We found that for both boys and girls, hair cortisol levels as an integrated index of chronic stress, were lower in children born very preterm compared to full-term. However, the relationship between neonatal pain-related stress (adjusted for multiple confounders associated with prematurity) and hair cortisol was affected by presence of the minor allele for *NFKBIA* rs2233409 in boys, but not girls. These findings are the first to reveal that long-term effects of early pain- related stress exposure on the HPA axis are associated with normal genetic variation affecting *NFKBIA*/IκBα–the major negative regulator of NF-κB activity. The present study extends our earlier work showing an altered trajectory of baseline salivary cortisol expression in very preterm-born infants from the same longitudinal cohort. These previous studies demonstrated upregulation of resting cortisol levels at 8 and 18 months corrected age in the the subset of preterm infants born at extremely low gestational age (ELGA, 23–28 weeks gestation). In the present study, we now found downregulation of hair cortisol overall in children born very preterm (both ELGA, 23–28 weeks and very low gestational age, VLGA, 28–32 weeks), in the same cohort at school age, suggesting persistent early programming. Cortisol levels in humans exposed to early adverse conditions can later show upregulation or downregulation depending on multiple factors including time since the stress occurred and the nature of the stress or trauma (. In particular, post-traumatic stress disorder is associated with a pattern of higher cortisol closer to the period of substantial stress, then lower cortisol later. Interestingly, *fetal* stress has been associated with low cortisol in adulthood, but only in a subset of individuals born low birthweight and post- mature (i.e. after 40 weeks gestation), whereas other subgroups of low birthweight adults displayed the opposite association between fetal stress and later cortisol concentrations. Whether among infants born very preterm or growth-restricted at full-term, the lack of prospective longitudinal studies limits understanding of the etiology of early life stress and later variation in direction of cortisol. Our findings suggest such developmental trajectories long-term to be complex and likely gender-specific, requiring a lot of further research to elucidate the factors involved. Relationships between dysregulation in cortisol levels and clinical manifestations are complex. A recent review of research on hair cortisol described contradictory findings on the direction of cumulative cortisol in relation to depression. In general, long-term associations between HPA axis dysregulation and clinical manifestations, and the pathophysiological mechanisms underlying variation in direction of these relationships, are not well understood. Our data are unique to our knowledge, as they arise from the only longitudinal cohort study of the developmental trajectory of HPA activity in children born very preterm compared to full-term, spanning infancy to school-age. In the only other study to compare cortisol levels in preterm versus full term children, high awakening salivary cortisol was reported in early adolescence in the preterm group, indicating that different information about HPA function may be revealed by measures of diurnal patterns of salivary cortisol compared to a more integrated measure of cortisol levels over time as evaluated in hair. Importantly, consistent with our hair cortisol findings in the present study, we found that the extent of pain-related stress adjusted for neonatal medical confounders was associated with reduced salivary cortisol levels on the test day, and salivary diurnal cortisol pattern on non-school days, primarily in boys (Grunau and colleagues unpublished data). Our findings that the relationship between neonatal pain-related stress and down regulation of cortisol at age 7 years converges across multiple independent cortisol measures speaks to the robustness of our findings. NF-κB belongs to a family of transcription factors that can orchestrate many stress and inflammatory processes. Activation of NF-κB represents a downstream effector for the response to stressful events and links changes in the activity of the neuroendocrine axis to the cellular response. An example of this link is the finding by Miller and colleagues demonstrating heightened expression of genes controlled by NF-κB in adults who experienced low socioeconomic status (SES) in early life. Genetic variation in the promoter region of *NFKBIA*, which alters the ‘tuning’ of stress and immune responsiveness, has been linked to alterations in susceptibility to infectious and inflammatory diseases, and a variety of cancers. In this study we identified a novel association in boys between *NFKBIA* rs2233409 genotype, hair cortisol and neonatal procedural pain- related stress. Intriguingly, rs2233409 lies in the putative binding site of octamer binding transcription factor-1 (Oct-1) that interacts synergistically with the glucocorticoid receptor (GR) to bind DNA. Disruption of Oct-1 binding due to the rs2233409 polymorphism would be predicted to reduce the anti- inflammatory effects of ligand bound GR, since DNA binding of Oct-1 strictly depends on GR binding. Our present findings that the minor allele (CT or TT) of *NFKBIA* rs2233409 may influence the long-term effects of neonatal pain-related stress provide evidence of an association only, and animal experimental work is required to establish a causal pathway. We have shown that in boys born preterm, the minor allele (CT or TT) of *NFKBIA* rs2233409 was associated with higher secretion of inflammatory cytokines, providing support to the idea that neonatal pain-related stress may act as a proinflammatory stimulus that induces long-term immune cell activation. Further research is needed to elucidate the functional implications of this immune system reprogramming effect. ## Gender Differences in the Effects of Early Stress on HPA Axis Function Importantly, boys with the minor allele (CT or TT) for *NFKBIA* rs2233409 displayed the greatest vulnerability to long-term effects of neonatal pain- related stress. In animal studies, early life stress can have sex-dependent long-term effects on the responsiveness of the HPA axis, with some studies reporting higher vulnerability or sensitivity in males to the early disturbances compared to females, while others report greater vulnerability or sensitivity predominantly in females. These discrepancies may be due to different types and timing of the stressors and suggest that males and females may either have different developmental windows of vulnerability and/or adapt differently to the early adverse environment. Human longitudinal studies on gender differences in HPA axis programming due to early adverse experiences are rare, and to our knowledge this is the first to investigate the relationship of early pain-related stress to HPA axis function comparing boys and girls born very preterm. A previous study by Jones et al., found that in 7–9 year old children, lower birth weight was significantly associated with a higher salivary cortisol response to the Trier Social Stress Test (TSST) in boys, but not in girls. In line with this, Elzinga et al. 2008 reported a blunted cortisol response to the TSST in subjects with early trauma, which was more prominent in men than in women. However, the direction of effects can vary, and in some studies no gender differences were found in neuroendocrine regulation after early stress or trauma (for review see:. Our findings extend these previous data, demonstrating downregulation of HPA activity in hair cortisol, which provides an integrated measure of cortisol levels over time. Taken together, early life stress appears to be more strongly associated with HPA axis function in males than in females, consistent with our current results of a negative association of early life pain-related stress and hair cortisol in preterm boys but not girls. Importantly, similar differences in prepubertal children have been reported previously. Thus variation in concurrent sex steroid concentrations cannot sufficiently explain the observed gender differences. Instead, evidence from animal studies suggests that there may be an important interaction between early stress and regulation of gonadal hormone production, suggesting that early adverse experiences may interfere with gonadal programming of a sexually dimorphic brain. However, little is known about disturbances of the normal hormonal milieu and surges due to premature birth and early stress exposure. It is clear that more research is needed to better understand the impact of gender on neuroendocrine function after early life stress in this population, and relationships between HPA axis regulation and clinical problems. We have recently reported that infants born very preterm exposed prenatally to chorioamnionitis with funisitis (inflammation of the umbilical cord) had a different pattern of cortisol levels at 18 months corrected age, compared to infants exposed to chorioamnionitis alone or no prenatal infection, suggesting that presence of a *fetal* inflammatory response may alter programming of the HPA axis. In the present study, in school-age children, we did not have the results of placental histopathology, which places a limitation on our current findings. However, this does not alter our conclusion that functional genetic variation of stress and inflammatory responses appears to play a role in altered cortisol expression in boys born very preterm. Another important limitation to the present study is that the sample size was too low to evaluate the role of factors such as sources of current child stress (beyond parent depression and trait anxiety), socioeconomic status, parent-child behavioral interactions, and childhood illness after the NICU. In fact, given the wide range of past and concurrent factors that might affect the expression of cortisol, it was noteworthy that the association between neonatal pain-related stress and hair cortisol persisted to school-age (albeit only in boys). # Conclusions Altered programming of the HPA axis indexed by cumulative cortisol levels, persists to school age in children born very preterm. In contrast to the elevated cortisol levels observed previously at 8 and 18 months, at age 7 years, cortisol was downregulated in preterm relative to full-term children. These findings suggest that these children may be experiencing some level of chronic stress over the long term, consistent with the pattern seen in post-traumatic stress disorder long after the stressor has ended. Importantly, this study is the first, to our knowledge, to reveal that functional genetic variation in the promoter of *NFKBIA*, which is a critical negative regulator of stress and inflammatory processes, plays a key role in long-term programming effects of neonatal pain-related stress on HPA function in children born very preterm. Furthermore, higher secretion of inflammatory cytokines was related to the minor allele (CT or TT) of *NFKBIA* rs2233409, consistent with the possibility that neonatal pain-related stress may act as a proinflammatory stimulus that induces long-term immune cell activation. The finding that these effects were gender- specific, with boys driving the association, has major clinical implications for understanding possible gender differences in vulnerability of these children. We thank the families that participated, and Gisela Gosse for co-ordinating the study, and Katia Jitlina, Amanda Degenhardt and Maria Kosovic for data collection. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: REG SET PL JW. Performed the experiments: ILC CMYC AFH ER ML. Analyzed the data: CMYC RB REG. Contributed reagents/materials/analysis tools: GK SVU PL. Wrote the paper: REG ILC CMYC SET JW SB.

# Introduction It is estimated that more than 25 million people worldwide have Alzheimer’s disease (AD), with these numbers expected to triple by the year 2050. Familial early-onset AD is well recognized as an entity but accounts for only about 2–3% of AD cases. FH is also a significant risk factor for, and predictor of late- onset AD – and studies suggests a 2 to 4-fold greater risk for AD in such individuals with a first-degree relative who has developed late-onset AD. Common gene polymorphisms (e.g. the ε4 variant of the APOE gene) account for about 50% of the heritability of late-onset AD and despite recent genetic findings of new candidate genes, there is still a significant unexplained heritability. If one assumes that the average person with AD has 3 living first-degree relatives (1 sibling, 2 children), then there would be some 75 million worldwide with a positive FH of AD. Subjects with a positive FH often participate in research studies as controls, thus making it particularly important to understand how FH affects biomarker phenotype. While previous multisite research has shown associations between AD biomarkers and specific genetic variations (such as APOE, APP, PSEN1, and PSEN2), in most research studies and in routine practice, clinicians usually rely on a simple “yes/no” of self-reported FH status. Prior studies of the effects of FH status on biomarkers have shown effects on PET glucose metabolism, hippocampal volumes and amyloid markers. If true, these findings have great significance since subjects with a positive FH are routinely enrolled in diagnostic and prognostic biomarker studies as “normal controls” and their inclusion might affect the accuracy of biomarker cut-points established for discriminating AD from controls. The Alzheimer’s Disease Neuroimaging Initiative (ADNI) is considered one of the more successful biomarker research studies and has been widely analyzed and reported. Its data forms the basis, in part, for newly proposed research diagnostic criteria for preclinical AD as well as MCI due to Alzheimer’s. A positive family history was not an exclusion for controls or MCI in ADNI nor its subsequent studies ADNI-GO and ADNI-2. Most ADNI subjects were recruited from the community using advertisements and referrals in a manner similar to that used for most therapeutic and biomarker trials in many disorders. The aims of our study were to systematically assess how subjects with a FH of AD differed from those without a FH in demographics, cognition, as well as neuronal and pathologic biomarker profiles, at the time of enrollment into a national biomarker study. Another aim was to examine whether the effect of FH was solely due to ApoE4 status. We also examined whether the effects of FH status on biomarkers differed across the spectrum of cognitively normal (CN) to MCI to mild AD to test the timing of such effects in relation to development of cognitive symptoms and clinical dementia. # Materials and Methods ## Subjects Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.ucla.edu). The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer’s disease (AD). ADNI subjects have been recruited from over 50 sites across the U.S. and Canada. The initial goal of ADNI was to recruit 800 adults, ages 55 to 90, to participate in the research and have approximately 200 cognitively normal older individuals be followed for 3 years, 400 people with MCI be followed for 3 years and 200 people with early AD be followed for 2 years. For up-to-date information, see [www.adni-info.org](http://www.adni-info.org). To read about the subject eligibility criteria for the ADNI database refer to the ADNI-1 Procedures manual. ### Ethics ADNI was approved by the IRBs of all participating sites including Duke University. All subjects and if applicable, their legal representatives, gave written informed consent prior to the collection of clinical, genetic and imaging data. Subjects selected for analysis were required to have information for all of the following available: age, gender, and family history of Alzheimer Disease; an Alzheimer’s Disease Assessment Scale Cognitive Subscale score (ADAS-Cog) and a Mini-Mental State Examination (MMSE) score obtained at the initial visit; *APOE* genotyping results; initial-visit 1.5 T MRI scans which were analyzed for ADNI by FreeSurfer software, version 4.4; a measure of estimated intracranial volume derived from any MR scan; and baseline values for CSF amyloid-beta 42 (Aβ42), total-tau (t-tau), and phosphorylated tau-181p (p-tau). We chose CSF Aβ42, t-tau, p-tau (a sensitive marker of amyloid and tangles) and MRI-hippocampal volume (a marker of hippocampal neuronal loss). We also looked at MMSE and ADAS, the two measures most often used in practice and clinical trials, respectively. We did not examine other measures to limit the number of comparisons. Not all subjects underwent CSF studies in ADNI and hence our sample was restricted to those that had CSF data. At the time of our data gathering, April 1st 2012, a total of 257 subjects in ADNI-1 met criteria for inclusion in the study. ## Family History and Clinical Diagnosis ADNI collected FH data using an interview with the subject and their study partner about the presence of Alzheimer’s disease in their parents or siblings and the site checked yes or no to parents and each individual sibling. The source of information was usually study partner for memory-impaired subjects and the subjects themselves for controls. A positive family history (FH+) was defined as having a parent or sibling–living or deceased–who had been diagnosed with AD. A negative FH (FH−) thus consisted of having no parents or siblings with a history of AD. Study participants with uncertain family history status were excluded from the analysis. 247 participants were excluded from the original ADNI database for having unknown or incomplete family history data before the addition of other variables. ## Cerebrospinal Fluid Collection and Assays CSF samples were obtained by lumbar puncture and examined for t-tau, p-tau181P, and Aβ42 as described previously. More detailed protocols can be found on the ADNI website. CSF proteins were used as continuous variables in the logistic regression and cut-off values for CSF signatures for AD (CSF t-tau/Aβ42\>0.39) were derived from a published autopsy verified correlative study of ADNI subjects. ## APOE Genotyping Genotyping of all subjects for *APOE* allele status was performed using DNA extracted from peripheral blood cells (ADNI-1 Procedures). ## MR Imaging Acquisition ADNI used 1.5T MP-RAGE T1-weighted MR images. All scans were performed using a standardized protocol specifically developed for ADNI, and which was tailored for use with each make and model of scanner used at the different data collection sites. More detailed information for the specific MR acquisition protocols and quality control methods for each type of scanner used can be found at <http://adni.loni.ucla.edu>. ## MR Volumetric Methods Hippocampal volumes (HV) were derived from volumetric segmentation of subject MR scans, which were performed with the Freesurfer image analysis suit. Freesurfer morphometric procedures have been demonstrated to show good test-retest reliability across scanner manufacturers and across field strength. Intra- cranial volumes (ICV) were derived from MRI scans (see adni.loni.ucla.edu and refer to the detailed ADNI 1 MRI Protocols for sequences and processing steps). ## Statistical Analysis We first ran unadjusted t-tests and chi-square analyses comparing demographics and biomarker data (CSF Aβ42, t-tau, ptau181p, and hippocampal volume, HV) between FH+ and FH− subjects within each diagnostic group. Biomarker data were log-transformed to normalize their distributions but results were essentially similar before and after log transformation. shows demographics and – show unadjusted biomarker data. We then ran a multivariate linear model, with age, baseline MMSE, gender, and ApoE4 as predictors to examine effect of FH on tau/Aβ42 and HV within each diagnostic group above and beyond the effects of ApoE4 ( and). We also examined the effect of FH status on intracranial volume (ICV) in each diagnostic group using multivariate linear models and also adjusted for ICV in models examining the effect of FH on HV. With chi-square, we tested the hypothesis that a positive FH status would be associated with a greater proportion of CN and MCI subjects with a CSF pathological signature of AD (CSF tau/Aβ42\>0.39). Finally, ROC curves were generated to examine whether the specificity and sensitivity of CSF Tau/Aβ42 ratio for distinguishing AD from FH− CN differed from that for FH+ CN. We also ran alternative models adjusting for education and without adjusting for MMSE. # Results The overall, prevalence of FH+ in subjects who volunteered for CSF in ADNI-1 (50.4%) was slightly higher than the prevalence of FH+ among all ADNI-1 subjects with FH data (42%). The prevalence of a FH+ status did not differ significantly between diagnostic groups (p = 0.36). Within each diagnostic group, there were no differences in key demographic or cognitive baseline characteristics by FH status. ## Effect of FH before Adjusting for ApoE4 depicts the unadjusted effect of FH on biomarker values. In univariate analyses, mean CSF t-tau/Aβ42 ratios were in general higher in FH+ subjects than in FH− subjects across all diagnostic groups but FH effects were strongest in MCI. Among MCI subjects, as shown in a–c, CSF Aβ42 was lower (p = .005), t-tau was higher (p = 0.02) and t-tau/Aβ42 ratio was higher (p = 0.002) in FH+ than in FH− subjects. Among CN subjects, there was no significant difference for CSF Aβ42 to be lower (p = 0.12) and t-tau/Aβ42 to be higher (p = 0.13) in FH+ than in FH− subjects. Among AD subjects, the effect of FH was also not significant for any comparison (p\>0.2 for all). Hippocampal volume (sum of left and right) did not significantly differ by FH status in any group. Additionally, there were no significant differences in ICV between FH+ and FH− in any group (p = 0.51, 0.28, 0.22 for AD, CN, MCI respectively;). ## ApoE4 Prevalence and Effects The ApoE4 allele was overrepresented in FH+ subjects (p = 0.0002) as expected. The % prevalence rate of E4+ was 42% CN, 68% MCI and 75% AD in FH+ subjects versus 21% CN, 42% MCI and 59% AD for FH− subjects, respectively. ## Effect of FH after Adjusting for ApoE4 After adjusting for age, gender, baseline cognition and ApoE4, the effects of FH+ status on CSF t-tau/Aβ42 ratio in MCI subjects remained significant (p\<0.036) but showed no significant difference in CN (p = 0.11). As with univariate analyses, the FH effect on t-tau/Aβ42 was again not significant in AD subjects. In these multivariate models, the effect of ApoE4 on t-tau/Aβ42 ratio was significant in CN (p = 0.0082) and MCI (p = 0.000) but not in AD (p = 0.13). The effect of FH on t-tau/Aβ42 ratio in MCI subjects remained significant (p = 0.036) in the model that adjusted for age, gender, E4 and education. The effect of FH on HV was not significant in any of the three diagnostic groups. Age and cognitive status had bigger effects than FH status on HV. The effect of ApoE4 on HV was not significant in CN (p = 0.86), became significant in MCI (p = 0.02) and turned nonsignificant in AD (p = 0.44). ICV did not differ by FH or ApoE4 in any diagnostic group but differed by gender. In another model that adjusted for age, gender, education, E4 and ICV, the effect of FH on HV was not significant in any of the diagnostic groups (p = 0.96, p = 0.66, p = 0.23, respectively) thus adjusting for ICV did not alter the findings. ## Prevalence of “Pathologic” AD Phenotype by FH Status More than twice as many FH+ CN (47%) exhibited “pathologic phenotype of AD” (CSF t-tau/Aβ42 ratio \>0.39) than FH− controls (21%) (p = 0.03). There was also a trend for a greater proportion of FH+ MCI subjects (82%) who exhibited this phenotype than in the FH− MCI group (66%) (p = 0.07). This was not statistically different in the AD group (95% vs 88%, p\<0.65). ## ROC Curves The sensitivity and specificity of CSF t-tau/Aβ42 ratio of 0.39 for separating AD from all CN was 92% and 66%, respectively. For separating AD just from the FH+ CN group sensitivity was 95% and specificity was 53%; for separating AD from the FH− group, sensitivity was 88% and specificity was 79%. The difference in sensitivity between FH+ and FH− groups was not significant (p = 0.65, chi- squared test) while the difference in specificity between FH+ and FH− groups was significant (p = 0.03). # Discussion Three main findings emerged from our study. We found significantly higher t-tau/Aβ42 ratios and a higher prevalence of the rate of a “pathologic t-tau/Aβ42 endophenotype” in FH+ (versus FH−) CN and MCI groups. There was an additional effect of FH on these markers above and beyond that of ApoE4 in MCI subjects and the model estimated suggests that this additional effect is about half of the size of the ApoE4 effect. We found no FH effects on CSF pathologic markers in AD. We also found no FH effects on neuronal loss marker (hippocampal volume) both before and after adjusting for intracranial volume. We also found no FH effects on ICV. These data extend findings from prior studies of FH effects – to the national ADNI sample and to MCI subjects. Our study also found that almost half of all normal controls with FH+ would have met research criteria for preclinical AD (based on CSF) at entry into ADNI but only about 20% of FH− subjects would have met such criteria. This result is also consistent with the view that a family history of AD does not reduce cognitive reserve, as there were no significant differences in cognitive test scores between FH+ and FH− groups. Rather, one can speculate that the risk of family history is probably mediated by earlier development of amyloid pathology or more rapid development of amyloid pathology with the same time of onset. Of prior studies examining FH effects, three are particularly relevant to our analyses. In a study of 269 cognitively normal controls, Xiong et al reported that FH status was linked to a decrease in CSF Aβ42, a finding that we extend by reporting a similar and even more robust change in MCI. Honea et al examined the relationship between biomarkers and parental history of all dementia types in the ADNI sample, thus potentially including also FH of vascular dementia, DLB or FTD or other etiologies. In their study, the rate of FH+ subjects was different from ours and unlike our study, the effects of FH on t-tau/Aβ42 ratio and t-tau effects in MCI did not reach significance. They did report a significant FH effect on Aβ42 in MCI consistent with our finding, but in their study the FH effects in MCI were not significant after adjusting for ApoE4, whereas ours remained significant. They also found pathologic signatures of AD in a smaller percent of CN than we did. Thus, their looser definition appears to have resulted in an underestimation of the effect of FH of AD. Andrawis et al found MCI subjects with positive maternal history of dementia had smaller baseline hippocampal volumes and greater 12-month atrophy rates. The effect of positive maternal history of dementia on hippocampal atrophy in MCI and AD was significant after controlling for age, ApoE4 genotype, and paternal history of dementia. Taken together, these studies along with prior studies showing potential FH effects on brain networks and glucose metabolism highlight the need to further examine FH effects on multiple biomarkers simultaneously. The mechanisms underlying the effects of FH status are not fully known but will likely vary depending on biomarker – ie genetic mechanisms underlying hippocampal volume changes are not likely to be identical to those underlying amyloid or tau processing. Prior autopsy, CSF and PET studies have linked ApoE4 to an amyloid phenotype and hippocampal changes, and studies have documented ApoE4 effects on greater neocortical amyloid-beta deposition and/or reduced clearance. However, E4 does not account for all of the variance and there is interest in determining the degree of unexplained heritability not accounted for by ApoE4 as well as the genes underlying such unexplained heritability. In our ADNI sample, the effect of adjusting for ApoE4 on pathologic markers was different in different diagnostic groups - the FH effect was considerably weakened in CN, remained significant in MCI subjects, and was not significant in AD. After adjusting for ApoE4, the remaining FH effect on CSF t-tau/Aβ42 was approximately half the size of the main ApoE4 effect. Thus, our data along with others confirms that there are as yet unidentified genetic factors embedded in FH status that influence pathology before the onset of dementia. Our sensitivity/specificity analyses also suggest that the presence of FH+ controls in an AD control group may significantly reduce the specificity of CSF pathologic biomarkers for separating AD from controls. It may be worth examining whether including FH+ controls may have reduced the accuracy of calculations in other tests, such as amyloid PET or plasma Aβ42. Likewise our data also suggests that companies planning registration studies of diagnostic biomarkers to detect AD pathology in at-risk subjects may wish to exclude FH+ controls to enhance their power for achieving the desired 80% or greater specificity. There are also some potential limitations of this study – by design ADNI’s sample size of controls and AD was relatively smaller than the MCI group, which may have limited power to detect small effects in controls. CSF data were only collected in a subset who agreed to volunteer, a selection bias that applies to most CSF biomarker studies. FH status was determined through interviews with subjects and informants, but may have been subject to a reporter bias and lack of informative pedigree (early death of relatives due to other causes, though this problem is less likely in the US due to longer life expectancies and greater awareness of dementia). Many respondents may not be well versed enough to know the difference between a dementia and AD. Because FH in most biomarker research and practice is usually collected only by simple history, our findings are relevant. We also did not distinguish between maternal and paternal inheritance and hence our data cannot be compared with findings that maternal family history may have greater risk for metabolic changes or hippocampal atrophy. Furthermore, given that there is a mitochondrial hypothesis providing an underlying biological mechanism for promoting disease on the maternal side we believe future studies should examine maternal versus paternal family history. However a recent longitudinal study of 108 middle-aged normal controls (of younger age than ADNI cohort) found that FH status predicted greater atrophy only within a posterior sub-region of the hippocampus but not in other gray matter regions, and that there was no effect of maternal versus paternal history. Differences in sampling, FH ascertainment, and biomarker methods could account for some of the discrepant findings. While the means differ significantly, the overlap in CSF data boxplots between FH+ and FH− MCI groups suggests that these findings may not be as robust a biomarker as one where the boxplots do not overlap at all - unfortunately no such biomarker exists. What do these phenotypic differences related to a positive FH in MCI mean for the subject? Other studies have linked CSF pathologic phenotypes with faster rates of future decline in CN and MCI subjects. By extrapolation, this would imply that the subset of FH+ MCI and CN subjects with abnormal biomarker phenotypes would decline faster than FH− subjects. Longitudinal data from ADNI and standardization of hippocampal sub-region analyses as well as CSF soluble amyloid oligomer assays may permit more definitive testing of the prognostic significance of FH differences on risk for decline. In summary, our study, derived from a large national biomarker cohort, documents that a positive family history of AD is associated with an abnormal beta-amyloid and tau endophenotype prior to the onset of clinical AD in mildly symptomatic subjects, and that there are genetic influences embedded within FH beyond that due to ApoE4 that are most obvious in the MCI cohort. Since CSF biomarkers correlate highly with cerebral neuritic beta-amyloid and neurofibrillary tangle changes, we also speculate that FH status is associated with earlier onset of preclinical pathologic AD. These findings have implications for the design of future research studies, heritability of AD and personalizing testing and care of at risk subjects. Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.ucla.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: <http://a dni.loni.ucla.edu/wpcontent/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf>. The ADNI was launched in 2003 by the National Institute on Aging (NIA), the National Institute of Biomedical Imaging and Bioengineering (NIBIB), the Food and Drug Administration (FDA), private pharmaceutical companies and non-profit organizations, as a \$60 million, 5-year public-private partnership. [^1]: The Alzheimer’s Disease Neuroimaging Initiative is funded by Amorfix Life Sciences Ltd.; AstraZeneca; Bayer HealthCare; BioClinica, Inc.; Biogen Idec Inc.; Bristol-Myers Squibb Company; Eisai Inc.; Elan Pharmaceuticals Inc.; Eli Lilly and Company; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; GE Healthcare; Innogenetics, N.V.; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC; Medpace, Inc.; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Servier; Synarc Inc.; and Takeda Pharmaceutical Company. PMD owns stock in Sonexa and Clarimedix, whose products are not discussed here. Jeffrey Petrella is on the neuroradiology advisory board of Janssen Alzheimer Immunotherapy. There are no patents, products in development or marketed products to declare. This does not alter the authors’ adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: PMD KRC JRP. Performed the experiments: JRP PMD. Analyzed the data: EJL KRC PMD. Contributed reagents/materials/analysis tools: JRP CAH KRC PMD. Wrote the paper: EJL PMD CAH KRC.

# Introduction In a supermarket individuals are confronted with a lot of different food products to choose from. Food companies anticipate on this decision process by marketing their products not only by directing the attention of potential consumers towards the sensory properties of a product (i.e. taste, smell and texture), but also by highlighting additional features of a food product related to specific non-sensory related characteristics of these products. Examples of the latter characteristics are the time a product most likely is consumed (e.g., *breakfast* drink) and the health benefits of that particular product (e.g., *low calorie* drink). Such marketing strategies are supported by scientific findings indicating that food choice is indeed based on a broad range of product characteristics, including sensory and non-sensory related. Recent studies focused on the mediating role of implicit associations on consumer choice. However, the relative importance of each of these factors implicitly affecting food choice remains elusive. Therefore, we set up the current study to differentiate between the association strength of different food product characteristics. Many food choices are made in the absence of the actual perception of a food’s sensory properties. Therefore, real life food choices highly rely on previous experiences of consumption of food products, which are stored in our memory system. The sight of food products triggers these memory traces. For example, when people are faced with or question to think of a food product, they have expectations about its taste based on memory from previous experiences with that particular or similar food product. According to the recognition heuristic of decision making, consumer choices are largely based on the recognition. This means that consumers prefer a match between their previous experiences (i.e. activated memory traces) and the sight of the food product when making food choices. Different associations affect food choice. First, descriptions of sensory properties, frequently applied in product marketing, facilitate adequate recall of previous consumption (e.g., *sweet* apple). Second, alongside of taste characteristics, studies have also shown that food choice is influenced by associations with the context in which a food product was previously consumed. For example, when food is categorized as “for *breakfast*”, it is more preferred in the morning than in the afternoon, while food categorized as “for *dinner*”, is more preferred in afternoon than in the morning. Furthermore, it was also found that people reported more post-meal hunger, for servings at inappropriate times. Third, food choice is affected by health beliefs. For example, dieting people assigned greater importance to nutrition information available on food products when evaluating healthiness. Analogous to this, it was found that people associated their healthy food choice with positive feelings as ways to promote well-being. Altogether, these studies show that different associations affect food choice. The extent to which these associations differentially affect food choice has recently attracted more attention. Taste was shown the most chosen as the first criterion in food choice. A direct comparison between taste and health associations showed that the taste of an apple was indeed a stronger predictor of individual dessert choice as compared to its assigned health score. Furthermore, food choices were guided by information about time of the consumption (i.e. “*breakfast* product”) only when people were yet unfamiliar with its taste. Overall, taste associations seem to be the strongest predictor of food choice in case of familiar food products. However, the differential importance of taste associations with respect to health and context associations remains elusive. Previous studies on food associations and food choice often used explicit behavioural measures of taste, health, or context associations. However, people are usually not or only to a limited extent aware that these learned associations affect the way they choose food products, namely through the generation of expectations. In the present study, we directly compared taste, health, and context associations with food using an *associative priming paradigm*. In this paradigm, participants are asked to make decisions about a target image, preceded by a prime word. Because we were interested in association strength and not so much in semantic relation, it was chosen to present the primes *masked*. The rationale behind this is that research has indicated that masked priming is affected by association strength whereas unmasked priming more by semantic relationship. In our study we used words as primes activating either taste (i.e. *sweet* and *salty*), health (i.e. *healthy*, *unhealthy*), or context associations (i.e. *breakfast*, *dinner*). Pictures of food and non-food items were used as targets. It was expected that taste primes would facilitate the categorization of food target more than health and context related associations. Furthermore, the prime-target relation could be either congruent (i.e. in accordance) or incongruent. We hypothesized that people will respond faster on congruent trials than on incongruent trials because of facilitated processing of the target by the prime. Participants responded significantly faster to strongly related pairs of (unconscious) prime words and target images of objects and animals compared to unrelated pairs. In accordance with these findings, we assumed that stronger associations would benefit more (i.e. faster and more accurate responses) from the effect of priming. Neuroscience adds value to understanding consumer decisions by unravelling the mechanism that are related to the observed choice. Such process knowledge enables us to make inferences beyond existing explicit behavioural findings, since these behavioural responses reflect a single end-product of different processes involved in information processing. Previous studies have related specific event-related potentials (ERPs) to associative memory processing. In the present study, we focus on the N400 component, found to be related to associative processing, because we consider this to be the stage of information processing where implicitly activated memory traces are compared to a perceived food product. It was, for example, found that the N400 amplitude was smaller in response to congruent compared to incongruent prime and target pairs. On top of that, high compared to low associative strength between prime and target resulted in a smaller N400 amplitude, as well. In sum, in the present study we used an associative priming paradigm in which participants identified food and non-food pictures preceded by food related prime words in order to examine implicit associations with food pictures. In addition to the behavioural measures we focussed on the N400 EEG component, reflecting associative memory. # Material and Methods ## Participants A total of 30 volunteers participated in the present study (15 males; M = 20 years, SD = 1 year). All the participants were recruited from at the faculty of Behavioral and Social Sciences of the University of Groningen. They were Caucasian, with good English language proficiency based on their participation in the English bachelor program of Psychology requiring English proficiency at the C1 level in the Common European Framework, roughly corresponding to TOEFL ITP sum score of at least 627, and a score 63 on the reading subscale. They had normal or corrected-to-normal vision. They had no history of an eating disorder or any other psychiatric, serious medical or neurological diseases. Also, none of the participants was on psychoactive or hypertensive medication. A total of four participants reported being vegetarian, and one participants reported being on a diet. Participants were given course credits in exchange for their participation. It was attempted to test all participants at least one hour after a meal, in the afternoon (at 14h or 15h). The local ethics committee of the faculty of Behavioral and Social Sciences of the University of Groningen reviewed and approved the present study. All participants signed informed consent. ## Apparatus The participants were tested individually in a dimly lit, sound-attenuated room. The experiment was done on a personal computer running Windows 7, with a refresh rate of 60 Hz. The task was fully programmed in Matlab (The Mathworks, Inc. 2014), which was also used to collect the behavioral data. EEG was recorded using 21 tin electrodes attached to an electrocap (ElektroCap International Inc., Eaton, Ohio, USA). The electrodes were placed according to the international 10–20 system. The amplifier was a REFA 8–72 (Twente-Medical Systems, Enschede, The Netherlands). An average reference was used. Sample frequency was 250 Hz. Two electrodes were placed at the mastoids and were used for off-line re-referencing of the EEG signal. An electrode placed on the sternum served as the participants ground. Four electrodes, placed at the left and right lateral canthi and above and below the right eye, were used to measure the Electro Oculogram (EOG). Data acquisition was performed using Brain Vision Recorder (version 1.03, BrainProducts GmbH, Munich, Germany). ## Stimuli In this study we used an associative priming paradigm. In this paradigm, the prime stimuli that were used were written words of either taste (i.e. *sweet* and *salty*), health (i.e. *healthy*, *unhealthy*), or context associations (i.e. *breakfast*, *dinner*). In addition, neutral primes (i.e. ‘XXXX’) of different lengths were used to control for the effect of word length of the different food related words. The independent variable for the primes was called *Modality*, and contained the levels ‘taste’, ‘health’, ‘context’, and ‘neutral’. The word categories of *Modality* were matched on word frequency using the online word frequency database of the Corpus of Contemporary American English (*taste* _{*mean (sweet & salty)*}: 24.277, *health* _{*mean (healthy & unhealthy)*}: 26.009, and context _{*mean (breakfast & dinner)*}: 25.777). As targets we used pictures of food and non-food items that could be either congruent or incongruent to the prime. To investigate what combinations of food and modality are considered congruent in the population we conducted a pilot study. The independent variable for the relationship between the target and the prime was called *Congruence*, and could be either “congruent” or “incongruent”. All the pictures used in this study were derived from Google. All the pictures had a white background. Consequently, the non-food pictures were chosen to resemble the matching food picture (see). Food and non- food pictures were matched on the color, size, orientation and amount of objects. There were in total six food categories (sweet, salty, breakfast, dinner, healthy, and unhealthy). For every category there were 15 food pictures and matching non-food pictures. Similarly, congruent and incongruent trials were equally balanced (i.e. 50/50). All the independent variables were within subject factors. μV/m² ## Task Each trial in the task began with a fixation cross (‘+’) varying between 800 and 1300 ms. After the fixation cross, a prime word was shown for 32 ms. To reduce the visibility of the prime, a backward mask of nine hash tags (‘#########’) was presented for 116 ms after the prime. The target picture in the center of the screen followed the presentation of the mask and stayed on the screen until response, but with a maximum of 5000 ms. To investigate the effects of *Modality* and *Congruence* we measured the reaction time and accuracy to both food and non-food target pictures. ## Procedure The participants were seated in one of the laboratory rooms facing a computer screen. They were first asked to fill out a questionnaire containing questions about their age, nationality, current feeling of hunger, diet, food restrictions, and if they were vegetarian. This questionnaire took approximately two minutes for the participants to complete. Thereafter, the main experiment started. During this part of the study, the participants were instructed to respond as quickly as possible if the presented picture contained either a food or a non-food object, by pressing a corresponding response mouse button with their left or right thumb. What button represented either food or non-food was counterbalanced between participants. The task was divided in three blocks of 150 trials. The blocks differed in which prime modalities were assessed. In the first block the primes sweet, salty, breakfast, dinner, and neutral were assessed. In the second block; breakfast, dinner, healthy, unhealthy, and neutral primes were investigated. The third block focused on sweet, salty, healthy, unhealthy, and neutral primes. Within each block, 32 trials per prime and 22 neutral were given (total 150 trials per block). The order of the blocks was counterbalanced across participants and within each block the trials were randomized. Between the blocks, the participants were allowed to take a rest. The time of rest was not pre-set; participants rested approximately one minute. After completion of this task, the participants were instructed to wait for further instructions from the experimenter. The whole task took about eight minutes to complete. After the associative priming task, craving for food was assessed. Explicit proxies of approach tendencies were collected for all food stimuli. Using visual analogue scales, food stimuli of the word-picture priming task were rated on liking at the moment of testing on a Visual Analogue Scale (VAS) using the question: “How much do you like this product?” which were answered on a scale (0–100) from “*don’t like*” (0) to “*like very much*” (100), which was answered on a scale (0–100) from *“not at all”* (0) to *“very much”* (100). Results are reported elsewhere. In total, the experiment took approximately 30 minutes. At the end of the experiment, the participants were debriefed. ## Data analysis ### Behavioral data—reaction times Due to technical issues, both behavioral and electrophysiological data of one participant was removed. Responses faster than 250 ms and slower than 1500 ms were excluded from the data analysis. Alongside these selection criteria, all incorrect trials (e.g., button press ‘food’ when a ‘non-food’ image is presented) were excluded from the analysis. Overall data selection resulted in the deletion of 929 observations (7% of total data). The data was fitted on linear mixed effect models with maximum likelihood (LME) using the *lme4* package in the open source statistical language R (version 3.1.2) (R Core Team 2012). These models were chosen because they deal well with repeated measures and missing data. Different models were build to test the effect of word length and priming effects of interest. Priming effects were analyzed for responses to food and non-food targets by a model including *Target* (two levels: food and non-food) and *Prime* (two levels: word (i.e. taste, health, context) and neutral) as fixed factors; the *Subjects* were used as a random factor (intercept). Furthermore, priming effects of interest were investigated in a separate model using responses to food targets, by entering *Modality* (three levels: taste, health, and context) and *Congruence* (based on prime and target combination; two levels: congruent and incongruent) as fixed factors; while taking into account inter-individual variability by adding *Subject* as a random factor (intercept). In order to control the latter priming effect for word length effects, the responses to food targets were compared between neutral primes of different length in a separate model including *Neutral* (four levels: 5, 6, 7, and 9 hashtags) as a fixed factor and *Subject* as a random factor. We report degrees of freedom, statistics, and p-values based on Satterthwaite’s approximations ANOVA. Statistical significance was evaluated at the 0.05 alpha level. ### Electrophysiological data—ERP latencies and amplitudes ERP data was processed using Brain Vision Analyzer 2 software (Brain Products, Munich, Germany). The EEG signal was filtered with a Butterworth high-pass filter of 0.5 Hz (24 dB/oct) and a low-pass filter of 15 Hz (24 dB/oct). Only correct trials with responses between 250 and 1500 ms were included for further analysis. The algorithm of Gratton, Coles, and Donchin (1983) was used to correct ocular movement artifacts. Further artifact removal was applied by removing segments with an absolute difference larger than 200 μV or a voltage step per sampling point larger than 50 μV. Baseline correction was applied from -350 until target onset. Epochs were averaged starting 350 ms before target onset and lasting until 1500 ms post-target onset, separately for each *Target*, *Prime*, *Modality*, and *Congruence* level. For confirmatory statistical analysis focused on the N400 priming effect, mean amplitudes (μV) were calculated in Brain Vision Analyzer using the amplitudes between 350–450 ms post-target onset from the average waveforms of individual participants. The averaged waveforms were grand averaged for display. Furthermore, we performed exploratory analysis on the frontoparietal electrodes based on previous evidence of priming. Comparable to the analysis of reaction times, ERP amplitudes were analyzed by means of linear mixed effect models, using the *lme4* package (39). We report degrees of freedom, statistics, and p-values based on Satterthwaite’s approximations ANOVA. Statistical significance was evaluated at the 0.05 alpha level. # Results ## Stronger focus on food compared to non-food images—N400 effect The analysis of reaction times (i.e. time needed to classify a target as food or non-food) showed that participants responded faster to food compared to non-food targets, both preceded by a neutral prime (e.g., ‘XXXXX’) (*t* _*Target* (1869) = 2.34, p =.012). Regarding the electrophysiological data corresponding to this effect, we observed that ERPs to food and non-food targets preceded by a neutral prime diverged clearly between 350 and 450 ms after target onset (*t* _*Target* (30) = -5.805, p \<.001). This effect was most pronounced on the parietal electrodes (i.e. P4, P7, and Pz), based on visual inspection of the current source density (CSD) maps. Therefore, we conclude the presence of a smaller N400 peak in response to food compared to non-food target images. ## Priming with food-related words facilitates associative processing of food images Interestingly, the effect described above was enhanced by the presentation of a word prime. Responses to food targets were faster following a word prime (e.g., ‘breakfast’) compared to a neutral prime (*t* _*Prime* (12580) = 4.745, p \<.001), whereas responses to non-food target images preceded by a word prime were similar as responses to non-food target images preceded by a neutral prime (*t* _*Prime* (12580) = 1.524, NS). The electrophysiological data demonstrated that this priming effect was reflected in a smaller N400 amplitude in response to a food targets following a food prime (e.g., ‘sweet’ or ‘breakfast’) as compared to a neutral prime (e.g., ‘XXXX’) on the parietal electrodes within the interval of 350–450 ms (*t* _*Prime*(90) = 2.70, p =.008). The priming effect was specific to food targets, since responses to non-food targets preceded by a word prime showed no difference in mean amplitude as compared to a neutral prime within this interval (*t* _*Prime* (90) = -0.43, NS). Thus, both behavioral and electrophysiological data showed priming of food associations. ## Taste associations stronger than health and context associations Subsequently, the next prime effects of interest focused on the associations with different food characteristics, namely differentiation of *Modality* (i.e. ‘taste’, ‘health’, and ‘context’) preceding the presentation of food target. We observed that the priming effect (i.e. faster response food target images following food-related word prime compared to neutral prime) was stronger for ‘taste’ compared to ‘health’ (*t* _{*Modality*: *taste-health*} (5313) = 1.94, p =.05) and ‘context’ (*t* _{*Modality*: taste-context} (5313) = 4.75, p \<.001). These effects were controlled for the effect of word length (*F* _*Neutral* (3,890) = 0.469, NS). Furthermore, the reaction time data revealed that the *Congruence* effect was different between primes (i.e. *Modality*) (*F*_{*Congruence x Modality*} (2,5313) = 6.20, p =.002). Remarkable, we observed that responses to congruent trials were slower as compared to incongruent trails for a ‘health’ prime (*t* _*Congruence* (5313) = -2.91, p =.003), and responses were not different between congruent and incongruent trials for ‘taste’ (*t* _*Congruence* (5313) = 1.22, NS) and ‘context’ (*t* _*Congruence* (5313) = 1.57, NS) primes. Compared to the reaction times, similar findings were observed in the electrophysiological data regarding the effect of *Modality* and *Congruence*. Results showed a smaller negative amplitude following a food target preceded by a taste prime compared to a context prime (*t* _*Modality* (150) = 3.24, p =.002) and a trend compared to a health prime (*t* _*Modality*(150) = 1.95, p =.053). This effect was most pronounced at Pz, following visual inspection of the current source density (CSD) maps. In contrast to the reaction time data, however, the electrophysiological data did not show a differential *Congruence* effect for the different primes. N400 amplitudes were similar across congruent and incongruent pairs for taste, health, context as well as context primes on the parietal electrodes (*F* _{*Congruence x Modality*} (2,150) = 1.18, NS). ## Food associations are frontally represented Following the absence of an interaction between *Modality* and *Congruence* on the parietal electrodes, we explored other electrodes that have previously been described to reflect the congruence effect. Indeed, we observed a right lateralized frontal congruence effect (i.e. max at FP2 electrode). More specifically, congruent trials showed a larger negative amplitude starting 130 ms after target onset as compared to incongruent trials (*F*_*Congruence* (1,31) = 7.76, p =.009). This effect was not specific to any word prime (i.e. *Modality*) (*F* _{*Congruence x Modality*} (3,210) =.07, NS). # Discussion ## General findings In the present study we explored food associations by means of an associative priming paradigm. We consider the current differentiation between implicit food associations like taste, health, and context relevant in three ways. First, understanding the neural mechanisms of food memory provides us with insights for adequate marketing of food products. Second, we included taste, health, as well as context associations, which allowed us to make direct comparisons within one study. Finally, this study unraveled the implicit mechanisms of food associations, whereas previous studies on food associations have primarily used explicit behavioural measures. We focused on reaction times and EEG event- related potentials related to associative memory processing, namely the N400 at parietal electrodes, because previous studies showed this to be the stage of information processing where implicitly activated memory traces are compared to a perceived food product. The current associative priming paradigm showed the plausibility to study factors affecting food choice without actual perception of the food product. Our findings reflected faster responses and a smaller parietal N400 peak in response to the sight of food compared to non-food. It has previously been shown that parietal neurons form bottom up “saliency maps” for quick selection of information in our environment. More specifically, the event-related potential N400 was smaller in response to strongly salient items. Based on our electrophysiological results, we consider that such “saliency maps” were constituted in the current paradigm by the repeated presentation of food information, thereby explaining enhanced processing of food compared to non-food images. These findings are in agreement with previous EEG studies suggesting increased motivational relevance and reinforcing properties of palatable food items to humans. In addition to salience, the current brain potentials reflect the role of associative memory in food choice. Besides the bottom up “saliency maps”, shifts of attention are thought to depend on “top down” signals derived from a current activated memory traces (e.g., finding a *sweet* and/or *healthy* apple). The results showed that activating associations of taste, health, and context leads to faster processing and a smaller parietal N400 peak in response to food, whereas processing of non-food items was not affected by priming. Priming thereby implicitly facilitates food choice based on the sight of food products. The electrophysiological data revealed that priming elicits a state-dependent change in associative food memory. Although there is increasing interest in how different factors are associated with food, how these “top down” signals differentially influence food choice remained elusive. For example, in a recent behavioral study on the use of simple descriptive food labels to promote healthy food choices, it was found that interventions that emphasize the taste of healthier foods are likely to be more effective at achieving healthier diets than those emphasizing health alone. Furthermore, it has been suggested that extra attention should be paid to the tastefulness of healthy food products. Following the current neuroimaging study we consider that a shift from explicit behavioral methods towards implicit neuroimaging methods, provides us with a more valid and detailed understanding of the mechanisms underlying such food choices. The current results point to the primary role of taste as a factor that directs consumers’ food choice. Thereby, we highlighted the potential role for associative food memory underlying the effects observed in behavioral consumer studies. We conclude that the affined behavioural and electrophysiological results reflect a stronger association with food for taste compared to non- sensory factors. In addition to priming studies showing enhanced processing for strong associations, it has been found that incongruent information can even hinder processing. In detail, high compared to low strength of associations resulted in a larger congruence effect, as reflected in a larger difference of N400 amplitudes at parietal electrodes. However, we did not observe this interaction with congruence in addition to the effect of priming. Following adequate power and good categorization accuracy of the target images, according to the different primes in both our pilot as well as actual study participants, we speculate that more complex mechanisms underlie the effect of incongruent information in food choice. It remains an ongoing challenge to facilitate healthy food choices. For example, it was found that health remains secondary to taste in the selection of corn chips. In line with this, consumer willingness to compromise on taste for health in the specific case of the functional foods category was considered a risky strategic option. Sensory aspects of food seems to be at the center of the development, maintenance and change of dietary patterns. So, in order to control or even counteract the effect of the ‘tasty = unhealthy intuition’, healthy foods should be marketed to be (more) tasty. It was suggested that efforts for promoting healthy eating behavior might benefit from an increasing attention towards memory principles in the development of interventions. This study provides a starting point for studying how subtle differences in food associations affect food choice. ## Conclusion The current study unraveled the implicit mechanism of food associations. We showed that taste associations are stronger related to food as compared to non- sensory associations like health and context. The modern day context of food choice requires more subtle choices among non-poisonous food products. This requires adequate “top-down” control using context associations in addition to the taste-conditioned approach-avoidance tendencies. The current method of associative priming combined with electroencephalography is a suitable measure to study such subtle differences among these “top down” signals affecting food choice. The recurring importance of taste in choosing healthy foods suggests that these concepts and associations may be promising targets for future marketing interventions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HRH JJ GJTH MML. Performed the experiments: HRH. Analyzed the data: HRH JJ. Contributed reagents/materials/analysis tools: HRH JJ MML. Wrote the paper: HRH JJ GJTH MML.

# Introduction Obesity is associated with an increased risk of numerous co-morbidities, such as type 2 diabetes, hypertension and cardiovascular diseases, collectively referred to as metabolic syndrome, and cancer. The mechanisms involved are being clarified, following the seminal discovery that adipose tissue, far from being a passive reservoir for the accumulation of lipids, is an endocrine organ that produces dozens of factors that regulate several aspects of organism homeostasis. Consistently, mounting evidence links the production of pro- inflammatory factors, by excess adipose tissue, to the development of a systemic chronic inflammatory state that contributes to the development of obesity- associated morbidities. Studies in mice carrying fat-specific disruption of the insulin receptor gene (FIRKO mice) demonstrated how the insulin signaling pathway plays a major role in the homeostasis of adipose tissue, while its subversion leads to pathological conditions. FIRKO mice display 50% reduction in adipose tissue, despite normal food intake, and are protected from age-related and hyperphagia-induced obesity, and obesity-related glucose intolerance. These metabolic changes correlate with extended longevity , thus corroborating the idea that reduced fat mass, even in the presence of normal or increased food intake, can reduce obesity-associated metabolic alterations and extend lifespan. The notion that insulin signaling in adipose tissue is critical in the regulation of lifespan received further support by findings that sirtuin 1 (SIRT1), encoded by the mammalian orthologue of SIR2 - a yeast life-extending gene, inhibits adipogenesis by repressing PPAR-γ, an insulin-dependent master regulator of fat cell development. It has been known for many years that caloric restriction (CR), with adequate nutrition, improves cardiometabolic health, prevents tumorigenesis, and increases life span in experimental animals. One relevant issue is whether CR exerts its beneficial effects *per se* or through the reduction of adipose tissue mass. Studies in FIRKO mice, support the latter possibility, while studies in genetic models of CR have not yet addressed the issue. Here, we report one such a model, eps8-null (Eps8KO) mice, which unexpectedly links actin dynamics to individual variations in bodyweight, metabolic status and longevity. Eps8 is a multimodular protein involved in actin remodeling, through several activities, including regulation of Rac, a pivotal GTPase in the control of actin dynamics, and direct interaction with actin. Through this latter property, Eps8 exerts both actin barbed end capping and actin bundling activities. We systematically analyzed Eps8KO mice to unmask possible defects pointing to phenotypes related to human disease. This analysis revealed that Eps8KO mice weigh less than WT mice and are partially resistant to aging and diet-induced obesity. Moreover, Eps8KO mice display a CR phenotype, accompanied by increased insulin sensitivity and an improved metabolic status, which are likely responsible for the increased lifespan observed in Eps8KO mice. This negative energy balance in Eps8KO mice is not caused by decreased food intake or increased energy expenditure, instead it correlates with decreased intestinal absorption and reduced intestinal microvilli length. Thus, the actin remodeler Eps8 is required for proper microvilli morphogenesis and loss of Eps8 in mice leads to altered intestinal function, improved metabolism and increased lifespan. # Results ## Eps8KO Mice Are Predisposed to Leanness Eps8KO mice displayed a significant reduction in bodyweight, when compared to wild-type (WT) mice, something that became much more evident during aging or high-fat diet-induced obesity. To gain insights into the nature of these phenomena, we used Soxhlet analysis to measure whole body fat content and lean body mass in Eps8KO mice. Reduced bodyweight was due to a reduction in both the fat and the lean mass in young and, more pronouncedly, in old Eps8KO mice. When the weight of individual organs was measured as a fraction of total body weight, however, we observed that for the majority of inner organs there was no significant difference between Eps8KO and WT mice. Instead, a significant reduction in inguinal, epididymal and scapular fat, relative to total body weight, was observed in Eps8KO mice. Importantly, total brain weight was unaffected in Eps8KO mice, which resulted in an increase of its fractional contribution to body weight, in older mice. We concluded that Eps8KO mice show a lean phenotype. In addition, results from the detailed measurements of organ weight are compatible with the possibility that these mice are functionally calorie-restricted. Mice subjected to alimentary CR, in fact, show an absolute reduction in both lean and fat mass, and a relatively higher proportional loss of fat mass, while the brain is the sole organ that does not show a reduction in weight. The possibility that Eps8KO mice are calorie-restricted will be further analyzed, and discussed, in a following section. ## Improved Metabolic Status in Eps8KO Mice Increased bodyweight correlates inversely with insulin sensitivity and overall metabolic status, which in turn affects the risk of type 2 diabetes and of cardiovascular diseases. Accordingly, Eps8KO mice displayed increased insulin sensitivity both at a young and an older age and after high fat diet; however, they displayed no increased glucose tolerance. Insulin levels were lower in Eps8KO mice at all ages and for all dietary regimes. Thus, in Eps8KO mice, reduced bodyweight both during aging- and diet- induced obesity correlates with improved insulin sensitivity. We next compared metabolic parameters of Eps8KO and WT mice. While only insulin and leptin levels were reduced in Eps8KO mice fed a normal diet (8% reduction in bodyweight compared to WT mice), after a high fat diet (20% reduction in bodyweight), Eps8KO mice additionally displayed significantly reduced glucose, triglyceride and cholesterol values. A similar improvement in blood parameters was observed in Eps8KO mice after an overnight fast. We concluded that Eps8KO mice display an overall improved metabolic status. ## Eps8KO Mice Display Reduced Fat Absorption One possible cause of reduced bodyweight is reduced food intake/assimilation. We did not detect significant differences in food intake in Eps8KO mice, either when fed on normal chow or on a high fat diet. Similarly, we did not detect alterations in the amount of feces produced by Eps8KO mice. Instead, determining the calorific value of the feces using a combustion calorimeter, we found that the feces of Eps8KO mice displayed a higher energy content than those of normal mice, both on a normal and on a high fat diet. This increase in fecal energy content was not due to altered intestinal permeability or intestinal transit time. Collectively, the above data suggest that Eps8KO mice absorb less. To investigate this possibility directly, we initially measured fat absorption, since increased fecal calorie content was more pronounced when Eps8KO mice were fed a high fat diet. Using the fecal dual isotope method in which the non- absorbable sitostanol serves as an internal normalizer, we observed a significant reduction in oleic acid absorption in Eps8KO mice. Next we measured intestinal absorption using the plasma dual isotope method in which plasma lipases were inhibited to allow accumulation of the absorbed lipids. Oleic acid and triolein (a triglyceride analog) absorption was equally reduced, indicating that the action of intestinal lipases is not impaired in Eps8KO mice. To address whether the absorption deficit was specific for fat or whether absorption of other nutrients was also impaired in Eps8KO mice, we used the everted sac model. Uptake of ¹⁴C-MDG, ¹⁴C-Gly-Sar was similar in WT and Eps8KO mice, while ¹⁴C-oleic acid uptake was reduced in intestinal sacs from Eps8KO mice. Permeability to ³H-mannitol was negligible throughout the experiment. Thus, while sugar and peptide absorption is normal, fat absorption is specifically impaired in Eps8KO mice. ## Energy Expenditure of Eps8KO Mice The above data show that reduced food assimilation, in particular fat, is associated with the lean phenotype of Eps8KO mice. However, increased energy expenditures might also contribute to this phenotype. Daily energy expenditure, determined as oxygen consumption at room temperature, was increased in Eps8KO mice. Next, we measured the metabolic rate at 30°C (TNZ, thermal neutral zone), where energy expenditure for thermal regulation is minimal. Under these conditions, we did not detect any significant difference between WT and KO mice. We also measured locomotor activity. On a 24 hr schedule, activity was not significantly increased in Eps8KO mice, although we noticed a trend towards increased activity during the light phase (*P* = 0.06), but not during the dark phase, in Eps8KO mice. This initial set of results indicated that there is increased energy expenditure in Eps8KO mice. However, this could not be immediately ascribed to increased locomotor activity or to increased metabolic rate when the need for thermogenesis was minimized at the TNZ. One possibility, to explain the increased energy expenditure is that Eps8KO mice invest more than WT in thermoregulation, possibly due to reduced insulation. To gain more insights into this issue, we measured core body temperature (Tb). In Eps8KO mice, there was reduced Tb during the light but not during the dark phase. In addition, we found that, in Eps8KO mice, core body temperature inversely correlated with body weight, while in WT mice no such correlation was observed. These results suggest that, in Eps8KO mice, energy resources are limited and individual mice allocate available energy resources preferentially either into fat storage or into thermogenesis. To test this hypothesis directly, we fed mice a high fat diet and measured core body temperature at the beginning and at the end of the experiment. As expected, under conditions of increased caloric intake, core body temperature in Eps8KO mice rose to levels of WT mice, while it remained unvaried in mice fed a normal diet. These results further support the hypothesis that limited energy resources are at the base of the reduced core body temperature in Eps8KO mice. Interestingly, reduced core body temperature is a phenotype also displayed by calorie-restricted mice, an observation that further supports the possibility that Eps8KO mice might be functionally calorie- restricted. ## Eps8KO Mice Are Calorie-Restricted To corroborate our hypothesis, we employed several approaches. Calorie- restricted mice show a characteristic gene expression profile, reflecting alterations in metabolic pathways due to reduced availability of nutrients. We analyzed the gene expression profiles in livers from Eps8KO and WT mice. Eps8 is not expressed in liver, thus differences in gene expression should reflect metabolic changes rather than the direct effect of the lack of Eps8. We found 20 genes differentially expressed in between WT and Eps8KO livers (cut-off ± 1.6 fold, *P*\<0.05). We validated, by quantitative PCR, nine genes (four upregulated and five downregulated in the Eps8KO/WT comparison), and found 100% concordance. We then compared this expression profile with a number of liver expression profiles published for the CR phenotype of mice. As shown in, there was a very significant overlap between the genes differentially expressed in livers of Eps8KO vs. WT mice and those differentially expressed in livers of CR vs. control mice. In addition, 15 of the 20 genes of our profile were concordantly and significantly differentially expressed in at least one of four published CR datasets (see also). It should be noted that in our expression profiling analysis the number of significantly differentially expressed genes was lower than in the published studies that we used for our comparisons. One obvious possibility is that differences in the profiling platforms employed in the various studies account for the differences. It should also be noted that Eps8KO mice represent a “chronic” condition of calorie restriction, in which adaptive changes might have occurred. In addition, Eps8KO are only moderately and selectively calorie- restricted (the absorption defect is fat-specific). The other studies reported conditions of acute (from 24 h to 8 weeks) and massive calorie restriction (starvation), in which metabolic and gene expression changes are more likely to be of a vaster magnitude. Regardless, the metabolic changes occurring in the livers of Eps8KO mice closely resemble those occurring under condition of CR. If Eps8KO mice were indeed functionally calorie-restricted they should display increased longevity. Indeed Eps8KO lived significantly longer than WT mice. The median survival of Eps8KO mice was increased by 26% (*P* = 0.005), mean survival by 37% (*P* = 0.00071) and maximum survival by 9% (*P* = 0.012). Lifespan was significantly extended both in male and female Eps8KO mice (–B). Increased lifespan in genetically modified mice has been linked to either a reduction in free radical production or alterations in the insulin signaling pathway. Moreover, Eps8 has recently been suggested to control ROS production directly. However, we did not detect altered ROS production in primary fibroblasts or in macrophages derived from Eps8KO mice. Similarly, we did not detect alterations in insulin signaling in fibroblasts or adipocytes devoid of Eps8. We concluded that the longevity phenotype of Eps8KO mice is not due to direct alterations of ROS production or insulin signaling, and it can therefore most likely be ascribed to the functional CR obtained in these animals. ## A Microvillar Morphogenetic Defect in Eps8KO Mice We searched for the possible causes of functional CR in Eps8KO mice. Since these mice absorb less fat, we directed our attention to possible intestinal alterations. In the mouse, Eps8 is expressed in the bowel, both in the small intestine and, to a lesser extent, in the colon. Histological examination of the small intestine did not reveal gross differences between Eps8KO and WT mice, and a morphometric analysis showed no differences in villus length, crypt height, or number of alcian-blue positive goblet cells. Similarly, no evident alterations were detectable in the large intestine of Eps8KO mice (data not shown). These results suggest that more subtle changes might be responsible for the phenotype of Eps8KO mice. We analyzed the subcellular localization of Eps8 in intestinal cells. We found that EGFP-Eps8, but not EGFP alone, colocalizes with F-actin to the apical membrane of differentiated intestinal Caco-2 cells, suggesting that Eps8 is localized to intestinal microvilli. Microvillar localization of Eps8 was confirmed by biochemical fractionation of the intestinal brush border membrane. Thus, we analyzed the morphology of intestinal enterocytes of Eps8KO mice at the ultrastructural level. Microvilli in Eps8KO enterocytes appeared disorganized and were significantly shorter than their WT counterparts. This was not accompanied by a general disorganization of the actin cytoskeleton as assessed by phalloidin staining for F-actin (data not shown). Since microvilli serve to augment the absorptive surface of the intestine, their reduction in Eps8KO mice is compatible with the absorption defect and with the calorie restriction phenotype that we observed in these animals. # Discussion In this study, we show that the genetic removal in mice of a regulator of actin dynamics, Eps8, leads to a complex phenotype characterized by leanness, improved metabolic status and increased lifespan. All these phenotypes most likely rest on the fact that Eps8KO mice are functionally calorie-restricted. Mechanistically, our data are compatible with a model in which the lack of Eps8 causes an alteration in microvillar morphogenesis and hence in the absorptive function of the intestinal brush border. The altered absorptive function of the intestine in turn, leads to a limited but constant reduction in energy intake, which is not compensated by increased food intake. The first result of the reduced energy intake is a reduction in bodyweight with consequent leanness. The reduction in weight is, in part, due to loss of adipose tissue. The concomitant reduction in insulation leads to a greater caloric demand for thermoregulation, which, when combined with the reduced energy intake, causes a CR-like state in the mice over time. As a consequence, insulin sensitivity is increased, glucose levels are lowered, and core body temperature decreases. These phenotypes are characteristic of an improved metabolic status that is most likely responsible for the increased lifespan detected in Eps8KO mice. While this scenario needs further experimental confirmation, first and foremost the reproduction of the phenotype in an intestine-specific KO strain, our results unexpectedly link a well-characterized regulator of actin dynamics to individual variations in bodyweight, metabolic status and longevity, and mark this process as a potential co-determinant in the pathogenesis of obesity and of its co-morbidities. ## Eps8 Is Essential for Proper Microvillar Morphogenesis Our findings identify the process of microvillar morphogenesis as the most likely initial step in the determination of the lean phenotype of Eps8KO mice. How does Eps8 control this process? Microvilli are highly dynamic structures undergoing constant remodeling through actin treadmilling. Consistently, even when microvilli have reached their final length and become bundled by cross- linkers, actin monomers are continuously added at the barbed end of the actin filaments, facing the microvilli tips, and dissociate from the pointed end. By controlling this process, actin binding and bundling proteins are crucial regulators of microvillar length and architectural organization, respectively. Eps8 is a multimodular actin remodeling protein, exerting both actin barbed end capping and F-actin cross-linking and bundling activity. The switch between these activities is regulated through interaction with its binding partners: association with Abi-1 activates Eps8's capping, association with Irsp53 its bundling activity. A further degree of regulation is exerted by phosphorylation regulating both Eps8's association with F-actin as well as its capping activity. Thus, Eps8 represents an actin remodeler that can be dynamically regulated to modulate its function. In C.elegans, the bundling activity of Eps8 is essential for microvilli formation, whereas the capping activity is dispensable. Whether the same holds true for mammals needs to be demonstrated. The fact that the capping activity of mouse Eps8 is 20-fold higher than the one of worm Eps8 might suggest that for mammalian microvilli morphogenesis not only the bundling but also the capping activity is needed. It is worth pointing out that in addition to Eps8 two other Eps8 gene family members, that display similar biochemical activities of Eps8, are expressed in the intestine, suggesting a further and unexplored level of complexity in the regulation of microvilli growth. Single and combined knockout mice for the various Eps8 family members will be needed to dissect the specific and redundant functions of the Eps8 family members in microvilli morphogenesis. How do other actin binding and bundling proteins fit into this picture? *In vitro* work in cell lines had suggested an important role for Villin and Espin, two F-actin bundling proteins, enriched in the intestinal brush border, in microvillus morphogenesis. Instead, neither knockout mice show any microvillar alterations, suggesting that, *in vivo* the loss of these two proteins is compensated by other F-actin bundlers. Instead, loss of Formin (Plastin 1), an F-actin bundler, which additionally, through its interaction with Keratin19, links the microvillar actin rootlets to the terminal web, leads to shortened microvilli and reduced transmembrane resistance in mice. Surprisingly, Formin knockout mice do not show any metabolic alterations. Since in these mice reduced microvilli length is accompanied by increased permeability, it is tempting to speculate that the two alterations balance each other, leading net to normal nutrient uptake. ## Determination of the Calorie-Restricted Phenotype of Eps8KO Mice Based on our data, we propose that the initiating event in the generation of the complex phenotype of Eps8KO mice is the reduction of the intestinal surface area available for nutrient absorption. However, we detected a specific fat absorption defect in Ep8KO mice, while sugar and peptides were equally well absorbed in comparison to WT mice. One possible explanation for the selectiveness of the defect might reside in the fact that sugars and peptides are transported across the apical surface of enterocytes via specific carriers. Such active transport mechanisms might compensate for the reduction of the transport area, either by increasing their rate or their concentration (although we did not measure these variables directly). Conversely, fatty acids enter via free diffusion, and thus a reduction in surface area would directly impinge on their influx rate. A number of issues require further discussion. First, why do Eps8 mice not compensate the reduced absorption by eating more? Food intake is centrally regulated and Eps8 is expressed in the brain, raising the possibility that alterations in the CNS are at the basis of the phenotype. Although, we did not directly investigate this possibility, it is worth noting that we have previously shown that neuronal Eps8 controls at least one type of behavioral response, i.e. that to ethanol. Other, not mutually exclusive, possibilities must also be contemplated. For instance, gut signals - both hormonal and nutrients - feed back to the brain to regulate food intake ; it is possible that the lack of Eps8 could impinge on the regulation of gut-derived signaling to control food intake. Finally, Eps8KO mice might not feel very hungry. The reduction in intestinal absorption is fat specific meaning that normal circulating levels of glucose and amino acids are still available to signal satiety. It should also be noted that, despite the fact that Eps8KO mice do not eat more than WT littermates in absolute terms, they do so, when food intake is normalized to bodyweight. This possibly reflects the fact that they react, at least partially, to lower levels of leptin and insulin. The fact that the differences that we observe are rather small raises another interesting point: is the absorption defect sufficient to explain the CR phenotype? The effect of CR on lifespan is observed at 20–40% reduction in food intake, a condition not likely mimicked by the absorption defect of Eps8KO mice. Thus, other factors possibly contribute to the negative energy balance. From our data, the prime suspect is heat loss, consequent to reduced thermoinsulation due to diminished fat mass, as also supported by the finding that Eps8KO mice invest more than WT littermates for thermoregulation. Reduced absorption and increased heat loss would lead to a slow, but progressive, negative energy balance responsible for the functional CR of Eps8KO mice. Finally, and compatibly with their CR phenotype, Eps8KO mice display improved metabolic status, which is the most likely cause of their increased lifespan. Indeed, we could not evidence any alterations in insulin signaling or in ROS production in primary cells derived from Eps8KO mice. This finding argues that Eps8 does not play a direct role in the classical pathways implicated in the regulation of lifespan, and that the longevity phenotype is likely a consequence of the functional CR of these mice. In further support of this possibility, we could evidence alterations in the gene expression profile of livers from Eps8KO mice that were similar to those detectable under conditions of CR. The absence of Eps8 expression in the liver definitely rules out a direct impact of Eps8 on biochemical pathways potentially leading to CR and increased lifespan, and further supports the notion that these phenotypes can be ascribed to the improved metabolic status of the KO mice. ## Implications Our findings raise the possibility that actin dynamics might play a previously unsuspected role in the co-determination of obesity and of its associated morbidities. Whether this correlation reflects a real occurrence in humans is presently a matter of speculation; however, based on our results, an investigation of polymorphisms in the *Eps8* gene, which might point to possible hypomorphic alleles, seems warranted. Such an analysis should be extended to other Eps8-family members, since in principle hypomorphic alleles of these genes might have a similar impact to that of *Eps8* on microvillar morphogenesis and on the ensuing phenotypes. Our data suggest that small changes in the daily energy balance are sufficient to gradually improve the metabolic state and prolong lifespan. This observation is important as it is generally thought that severe caloric restriction of 20–40% is necessary to achieve such effects. Our findings might open perspectives for much more feasible therapeutic strategies than the rigorous dieting regimens that few people are able to follow. # Materials and Methods ## Animal Housing, Diet and Aging Mice were housed on a 12-h light/dark cycle and had *ad libitum* access to water and food. Eps8KO mice were backcrossed for ten generations into C57BL/6 mice and then used for the aging experiments. All other experiments were performed after an additional 8 generations of backcross. Both mice from heterozygous crosses (diet, metabolism, gene expression profile, everted sac) and homozygous F2N18 colonies (all remaining experiments) were used. Male mice were used unless otherwise indicated. For the diet experiments, mice were single housed at 10 weeks of age. After one week of habituation mice were placed either on a normal chow (Harlan Teklad Global 2018) or on a high fat diet (60% fat, D12492, Research Diets Inc.) and food intake and bodyweight were monitored for 10 weeks, twice weekly. All experiments were performed in accordance with the guidelines established in the IFOM-IEO Campus Principles of Laboratory Animal Care (directive 86/609/ECC). ## Analysis of the Metabolic Status For glucose tolerance tests, mice were fasted overnight for 16 h and then injected intraperitoneally with 2 g/kg body weight glucose. For insulin tolerance test (ITT), mice were fasted for 5 h and then injected with 0.4 U/kg body weight (for 3 month old mice) or 0.5 U/kg body weight (for 6 and 12 month old mice) of human insulin (Eli Lilly) into the peritoneal cavity. Overnight fast was chosen for GTT since mice are night active and most of the food intake occurs during the night.; for ITT a 5 hrs fast was chosen because prolonged fasting causes the blood glucose values to drop too much after insulin administration. Blood glucose values were determined by an automatic glucose monitor (Glucotrend 2, Roche) and Accu-Chek active bands (Roche). Plasma parameters reported in were measured as described in. ## Analysis of Intestinal Absorption Everted intestinal sacs (1–1,5 cm length) were prepared from the upper part of the small intestine of WT and Eps8KO mice under constant oxygenation as described. Details are in. For fecal dual isotope experiments, mice were starved overnight and then received a gastric bolus of 100 µl cornoil containing 1 µCi ¹⁴C-Oleic Acid and 1 µCi non-absorbable ³H-Sitostanol. Feces were collected for 4 days every 24 hrs. Samples were homogenized in 3 M KOH in 60% EtOH and incubated over night at 60°C under shaking to solubilize fatty acids. Ten ml Hionic Fluor (Perkin Elmer) was added to 100 µl aliquots, left in the dark at room temperature for 4 h and then counted using a liquid scintillation counter. Absorption was calculated as: For plasma dual isotope experiments, mice were starved overnight, injected with 500 mg/kg Tyloxapol to inhibit plasma lipases and after 10 min given an intragastric bolus of 100 µl cornoil containing 2 µCi ¹⁴C-Oleic Acid and 2 µCi Triolein. Blood was sampled at the indicated time points and 10 µl were solubilized in 100 µl Isopropanol/Soluene-350 (1∶1) for 30 min. Ten µl 30% peroxide was added and samples were incubated overnight at 37°C in tightly capped vials. Samples were cooled to room temperature and 10 ml Hionic Fluor (Perkin Elmer) was added. Samples were left in the dark at room temperature for 4 h and then counted using a liquid scintillation counter. *In vivo* intestinal permeability was measured as described in An et al.. Intestinal transit time was measured as described in Friebe et al.. Details are in. ## Analysis of Energy Expenditure For the measurement of oxygen consumption and carbon dioxide production, mice were placed inside a climate chamber. Gas concentrations were measured by sucking compressed air through custom made metabolic chambers. Further details are in. Core body temperature (T_b) and activity were monitored using implanted thermosensitive transmitters. For the determination of total daily energy expenditure (DEE) at room temperature mice were kept in metabolic cages with food and water ad libitum. Red mouse igloos (Plexx, NL) were offered as shelters. The measurement started in the afternoon of day one and was terminated in the morning of day three. This procedure enabled a recording of a complete data set of resting (photophase) and activity (scotophase) during a period of 24 consecutive hours. For the determination of the thermoneutral zone, metabolic rate was monitored within a wide range of ambient temperatures from 6 to 34°C. Details are described in. ## Histology and Ultrastructural Analysis Intestinal segments from WT and Eps8KO mice were fixed in 10% buffered formalin, processed and embedded in paraffin. Five µm tissue sections were stained with H&E or Alcian Blue. Morphometric analysis was with the image analysis software Image-Pro Plus (version 4.5, Media Cybernetics, Silver Spring, USA). Ultrastructural analysis was essentially as in Croce et al.. Briefly, intestinal pieces of 1×1 mm², collected from the proximal duodenum of 4 WT and 4 Eps8KO male mice of 6 months of age, fed ad libidum on standard chow. Samples were first fixed in 2.5% glutaraldheyde for 2 h at room temperature, postfixed in osmium tetroxide for 2 h and then in uranyl acetate for another hour. Subsequently, samples were dehydrated through a graded ethanol series and propylene oxide and embedded in resin (Poly-Bed; Polysciences, Inc., Warrington, PA) overnight at 42°C and 2 days at 60°C. Ultrathin sections (50 nm) were collected, stained with uranyl acetate/lead citrate and observed under an electron microscope (model CM10 or Model G2 Tecnai; Philips, Eindhoven, The Netherlands). Morphometric analysis was performed arbitrarily in ten different regions per tissue block in which the intestinal tight junctions were well preserved as a read-out of the intactness of the tissue. A total length of 70 µm intestinal surface per animal was analyzed. ## Statistical Analyses Means were compared between WT and Eps8KO mice using two-tailed Student's *t* test. Values in the text are means ± S.E.M. Differences were considered significant at *p*\<0.05. ## Cell Lines and Cellular Biochemistry Preparation of PEFs and macrophages, and conditions for the cultivation of 3T3-L1 and Caco-2 cells (obtained from ATCC-LGC) are in. Preparation of brush border membranes, determination of ROS, and IB analysis are also in. ## Gene Expression Analysis Glass cDNA-chips were produced as recently described. A full description of the approximate 21.000 probes on the microarray is available in the GEO database (GPL4937). The expression data have been submitted to the GEO database (GSE14454). For preparation of total RNA, individual organs were thawed in a buffer containing chaotropic salt (RLT buffer, Qiagen) and homogenized using a Polytron homogenizer. Total RNA from individual samples was obtained according to the manufacturer's protocols using RNeasy Midi kits (Qiagen). Microarray slides were hybridized and processed as described previously. The TIGR Microarray Data Analysis System \[TM4)\] was used for normalization \[MIDAS; \] and identification of genes with significant differential regulation \[SAM, Significance Analysis of Microarrays; \]. Expression data were normalized by performing a total intensity normalization to transform the mean log₂ ratio to zero. To eliminate low-quality array elements several filtering methods were applied. They included: background checking for both channel with a signal/noise threshold of 2.0, one bad tolerance policy parameter and flip dye consistency checking. Validation was performed by quantitative PCR using Roche chemistry on selected genes. ## Gene Expression Meta Analysis Microarray gene expression data of Pohjanvirta et al. (GSE9121, Affymetrix RAT230 2.0 chip), and Bauer et al. (GSE858, custom two channels cDNA microarray) were downloaded from Gene Expression Omnibus (GEO, [www.ncbi.nlm.nih.gov/geo/](http://www.ncbi.nlm.nih.gov/geo/)). Normalized data were log2 transformed and analyzed using GeneSpring GX 7.3 (Agilent Technologies, USA). Statistical analysis was performed using a two-samples parametric Welch's t-test (variance not assumed equal) followed by multiple testing correction (Benjiamini and Hochberg False Discovery Rate). All genes with a *P*-value less than 0.05 were considered to be significantly regulated. In the Bauer et al. study, a direct design for two colors cDNA microarray experiment was employed (i.e. the treated versus control condition were analyzed directly on the same chip) and the sole gene expression profile of control mice, as in the Pohjanvirta et al. study, was not available. Therefore, in order to analyze this dataset consistently with the analysis of Pohjanvirta et al. (i.e. two-samples Welch's t-test with multiple test correction), we compared the 48 h starved group with the 48 h sugar-supplemented group. # Supporting Information We greatly appreciate the help of Pietro Transidico for confocal analysis and Laura Tizzoni and Valentina Dall'Olio for genotyping. We thank the other members of the German Mouse Clinic for comprehensive phenotyping of the Eps8KO mice and fruitful discussions. We thank Pascale R. Romano for critically editing the manuscript and Enrica Migliaccio for stimulating discussions. [^1]: Conceived and designed the experiments: AT JR RE BR MH ES CT PPDF NO. Performed the experiments: AT CBES FZ MCG BP NE BR CM MH ES NO. Analyzed the data: AT CBES FZ FB MCG BP JR RE NE BR CM MH ES CT NO. Contributed reagents/materials/analysis tools: HF VGD GS. Wrote the paper: GS PPDF NO. Provided technical supervision of the mouse clinic: HF. Provided scientific administrative supervision of the mouse clinic: VGD. Provided supervision of microarray experiments: JB. Provided supervision of metabolism experiments: MK. Provided supervision of blood chemistry experiments: EW. Provided supervision of the mouse clinic: MHA. [^2]: Current address: KtedoGen srl, Milan, Italy [^3]: Current address: Department of Physiology, University of Fribourg, Fribourg, Switzerland [^4]: Current address: Sanofi-Aventis GmbH, Frankfurt am Main, Germany [^5]: Current address: Department of Medical Genetics, Cedars-Sinai Medical Center, Los Angeles, California, United States of America [^6]: Current address: Animal Facility, Max Planck Institute of Biochemistry, Martinsried, Germany [^7]: RE was employed by Sanofi-Avensis at the time the manuscript was written. Sanofi-Avensis was not involved in study design, data analysis/interpretation or preparation of the manuscript. Sanofi-Avensis has no financial or intellectual interest in any of the work described. The authors fully adhere to PLoS ONE policies on sharing data and materials.

# Introduction Schnitzler syndrome (SS) is a rare inflammatory condition characterized by urticarial-like rash, fever and monoclonal IgM gammopathy. A role of interleukin (IL)-1β has been suggested, and IL-1 receptor antagonist anakinra was tried successfully in SS on the basis of its efficacy in treating some hereditary autoinflammatory syndromes, especially cryopyrin-associated periodic syndrome. This supported the hypothesis that the inflammasome could play a crucial role in these diseases. Studies on cytokines production in SS patients are scarce and *in vitro* and *in vivo* effects of anakinra on their production are not known. We report 2 cases of SS successfully treated with anakinra. We determined, before and during treatment with anakinra, IL-1β, IL-6 and TNFα plasma levels and *in vitro* production by PBMCs of these cytokines with and without stimulation by lipopolysaccharide (LPS) and with and without *in vitro* anakinra. # Methods ## Cytokines Concentrations Assays Heparinized blood samples were obtained from two SS patients before and one month after starting subcutaneous 100 mg/day anakinra (*in vivo* condition) and from six healthy subjects (HSs) between January 2009 and January 2011. The blood sample obtained after one month of treatment with anakinra was taken 24 hours after the last anakinra injection and just before the next one. SS patients and healthy subjects gave written informed consent to take part in the study, which was approved by our IRB (Comité de Protection des Personnes, Nord). PBMCs were isolated by density centrifugation on Ficoll-Paque PLUS (GE Healthcare, Uppsala, Sweden). Cells were maintained in 24-well flat-bottomed plates with a final density of 1×10⁶/ml in a volume of 1 ml RPMI 1640 including 10% fetal bovine serum or patient plasma, 1% penicillin/streptomycin (Life Technologies, Cergy Pontoise, France) PBMCs were incubated with or without 100 pg/ml LPS (Sigma-Aldrich, Lyon, France) and with (“*in vitro*” anakinra condition) or without 500 ng/ml anakinra (SOBI). After 6 and 16 hours, supernatants were collected and stored at −80°C. Concentrations of IL-1β, IL-6 and TNF-α in plasma and supernatant were measured by enzyme-linked immunosorbent assay (Life Technologies, Cergy Pontoise, France). Results are expressed as the mean of triplicate measures of cytokine concentration (± standard deviation). ## RNA Quantitative Real-time qPCR PBMCs of 3 healthy subjects and of the second patient with SS, obtained before and 1 month after the start of anakinra, were kept at 80°C in microcentrifuge tubes. Extractions of RNA in these unstimulated PBMCs were performed using GeneJet RNA Purification Kit (Fermentas, K0731). Analyses were performed using AbiPrism 7500 Lighcycler system using 0.25 µM concentrations of specific primers and SYBR Green II PCR Master Mix (Qiagen). Specific primers for: HBMS forward (5′-GGCAATGCGGCTGCAA-3′) and reverse (3′-GGGTACCCACGCGAATCAC-5′); IL-1ß forward (5′-AAA-CCTCTTCGAGGCACA-AG-3′) and Reverse (3′-GTTTAGGGCCATCAGCTTCA-5′); IL-6 forward (5′-GGTACATCCTCGACGGCATCT-3′) and Reverse (3′- GTGCCTCTTTGCTGCTTTCAC-5′) were used for amplification in duplicate assays. PCR amplification was performed to control for sample loading and to allow normalization between samples. Quantitative comparison was obtained through the comparative Ct method. # Results ## Description of Patient 1 A 54-year-old male was referred to our institution with a 5-year history of recurrent episodes of urticarial lesions associated with fever up to 39°C. He also complained of relapsing arthralgia and bone pain. At his admission, laboratory investigations showed neutrophilic leukocytosis at 12×10⁹/L, and increased C-reactive protein (CRP) (6.0 mg/dl, N\<0.3). IgM-κ monoclonal gammopathy was also found. Tests for autoimmunity, infections and malignancy were negative. Based on these findings, we diagnosed SS. Anakinra was started at a daily dose of 100 mg. Within 48 hours, the patient had a dramatic clinical/biological response, with the resolution of skin symptoms and bone pain and normalization of CRP. During the following year, the patient experienced no episode of urticarial lesions and CRP was normal under treatment. ## Description of Patient 2 A 77-year-old female was referred to our institution with a similar history than patient 1, i.e. a 6-year history of recurrent episodes of urticarial lesions associated with fever up to 40°C. Her medical history included a severe osteoporosis. She also experienced relapsing bone pain. Laboratory investigations showed neutrophilic leukocytosis at 17×10⁹/L, and increased CRP (14.0 mg/dl, N\<0.3). IgM-κ monoclonal gammopathy was also found. Tests for autoimmunity, infections and malignancy were negative. Based on these findings, we diagnosed SS. Anakinra was started at a daily dose of 100 mg. Within 48 hours, the patient had a dramatic clinical/biological response, with the resolution of skin symptoms and bone pain and normalization of CRP. During the following year, the patient experienced no episode of urticarial lesions and CRP was normal under treatment. ## Production of Cytokines before Treatment, with and without *in vit*ro anakinra Spontaneous production of IL-1β, IL-6 and TNF-α by PBMCs was similar in the patients and the healthy controls and was almost undetectable ( baseline level), as were plasma levels of cytokines (data not shown). Stimulation of the patients PBMCs with their own plasma induced a negligible secretion of cytokines (data not shown). Before the first administration of anakinra, 6 and 16 hours of stimulation with LPS caused a larger release of cytokines by PBMCs from the patients than from the healthy controls but with different patterns between the patients. In patient 1, all 3 cytokines were released at a very high level when compared to healthy controls. In patient 2, although IL-1β and TNF-α levels were higher than in healthy controls, they were lower than in patient 1 and the peak of production was delayed (6 hours for patient 1 and at least 16 hours for patient 2). Conversely, IL-6 production was quite similar between both patients and much higher than in healthy controls. Adding anakinra *in vitro* reduced LPS-induced secretion of IL-1β and TNF-α after 6 and 16 hours in both patients but had no obvious effect on IL-6 in patient 1 while it decreased for more than 50% IL-6 production in patient 2. ## Production of Cytokines after 1 Month of anakinra Treatment with or without *in vitro* anakinra One month after the initiation of *in vivo* anakinra treatment, the release of cytokines was dramatically decreased both after 6 and 16 hours of stimulation with LPS in both patients and almost normalized in patient 2. The most impressive decreases were observed for LPS-induced IL-1β production after 16 hours of stimulation in patient 1 which was close to healthy controls, and for LPS-induced IL-6 production after 16 hours of stimulation in patient 2 which was also close to healthy controls. LPS-induced IL-1β release observed after 6 hours of stimulation for patient 1 and 16 hours for patient 2 was similar to the release, which was observed after 6 hours with *in vitro* anakinra before treatment in patient 1 and 16 hours in patient 2. When the patients PBMCs obtained after 1 month of anakinra treatment were incubated with additional anakinra *in vitro*, we did not observe any additional effect on cytokine production. ## Gene Expression of IL-1β and IL-6 before and after 1 Month of anakinra Before the start of anakinra, spontaneous gene expression of IL-1β, but not of IL-6, in unstimulated PBMCs was increased in the patient with SS when compared to healthy subjects. After 1 month of treatment with anakinra, gene expression of IL-1β in unstimulated PBMCs of the patient with SS was not different from the healthy subjects. Gene expression of IL-6 remained unchanged and similar to the healthy subjects. # Discussion In accordance with previous studies, we showed that plasma levels as well as spontaneous production by PBMCs of IL-1β, IL-6 and TNF-α were similar in the SS patients and the healthy controls. These results contrast with the frequent elevated spontaneous IL-1β production in some other autoinflammatory syndromes like the cryopyrin-associated periodic syndrome, and suggests that plasma level determination of these cytokines are not useful in the diagnosis and the follow- up of SS patients. Most studies showed that LPS alone was sufficient to overactivate IL-1β secretion in SS, suggesting an abnormal activation of the intracellular processes leading to production and secretion of inflammatory cytokines, perhaps through inflammasome-dependent mechanisms. Conversely, one study suggested that in some SS patients LPS could not increase the inflammatory cytokines production. Our study adds some line of evidence that LPS does induce an abnormal release of proinflammatory cytokines IL-1β, IL-6 and TNF-α by PBMCs in SS, with different kinetic patterns and magnitude according to the patients. In patient 1, all the 3 cytokines were produced at a high level after 6 hours of LPS stimulation. In patient 2, cytokines production after 16 hours of LPS was higher than after 6 hours and was very high for IL-6. These interindividual variations in cytokine responses observed in our patients had to be expected and may be explained by polymorphisms as well as gender and age differences. The high production of IL-6 after LPS stimulation is of particular interest since a recent study reported a complete remission after anti-IL-6 treatment in 3 patients who failed anti-IL-1β treatment. We observed that stimulation of the patients PBMCs with their own plasmas induced a negligible secretion of cytokines. This argues against the existence of a stimulatory factor in the plasma in SS. Although the spontaneous production of IL-1β and IL-6 by unstimulated PBMCs of SS patients was very low and not different from healthy subjects, we show that spontaneous IL-1β but not IL-6 gene expression was markedly increased in unstimulated PBMCs of SS patient when compared to healthy subjects. This spontaneous high gene expression of IL-1β in unstimulated PBMCs in SS could explain why LPS alone can trigger an efficient release of cytokine. Indeed, these findings suggest that PBMCs in SS patients have already undergone a first priming from extrinsic or intrinsic origin that allows substantial IL-1 β release in the absence of the ‘second hit’ usually required to trigger efficient release. The fact that the spontaneous gene expression of IL-6 in PBMCs was similar in SS patient and healthy subjects while IL-6 production after LPS stimulation was much higher than in healthy controls suggests that IL-6 production could be dependent on IL-1 β stimulation, by an amplification loop process. We report here for the first time the effects on LPS-induced cytokines release of anakinra both *in vitro* and after 1 month of subcutaneous *in vivo* administration in two SS patients. The direct and rapid *in vitro* effect of anakinra on IL-1β suggests that there is an inhibition of the known autostimulatory loop of IL-1β production. In keeping with this hypothesis, we show that the spontaneous gene expression of IL-1β in PBMC after 1 month of anakinra was normalized in SS patient 2. IL-1 trap (an inhibitor of IL-1) has also been shown to reduce IL-1β, IL-6 and TNF-α production by LPS-stimulated PBMCs in SS. These *in vitro* results suggest that inhibiting the IL-1 autostimulatory loop by blocking IL-1 receptors, can also inhibit TNF-α production in SS. Patterns of response differed between the 2 patients for IL-6 as its production was markedly decreased after *in vitro* anakinra in patient 2 and less impressively in patient 1. After 1 month of anakinra treatment, we observed a dramatic decrease in IL-1β production. For patient 1, after 16 hours of LPS stimulation, IL-1β release was ten-fold lower than IL-1β release observed at baseline and within range of the values observed in the healthy controls, as was IL-6 production. In patient 2, IL-1β, IL-6 and TNF-α production after LPS stimulation was similar to the healthy controls, suggesting that anakinra had restored a normal phenotype of LPS response in SS patients. Interestingly, dramatic decrease of IL-6 production after 1 month of anakinra treatment was confirmed by normal CRP values in both patients. In cryopirin associated periodic syndromes, *in vivo* anakinra also decreased the production of IL-1β by LPS-stimulated PBMCs, but *in vitro* data are lacking. The release of IL-6 and TNF-α was lower than with anakinra *in vitro* before treatment. This suggests that IL-6 and TNF-α production is also dependent on the long-term action of IL-1β on PBMCs rather than the immediate effect of IL1 receptor blockade. We also have to discuss here why we chose to obtain the blood sample 24 hours after the last anakinra injection and not just after. The aim of our study was to assess the long term and not the acute effect of *in vivo* anakinra on cytokines production in SS. By showing that even 24 hours after the last injection, LPS-stimulated cytokines production was improved in our patients (and even almost normalized in one SS patient), we suggest that anakinra is able to restore an almost normal PBMC phenotype in SS. Altogether, these results could explain the dramatic improvement of the clinical manifestations after the start of anakinra. When the patients PBMCs obtained after 1 month of anakinra *in vivo* were incubated with additional anakinra *in vitro*, we did not observe any additional effect on cytokine production Whether this result could translate in indications of uptitrating or not the dose of anakinra needs further study. Our study has some limitations. It involves a limited number of patients. However, SS is a very rare condition and studies focusing on cytokine production have included between 1 and 3 patients. To better understand the mechanisms of abnormal LPS-induced cytokine productions, it would also have been interesting to examine the levels of IL-1β in cells lysates to determine whether elevated pro-IL-1 levels were readily present in patient’s PBMCs leading to higher cytokine release upon inflammasome-dependant activation. The use of ultrapure LPS and/or other inflammasome- activators could also allow them to better understand the mechanisms involved in the enhanced release of IL-1β by patient’s PBMCs. IL-18 and IL-1α would also have been interesting to focus on. Our study confirms the pivotal role of IL-1β in SS and shows that the marked efficacy of anakinra on clinical symptoms is associated with a dramatic decrease of abnormal IL-1β, IL-6 and TNF-α production by LPS-stimulated PBMCs. Whether *in vitro* and *in vivo* effect of a treatment on cytokine production in SS could help anticipating its efficacy or failure warrants further studies. We are indebted to Pr Jacques Bienvenu for technical support. Nicholas Barton, a medical writer, provided editorial assistance to the authors during preparation of this manuscript. [^1]: The work was supported by a research grant from Swedish Orphan Biovitrum. This does not alter the authors’ adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: DL OR VDL EF CH VS EH ML PYH SD. Performed the experiments: VDL EF CH. Analyzed the data: DL OR VDL EF CH VS EH ML PYH SD. Contributed reagents/materials/analysis tools: VDL ML SD EF CH. Wrote the paper: DL VDL SD EH CH.

# Introduction Jaborandi is the vernacular name of several species of medicinal plants belonging to the families Piperaceae and Rutaceae that are native to Brazil and neighboring countries. In Brazil, the genus *Pilocarpus* Vahl (Rutaceae) comprises 15 species, 12 of which are endemic; most are found in the eastern part of the country at the center of the genus’ genetic diversity. According to the literature,*Pilocarpus* species contain many secondary metabolites, especially alkaloids. Many alkaloids have been identified, namely pilocarpine, pilosine, anhydropilosine, 3-nor-8(11) dihydropilocarpine, pilosinine, isopilocarpine, pilocarpidine, isopilocarpidine, isopilosine, epiisopilosine, epiisopiloturine, 13-nor-7(11)-dehydro-pilocarpine, *N*,*N*-dimethyl-5-methoxy- triptamine, *N*,*N*-dimethyl-triptamine, plastydesmine, (1*H)*-4-methoxy-2-quinolone, and dictamine. *Pilocarpus microphyllus*Stapf ex Wardleworth is native to the northern and northeastern regions of Brazil and grows in eastern Pará, northwestern and northern Maranhão, and Piauí. Within the genus, leaves of this species have the highest accumulation of pilocarpine (PIL) content, which can vary from 0.5% to 1%. It is one of the most important Brazilian medicinal species because PIL is used for treating glaucoma and xerostomia. Brazil is at present the only supplier of this ingredient for the international pharmaceutical industry,exporting tons of PIL hydrochloride and PIL nitrate every year. The species is cultivated as a crop in Maranhão and Piauí, although it is still harvested from wild populations in some localities. However, since 2008, *P*. *microphyllus* has been listed as an endangered species in the Brazilian flora.Propagation of cultivars for research and industrial applications is therefore of fundamental importance for the species’ biological conservation and for reducing pressure on wild populations. Many low-income communities depend on harvesting *P*. *microphyllus* during a particular season of the year, and companies use the species for industrial PIL extraction, so more information about the seasonality of alkaloid contents will benefit both groups. For many years, the biological activity of most of the alkaloids found in *P*. *microphyllus*, apart from PIL, remained largely unknown. However, another alkaloid, epiisopiloturine (EPI), has been of interest in the scientific community. EPI wasfirst identified in 1978 and is now considered a promising alkaloid for combatting schistosomiasis. The anti-inflammatory and antinociceptive activities of EPI have been characterized, and chemical parameters have been improved and developed for industrial-scale isolation and spectroscopic characterization. Our group is conducting other studies, such as the nanopharmaceutical application of EPI in liposome systems, and thermal characterization and preformulation of the prototype EPI with pharmaceutical excipients. EPI is obtained from the industrial biomass waste from industrial PIL production, so its pharmaceutical application results both in environmental benefits and an increased economic importance of the species. The principal aim of the present work was to identify, quantify, and evaluate seasonal changes in the two main imidazole alkaloids, PIL and EPI, in three populations (S01, S02, and S03) of cultivated *P*. *microphyllus* in the state of Piauí, Brazil, over one year, including the dry and rainy seasons. Morphological and molecular characterizations of the same populations were correlated with the alkaloid profiles to investigate the genetic diversity of cultivated collections of *P*. *microphyllus* in Piauí. # Materials and methods ## Plant material Samples of cultivated *P*. *microphyllus* were obtained from the collection maintained at the Anidro do Brasil Extrações S.A. farm (3°6′S, 41°47′W), which is a plantation situated in the municipality of Parnaíba, Piauí state, Brazil. The company gave the permission to conduct the study on its site and it has the Certificate of Regularity—CR fromBrazilian Environment Authority IBAMA (Brazilian Institute of Environment and Renewable Natural Resources).Voucher specimens were collected and identified by Dr. Ivanilza Moreira de Andrade (Department of Biology, Federal University of Piauí) and deposited in the herbarium of the Parnaíba Delta (HDELTA) at the Federal University of Piauí (UFPI), Campus Ministro Reis Velloso, Parnaíba, Piauí, Brazil, under the numbers 2866 (S01), 2869 (S02), and 2874 (S03). The names of botanical taxa and their authors follow the*Flora do Brasil* 2020 list; the description and illustrations were made from samples collected during the study. ## *Pilocarpus microphyllus* sampling and harvest Fifteen adult plants between 0.5 and 2.0 m tall were selected and identified according to their leaf color and general morphology. The plants were categorized into three groups, each with five morphologically similar specimens: S01 (plants 1–5), S02 (6–10), and S03 (11–15). S01 represented the “jaborandi green line,” a form of *P*. *microphyllus* informally recognized as distinct within the jaborandi extractive industry. The other two groups (S02 and S03) were designated the “jaborandi traditional line.” Samples for chemical analysis were harvested between the 25^th and 30^th days of each month over one year. Young branches were harvested with pruning shears. The material was dried in the sun until water content measured by an OHAUS^® MB45 moisture analyzer was less than 15%. ## Alkaloid extraction from *P*. *microphyllus* leaves Dried powdered *P*. *microphyllus* leaves (5 g) were extracted with chloroform in an alkaline solution of 10% ammonia hydroxide at pH 12. This mixture was stirred for 30 minutes (Orbital Shaker, Nova ética^® 109 model), filtered with cotton, and partitioned with a 5% sulfuric acid solution. The acid solution was collected and the leaves were re-extracted and partitioned again. The alkaloid-rich acid solutions were homogenized and analyzed by HPLC. ## High performance liquid chromatography (HPLC) analysis The alkaloid-rich acid solution was diluted (1:10) with the mobile phase (potassium phosphate, 5% KH₂PO₄, pH 2.5), filtered with a 0.45 μm pore membrane, and analyzed by HPLC (LaChron Elite^®, L-2000 system; Merck–Hitachi, Tokyo, Japan). The column was a Merck/Lichrospher^® 60 RP, select B, 5 μm, 250 × 4mm, with a flow rate of 1 mL/min and an injection volume of 20 μL. The oven was set to 50°C and a UV detector was used at 216 nm. External standards were used to identify and quantify the alkaloids. All solvents used in HPLC analysis were from Merck KGaA (Darmstadt, Germany). ## Alkaloid standards PIL was isolated, purified, and provided by Anidro do Brasil Extrações S.A. Company, an international supplier of this chemical. EPI was isolated and purified at the Federal University of Piauí (UFPI), using methods reported in our previous studies. PIL and EPI were dissolved in acetonitrile-formic acid 1% (100 μg/ mL) and analyzed by LC-MS(AmaZon SL system, Bruker Daltonics; Bremen, Germany)to confirm their structure and purity prior to their use as standards. The conditions used for mass spectrometry detection were as follows: electrospray under positive mode and nebulizing gas flow at 2.5 L/min, interface voltage at 4.5 kV, heat block temperature at 230°C, and helium as the collision induced dissociation gas at 17 kPa.All solvents used in LC-MS analysis were HPLC-grade solvents from Merck KGaA (Darmstadt, Germany).Nuclear Magnetic Resonance spectroscopy (NMR) was done to confirm EPI alkaloid and its three stereoisomers structures. NMR experiments were carried out using a Bruker Avance III 600 HD spectrometer,operating at 150.92 MHz for carbon and 600.13 MHz for protons; it was equipped with 5 mm Prodigy CryoProbe and pulse gradient units, which are capable of producing magnetic field pulsed gradients of 50 G cm^-1 in the z-direction. The NMR spectra were obtained at atemperature of 300 K in deuterium oxide (D2O) or methanol-d4; the chemical shifts were referenced withsodium trimethyl-silyl -\[2,2,3,3-d4\]-propionate (TSP) or tetra-methylsilane (TMS), respectively. Standard 1D ¹HNMR experiments, i.e. using 30° pulses, an acquisition time of 2.7 s, a relaxation delay 1 s, and 16 transients of a spectral width of 9600 Hz, were collected into 64 K time domain points. 1H NMR experiments for the samples in water solution were performed with water suppression, using excitation sculpting with gradients, and an acquisition time of 1.7 s, a relaxation delay of 2 s, and with 16transients of a spectral width of 10000 Hz, and were collected into 32 K time domain points. ## Environmental conditions All the sampled plants were cultivated in the same locality, under the same agronomic, irrigation, and climatic conditions (temperature, humidity, and rainfall). Rainfall was measured daily in mm with a calibrated rain gauge. The temperature (°C) and humidity (%) were recorded with a calibrated digital thermo hygrometer (Cole-Parmer^®, EW-90080-03 model) during harvests. The harmonic means were calculated for each parameter, taking into consideration the dry and rainy seasons. ## Morphological quantitative variables The following 11 morphological characters were measured in cm in the 15 plants: overall length and width of each parameters, imparipinnate leaf, length of the petiole, number of leaflets per leaf, length and width of the terminal leaflet blade, length and width of the terminal leaflet petiolule, length and width of one of the most distal pair of lateral leaflets, and number of secondary veins in the leaflets. ## Molecular markers The DNA extraction and inter simple sequence repeats (ISSR) analysis were carried out on samples from the same 15 individual specimens measured for morphology. DNA was extracted from silica dried leaf material, following a protocol modified for microtubes, using autoclaved sand and modified 2% cetyl trimethyl ammonium bromide (CTAB) buffer to disrupt the cell membranes. Secondary compounds were isolated with a 24:1 mixture of chloroform to isoamyl alcohol and precipitated with isopropanol. The pellet was cleaned three times with 70% ethanol. The pellet was dissolved in 100 μL of Tris- ethylenediaminetetraacetic acid (EDTA) buffer (10 mM Tris, 1 mM EDTA). DNA quality was checked through gel electrophoresis using 1% agarose with TBE buffer (Tris-borate EDTA) and ethidium bromide at 100 V for 30 min. Eleven polymorphic ISSR primers used were from previous studies. The samples were amplified by the polymerase chain reaction (PCR) using a modified version of the protocol. The DNA extracts were diluted 10-fold to optimize amplification. The PCR conditions were carried out with a final volume of 10 μL, containing 1.0 μL of DNA extract, 7.5 pmol of each primer, and 5 μL of TopTaq^™ Master Mix (Qiagen Biotechnology). The Esco^® Swift^™ MaxPro thermocycler program included an initial denaturing (pre-melt) at 94°C for 90 s, followed by 35 denaturing cycles at 94°C for 40 s, annealing at 47°C for 45 s, and extension to 72°C for 90 s, with an additional final cycle of 94°C for 45 s, 44°C for 45 s, and 72°C for 10 min. The amplified fragments were analyzed by electrophoresis in a 1.5% agarose gel with SB buffer (10 mM sodium hydroxide, pH 8.5 with boric acid) for 2 h at 100 V, stained with ethidium bromide (1.0 mg/L for 30 min), and destained in distilled water for 5 min. The agarose gel was visualized under UV light and a digital image was obtained using a photo documentation system (L-PIX, Loccus Biotecnologia). For each gel run, a negative (water) control was used to detect any contamination problems. Five to ten percent of the quantifications were randomly repeated in all cases to guarantee repeatability of the bands observed. The gels were analyzed using GelCompar II^® version 5.0 (Applied Maths NV, Saint-Martens-Latem, Belgium) to align the bands according to marker and to identify the fragments (200–1500 base pairs). The results were used to construct a genetic binary matrix in which each cell was assigned with presence (1) or absence (0) of a given fragment (band). This matrix was used for further analysis. ## Data analysis ### Chemical analysis All alkaloid assay data were expressed in percentages (%, w/w) as mean values ± standard deviation. The chemical results were checked for normality and homoscedasticity using the Shapiro-Wilk and Bartlett tests, respectively. When the assumptions were not met, the nonparametric Kruskal-Wallis test was used, followed by the Dunn test for multiple comparisons. The Tukey test was used to evaluate the statistical differences within and among samples. Differences were considered significant when *p*\< 0.05. All statistical analyses were performed with the ExpDes.pt package in R software version 3.2.3, and with SAS 9.3 software. Graphics were prepared using R and OriginPro 8.5. ### Morphological analysis Principal coordinate analysis (PCoA), using Euclidean distances between individuals computed from the matrix of 11 quantitative morphological variables, was used to explore the similarity relationships between individuals. A principal component analysis (PCA) was used to visualize the major trends of morphological variation in the data set. The “broken stick” test was applied to determine which principal components were significant in the analysis. Linear discriminant analysis (LDA) was used to analyze the same quantitative morphological matrix but with the fifteen individuals classified into two groups, one consisting of the "green" population S01 and the other consisting of the two "traditional" populations S02 and S03. For this analysis the columns of the data matrix (the variables) were standardized to zero mean and unit variance. All analyses were conducted with PAST version 2.02. ### Population genetic variability and structure The binary ISSR matrix of 15 rows and 111 columns was analyzed using GenAlEx 6.5 software to estimate the genetic variability within and among the three populations. Values for the following parameters were recorded: number of loci (*N*), number of exclusive loci (*NE*), proportion of polymorphic loci, mean expected heterozygosity (*He*), and Shannon Index (*I*).Analysis of molecular variance (AMOVA), as implemented in GenAlEx 6.5 based on estimation of the parameter PhiPT, was used to verify the degree of genetic structure within and among the three groups; 999 permutations were used for the significance tests. ### Morphological and molecular analyses A matrix of pairwise Gower distances between the 15 individuals, computed from the combined matrix of 11 quantitative morphological variables and 111 molecular ISSR markers, was analyzed with PCoA to investigate the similarity relationships between the individuals. Gower’s index is a dissimilarity measure that can combine binary and quantitative variables, as implemented in PAST version 2.02. # Results ## Alkaloid standards PIL and EPI structures were confirmed by LC-MS and compared with data from the literature: PIL(3S,4R)-3-ethyl-4-\[(1-methyl-1H-imidazol-5-yl)methyl\]oxolan-2-one; and EPI( 3R,4R)-3-\[(S)-hydroxy(phenyl)methyl\]-4-\[(1-methyl-1H-imidazol-5- yl)methyl\]oxolan-2-one. The HPLC chromatography profile of the alkaloid standards used in the seasonal study is showed in.The EPI alkaloid and its three stereoisomers showed in. had their structures confirmed by NMR. The NMR data results of these alkaloids are show in. ## Environmental conditions No precipitation was recorded during the dry season from August to December (0 mm). In the rainy season (January–July), there was recorded precipitation (109.69 mm). During the harvest in the dry season, the recorded mean temperature and humidity were 32.42 ± 2.03°C and 54.08 ± 9.66%, respectively. In the rainy season, the recorded mean temperature and humidity were 33.75 ± 0.95°C and 58.97 ± 5.47%, respectively. ## PIL seasonality The chromatogram profiles showed that PIL was the primary imidazole alkaloid in the cultivated samples. PIL content exhibited a significant variation between the groups (S01, S02, S03) and months according to the statistical analysis. It varied during the year and according to season and month, except in September, when the PIL contents for the three populations were the same.The PIL content of the three groups demonstrated normality (*p* = 0.9278) and homogeneity of variance (*p* = 0.3128) and was analyzed by time-divided installments by months. Differences among the three groups were compared using Tukey’s test, which verified that PIL content in all pairs of groups was significantly different at 5% (*p* = 0.000). The PIL content of S01 was highly significantly different from S02 and S03 using Tukey’s multiple comparisons of means. S02 and S03 also showed significant differences in PIL content, although they had similar chromatographic profiles. The S01 sample had an adequate PIL content for industrial extraction (\> 0.500%) during all months, with higher contents in January, February, September, and November. PIL levels remained constant during March, April, and May, and decreased to their lowest levels, nearly 0.500% w/w, in June and July. S02 showed a significantly high PIL content only in September, and content decreased to \< 0.500% in October, December, and from February to July. S03 had a significantly high PIL content in August and September. From October to February, the PIL content was adequate for industrial extraction, while from March to July it decreased (\< 0.500%), as with S02. The Tukey test did not show significant differences in the PIL content of S01 samples between the dry and rainy seasons (*p* = 0.5637). S02 and S03 showed significant differences between PIL content by season, as shown in (*p* = 0.0149 and *p* = 0.0030, respectively). The mean PIL values in S02 and S03 were not suitable for industrial extraction during most of the rainy season ( and). ## EPI seasonality The EPI content data did not meet assumptions of normality and variance homogeneity, so the nonparametric Kruskal-Wallis test was used to evaluate differences between the means, followed by Dunn’s test for multiple comparisons at the 5% significance level. EPI content differed significantly (*p* = 0.00037) among the three groups. Dunn’s multiple comparison test showed that S01 was different from S02 (*p* = 0.0000) and S03 (*p* = 0.0017). However, no significant difference in EPI content was found between S02 and S03 (*p* = 0. 0943). ## Molecular analysis ### Intrapopulation diversity and genetic relationships The 11 ISSR markers generated 111 loci (*N*) from S01, S02, and S03, with 8–12 obtained per primer pair (average of 10.9 loci/primer pair). The mean percentage of polymorphic loci (*P*) was 57.06%. The population with the greatest genetic variability was S03 (*P* = 67.56%, *He* = 0.274, *I* = 0.400), followed by S01 (*P* = 53.15%, *He* = 0.203, *I* = 0.302). All populations showed at least one exclusive (private) band. ### Genetic differentiation between populations The AMOVA showed that most of the genetic variation (91%) was expressed within the populations, with only 9% between them. Nevertheless, the permutation test showed a significant difference in among-population variance The pairwise population PhiPT values (lower hemimatrix) showed significant differences between S01 and S02 plus S03 (upper hemimatrix shows *p*-values). ### Morphological characterization of jaborandi (*P*. *microphyllus*) The following description of *P*. *microphyllus* combines the characteristics observed in the plants used in this study: shrub about 40 cm tall; stem winged, ridged and pubescent; stipule pubescent, 1–2 mm long; leaves alternate, 2.0–12.7 × 2.1–7.5 cm, imparipinnate, rarely paripinnate, chartaceous, petiole 0.3–3.7 cm, canaliculate, pubescent; winged, canaliculate, olive green, pubescent,1.3–11.9 cm; rachis1.0–15.5 cm, pubescent; leaflets 1–11, 1.1–8.0 × 0.9–8.0 cm, opposite, sessile except for the terminal leaflet, dark green adaxially and paler green abaxially, leaflet blade chartaceous, glabrous, elliptic to narrowly ovate, apex rounded to emarginated or retuse, base asymmetric to attenuate, rachis margin entire; leaflet venation brochidodromous, midrib prominent adaxially, planar or slightly prominent abaxially, secondary veins 6–13; terminal leaflet blade ovate, elliptic, apex rounded to emarginated or retuse, base attenuate; pellucid punctate glands present on the leaflets. Inflorescence a raceme, 13.5–39.2 cm long with small greenish-yellow flowers from March to July; fruit a dehiscent white capsule. The individuals of the three groups overlapped substantially when the first two principal components (54.8% of variance) derived from the 11 morphological variables were plotted, although principal components two and three (35.5% of variance) partially separated S01 from the other two groups. The scree plot and broken stick test (in PAST) showed that the first four components (82.7% of variance) were significant in the PCA. The LDA showed a clear separation between the green (S01) and traditional (S02 and S03) lines along the discriminant axis, but the Hotelling *t*^*2* test was not significant (*t*^*2* = 74.1, *F* = 1.6, *P* = 0.4) because of the small sample size. The variables contributing most to the separation were as follows: the green line had more leaflets and the terminal leaflet blade was wider with a longer and narrower petiolule, while the traditional line had a longer petiole and the petiolule of the terminal leaflet was shorter and broader. ## Correlation between morphological and molecular data The PCoA of ISSR markers and quantitative morphological variables clearly separated S01 (green line) from S02 and S03 (both traditional line) along the first two principal coordinate axes, as shown in. These axes represent 36.8% of the total variance. # Discussion This study combined several different methods to assess morphological, genetic, and chemical variation in *P*. *microphyllus*. Multivariate morphometric studies, which use statistical methods to explore the morphological variability among specimens, are frequently used to identify phenotypic differences in within species. ISSR markers can reflect variation among and within small populations at relatively low costs because they are characterized by a high degree of polymorphism. *Pilocarpus microphyllus* has been previously studied using random amplification of polymorphic DNA (RAPD) to assess germplasm genetic variability. The authors found no correlation between RAPD results for samples from wild populations and an improved genotype from Merck, which had higher dry matter and leaf area. Sandhu et al. found no correlation between RAPD markers and PIL content in 20 genotypes of *P*. *microphyllus* from the state of Maranhão (Brazil). In contrast, our results showed clear genetic and phenotypic differences between the two cultivated lines, including alkaloid content. ISSR markers are known to be highly polymorphic in *P*. *microphyllus* and will be informative for future studies on the genetic diversity of wild populations. The morphological and molecular data indicated significant diversity in the *P*. *microphyllus* lines, despite the small sample sizes used in this pilot study. These findings indicate that further studies of wild populations, as well as more detailed data from currently cultivated lines, will have important applications for industrial extraction of the alkaloids and conservation of natural populations. Further investigation of genetic and phenotypic variability will be important for crossing genetically divergent parents to produce hybrids with a higher heterozygosity. The major findings of this study are that PIL and EPI alkaloid contents vary by month in the dry and rainy seasons, and that in the three groups of *P*. *microphyllus* plants, alkaloid content is correlated with genetic and morphological diversity.The abundance profiles of PIL and EPI were quantified and evaluated throughout one year to assess the effects of the dry and rainy seasons. Molecular and morphometric analyses identified the genetic and morphological differences among the cultivated plants comprising green and traditional lines. Previous studies have shown that agronomic and environmental conditions can affect alkaloid contents, and in *P*. *microphyllus*, mineral, salt, and oxygen stresses affected the PIL content. In the present study, plants were cultivated under the same environmental and agronomic conditions, and the humidity, temperature, and rainfall were monitored. All samples were harvested in the same way, using juvenile material that has been shown to yield the highest PIL levels.Northeastern Brazil has only two well-defined seasons: the dry season between August and December when rainfall is rarely recorded, and the rainy season between January to July when high precipitation is usually recorded. For industrial use of *P*. *microphyllus*, only plants with a PIL content greater than or equal to 0.500% are considered suitable for extraction, and most wild plants have this level. PIL was the primary alkaloid found in *P*. *microphyllus*, with the highest accumulation in the leaves, corroborating a previous report. In this study, PIL content varied throughout the year in all samples. The lowest levels were recorded in the rainy season and the highest levels in the dry season, suggesting that rainfall has a negative influence on PIL contenton S01 and S02 populations. An exception was found in jaborandi green line population, when rainy season just arrived there was an increasing of PIL content in January and February months but from March to July it has a tendency to declineeven if this evidence was not statistically significant.A previous study reported that random samples of *P*. *microphyllus* harvested in the dry season showed greater PIL production. These results confirm that PIL content responds to environmental conditions, especially rainfall. Local suppliers do not collect this species in the rainy season, presumably owing to the difficulty of sun drying the leaves and the rejection of raw materials with low levels of PIL below the specification for extraction. This reduction in PIL content had been previously observed in natural populations of *P*. *microphyllus* and was recorded in the cultivated lines used in this study. Another important finding of this study is the high PIL content of the jaborandi green line (S01). This form is highly suitable for industrial PIL extraction because the PIL content is close or above the industrial specification limit of ≥ 0.500% in all months. The species is officially listed as endangered by IBAMA, the Brazilian environmental regulatory body, but the present study indicates that the remaining natural populations of *P*. *microphyllus* could be protected by cultivating specific lines for extractive purposes, lessening pressure on wild plants. EPI is another alkaloid from *P*. *microphyllus* and has recently attracted attention within the scientific community because it has *in vitro* and *in vivo* schistosomicidal activities. It is a promising molecule for combatting a neglected disease that affects millions of people around the world, especially in underdeveloped countries. EPI has an important effect against young adult *Schistosoma mansoni* parasites and inhibits egg laying. The extraction, purification, and isolation of EPI on an industrial scale, and the spectroscopic structural characterization, have been studied. Other *in vivo* studies have investigated EPI’s anti-inflammatory and antinociceptiveactivities, which might aid treatment of liver granulomas formed by *S*. *mansoni* eggs. However, the seasonal content profile of EPI has not been studied previously. This alkaloid was found in significantly lower amounts in S01 in all months, indicating that the green line is not the best source for EPI extraction. In contrast, the traditional line represented by S02 and S03 had considerably higher EPI content and would be suitable for extraction. The jaborandi traditional line should be studied further for the industrial extraction, purification, and isolation of EPI. Despite the high EPI contents recorded from S02 and S03, only in August (S02) or September (S03) did values reach the threshold required for industrial extraction (\> 0.500%), suggesting further genetic improvements or selection studies are required to enhance EPI yield. The molecular and morphological analyses differentiated the green line individuals (S01) from those of the traditional line (S02, S03). In addition, the jaborandi green line has bright green leaves and a greater degree of branching, whereas the jaborandi traditional line has darker green leaves and sparser branching. The chemical data also indicated differences between the lines, with alkaloid profiles that were similar in S02 and S03 but rather different in S01 (Figs and ; Tables). These results indicate that two distinct forms of *P*. *microphyllus* are present in the studied plantation, perhaps deliberately selected from natural populations by early collectors. More studies are needed to fully characterize these two distinct forms of *P*. *microphyllus*. This study shows the first correlation among the chemical, morphological, and molecular profiles of *P*. *microphyllus* and highlights the potential benefits of a multidisciplinary approach, in which the seasonal content of industrially important alkaloids can be linked to population-level genetic diversity and morphological variation. Furthermore, this study will allow better selection and development of jaborandi cultivars, with a focus on maximizing the alkaloid content. Better cultivars will reduce collecting pressure on wild populations and support year-round production of two important alkaloids. # Supporting information The authors are grateful to the Laboratory of Cell and Biology Molecular (LAMOL) of the Federal University do Piauí and Laboratory of Molecular Plant Systematics in the State University of Feira de Santana (UEFS) for analysis facilities. We thank the Anidro do Brasil Extrações SA and Phytobios Research Development Innovation LTDA (Centroflora Group) for the PIL standard, samples of the *Pilocarpus microphyllus* specimens from its farm, and the species photographs. We also thank Dr. Simon Mayo from Royal Botanic Gardens, Kew, for his valuable contributions to the morphometric and molecular analyses. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** DFL JRSAL IMA. **Data curation:** DFL JRSAL IMA IM. **Formal analysis:** DFL JRSAL LIL JAR IMA LGG CV LM LMCV IFSA AGB JC IM. **Funding acquisition:** JRSAL IMA MBPPO IM. **Investigation:** DFL LIL JAR LGG CV LM LMCV JC IM. **Methodology:** DFL LIL JAR LGG CV LM LMCV IFSA AGB JC IM. **Project administration:** JRSAL DFL IMA. **Resources:** JRSAL IMA IM. **Supervision:** JRSAL IMA MBPPO IM. **Validation:** DFL JRSAL IMA. **Writing – original draft:** DFL JRSAL JAR IMA LGG CV LM LMCV IFSA AGB JC MBPPO IM. **Writing – review & editing:** DFL JRSAL IMA IM. [^3]: ‡ These authors also contributed equally to this work.

# Introduction Hundreds of millions of people are infected with the common soil-transmitted helminths (STHs), namely hookworms (*Ancylostoma duodenale* and *Necator americanus*), *Ascaris lumbricoides* and *Trichuris trichiura*, many by multiple species concurrently. *Taenia* spp. infections are also widespread. STHs and taeniasis/cysticercosis belong to the neglected tropical diseases (NTDs) and are responsible for mainly chronic and often inconspicuous morbidity. Iron- deficiency anemia, malnutrition, and impaired physical and cognitive development have all been attributed to STH infections. *Taenia solium* cysticercosis is a major cause of epilepsy and other neurological disorders in developing countries. The current strategy for STH control in highly endemic areas focuses on morbidity control through large-scale administration of single-dose anthelminthics to at-risk populations, particularly school-aged children. Due to the zoonotic nature of taeniasis/cysticercosis, its control must also include the veterinary sector,. At present, only four drugs are recommended by the World Health Organization (WHO) for treating STH infections. The global STH control relies on two of them – albendazole and mebendazole – both benzimidazole carbamates. Albendazole and mebendazole display a broad spectrum of activity and are administered orally, usually at a single dose of 400 mg and 500 mg, respectively. Children below the age of 1 year and pregnant women in the first trimester of pregnancy are not eligible for treatment. Albendazole and mebendazole have been extensively used worldwide for more than 30 years, both as stand-alone treatments and, more recently, in combination with other drugs, e.g., praziquantel (against schistosomiasis and food-borne trematodiasis) or ivermectin (against lymphatic filariasis),. Surprisingly though, only few clinical trials compared the efficacy of albendazole and mebendazole against STHs. Rather, availability, cost, drug donation programs, and policy instead of the local parasite spectra and evidence determine the choice of which anthelminthic drug is deployed. Justification for the indiscriminate use of either drug is derived from high egg reduction rates (ERRs) achieved with both albendazole and mebendazole, and the assumption that morbidity is a function of infection intensity. However, a recent meta-analysis of randomized placebo-controlled single-dose drug efficacy trials pointed to a marked superiority of albendazole over mebendazole against hookworm, high efficacy (in terms of cure rate \[CR\]) of both drugs against *A. lumbricoides*, and disappointing efficacy of either drug against *T. trichiura*. Few data are available regarding ERRs. The aim of this randomized controlled trial was to assess the efficacy of standard single-dose *versus* triple-dose oral albendazole and mebendazole against hookworm and other STH infections in a highly endemic but virtually benzimidazole-naïve population in the People's Republic of China (P.R. China). # Methods The protocol for this trial and the supporting CONSORT checklist are available as supporting information; see and. ## Study Area, Study Period, and Participants The study was conducted between October and December 2008 in Nongyang, a village located in Menghai county, Yunnan province, P.R. China. Details of the study area, population and epidemiological characteristics, including the prevalence of STHs, *Taenia* spp., and intestinal protozoa, have been described before. The local prevalence of each *A. lumbricoides*, hookworm, and *T. trichiura* exceeded 85% in a survey conducted in 2006. Upon completion of the 2006 survey, compound mebendazole (mebendazole 100 mg/tablet+levamisole hydrochloride 25 mg/tablet, 2 tablets per day for 3 consecutive days) was distributed to the village population. No further interventions took place until the present study. ## Ethics The study was approved by the Ethics Committee of Basel (no. 294/08) and the Academic Board of the National Institute of Parasitic Diseases, Chinese Center for Disease Control and Prevention in Shanghai (no. 2008091701). The trial was registered with Current Controlled Trials (identifier: ISRCTN47375023). The study objectives and procedures were discussed with the village head, village committee, and local health care officials who informed the residents. Individuals who were interested to participate signed an informed consent form in Chinese (parents or legal guardians in case of minors aged 5–17 years). Upon study completion, albendazole was provided for treatment of study participants found to be infected at evaluation, drop-outs, sick individuals upon recovery, and pregnant women once beyond the first trimester. ## Interventions, Trial Medication, and Outcome Measures The trial was designed as a community-based open-label, outcome assessors- blinded randomized controlled trial with four arms: (i) single-dose albendazole (400 mg), (ii) single-dose mebendazole (500 mg), (iii) triple-dose albendazole (3×400 mg, given over 3 consecutive days), and (iv) triple-dose mebendazole (3×500 mg, given over 3 consecutive days). No placebo drugs were given to individuals assigned to single dose treatment (open label). Albendazole (Zentel®; lot no 08060407) was commercially obtained from Sino- American Tianjin SmithKline and French Laboratories Ltd., a Chinese joint venture of GlaxoSmithKline Plc. Mebendazole (Vermox®; lot nos. 8CL4F00 and 7CL8900), produced by Johnson & Johnson/Janssen-Cilag S.p.A., was provided by the WHO regional office in Hanoi, Vietnam. The primary outcome considered was CR against hookworm 3–4 weeks following dosing. Changes in hookworm infection intensity, as determined by ERR, and efficacy against *A. lumbricoides* and *T. trichiura* served as secondary outcomes. Additionally, the effects of all four treatment regimens on *Taenia* spp. were assessed. ## Eligibility Criteria and Sample Size Eligible for inclusion were all residents of Nongyang aged 5 years and above. The following exclusion criteria were applied: presence of diagnosed or perceived chronic disease or other conditions likely to interfere with anthelminthic treatment (e.g., hypersensitivity to anthelminthics), pregnancy (verbally assessed at enrolment and again before treatment), recent history of anthelminthic treatment, and participation in other trials (within 1 month). The intended sample size at enrolment was 370 individuals, based on the following assumptions: a total of 176 individuals (44 in each of the four treatment arms) would be needed to detect differences in the CR following different treatments for the cure of hookworm infections with 80% power using a 2-sided statistical test with an α-level of 0.05 and CRs of albendazole and mebendazole against hookworm infections of 75% and 45%. According to Keiser and Utzinger, the respective CRs are 78% and 23%; the higher estimate for the CR of mebendazole was employed in order to include a safety margin. The local prevalence of hookworm infections was assumed to be 60% and compliance was estimated to be 80%. Recruitment was to be stopped once 400 individuals had been enrolled. ## Field and Laboratory Procedures Families were contacted in batches of 20–30 (∼80–120 potential participants) based on family registry numbers. Interested family members were invited to the local primary school for further information and enrolment. No monetary compensation was offered for participation. Participants answered a short questionnaire investigating demographic and health-related issues, and were given a stool collection container labeled with a unique identifier and their full name. The ability of all study participants to recognize their collection container was determined, and the importance of using the own receptacle emphasized. Each morning, filled containers were collected, and a new container handed out with the aim to obtain two stool samples from each participant. Stool samples were forwarded to a nearby laboratory and processed on the collection day. First, samples were visually inspected for adult *A. lumbricoides* and *Taenia* spp. proglottids. Second, two 41.7 mg Kato-Katz thick smears were prepared from each sample. Depending on the ambient temperature and considering over-clearance of hookworm eggs, slides were read within 30–90 min of preparation. At least 5% of the daily diagnoses were cross-checked by the principal investigator. Procedures for the evaluation of the treatment efficacy commenced 3 weeks post-treatment, lasted 2 weeks, and involved all participants given at least one drug dose. The same approach was adhered to as during the baseline survey. ## Randomization All participants who had submitted at least one stool sample during the baseline survey were randomly assigned either to the albendazole or the mebendazole arm of the study. In an independent randomization step, single or triple dose treatment using two computer-generated random sequences of 0 and 1 which were aligned with the list of participants in ascending order of their identification numbers. The eligible individuals were neither stratified by age nor sex before randomization. ## Drug Administration For each day of treatment, an envelope of the type locally used to hand out drugs was labeled with the name, identification number, and number of treatment, loaded with the appropriate drugs, and sealed. The distribution teams directly observed drug intake after asking about acute health problems and pregnancy status. Study participants had been reminded not to drink alcohol on treatment days and to report emerging health problems to the study physician (a medical doctor from a nearby hospital who visited the village each morning after drug distribution), any member of the research team, or the head of the village. On the second morning – 36 hours after the first dosing – all participating households were visited and participants actively solicited to report any potential adverse events. Reported health problems were classified by the study physician and graded by severity according to a pre-defined scale. ## Statistical Analysis Data were double-entered in EpiData version 3.1 (EpiData Association; Odense, Denmark) or Microsoft® Excel 2002 (Microsoft; Redmond, USA). After removing discrepancies, the datasets were aligned, and the accuracy of the merged database verified against the original data through random cross-checking. All analyses were performed on a per-protocol basis. Only participants with complete datasets were included. Baseline and post-treatment prevalences were estimated, and CRs determined for each study arm. The extent of prevalence reductions and differences in CRs between groups were explored, using a 2-sided 2-sample test of proportions, which tests the equality of proportions using large-sample statistics. For each participant, the species-specific helminth infection intensity at baseline and at treatment evaluation was calculated and expressed as eggs per gram of stool (EPG), based on the arithmetic mean of the quadruplicate Kato-Katz thick smear readings, multiplied by a factor 24. Arithmetic and geometric means and ERRs were calculated according to Montresor et al.. Confidence limits for the ERR were calculated using a bootstrap re-sampling method with 2000 iterations. Significant treatment group differences were defined by non-overlapping 95% confidence limits. For all tests, a *p*-value of 0.05 was considered the limit of statistical significance, and 95% confidence intervals (CIs) were calculated as appropriate. Statistical analyses were done in STATA version 10.1 (StataCorp LP; College Station, USA), bootstrap confidence intervals were calculated using R 2.9.1. # Results ## Participant Flow and Baseline Characteristics As detailed in, at least one stool sample was available from 378 people who were randomly assigned to one of the four treatment arms. Among them, 314 (83%) could be included in the final analysis. The composition of all four groups with regard to sex and age was comparable and baseline prevalences of *A. lumbricoides*, *T. trichiura*, hookworm and *Taenia* spp. were 90%, 75%, 73% and 11%, respectively, with no differences among the four treatment arms. ## Efficacy Against Hookworm and Other STHs A single dose of albendazole cured 69% (95% CI: 55–81%) of the hookworm infections, while single-dose mebendazole only cured 31% (95% CI: 20–45%), significantly less. Triple doses of either drug were significantly more efficacious than single-dose regimens, but the difference between the two drugs persisted: triple-dose albendazole cured significantly more hookworm infections (92%, 95% CI: 81–98%) than triple-dose mebendazole (58%, 95% CI: 46–71%). Triple-dose mebendazole exhibited the highest reduction in *T. trichiura* prevalence (CR: 71%), followed by triple-dose albendazole (56%). Single dose applications were found to be significantly less efficacious (mebendazole: 40%, albendazole: 34%). In both cases, the differences between drug-specific CRs were not statistically significant. As expected, both albendazole and mebendazole cleared most of the *A. lumbricoides* infections with observed CRs ranging between 93% and 97%. The efficacies of albendazole and mebendazole were comparable. Triple-dose treatment tended to be slightly more efficacious than single-dose treatment, but the difference was not statistically significant. For *Taenia* spp., a single dose of either drug cured about one half of the infections; triple-dose administration cured all infections. and (and in greater detail the) show the baseline EPGs and changes following treatment. In general, the efficacy regarding ERRs followed a similar pattern as that of CRs. Albendazole outperformed mebendazole in terms of hookworm ERR, whereas mebendazole tended to be more efficacious against *T. trichiura*. Triple-dose regimens exhibited significantly higher ERRs against both parasites. All treatments resulted in ERRs\>99.9% against *A. lumbricoides*. The median hookworm egg count in the 228 infected participants was 84 EPG at baseline and 30 EPG in those 92 still infected after treatment. The administration of three doses of albendazole resulted in the highest ERR against hookworm (99.7%; 95% CI: 99–99.9%). Single-dose albendazole with an ERR of 97% (95% CI: 95–99%) performed as well as triple-dose mebendazole (96%, 95% CI: 93–98%). A single dose of mebendazole resulted in an ERR of only 84% (95% CI: 73–90%). For *T. trichiura*, the administration of triple doses resulted in an ERR of 97% for mebendazole, and 94% for albendazole. With ERRs of 83% and 77%, respectively, single doses performed significantly worse. ## Adverse Events Thirteen study participants (4.1%) reported between one and five adverse events following drug administration, mostly in the morning of the third drug distribution day (about 12 hours after the administration of the second dose, if given) and upon active questioning. Four of these individuals were treated with a single dose (3 with mebendazole, 1 with albendazole) while the remaining nine were treated with triple mebendazole (*n* = 5) or triple albendazole (*n* = 4). One symptom was reported by nine individuals, two symptoms by two individuals (1 treated with triple albendazole, 1 with triple mebendazole), three symptoms by one individual (triple mebendazole) and five symptoms by one individual (triple mebendazole). Adverse events included headache (*n* = 3; all mebendazole), abdominal cramps (*n* = 3; 2 mebendazole, 1 albendazole) and the closely related “full stomach” (*n* = 2; mebendazole), and waist pain (*n* = 1; albendazole). Two individuals each reported vomiting, including production of *A. lumbricoides* worms (1 albendazole, 1 mebendazole), diarrhea (2 mebendazole), fatigue (1 albendazole, 1 mebendazole), and chills (2 mebendazole). Vertigo (albendazole), throat pain (albendazole), fever (mebendazole), and a swollen face (mebendazole) were each reported once. None of the study participants requested medical interventions as adverse events were mild and self-limiting. More women than men reported adverse events (ten women among whom four treated with albendazole and six treated with mebendazole *versus* three men; *P* = 0.046) but there was no significant association between the report of adverse events and age, drug, or number of treatments according to the Fisher's exact test. # Discussion This randomized controlled trial comparing the efficacy of single and triple dose albendazole and mebendazole confirmed that single oral albendazole is more efficacious than mebendazole against hookworm infections. It also corroborated that triple-dose regimens result in significantly higher CRs than recommended and widely used single-dose regimens. A single dose of mebendazole only cured 31% of the hookworm infections, while the highest CR, after triple albendazole, was 92%. Even triple administration of mebendazole was less efficacious than a single dose of albendazole. Keiser and Utzinger's meta-analysis estimated a CR of only 15% after single-dose mebendazole, and a value comparable to that found in the present study after single-dose albendazole (present study: 69%, meta- analysis: 72%). With regard to ERRs, all four drug regimens resulted in significant reductions among those infected at baseline. A triple dose of mebendazole was significantly more efficacious than a single dose. The number of *T. trichiura* infections in each treatment arm was significantly, though only moderately reduced, in line with previous findings. As expected, triple doses resulted in higher CRs than a single dose regardless of the drug. Worryingly, the highest CR observed was only 71% following triple-dose mebendazole. Single and triple doses of mebendazole resulted in higher ERRs than the respective number of albendazole administrations. With regard to *A. lumbricoides* infections, high CRs were observed for both drugs even at a single dose; observations that are in line with systematic reviews and meta-analysis. Attention was paid to enhance the sensitivity of STH diagnosis by examining multiple Kato-Katz thick smears before and after drug administration. The low number of “new” infections found at treatment evaluation indicates that a high sensitivity had been achieved despite the rather low density of hookworm and *T. trichiura* eggs. Because of the low *Taenia* spp. prevalence and since the study was not designed to evaluate treatment efficacy against this parasite, the respective results should be interpreted with caution. The conventional indicator for the successful cure of *Taenia* spp. infections – i.e., recovery of the scolex – is no definitive proof whenever individuals harbor several worms, and is difficult to perform outside an institutional setting. We focused on the presence of proglottids and eggs. An open-label trial design was adhered to due to the complexities and high cost for implementing a double-blind trial in a field setting. We are confident that this did not negatively impact on the validity of the results since outcome assessors were blinded. One individual assigned to the triple albendazole group switched to the single-dose group, and in two instances the drug assignment was changed between members of the same family due to an initial mix-up. We used logistic regression to assess if our results were sensitive to the potential effect modifiers age and sex. Age was treated as a categorical variable (categories as in) and also as a continuous variable (in years). None of the analyses showed noteworthy differences between the crude and adjusted models with respect to the point estimates or CIs of the odds ratios. The sole exception was the treatment regimen (single dose *versus* triple dose) for which adjustment for sex and age showed stronger effects for both drugs in the case of *T. trichiura*. The susceptibility of the two human hookworm species to albendazole is known to be unequal, with CRs for the more pathogenic *A. duodenale* higher than that for *N. americanus*. Both hookworm species are endemic in P.R. China but the locally predominant species probably is *N. americanus* according to a polymerase chain reaction (PCR)-based species identification performed in a neighboring area. Multiple-species intestinal helminth infections are common but no associations between species have been found and the high prevalence of multiparasitism in the study population is unlikely to diminish the validity of the findings for other settings. Two additional observations are worth discussing. First, the *A. lumbricoides* CR did not differ significantly (p\>0.05) between infection-intensity classes as defined by WHO. Second, the baseline prevalence of *A. lumbricoides* and hookworm was higher among females than males. At evaluation, the difference persisted for hookworm, but had disappeared for *A. lumbricoides*, probably owing to the high CR against the latter parasite. In the case of *T. trichiura*, comparable prevalences were found for males and females at baseline, but treatment with either drug reduced the prevalence in males more markedly than in females. The raw data of our randomized controlled trial is provided as supplementary files. In the spirit of trial registration prior to conducting clinical research, of open-access publishing, and of evidence-based medicine, we believe that others might find our data useful (e.g. for subsequent meta-analysis of drugs used against STHs). We hope that other clinical investigators and research groups will follow our example. In conclusion, single-dose albendazole and mebendazole are highly efficacious against *A. lumbricoides*, albendazole is superior to mebendazole for treating hookworm, and mebendazole slightly outperforms albendazole with regard to treating *T. trichiura*. To achieve high CRs against hookworm and *T. trichiura* infections, triple dose regimens should be considered. Yet, for *T. trichiura*, even triple doses only resulted in the cure of a bit more than half of the infections, a result corroborating previous reports. Triple-dose treatment is commonly deemed unfeasible in the context of large-scale drug administration programs based on logistical and organizational considerations, an issue which needs careful attention. To justify rolling out triple dose treatment, the additional efforts and costs required to do so must be weighed against the benefit, i.e., the higher treatment efficacy, and hence the prevention of harm. From a patient perspective, triple dose treatment appeared acceptable in the present study. Our findings therefore underscore the need for discovery and development of novel drugs for the management of trichuriasis. Until new drugs become available, it is recommended to investigate ways to boost the efficacy of existing anthelminthics, including combination therapy (e.g., albendazole or mebendazole plus ivermectin), and multiple dosing. The higher efficacy of triple doses for treating *Taenia* spp. infections further tips the balance in favor of triple dose schedules in certain areas. With regard to large-scale interventions, the present results call for a more nuanced approach than the standard single-dose mono-drug distribution. Indeed, our findings emphasize the need for careful assessment of the locally endemic STHs, and the adaptation of the employed anthelminthic drug regimens to the prevailing situation. In populations primarily parasitized by *A. lumbricoides* and/or hookworm infections, single or – in case of a high prevalence or high-intensity hookworm infections – triple-dose albendazole might suffice. Mebendazole treatment with one or better three doses should be adopted in areas with a high prevalence of *T. trichiura* (and possibly *A. lumbricoides*), but a lower number of hookworm infections. In areas where all three species are co-endemic, alternation between albendazole and mebendazole as well as co-administration of different anthelminthic drugs should be considered. # Supporting Information We thank the field and laboratory team for its commitment, and the study participants for their cooperation. We are grateful to the WHO office in Hanoi for providing the mebendazole tablets free of charge. [^1]: Conceived and designed the experiments: PS JU J-XC X-NZ. Performed the experiments: PS Z-WD J-YJ HZ. Analyzed the data: PS JU JH. Contributed reagents/materials/analysis tools: PS JU Z-WD J-YJ HZ JH. Wrote the paper: PS JU JH. [^2]: PS is supported by the Novartis Foundation through a personal stipend. This does not alter the authors' adherence to all the PLoS ONE policies on sharing data and materials.

# Background Globally, under-five children are the most vulnerable segment of the population. Under nutrition is one of the most public health burdens in developing countries. It includes stunting, wasting, and under-weight and is determined through measurements of height, weight, and age. Stunting has been defined as a child who is too short for their age. Globally, stunting affected about 149 million under-five children in 2018. Of these 55% of stunted children lived in Asia and 39% lived in Africa. Currently, in Ethiopia, the prevalence is estimated to be about 38% among under-five children. Wasting has been also defined as a child who is too thin for their height. In 2018, worldwide wasting contributed to threaten the lives of an estimated 49 million under-five children. Of these, more than two-thirds of all wasted children are found in Asia and more than one quarter found in Africa. The latest recent study in Ethiopia showed that the prevalence is estimated to be about 10% among under-five children. Similarly, under-weight refers a child who is too small for his/her age which is the weight for age \< -2 standard deviation(SD) of the WHO Child Growth Standards median. Consecutively, the recent report in Ethiopia indicated that \`the prevalence is estimated to be about 24% among under-five children. The results of impaired growth and development in children can be life long and reduce academic performance and the ability to contribute to the nation. Many factors contribute to child under-nutrition including inadequate diet and poor infant and young child feeding (IYCF) practices, a high burden of infectious disease, and a lack of basic infrastructure to enabling access to clean water, sanitation, and health services. Different studies have been conducted to explore the spatial distributions of communicable diseases. The application of spatial analysis to non-communicable diseases has become a common practice to date. Consecutively, previous studies revealed that spatial variation of under-nutrition among under-five children. Similarly, studies conducted in Ethiopia have also showed the spatial variation of childhood stunting. However, spatial studies on under-nutrition status of children are limited and lack modeling of spatial relationships between the identified clusters of under-nutrition and its predictors. Thus, understanding the area-based heterogeneity and factors affecting under- nutrition is a footstep for evidence-based decision-making in under-nutrition prevention and control programs. In addition, detecting spatial variation is also useful to recognize gaps in the performance of program on childhood nutrition that could not be identified through the routine monitoring of the nutritional status of children. Hence, this study aimed to explore the geographical variation of under-nutrition and its predictors among under-five children in Ethiopia using geographically weighted regression analysis. # Methods ## Study setting Ethiopia is found in the horn of Africa covering 1,104,300 km² and ranks 10^th in Africa in land coverage. Ethiopia is a country with a great geographical diversity ranging from peaks up to 4,550 m above sea level down to a depression of 110 m below sea level. Ethiopia has nine administrative regions (Tigray, Afar, Amhara, Oromia, Somalia, Benishangul Gumuz, Gambella, Somalia, Harari and SNNPR) and two city administrations namely Diredawa and Addis-Ababa. The country is divided into 68 zones, 817 districts, and 16,253 *kebeles* based on the report of 2010 Ethiopian fiscal year. Contextually, it is categorized as agrarian, pastoralists and city-based population. It has a total of 104,957,000 populations, of which 36,296,657 were women. Majority of the population about, 83.6% living in rural areas and 16.7% of the population reside in urban areas. The average household size in national level is 4.7 persons. The country has fertility rate of 4.6, infant mortality rate (per 1,000 live births) of 48, and child mortality rate (per 1,000 live births) of 67 children deaths. We used the 2016 EDHS data for this study. The EDHS waive was conducted from January 18 2016 to June 27, 2016. ## Population All children aged 0 to 59 months living in Ethiopia were considered as a source population and the study population includes all under-five children in the selected Enumeration areas during the EDHS data collection. A total of 9,384 children aged 0 to 59 months who full fill the inclusion criteria were considered for the entire analysis. To select study participants, a stratified, two-stage cluster sampling technique was employed for the 2016 EDHS. Enumeration areas (EAs) were the sampling units for the first stage and households were the sampling unit for the second stage. In the 2016 EDHS, a total of 645 EAs (202 urban and 443 rural) were selected with a probability proportional to EAs size (based on the 2007 housing and population census) and independent selection in each sampling stratum. Of these, 18,008 households and 16,583 eligible women were included. The detailed sampling procedure was presented in the full EDHS report. The 2016 EDHS spatial data (latitude and longitude coordinates) was used for this study. It was taken from the selected enumeration areas during the data collection period. From a total of 645 clusters included in the 2016 EDHS, 23 of them had no latitude and longitude coordinates. The location data were accessed through the web page of the measure DHS Program after being authorized for utilization of the data. ## Variables of the study The outcome variable used for this study was under- nutrition which includes wasting, stunting, and underweight. Independent variables of the study include mothers’ age, mothers’ educational level, fathers’ educational level, marital status, sex of children, type of toilet facility, drinking water source, distance from the health facility, residence, wealth index, and family size. Categorization of variables like education level of respondents, educational level of fathers, and family size were based on different literatures. ## Operational definition ### Under-nutrition Defined as the type of malnutrition which includes stunting, wasting, and underweight among under-five children. ### Underweight Defined as children who have \<-2 SDs below the mean weight for age of the National Center for Health Statistics and the World Health Organization (WHO) reference population. ### Stunting Defined as children who have low height/length for age Z score \<-2 SDs of the median value of the WHO Child Growth Standards median aged 6–59 months. ### Wasting Defined as the children who have low height/length-for-age-Z score \<-2 SDs of the median value of the WHO Child Growth Standards median aged 6–59 months. ### Improved toilet facility A toilet facility includes any non-shared toilet of those types: flush/pour flush toilets to piped water systems, septic tanks and, and pit latrines; ventilated improved pit (VIP) latrines, pit latrines with slabs, and composting toilets. ### Unimproved toilet A type of toilet facility that includes a bucket or a toilet that flush’s to elsewhere (in or nearby the household environment), pit latrine without slab/open pit, no facility/bush/field, bucket toilet, hanging toilet/latrine, no facilities. ### Improved source of drinking water Drinking water includes water piped into the residence, from piped water, a public tap, standpipes, tube wells, water from a borehole, a protected wells and spring, rainwater, and bottled water. ### Unimproved source of drinking water Drinking water from unprotected wells or springs, water from a vendor or tanker- truck and surface water (including rivers, dams, lakes, ponds, streams, canals, and irrigation channels. ### Anthropometric measurements of weight and height These measurements during the EDHS 2016 were taken with standardized and calibrated measuring tools after oral consent was obtained from their mothers. Accordingly, weight and height measurement of children was carried out from the selected enumeration areas. During data collection children less than 24 months were measured for height lying down and children greater than 24 months were measured while standing using a short measuring board. Weight measurements were taken with lightweight SECA mother infant scales with a digital screen designed and measured under the guidance of UNICEF. Details on anthropometric measurement were found on the EDHS report. ## Data management and analysis After accessing the data from the MEASURE DHS website data extraction, data weighting, data cleaning, recoding, Descriptive and summary statistics were held using STATA version 14.1 software. ## Spatial analysis The spatial autocorrelation (Global Moran’s I) statistic was held in order to assess the pattern of wasting, stunting, and underweight whether it was dispersed, clustered, or randomly distributed in the study area. Details about spatial autocorrelation is published everywhere. Local Moran’s I measure positively correlated (high-high and low-low) clusters and outliers. The statistically determination of cluster outlier is published everywhere. Gettis-ord Gi\* statistics had been calculated to measure how spatial autocorrelation differs through the study location by computed Gi\* statistics for each area. Z-score was calculated to ensure the statistical significance of clustering and the p-value calculated the significance *p*-value\<0.05 at 95% CI if the Z-score is between -1.96 and +1.96, the *p*-value must be greater than 0.05 and vice versa. If the *p*-value is less than -1.96 it is declared as a cold spot and if greater than +1.96 it is declared as hotspot areas. ## Spatial scan statistical analysis It tests the presence of statistically significant spatial clusters of wasting, stunting, and under-weight among under-five children using Kuldorff’s SaTScan version 9.6 software. Children who had been wasted, stunted, and underweight were considered as cases and children who had normal nutritional status as controls to fit the Bernoulli model. Spatial cluster size \< 25% of the population was used, as a higher boundary, which allowed both small and large clusters detection. The primary and secondary clusters were identified and assigned *p*-values and ranked based on their log likelihood ratio test. ## Spatial regression analysis Spatial regression has both local and global analysis techniques. Therefore, first, we had handled global geographical regression models and then local geographical analysis in order to ensure the variability of coefficients across each enumeration areas. Then we have checked the assumptions using exploratory regression with the respective tests. The normality assumption was checked for residuals using Jarque-Bera test. As residuals are not spatially auto- correlated, confirming the koenker Bp test was done to check if the model under gone for geographically weighted regression or not. Geographically weighted regression was executed using GWR version 4.0 software. We had also checked the six checks which recommended for a model undergone for spatial regression. These are coefficients have the expected sign, no redundancy among model explanatory variables, coefficients are statically significant, strong adjusted R² values and the above two conditions stated before. Variables that have a *p*-value less than 0.05 are selected and described based on their coefficients. ## Ethical consideration Permission to use the dataset has been granted by the Measure DHS program through legal registration. EDHS (2016) data was used which is available on the public domain through the Measure DHS website ([www.measuredhs.com](http://www.measuredhs.com/)). Accordingly, the investigators had requested permission to use the data set on the Measure DHS website about the geographical variation of wasting, stunting, and underweight among under-five children in Ethiopia in May, 2020. # Results ## Socio- demographic and other characteristics of respondents A total of 10,640 under-five children were included in the 2016 EDHS survey. Of these, 1872 under-five children were dropped since valid or complete data were not obtained during data collection. Thus, after weighting the data a total of 9384 under-5 children were included in the analysis. The median age of children were 28 months with IQR = 31. In addition, the mean height of mothers was 158 centimeters (cm) with SD = 6.7. About 6183 (65.9%) of mothers had no formal education and 8363 (89.1%) were rural residents. About, 8471(89.9%), of respondents had used improved drinking water and about, 5278(56.25%) of respondents had used unimproved toilet facility. ## Nutritional status The prevalence of stunting, wasting and under-weight among under-five children in Ethiopia was 3598 (38.30%) (95% CI: 37.34–39.30), 949(10.10%) (95% CI: 9.51–10.73) and 2192 (23.54%) (95%CI: 22.70–24.40), respectively. The study showed variation of under- nutrition across regions of Ethiopia. Consecutively, stunting was lowest in Addis Ababa (15%) and highest in Amhara region (46%), wasting is lowest in Addis Ababa (3.4%) and highest in Somali region (20.9%) (46%), and under-weight is lowest in Addis (4.4) and highest in Afar (37.8). ## Spatial variation of under-nutrition in Ethiopia The spatial distribution of under-nutrition (wasting, stunting, and under- weight) was clustered at zonal level. Hence, the global Morans I index value was 0.363 (*p*-value \< 0.001), 1.072 (*p*-value \<0.001) and 0.879 (*p*-value \<0.001) for wasting, stunting, and under-weight respectively as shown below in and. The highest case distribution of stunting was spatially clustered in the Amhara region (North Gondar, South Gondar, Waghmira, and East Gojam zones), Afar region (Zone 1), and Oromia region (Guji, Borena, and West Arsi zones). Zone 4 of Afar region, Afder, Liben, Shinale, Degahbur, and Warder Zones of Somali region, Neuer zone of Gambella region was highest in wasting children case distribution. The highest case distribution of under-weight was spatially clustered in the Amhara region (Waghmira and North Wollo zones), Afar region (Zone 2 and Zone 3), and Somali region (Liben zone). ## Cluster and outlier zone detection for under-nutrition Local Moran’s I analysis result of wasting, stunting, and underweight revealed that there were significant outliers. Accordingly, high outliers for underweight were detected in Shinale and Liben zones of the Somali region, Zone 3 of Afar region. Similarly, Shinale and Liben zones of the Somali region, Zone 3 of the Afar region, Guji, and West Shewa of the Oromia regions were detected as high outliers for wasting. ## Hot spot and cold spot zones for under-nutrition in Ethiopia We conducted a hot spot analysis using Gettis-Ord Gi\* statistics. GIZ-score is computed to determine the statistical significance of clustering, and the *p*-value is computed for the significance. When the Z-score increases (+/−), its significance level increases. Statistical output with high Gi\* indicates “hotspot” whereas low Gi\* means a “cold spot”. As shown in, dark red colors show significant (*p*-value \< 0.001) clusters of Under-nutrition (risk areas), whereas, dark blue colors show significant (*p*-value \< 0.001) non-risk areas. The more clustered red and blue colors indicate more Under-nutrition risk and non-risk areas, respectively. Hotspot analysis enables the detection of both extremities either high or low wasting, stunting, and underweight coverage zones. Accordingly, hot spot (high risk) regions for stunting were detected in the Amhara region (West Gojam, Awi, South Gondar, and Waghmira zones). Somali region (Afder, Gode, Korahe, Warder Zones), Afar region (Zone 2), Tigray region (Southern zone), and Amhara region (Waghmira zones) was detected as hot spot zones for wasting. Amhara region (South Wollo, North Wollo, Awi, South Gondar, and Waghmira zones), Afar region (Zone 2), Tigray region (Eastern zone, North Western zone, Central zone, Southern zone, and Mekele special Zones), and Benshangul region (Metekel and Assosa Zones) were hot spot zones for underweight. On the other hand, North Shoa Zones of the Amhara region, Zone 3 of Afar region, Arsi and East Hararge of Oromia region, Shinale and Jijiga zones of the Somali region Dire Dawa city administration, Harari city administration cold spot zones for stunting. Arsi, East Shoa, and South West Shewa Zones of the Oromia region, zone 3 Zone of Afar region, Gurage and Silte Zones of SNNPR for wasting were considered as cold spot zones. Amhara region (North Shoa Zone), Afar region (Zone 3), Oromya region (Arsi, Horo Gudo Wellega, North Shoa, East Shoa, West Shoa, and South West Shoa Zones), SNNPR (Gurage and Silte Zones), and Addis Ababa city administration were detected as cold spot areas for childhood underweight. ## Spatial clustering of under- nutrition in Ethiopia Spatial scan statistics identified primary and secondary clusters of wasting, stunting, and underweight using the maximum spatial circular windows ≤ 25% of the total population. Accordingly, spatial scan statistics identified (one primary and one secondary cluster) for stunting, (one primary and three secondary clusters) for wasting, and (one primary and three secondary clusters) for under-weight. The primary cluster for stunting (LLR = 466.3, P\<0.001) was centered at (14.033876N, 37.105923 E) with 578.81 km radius and a relative risk (RR) of 1.44. It incorporates all zones of Amhara, Tigray, Afar, and Benishangul regions. Kelem Wollega, Horo Gudu Wollega, West Shoa, South West Shoa, and North Shoa Zones of the Oromia region were also primary clusters for stunting. The primary cluster for wasting (LLR = 215.9, P\<0.001) was centered at (7.650693 N, 47.007919 E) with 819.42 km radius and a relative risk (RR) of 1.8. It surrounds all zones of Somali region, (Bale, East Harerege, West Harerege, and Arsi) Zones of Oromia region, and (Zone 1, Zone 3, and Zone 5) of Afar region. The primary cluster for under-weight (LLR = 402.9, P\<0.001) was centered at 14.033876N, 37.105923 E about 578.81 km radius with relative risk (RR) of 1.57. It surrounds all zones of Amhara, Tigray, and Afar regions. In addition, Kelem Wollega, Horo Gudu Wollega, West Shoa, South West Shoa and North Shoa Zones of the Oromia region also contains primary clusters for underweight. ## Spatial regression analysis of the determinants of childhood under-nutrition ### Ordinary least square After checking spatial regression assumptions for stunting, wasting, and underweight using exploratory regression, an ordinary least square analysis was carried out. Outputs from the spatial regression analysis revealed that, residuals of spatial relationship are uncorrelated (, Tables –) and there was no multi-collinearity among explanatory variables (Tables –). To handle geographical weighted regression, the koenker Bp test in the ordinary least square analysis could be significant, which reveals the difference of coefficients across enumeration areas. In this study, koenker Bp test was found significant hence executing geographically weighted regression is recommended. Accordingly, we have executed geographically weighted regression and determined the local coefficients of each independent variable. ### Ordinary least square analysis for factors associated with under-nutrition In ordinary least square analysis households used unimproved toilet facility, father completed primary and secondary education and has female children were factors significantly associated with stunting. Respondents who had used unimproved toilet facility and being a female child increases stunting by 0.186728 and 0.133730 times. A unit increase for father’s primary and secondary education stunting decreases by 0.127514 and 0.206403 times respectively. A unit increase for respondent’s l who had used unimproved toilet and having children 8 and above increases wasting by 0.059517 and 0.085499 times. In contrast, a unit increase for father had primary education and mothers aged 35–49 years decreases wasting by 0.071297 and 0.115005 times. Similarly, in ordinary least square analysis unimproved toilet, mother had primary education, father has secondary education, and mothers aged 35–49 years was significant predictors of under-weight. Accordingly, a unit increase for respondent’s households who had used unimproved toilet increases underweight by 0.134182 times. In contrast, a unit increase for mother’s primary education, father’s secondary education, and mother’s aged 35–49 years decreases underweight by 0.094737, 0.100582 and 0.158624 times respectively. ## Geographically weighted regression of under-nutrition The result of geographically weighted regression identified different variable coefficients for the identified variables on ordinary least square analysis. For stunting higher coefficients of unimproved toilet of households were detected in all parts of Amhara region, most parts of benshangul-gumz region, western parts of gambela region, western and eastern parts of SNNPR, and western oromia region. Similarly, higher coefficients for father has primary education were detected in central Tigray region, North Eastern parts of SNNPR and Central Oromia region. higher coefficients for father have secondary education was detected in Central Tigray, Eastern parts of SNNPR, most parts of Gambela region and Western Oromia, and higher coefficients for mother’s primary education was detected in Central, Eastern and Southern Tigray region, Eastern Amhara region and Northern SNNPR. For wasting higher coefficients of unimproved toilet of households were detected in eastern Tigray, Eastern Amhara, Eastern parts of SNNPR region and Western parts of Gambella region. Consistently, higher coefficients for father has primary educational level were detected in most parts of Tigray region, Eastern parts of SNNPR region, Harari region and Diredawa city administrative, higher coefficients for mothers aged 35–49 years old were detected in most parts of Tigray region, Eastern parts of Somali region, Harari region and Dire-dawa city administrative. Similarly, for under-weight higher coefficients of unimproved toilet of households were detected in all regions of the country except Oromia region, higher coefficients for mother had primary education was detected in Tigray region, Gambela region, and Diredawa city administrative. higher coefficients for father have secondary education was detected in Eastern Tigray, central and Eastern Amhara region and Addis Ababa city administrative. # Discussion Under-nutrition remains a significant problem in Ethiopia, even efforts have been exerted to reduce the problem. In this study the prevalence of under- nutrition (stunting, wasting and underweight) was 38.30%, 10.10% and 23.54%, respectively. The prevalence of under-nutrition (stunting, wasting and under- weight) was also varying across Ethiopian administrative regions. The burden of under-nutrition is still very high and not evenly distributed in the country. The reports of EDHS also ensure a very stagnant reduction of under-nutrition (stunting, wasting and underweight). This finding for stunting was lower than studies conducted in Ethiopia, in Ghana, in Zambia, and in Cameroon. But it was higher than studies conducted in Tanzania, in Kenya, in Ghana, and in Senegal. This finding was also in line with studies conducted in different parts of Ethiopia previously. This finding for wasting is lower when compared with studies conducted in Ethiopia, in Senegal, and Cameroon. But it is in line with studies conducted in Ghana, Haromaya Ethiopia and higher than studies conducted in Tanzania, Northern Ghana, rural Ethiopia. Finding from this study for under-weight is also lower when compared with studies conducted in Ethiopia, and Cameroon. But the finding was higher than studies conducted in Ethiopia, Tanzania, and Northern Ghana. The find was consistent with studies conducted in Ethiopia. The discrepancy might be due to; those pocket studies in Ethiopia might not use adequate representative sample size. In addition, it could be due to health system structure and the health policy of Ethiopia, in which different studies conducted in Ethiopia revealed that, the distribution of health care facility and health care professionals across the country is not evenly distributed. The finding showed that under-nutrition indicators was clustered spatially at the zonal level. Getis-Ord spatial analysis detected hot spot, cold spot and outliers’ zones in Ethiopia. Furthermore, the detection of hot spot zones using Getis-Ord analysis was assessed using sat scan analysis and reports confirmatory findings for the analysis. In this study the spatial distribution of stunting was clustered at the zonal level. Different studies also support the presence of geographical variation for stunting. A study conducted on geographical variation of stunting found a significant clustering of stunting across Ethiopia zones. The study conducted in India demonstrates a significantly higher value of Moran I (0.65) that suggests a high level of spatial clustering of childhood stunting in zones. Thus, a total of 159 hot spots zones mostly from the central and eastern parts of India. In this study the spatial distribution of wasting was clustered. Different studies also revealed the presence of geographical variation for wasting. A study conducted in Ethiopia reported that significant clustering of stunting across Ethiopian zones. Thus, Liben, Afder and Shinile administrative zones from Somali Regional State and Zone 1, Zone 3 and Zone 4 administrative zones from Afar Regional State, Borena, East and West Harargie from Oromiya Regional State were identified as hot spot zones for wasting. A study conducted previously reported that significant clustering of wasting across regions. Thus, Afar and Somali regions were identified as hot spot areas for wasting. A study conducted on spatial clustering of stunting and wasting in Meskane Mareko District in Gurage Zone found significant clustering of wasting in Diram and Bati Lejano *kebeles* of the district. The current study revealed the spatial distribution of underweight was clustered at Zonal level of Ethiopia. Different studies also revealed the presence of geographical variation for underweight. All administrative zones of Amhara region, Tigray, Afar, Ben. Gumz regional state administrative zones and East Wellega and East Shewa, North Showa administrative zones, from Oromiya Regional State, Liben, Afder, Borena and Gode administrative of Somali regional state were identified as hotspot zones for underweight. A study conducted previously found a significant clustering of underweight across Ethiopia regions. Since Afar and Somali regions were identified as hot spot areas for underweight. A study conducted in India also identified clusters of underweight, 146 districts (23% of all the districts) were observed as of high underweight. The most convenient explanation for this spatial variation of under-nutrition could be as a result of geographical variation in the country which ranges from 4550 meters above sea level to 110 meters below sea level. Consequently, infrastructure differences like: road, electricity, water, the distribution of health facilities, and health care professionals across regions. In addition, there is a difference in culture, socio-demographic characteristics of mothers and fathers, attitude and knowledge difference of the society towards under- nutrition across different regions. Overall this form of difference in Ethiopia could came up with inequalities of under-nutrition indicators across different parts of the country. This implies that, geographic-based nutritional interventions, mainly mobilizing additional resources could be held to reduce the burden childhood under-nutrition in hot spot areas. In addition, governmental and non-governmental organization working in the identified hot spot areas could be routinely monitored and evaluated nutritional programs in order to improve nutritional status of the children with a special emphasis. In spatial regression analysis, different statistically significant predictor variables of all forms of under-nutrition were identified. had unimproved toilet, father had primary education; father had a secondary education and has female children were significant predictors for stunting. Similarly, respondents had unimproved toilet, father has primary education, having children 8 and above and mothers aged 35–49 years were significant predictors for wasting. Further respondents had unimproved toilet facility, mother has primary education, father has secondary education and mothers aged 35–49 years was identified as significant predictors for underweight. Different studies conducted in Ethiopia and abroad also revealed the existence of considerable significant difference of under-nutrition predictors across a geographical area. The education level of mothers’ and the fathers was negatively correlated with under-nutrition indicators (stunting, wasting and underweight) and its coefficients significantly varied across regions of Ethiopia. This might be due to the reason that there is considerable infrastructure difference among regions which in turn inhibits the educational status of the society. Education is a key tool to acknowledge nutritional status of children and to solve problems associated with under-nutrition. Those who are educated are in a better advance than who are not educated in most occasion. Thus, when education level increase under-nutrition could be decreased. This implies that, Education of parents could be encouraged using different strategies like organizing mothers at village level and assigning community educators, since most of the mothers are aged enough and are not eligible to have formal education. In addition, young mothers could be encouraged to be involved in formal education. Furthermore, the life style of the community could be improved. In the context of unimproved toilets; on most occasion those households having unimproved toilets could be linked with the poor wealth status of the household and lower education level of the respondents this in turn could be influence the nutritional status of the children. This is because of the reasons that unimproved toilet could increase under-nutrition of the children. This implies that, the government should focus on the construction of independent toilet for each household. Hygiene, sanitation and utilization of household toilets could be improved by frequent education using different agents other than health extension workers. In addition, Hygiene and sanitation as a course could be incorporated in formal education of Ethiopia. In the context of households have 8 and above members; mothers who have children in the past enable to understand the possible components of adequate nutrition of a child through frequent exposure and experience. In addition, when household had a greater number of members, they could assist in the preparation of foods for their mothers that enable the mother to afford varieties of food for their children easily. This implies that, having greater number of family members could bring the opportunity of caring a child and improve the nutritional status of a child. In the context of mothers aged 35–49 years old; when the age of the mother increases, she could have experiences in caring a child in different occasions. Thus, mothers getting older could acknowledge the nutritional status of their children easily. Since they are well experienced in doing so. This implies that, experience has paramount importance in appreciating and solving a certain problem. For stunting the analysis of hot spot and geographically weighted regression fit with higher coefficients of unimproved toilet of household’s areas identified by GWR. In those areas which has higher coefficients for unimproved toilet of household’s also has hot spots of stunting, in those areas which has higher negative coefficients for father has primary and secondary education also has hot spot of stunting. For wasting the analysis of hot spot and geographically weighted regression fits with higher coefficients of unimproved toilet of household’s areas identified by GWR. In those areas which has higher coefficients for unimproved toilet of household’s also has hot spots of stunting, in those areas which has higher negative coefficients for fathers has primary education also has hot spot of wasting. Similarly, in areas which have higher negative coefficients for mothers aged 35–49 years old has also hot spot zones for wasting. For underweight, the analysis of hot spot and geographically weighted regression fit with higher coefficients of unimproved toilet of household’s areas identified by GWR. In those areas which has higher coefficients for unimproved toilet of household’s also has hot spots of under-weight, in those areas which has higher negative coefficients for fathers and mother has primary education also has hot spot of under-weight. Over all the analysis of hot spot areas using Getis-Ord spatial auto correlation cluster and outlier and spatial scan statistics fits with the analysis of spatial regression in somewhere and vary also somewhere. This study supports the existing knowledge on the influence of infrastructure coverage, and geographic features on under-nutrition across the country. This study contributes to identify specific hot spot zones throughout the country and factors significantly affect under-nutrition, which is very important for intervention. Even though there is documented influence of factors like infrastructure coverage and geographic features it is difficult to understand specific hot spot zones for under-nutrition in which this study could brought a solution. The findings of the present study are crucial for zonal level planning for child health. Regional pattern and clustering of indicators used in the study suggest a need to strengthen and continue zonal focused programs. ## Strength and limitation of the study The study findings can be generalized to all under-five children in Ethiopia. Moreover, the use of Geographic Information System (GIS) and Sat Scan statistical tests helped to detect similar and statistically significant high- risk clusters/hotspots of under-nutrition. Additionally, use of geographic weighted regression analysis helps to show the real impact of predictors at each specific geographic area. Furthermore, this study had used geographically weighted regression analysis that could enables to determine local coefficients a step advance from ordinary least square analysis. In addition, geographically weighted regression analysis improves model performance by employing a spatial weighted function and addresses spatial heterogeneity and also has an advantage of having less biased predicted coefficients compared to spatial lag and spatial error models. Even though spatial lag and spatial error models enable to addresses both spatial heterogeneity and spatial homogeneity, spatial heterogeneity and spatial homogeneity are unique, not mutually exclusive, features of spatial data. Therefore, this model might bring biased coefficients. As to limitation, location data values were shifted 1-2km for urban and 5km for rural areas for data confidentiality reasons since it didn’t show exact case locations. The data used for analysis were not taking into account seasonal variations of under-nutrition. In addition, the study did not do an adjustment for covariates during estimation of the spatial epidemiology of child malnutrition. The missed data may also affect the true estimates of the analysis. Furthermore, there may be false inclusion and exclusion of SaT Scan clusters. # Conclusion Our study showed that the distribution of under nutrition among under-five children was clustered at zonal level in Ethiopia. It had a geographical variation across regions of Ethiopia. Hot spot zones for stunting were detected in West Gojam, Awi, South Gondar, and Waghmira zones). Similarly, Somali region (Afder, Gode, Korahe, Warder Zones), Afar region (Zone 2), Tigray region (Southern zone), and Amhara region (Waghmira zones) was detected as hot spot zones for wasting. Furthermore, hot spot zones for under-weight was detected in Amhara region (South Wollo, North Wollo, Awi, South Gondar, and Waghmira zones), Afar region (Zone 2), Tigray region (Eastern zone, North Western zone, Central zone, Southern zone, and Mekele special Zones), and Benshangul region (Metekel and Assosa Zones). In spatial regression analysis, different statistically significant predictor variables of all forms of under-nutrition were identified. Had unimproved toilet, father with primary and secondary education, and has female children were significant predictors of stunting. Consistently, has unimproved toilet, father has primary education, having children 8 and above, and mothers aged 35–49 years were also significant factors of wasting. In addition, has unimproved toilet, mother has primary education, father has secondary education, and mothers aged 35–49 years were significant predictors of underweight. Thus, geographic based nutritional interventions mainly mobilizing additional resources could be held to reduce the burden childhood under-nutrition in hot spot areas. In addition, improving sanitation and hygiene practice, improving the life style of the community, and promotion of parent education in the identified hot spot zones for under-nutrition should be more emphasized. # Supporting information We are grateful to the MEASURE DHS program that provides permission with data access authorization so as to enable us to conduct the study. DHS Demographic and Health Survey EDHS Ethiopian Demographic and Health Survey HAZ Height for Age Z-score ICF International Classification of Functioning, Disability, and Health IQR Inter Quartile Range LLR log-likelihood ratio SD Standard Deviation UNICEF United Nation International Children’s Fund WAZ Weight for age Z-score WHO World Health Organization WHZ Weight for Height Z-score EFY Ethiopian Fiscal Year FMOH Federal Ministry of Health GWR Geographic Weighted Regression OLS Ordinary Least Square 10.1371/journal.pone.0248156.r001 Decision Letter 0 Goli Srinivas Academic Editor 2021 Srinivas Goli This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 Nov 2020 PONE-D-20-27277 Geographical clustering of under-nutrition and its predictors among under-five children in Ethiopia: Geographically weighted regression Analysis PLOS ONE Dear Dr. sharew, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== Kindly look at the comments of all four reviewers and revise and resubmit your paper. In particular, I invite your attention to the second reviewer comments. While submitting the revised version, please format the paper according PLOS One Stylistic guidelines.  ============================== Please submit your revised manuscript by Dec 27 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Srinivas Goli, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. We suggest you thoroughly copyedit your manuscript for language usage, spelling, and grammar. If you do not know anyone who can help you do this, you may wish to consider employing a professional scientific editing service. Whilst you may use any professional scientific editing service of your choice, PLOS has partnered with both American Journal Experts (AJE) and Editage to provide discounted services to PLOS authors. Both organizations have experience helping authors meet PLOS guidelines and can provide language editing, translation, manuscript formatting, and figure formatting to ensure your manuscript meets our submission guidelines. To take advantage of our partnership with AJE, visit the AJE website ([http://learn.aje.com/plos/](about:blank)) for a 15% discount off AJE services. To take advantage of our partnership with Editage, visit the Editage website ([www.editage.com](http://www.editage.com)) and enter referral code PLOSEDIT for a 15% discount off Editage services.  If the PLOS editorial team finds any language issues in text that either AJE or Editage has edited, the service provider will re-edit the text for free. Upon resubmission, please provide the following: The name of the colleague or the details of the professional service that edited your manuscriptA copy of your manuscript showing your changes by either highlighting them or using track changes (uploaded as a \*supporting information\* file)A clean copy of the edited manuscript (uploaded as the new \*manuscript\* file) 3\. We note that Figures 5, 7, 8, 9, 10, 11 and 12 in your submission contain map images which may be copyrighted. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For these reasons, we cannot publish previously copyrighted maps or satellite images created using proprietary data, such as Google software (Google Maps, Street View, and Earth). For more information, see our copyright guidelines: <http://journals.plos.org/plosone/s/licenses-and-copyright>. We require you to either (a) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (b) remove the figures from your submission: a\. You may seek permission from the original copyright holder of Figures 5, 7, 8, 9, 10, 11 and 12 to publish the content specifically under the CC BY 4.0 license.  We recommend that you contact the original copyright holder with the Content Permission Form ([http://journals.plos.org/plosone/s/file?id=7c09/content- permission-form.pdf](about:blank)) and the following text: “I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 ([http://creativecommons.org/licenses/by/4.0/](about:blank)). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.” Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission. In the figure caption of the copyrighted figure, please include the following text: “Reprinted from \[ref\] under a CC BY license, with permission from \[name of publisher\], original copyright \[original copyright year\].” b\. If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only. The following resources for replacing copyrighted map figures may be helpful: USGS National Map Viewer (public domain): [http://viewer.nationalmap.gov/viewer/](about:blank) The Gateway to Astronaut Photography of Earth (public domain): [http://eol.jsc.nasa.gov/sseop/clickmap/](about:blank) Maps at the CIA (public domain): [https://www.cia.gov/library/publications/the- world-factbook/index.html](about:blank) and [https://www.cia.gov/library/publications/cia-maps- publications/index.html](about:blank) NASA Earth Observatory (public domain): [http://earthobservatory.nasa.gov/](about:blank) Landsat: <http://landsat.visibleearth.nasa.gov/> USGS EROS (Earth Resources Observatory and Science (EROS) Center) (public domain): [http://eros.usgs.gov/#](about:blank) Natural Earth (public domain): [http://www.naturalearthdata.com/](about:blank) 4\. Your ethics statement should only appear in the Methods section of your manuscript. If your ethics statement is written in any section besides the Methods, please delete it from any other section. Additional Editor Comments: Kindly look at the comments of all four reviewers and revise and resubmit your paper. In particular, I invite your attention to the second reviewer comments. While submitting the revised version, please format the paper according PLOS One Stylistic guidelines. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes Reviewer \#3: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Reviewer’s comments Full title: Geographical clustering of under-nutrition and its predictors among under-five children in Ethiopia: Geographically weighted regression Analysis Manuscript number: PONE-D-20-27277 Corresponding author: Mequannent Melaku sharew, Mph in health informatics Wollo University Desse, ETHIOPIA Comments and questions I appreciate the authors for addressing such highly contextually and culturally affected public health issues. It so concerning that under-nutrition is continued to be a public health problem in Ethiopia. Therefore, evidence regarding the distribution of under-nutrition with its respective determinants is important for decision-making in dealing with its prevention and control program. The paper is well organized and presented. The study presents the findings of original research in the area of public health. And the results have not been published elsewhere. Having adequately performed relevant analyses with sufficient detail; it presents coherent and data-oriented conclusions. The article is presented in an acceptable level English language standard and reporting guideline. Additional comments and questions The author would respond to some of the following questions and comments. 1\. Title: “Geographical clustering of under-nutrition and its predictors among under-five children in Ethiopia: Geographically weighted regression Analysis”. Since the data for this study comes from the DHS source; it’s good to indicate it in the title. So, the author would restructure the title as…; “Using Geographically weighted regression Analysis to cluster under-nutrition and its predictors among under-five children in Ethiopia: Evidence from demographic and health survey”. 2\. Abstract: Line 30: Methods: in the method section of the abstract; the author should indicate that the data comes from DHS. Indicating it is very important as it adds quality to the paper (data from a large population) and attracts the readers at a first impression. 3\. Introduction: page: Line 99-10: “Geographical variation of under-nutrition can delay control and elimination of malnutrition which is the underline cause of most childhood disease that in turn brings a high proportion of under-five mortality.” This sentence is not clear or it seems incomplete. It would good for the author to rephrase it? 4\. Methods and materials: Study setting: Line 110: the author would briefly discuss information about study settings or Ethiopia. E.g about the nine regional states of Ethiopia and other relevant information to under-nutrition such as urbanization, economic and agricultural practices, etc. 5\. Discussion: Line 374 and the Implications of the study: Line 510: It gives more sound meaning if the author could combine discussion points with implications. The author tried to compare the findings of the study with other previous pieces of evidence and discuss discrepancies as well. That’s good. But it would be more informative if the author could also discuss the implications (policy, methodological and practical implications) for each point of discussion and explanations. in so doing, the author would remove the section “Implications of the study in Line 510”. 6\. Question: It’s clear that the distribution of under-nutrition in Ethiopia was indicated in DHS 2016. And the determinants as well. My question is; what specific value (scientific or methodological) did your study added to the existing body of evidence? Would you please indicate it clearly in your discussions? Keep in mind my comments in \#5 above. Reviewer \#2: The manuscript written in an organized and impressive way. There are minor comments I attached and author has to come up with justification for the comments. I attached the comments in word format for details Reviewer \#3: The author need to address the all the comments properly. The article can be a good addition in the field of spatial demography provided that author substantially revise the manuscript as per the suggestions. Reviewer \#4: 1. In the abstract part line 31, says that a stratified two stage cluster sampling was used for to include clusters. however, the stratification and stages of sampling has not available in the manuscript. So it is better you will add it in your manuscript. 2\. Your manuscript have no page numbers. You have to add it. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Fira Abamecha Reviewer \#2: No Reviewer \#3: No Reviewer \#4: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0248156.r002 Author response to Decision Letter 0 27 Dec 2020 dear editors, this are some of my answers for your comment Thank you, dear editor, for the comment. We have written the manuscript according to PLOSE ONE format using the guideline. We have revised the language usage, spelling, and grammar of the manuscript. We have removed the ethics statement that found in the declaration section. All the figures you noted (5, 7, 8, 9, 10, 11 and 12) was produced using shape file from open Africa website (<https://africaopendata.org/dataset/ethiopia- shapefiles>) which freely available website and enumeration area shape file from DHS website(<https://dhsprogram.com/data/dataset_admin/login_main.cfm?CFID=12423 68&CFTOKEN=c892a8da9855f981-8D71EDAA-BF68-5950-1D6BA7F0BB37D5E8>) and the we have produced the figures by ArcGIS version 10.7 software. Permission letter to use the Ethiopian enumeration area shape file and non-spatial data is attached as supplementary information(S5). For this reason, we have incorporated the source of Ethiopia shape file URL in each figure. 10.1371/journal.pone.0248156.r003 Decision Letter 1 Goli Srinivas Academic Editor 2021 Srinivas Goli This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Feb 2021 Using geographically weighted regression Analysis to cluster under-nutrition and its predictors among under-five children in Ethiopia: Evidence from demographic and health survey PONE-D-20-27277R1 Dear Dr. sharew, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, Srinivas Goli, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): All comments have been addressed Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#2: All comments have been addressed Reviewer \#4: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#2: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#2: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#2: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#2: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#2: Thank you Authors. The comments I raised were adequately addressed and in a coherent way. The research is well edited and addressed issues I raised in the previous review process. The only point I have is that figures given are too many and if there is a way to shorten it, the paper will be clear for the readers of the article. Reviewer \#4: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#2: **Yes: **Adem Abdulkadir Abdi Reviewer \#4: No 10.1371/journal.pone.0248156.r004 Acceptance letter Goli Srinivas Academic Editor 2021 Srinivas Goli This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 7 Apr 2021 PONE-D-20-27277R1 Using geographically weighted regression analysis to cluster under-nutrition and its predictors among under-five children in Ethiopia: Evidence from demographic and health survey Dear Dr. Melaku: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Srinivas Goli Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Concerns over the increasing demands on primary and urgent care services, and the rising costs of healthcare, have driven policymakers to capitalise on the potential of community pharmacies to provide medicines-related and public health services beyond medicines supply. This builds on the concept of ‘pharmaceutical care’ whereby community pharmacists utilise their increasingly clinical training and skills to deliver services such as medication reviews, minor ailments services, support for the self-care of long-term conditions and healthy lifestyle services (e.g. smoking cessation, weight management). In the United Kingdom (UK), the solution that pharmacy may offer to the financial and workload problems facing the National Health Service (NHS) has led to the introduction of healthcare policies which advocate closer involvement of clinical pharmacists and community pharmacy in the organisation and delivery of primary healthcare services. In England, this includes the introduction of a national scheme to employ pharmacists in general practices, an integration fund to support collaborative working with community pharmacies and the creation of new models of care including multispecialty community providers which will integrate primary and community providers, including pharmacy, to serve and improve the health of local populations. This echoes findings published in the Royal Pharmaceutical Society commission on future models of care, *Now or Never*, which called for “a national primary care strategy that embraces the potential of pharmacy alongside that of general practice and nursing, and bold and imaginative commissioning that supports new models of integrated care.” (p47) However, for the expansion of community pharmacy’s contribution to be a success, a greater understanding of its organisational context is required. Community pharmacies range from single-handed ‘independent’ pharmacies to large national or multi-national chains or ‘multiples’ and supermarkets. They are for-profit organisations delivering medicines-related healthcare alongside the sale of health and non-health related services and products. They are somewhat unusual amongst healthcare providers in that they depend upon income from a range of sources, both retail and healthcare, and are thus subject to business *and* healthcare policy drivers, regulations and pressures. The expansion of healthcare and other public service provision into the private sector has been increasing across Europe and other developed countries, driven by a desire to increase patient choice and access to services whilst improving efficiency. However, concerns have been raised over the quality and safety of patient care and the motivations of managers in private sector organisations where there is a need to balance delivery of healthcare with generation of profit. In England, community pharmacy dispensing, medicines-related healthcare and public health services are provided under contract to the publicly-funded NHS (responsible for commissioning the vast majority of primary, secondary and tertiary healthcare services both nationally (NHS England) and locally (clinical commissioning groups; CCGs)) and local authorities (responsible for commissioning public health services at a local level), primarily on a fee-for- service basis. The current contractual framework for community pharmacy was introduced in 2005 to place greater emphasis on the delivery of extended services alongside more traditional supply functions. It introduced three service tiers: nationally commissioned *essential* (e.g. dispensing) and *advanced* (e.g. medicines use reviews (MURs)) services, and *locally commissioned* (medicines-related healthcare and public health) services. However, the associated rise in the *quantity* (range and volume) of services provided has been associated with a worrying increase in pharmacists’ workload and claims that excessive pressure to meet contractually- incentivised business targets, particularly in the larger pharmacy chains, risks compromising service *quality*. As part of a wider, mixed-methods study of clinical productivity in English community pharmacies published in full elsewhere, this qualitative investigation sought to explore the organisational and extra-organisational factors associated with the quantity (range and volume) and quality of services provided. # Method This paper presents findings from a series of semi-structured face-to-face and telephone interviews conducted between November 2014 and April 2015. Research ethics approval was obtained from the UK National Research Ethics Service (13/WM/0137) and endorsed by the University of Manchester Research Ethics Committee (13025). ## Setting The study was conducted across nine socio-economically diverse geographical areas of England, purposively selected to cover locations in the north and south, rural, urban and suburban areas and differing degrees of deprivation. All community pharmacies in these primary care administrative areas, bar those owned by four non-participating national chains, were invited to take part in a survey of organisational characteristics. Of those responding (227/800 (34.6%)), 39 (from a stratified random sample) participated in a patient survey, and all interviewees were drawn from these 39 pharmacies and the NHS commissioning bodies (CCGs and NHS England area teams) operating across the nine study sites. ## Sample Interviewees were selected purposively to include those directly involved in the commissioning and provision of community pharmacy services: at least one pharmacy commissioner (from NHS England and clinical commissioning groups (CCGs)) and a cross-section of frontline and superintendent (appointed to represent the pharmaceutical aspects of a retail pharmacy business) pharmacists from pharmacies of differing ownership types (independent, small/medium and large chains) from each geographical area. ## Recruitment Participants were contacted by email, followed up by telephone to discuss the study, before obtaining written informed consent to participate. Where individuals chose not to participate, attempts were made to recruit a similar replacement (by study area; pharmacy ownership type) until all available options were exhausted. ## Interview content and process Interviews, conducted by SJ, TF and FB (none of whom had pre-existing relationships with any participant), took a broadly phenomenological approach, with topic guides specific to each stakeholder group (commissioner, pharmacist, superintendent pharmacist) developed from the study aims and research literature. Lines of questioning explored definitions of quality, the relationship between the quantity and quality of service provision in community pharmacies, opportunities and barriers to maximising clinical productivity and the mechanisms by which different organisational characteristics may help or hinder this objective. A prompt sheet listing the organisational factors of interest was sent in advance. All interviews were audio recorded, with consent, and transcribed verbatim. ## Data analysis Interview data were thematically analysed, supported by NVivo 10. A framework approach was adopted, with analysis involving five steps: familiarisation; developing a thematic framework; indexing; charting; and mapping and interpretation. Two researchers (TF and FB) undertook the first four steps collaboratively, developing the thematic framework through independent familiarisation with different sets of interview transcripts, followed by close discussion and agreement of identified themes. Where consensus could not be reached, a third researcher (SJ) was brought in. The agreed thematic framework was applied and charted by TF and FB and the final stage of mapping and interpretation was undertaken by SJ. The face validity of the findings was examined during a stakeholder workshop held in July 2015 attended by service commissioners, community pharmacy representatives and service users. # Results ## Interviews conducted Forty interviews were conducted face-to-face (21) or by telephone (19), lasting 33–97 minutes. Ten participants were service commissioners and 30 were pharmacists/superintendent pharmacists. No substantive differences were found between the data elicited from face-to-face and telephone interviews. The findings from pharmacist and service commissioner interviews are presented together under the main themes: definitions of quality, organisational and extra-organisations characteristics associated with the quality and quantity of service delivery. ## Definitions of quality Interviewees were asked how they would define service quality in community pharmacy, specifically in relation to dispensing services and MURs. MURs were one of the advanced services introduced in 2005 and involve a pharmacist consultation to improve a patient’s understanding of and adherence to their medicines. For dispensing, speed and accuracy were the most commonly cited elements of service quality. Most frontline and superintendent pharmacists believed that speed of dispensing, and thus limited waiting time, was valued by patients above anything else. However, for pharmacists themselves, and for many service commissioners, accuracy was paramount. Additionally, clinical aspects were considered by a number of pharmacists and commissioners to be an important element of quality either through counselling or the clinical check. > “*I would want a really strong cognitive element at the beginning in > terms of the clinical check so people really thinking about what > they’re doing rather than necessarily looking at prescriptions > thinking “have we got this in stock?” Thinking critically about that > prescription and at the end of the process making sure that patients > know what they’re taking and why they’re taking it and recognising > that there is somebody that they can call upon should they have any > problems. So there’s the whole of the patient counselling piece at the > end which I think is often forgotten about in the spirit of getting > prescriptions done as quickly as possible to meet customer demand*.” > *\[Superintendent Pharmacist 3\]* Less often mentioned but considered to be additional domains of dispensing quality were the importance of maintaining stock levels, to prevent patients having to wait or return, and offering good customer service. In relation to the MUR service, quality was defined by most interviewees in terms of outcomes for patients, e.g. increasing knowledge and understanding of medications, improving adherence, addressing side-effects, improving clinical outcomes and quality of life, and providing reassurance. > “*I only find them \[MURs\] worthwhile when you have an outcome at the > end of it. So if, say for example, they've been experiencing a > side-effect and they haven't linked it to a particular drug, or you've > actually offered something more to them to either improve their > regime, or to resolve a problem that they've been having, or there's > been some kind of positive outcome where they either feel more > reassured about taking the medication or there's been some kind of > change, or something else that you have to offer which has helped in > any kind of way—then I find them worthwhile*.” *\[Pharmacist 38\]* Targeting of MURs to patients most likely to benefit, e.g. those taking several medicines or people with certain long-term conditions including respiratory and cardiovascular disease, was mentioned by several interviewees as an important element of quality. Many perceived that taking a patient-centred approach also contributed to a high quality MUR, utilising communication skills and tailoring the consultation to a patient’s needs. Where speed was seen an important element of quality dispensing, allowing adequate time was perceived as important for MURs to be of benefit. Less frequently mentioned were the provision of healthy living advice, integration with other pharmacy or general practitioner (GP) services and for the patient to be engaged with the process. ## Organisational characteristics associated with the quality and quantity of service delivery ### Staffing and skill-mix All pharmacists/ superintendent pharmacists and most service commissioners mentioned staffing and skill-mix as an important, possibly the most important, organisational factor influencing the quantity and quality of community pharmacy services. This related to overall staffing levels, skill-mix, training, teamwork, delegation and continuity of staff. Insufficient staffing levels were often reported with implications for service quality (increased waiting times, decreased clinical input, increased risk of dispensing errors) and service quantity (dispensing prioritised over MURs and other services). > “*I think the main thing is staffing. It’s because margins are being > cut and employers, owners, can’t afford to employ more staff, so > everybody’s under pressure and the things that go are the services. > You can’t take ten minutes out of your day to do an MUR if you’ve got > piles of prescriptions that need checking*.” *\[Pharmacist 41\]* For many commissioners, continuity of pharmacy staff and turnover was perceived as a particular problem to the reliability and quality of service delivery. > “*With the pharmacy services, it’s often the individual pharmacist > that’s either keen or not keen. And so we very often have the > experience of phoning or going into a pharmacy and saying, “Oh hello, > can I ask you if you’re providing the minor ailments service today?” > for example. And they’ll say, “Oh no, we’ve only got a locum on > today”, or, “Oh, well, we used to but our regular manager’s gone off > on maternity leave and we’re covered by temporary staff now*”” > *\[Commissioner 1\]* More commonly discussed by pharmacist and superintendent interviewees than staffing levels was the skill-mix of the pharmacy team and its influence on the quality and quantity of pharmacy services. A key enabling factor was the support of a trusted and competent team, each trained to an appropriate level, to free up the pharmacist to be more patient-facing and provide more clinical and higher quality services, spending more time with patients. It was suggested that this would be more likely to improve patient outcomes (e.g. adherence) and reduce waste. Many interviewees (both pharmacist and commissioner) emphasised the role of accuracy checking technicians (ACTs; pharmacy technician or other member of the pharmacy team qualified to conduct the final accuracy check on dispensed prescriptions) in the dispensing process as a successful way of freeing up pharmacists’ time for clinical services, extending the range offered, volume delivered and also the quality of those services (e.g. time spent with a patient for an MUR). > “*I think a big thing is if you can get an ACT…it frees up your time > incredibly. You have a dispenser and an ACT, you’ve got nothing to > worry on the dispensing side, then you’ll become fully > patient-focused…you can offer other services, diabetes screening and > blood pressure monitoring*.” *\[Pharmacist 15\]* Others suggested that the ability to employ a second pharmacist (most pharmacies operate with only one pharmacist) would provide the ideal solution to pharmacies looking to expand their range of services. However, the financial barriers to this were recognised. > “*I would really like to see two pharmacists in every premises, so > that they could cover for each other and provide the clinical services > that I would want to see from community pharmacy. I don’t think that’s > going to happen until it’s part of the national contract. And perhaps > the national contract has to move away from an item-based fee basis, > onto more of a population, wellbeing basis*.” *\[Commissioner 10\]* ### Workload and its management Most interviewees perceived that workload and time constraints were a ubiquitous barrier to both the range and quality of services provided. Several factors were thought to have contributed to increasing demands on pharmacists’ time including: increasing (and fluctuating) dispensing volumes; reduced staffing; the growing range of pharmacist-led services; an increasing regulatory and administrative burden; and stock shortages. > “*It’s going to get to tipping point, I think, where there’s so much > pressure on the teams. And, not just the pharmacist time, the > dispensers I’ve worked with, I can’t say enough good things about > them…sometimes they’re in the back and they’re in tears because > they’re that stretched. They know they’re working at absolute capacity > and it’s still not enough, and they still can’t get through the > work*.” *\[Pharmacist 29\]* High dispensing volumes limited both the quality and quantity of services delivered in a number of ways. It reduced the time available for counselling and follow-up of patients, increased waiting times and threatened accuracy. Because of the reactive nature of dispensing, this was often prioritised over other services, reducing not only service volume but also quality, e.g. by only offering MURs to less complex cases or reducing the time spent in consultation. A number of workload management strategies were described by pharmacist interviewees. These included: appointment systems for MURs and other services, mechanisms for handling dispensing of repeat prescriptions, pharmacy level procedures, and the use of technology. Service commissioners endorsed the need for pharmacies to adopt better workload management systems. > “*Everybody gets lulls in the day…so it might be that if you're quiet > from two 'til three before the sort of late afternoon surge kicks in, > then it might be that if, you were offering …a pre-bookable service, > like health checks…you would say to your staff “well, look, that's a > really good slot.” That's what the companies are good at doing…they > literally look at the till receipts, so they can see the surges in > business…and the dispensing flow and everything, and they target the > services around those dips in walk-ins*.” *\[Commissioner 3\]* Whilst some pharmacists were able to use an appointments system for managing competing workloads, others felt that this was not workable because of the unpredictable nature of their walk-in business. A number of methods of organising the more predictable workload associated with repeat prescription dispensing were also described by pharmacist interviewees, including the use of collection services, off-site dispensing and pharmacy-level procedures for managing dispensing workflow, all of which helped some pharmacies free up time to maximise both the quantity and quality of service provision. ### Pharmacy ownership and organisational culture A number of opportunities and barriers to service quality and quantity were identified by interviewees in relation to the type of organisation a pharmacy belonged to (e.g. large multiple, supermarket, smaller chains and independents). Many interviews highlighted the central role of organisational culture, or “the way we do things around here,” describing it in terms of the extent to which business targets (or quantity) were prioritised over service quality, the value the organisation (head office, the superintendent pharmacist or pharmacy owner) placed on investing in staffing and skill-mix, and management style and structure. Most pharmacists interviewed reported the existence of service targets to help maximise service volume. Some recognised that targets could be helpful in ensuring that a range of services was provided. However, where the culture of the organisation was one where the pressure to meet targets was perceived as excessive or as prioritising profit over meeting patient needs, this was viewed as detrimental to service quality. > “*Everybody I know thinks it's quantity, not quality. If you don't set > targets, maybe nothing will ever get done. But setting targets just > creates rubbish; you just end up with rubbish being done to just earn > some money*.” *\[Pharmacist 34\]* Although it was recognised that all pharmacy organisations placed a certain degree of pressure on pharmacists to maximise productivity, a number of interviewees perceived this pressure to be excessive in some large multiples. Reported managerial strategies to enforce targets varied from collaborative, supportive and encouraging to autocratic, discouraging and humiliating. There were reports of daily pressure from some area managers (responsible for overseeing pharmacy branches in a locality) to hit targets and, in extreme cases, bullying. Conversely, in pharmacies where the culture supported investment in staffing, skill-mix and training, benefits were seen in terms of both the quality and quantity of services provided. > “*Skill mix we very much believe in, that’s why we have ACTs, and we > train staff up as much as we can, as long as we feel that they’re > going to be able to practise their new role and use those skills. We > pretty much believe that the majority of our staff…should be at least > pharmacy assistant trained. And, we’re quite happy to use our ACTs, > such to enable pharmacists to carry out the services, and that’s the > philosophy of the company*.” *\[Superintendent Pharmacist 3\]* Some interviewees perceived that, staffing levels were not increasing in line with increasing workloads and, in some cases, were being pared back, particularly in large multiples. It was felt that, in such cases, the pharmacy team’s capacity to provide additional services was compromised, and patient safety might be at risk. Independent pharmacies were viewed by some to be more willing to increase staffing levels which helped engender a better patient experience. > “*\[Patients\] said, we’re treated with courtesy, we’re treated with > respect, we’re dealt with promptly and it’s all the sort of things > that the patients value. They’ve always got time for you. There’s > always a pharmacist there, so I can quickly talk to them…despite the > sheer volume of prescriptions that they deal with. Definitely the > independents have that ability because they always have more staff, > whereas the multiples are very much it’s down to that n^th > degree of the staffing*.” *\[Commissioner 1\]* Yet others felt that a better quality of service could be delivered by large multiples because of investment in training for pharmacists and staff. A small number also perceived that this was augmented by the level of support for the pharmacy team from head office. The role of management style and structure in supporting service delivery was also highlighted in relation to the extant culture of the organisation. For example, having a non-pharmacist area manager, who could be perceived as unreceptive to pharmacists’ problems, was sometimes seen as detrimental to productivity; having a non-pharmacist store manager, conversely, could be seen to be beneficial by taking on pharmacy management and administrative tasks and freeing up pharmacists’ time. ## Extra-organisational factors associated with the quality and quantity of service delivery ### Patient and population characteristics and expectations The demographic of the local population was perceived to influence the range and volume of services community pharmacies could offer. For example, a number of pharmacists highlighted that being situated in an area with a large proportion of older residents limited both the number of MURs they could conduct and the range of other services they could offer. Although it was recognised that domiciliary MURs could be conducted, regulatory and organisational barriers often meant that they were not. > “*… unfortunately, those people that…most need our services, are quite > often housebound patients that you never see in the pharmacy…things > like MURs are completely useless, because you’re not seeing the > patient*.” *\[Pharmacist 3\]* There was also the perception that older people were more likely to visit their GP for services that could otherwise be provided by the pharmacy (e.g. flu vaccinations; glucose monitoring) either through preference or service restriction. A number of pharmacists and commissioners perceived that public perceptions and expectations of community pharmacy–of the services available and pharmacists’ roles–could influence both the volume and quality of services. Although it had been ten years since the introduction of MURs and an extended range of other medicines optimisation and public health services from community pharmacies, public perception that pharmacies are only there to dispense prescriptions remained an important perceived barrier. ### Community pharmacy–general practice relationships The strength of a pharmacy’s relationship with its local GP surgery(ies) was cited by a large number of pharmacists, superintendents and all commissioners as an important factor influencing the quality and quantity of community pharmacy services. Positive relationships were seen to help nurture interdisciplinary practice, foster closer working around patients, increase effective signposting and improving communication. > “*I also think that good communications and relationships…makes a > huge, huge difference to the working lives of pharmacists and GPs > \[…\] So the wider working relationships are as important as the staff > to underpin that service*.” *\[Commissioner 5\]* Many pharmacists believed that good relationships with local GP surgeries enabled prompt resolution of issues, for example problems with prescriptions, clinical interventions and stock shortages. This had a positive effect on service quality, with the opposite effect where poorer relationships existed. > “*Sometime we have problems, you know, with the GP …the receptionist, > but I suppose that everybody’s got the same problem. The way you’re to > trying to help a situation, they will hinder it basically. That causes > a massive problem with the flow of things*.” *\[Pharmacist 28\]* A number of pharmacists felt that positive relationships encouraged GPs to initiate contact with pharmacists, for example, to request advice and follow-up recommendations from MURs. Indeed, some perceived that effective working relationships were optimised where local GPs had developed trust in the pharmacist. Good working relationships could therefore help increase pharmacy service volume and range through referral or signposting of patients. Commissioners also spoke about the importance of GP endorsement of pharmacy services, to increase patient uptake. > “*If a GP said to a patient, “Oh, by the way the pharmacy down the > road offers this really good service called an MUR and I think it’s > really important that you go and talk to them about your inhaler > technique”, the patient is far more likely to do that… if you get that > endorsement from GPs around pharmacy services I think that carries a > lot of weight with patients*”. *\[Commissioner 6\]* However, for some pharmacists, referral or signposting to a pharmacy was believed to be, at best, selective for some services, with competition between pharmacies and general practices to provide some services encouraging silo behaviours. > “*…pharmacy/GP integration is something that needs to be focused > upon…The problem is… they’re two competing businesses…there are > certain cross-over areas, such as flu vaccinations and things like > that….If we work together…pharmacists can help the GPs to achieve > their targets…\[and\] at the same time…also enable the pharmacist to > provide another service*.” *\[Pharmacist 3\]* ### Commissioning and contractual arrangements As community pharmacies are private businesses required to make a profit, many of the pharmacists and superintendents interviewed perceived that the level of remuneration for services was insufficient for investment in staffing and infrastructure (e.g. consultation rooms/IT) necessary to be able to deliver these services. > “*To enable you to do MURs you have to invest in a consultation > room…Then you need…a second PMR \[patient medication record\] system > to use in that room. In order to free the pharmacist up you need a > team of better trained counter staff and dispensary assistants. When > you add up all those costs you would be much better off to forget the > MURs … the economics don't work*.” *\[Superintendent Pharmacist 4\]* Other contract-related issues seen as having a detrimental effect on clinical productivity included the short commissioning cycle operated by NHS commissioners, seen as a barrier to investment in staffing levels and training. > “*\[Pharmacies\] don’t know whether it’s worthwhile doing \[services\] > because two years down the line they mightn’t get paid for it, or they > might be expected to do it without the same reimbursement*.” > *\[Pharmacist 35\]* Some interviewees suggested changing the basis of remuneration from fee-for- service to service quality or outcomes, e.g. having a pharmacy quality and outcomes framework (QOF) in line with general practice. It was suggested that this could incentivise an increase in the quality of community pharmacy services, and also alleviate pressure on general practice workloads. > “*A QOF for pharmacy would be good… \[The pharmacy contract\] doesn’t > appear to be aligned particularly well to the GP contract so we’ve not > necessarily got two professions working in the best interests of > patients. I think a root and branch review of the pharmacy contract > would be a good start, and start to recognise the clinical value that > pharmacists add through their interaction with patients*” > *\[Superintendent Pharmacist 3\]* Lastly, at the time this study was conducted, the NHS in England had just undergone a substantial re-organisation leading to the fragmentation of commissioning responsibilities for community pharmacies between NHS England (essential and advanced services), CCGs (locally commissioned medicines-related healthcare services) and local authorities (locally commissioned public health services). This not only caused confusion amongst many pharmacy interviewees and increased bureaucratic processes but was also associated with some service decommissioning. This reduction in service provision, by reducing service accessibility to patients, was also seen as lowering the quality of NHS services. > *“Most of the public health initiative type services*, *have gone onto > the local authority tendering format which has put a much increased > burden bureaucratically on community pharmacies…there are a number of > pharmacies that have said*,*…“That’s too much like hard work to > complete all of that paperwork …*. *Actually I’m busy enough as it is; > I’m not going to bother*.*”… That leads to poorer quality of service > because there’ll probably be fewer pharmacies delivering certain > services and therefore you haven’t got the accessibility for the > service that was previously available*.*" \[Superintendent Pharmacist > 5\]* # Discussion The findings presented here are based on interviews with community pharmacists, superintendents and commissioners as part of a larger study into clinical productivity in community pharmacy. These findings have highlighted the implications of increasing workloads in community pharmacy and the importance of adequate staffing and skill-mix to support service expansion alongside increasing dispensing volumes. Organisational culture plays a central role in determining productivity in relation to the extent to which it engenders staff and team development or prioritises business targets over service quality. Extra-organisationally, the characteristics of the local population and patient expectations, as well as working relationships with local GPs and (lack of) integration with the services they provide, can limit the uptake of services. The influence of the nature of the contractual framework and commissioning processes on the quantity and quality of service provision has also been highlighted. Although not all of these issues are new, this paper provides the first comprehensive insight into the organisational and extra-organisational factors associated with the quantity (volume and range) and quality of services provided by community pharmacies and an exploration of their mechanisms of action. These findings thus offer explanatory insights for the outcomes of a linked quantitative investigation, which concluded that whilst a pharmacy’s dispensing volume was positively associated with local population need, the volume of MURs delivered were driven more by corporate ownership and that, whilst levels of staffing and skill-mix were associated with dispensing volume, they were not associated with levels of MUR provision. This study has some important limitations. Whilst socio-economically diverse, the geographical coverage of study sites was limited to those selected for the wider study. As with most qualitative studies, the findings cannot be said to be generalisable to the wider population. Nonetheless, the sample was purposively selected to cover the full range of pharmacy ownership types and the sample size was sufficient for the requirements of data saturation to be met. Furthermore, our sample provided perspectives from front-line community pharmacists, superintendents and commissioners. The problem of increasing workloads in UK community pharmacy since the introduction of the 2005 contractual framework has been well documented, particularly in relation to the impact on pharmacists’ well-being and the potential detriment to patient safety. This problem is not unique to the UK, with workload implications of pharmacists’ role expansion and workforce shortages reported as having an impact on levels of job stress and commitment, patient safety and the adoption of new services in several countries, particularly the United States (US).\[–\] The central importance of adequate staffing, appropriate skill-mix and teamwork in supporting this increasing workload and the quality and safety of community pharmacy services was emphasised throughout these interviews. In the absence of any growth in funding for community pharmacy services, or indeed the threat of funding cuts, pharmacies will need to rethink how they can remain competitive, viable and make increasing contributions to medicines supply and public health services. Whilst staffing levels are being cut, this could become unmanageable in light of already excessive workload pressures and job stress amongst community pharmacy staff. It will therefore be necessary for community pharmacy organisations to look more closely at skill-mix and opportunities to deploy staff effectively to support the expansion of services. In addition to often a single pharmacist, the community pharmacy team includes pharmacy technicians, medicines counter assistants, dispensing assistants, and accuracy checkers. Although in some European countries (e.g. Denmark, the Netherlands) pharmacy technicians routinely undertake dispensing without pharmacist supervision, in the UK and US pharmacists are required to either undertake or supervise different parts of the process. Some research exists which supports the expansion of the roles of pharmacy technicians and other support staff in community pharmacy which would help free up pharmacists’ time for clinical roles. Community pharmacies are long-standing private sector organisations, and the expansion of services provided by them offers a good exemplar to explore the implications of a more widespread move towards a mixed economy of healthcare. There may be a risk in some organisations for an organisational culture to prevail where profits are prioritised over patient care, and service quantity over quality. Indeed, the quantitative research linked to the interviews reported here suggests that the delivery of some community pharmacy services may be driven more by organisational factors than local population need. There is a trend towards the corporatisation of healthcare (the use of market mechanisms, growth in for-profit provision, and increasing size of provider organisations), in pharmacy and also in primary care provision more widely. These findings may therefore have implications for the way in which such services are commissioned, to ensure that service quality is incentivised alongside quality. A contract which offers remuneration only on a fee-for-service basis appears, from these findings, to incentivise quantity over quality. Moreover, the short term nature of local service commissioning does little to encourage growth. Healthcare administrations need to better understand the nature of private sector provider organisations when implementing change, so that the commissioning of publicly-funded healthcare services from private sector organisations, such as community pharmacies, incentivises quality as well as quantity of service provision by offering remuneration on the basis of process and outcome quality measures, although these remain to be defined. Moreover, commissioning and remuneration processes should be integrated within healthcare systems to encourage collaborative working, in this case between community pharmacies and general practice. Rewarding joint working through integrated commissioning could encourage the signposting of patients to community pharmacy services by GPs, for example, which is known to encourage service uptake. # Conclusion Through identifying the key drivers and barriers to service expansion in community pharmacy, this study offers valuable insights both for community pharmacy organisations and for service commissioners. Exploration of the organisational context has highlighted for pharmacies the importance of staffing and skill-mix and engendering a supportive, team-building culture. For healthcare administrations, it has suggested a need to better understand the organisational culture within private sector providers to ensure the quality of services commissioned outwith the public domain. Moreover, the importance of extra-organisational factors such as relationships with local general practices and contractual arrangements has implications for future contractual and remuneration arrangements. # Supporting information We would like to thank all study participants and other pharmacy stakeholders who helped to facilitate this study. We would also like to acknowledge the involvement of Professor Karen Hassell, California Northstate University, who contributed to study conceptualisation and methodology, funding acquisition, and the early stages of project administration. [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Health Services Research Centre, Alliance Manchester Business School, The University of Manchester, Manchester, United Kingdom [^3]: Current address: NHS England (Cheshire and Merseyside), Liverpool, United Kingdom

# Introduction Experimental evolution is a research approach which is increasingly used to analyze the functioning of fundamental evolutionary mechanisms (mutations, selection, drift) in real time in controlled conditions. Several experimental evolution studies have shown the ability of model organisms, such as the fruit fly *Drosophila melanogaster*, to adapt quickly to various adverse conditions. It is usually assumed by default that the observed adaptation to a new environment is explained by changes in the gene pool of the experimental population, or by more transient epigenetic changes. Meanwhile, it is known that microbiome plays an important role in the lives of many animals; it can be transmitted from parents to offspring and influence the fitness of macroorganisms in particular environmental conditions. Thus, within the framework of the so-called “hologenomic theory of evolution,” which is rapidly gaining popularity among evolutionary biologists, it is proposed that the basic unit of selection is not a separate organism, but a holobiont, i.e. a system consisting of a macroorganism and its microbiome. *D*. *melanogaster* is a well- studied species with a relatively simple microbiome and well-developed genetic tools. It is considered a good model for investigating host–microbe interactions. Until very recently, there was almost no direct experimental evidence that the observed increase of fitness of model animals in evolutionary experiments can be explained to some extent by changes in the microbiome, rather than by changes in the macroorganism itself. Nonetheless, many facts are compatible with this assumption. For instance, experiments on “artificial speciation” of aphids performed in the mid-20th century have shown that transfer to a different plant species can result in rapid emergence of partial reproductive isolation in these insects. At that time, this unexpected result was not properly explained. Later, however, such obligate symbionts of aphids as *Buchnera* were discovered; these bacteria provide aphids with adequate nutrition. It is likely that adaptation of aphids to the new host plants, as well as some other adaptive abilities (e.g., temperature optimums), depend on the evolution of the endosymbionts. Moreover, some experimental data implies that microbiome probably can influence mate choice, which, in turn, can facilitate rapid development of partial behavioral isolation, although recent researh casts some doubt on the validity of these conclusions. The ability of *D*. *melanogaster* to adapt quickly to adverse conditions, including food substrates with high NaCl concentrations, makes this species an appropriate object to study mechanisms of adaptation. Although *D*. *melanogaster* is not typically found on high salt substrates in nature, adaptation of this species to salty food proved to be a convenient experimental model; moreover, there are numerous salt-adapted species in other Diptera families, e.g., Ephydridae. Salt concentrations exceeding 2% are a negative factor for the wild type *D*. *melanogaster*, and concentrations higher than 4% can be fatal to larvae and adults. Nevertheless, several evolution experiments have demonstrated that laboratory lines of *D*. *melanogaster* are able to adapt to NaCl concentrations as high as 6–8% over several dozen generations, given that salt concentration increases gradually. Moreover, the evolved tolerance to salty food can result in higher reproductive efficiency of the flies not only on the salty substrate, but also on the standard, favourable food substrate. These results do not contradict the assumption that flies' adaptation to salt is due to changes in microbiome. Such changes may probably act as a broad-scale adaptation, simultaneously increasing the fitness of the flies on various food substrates, although this possibility has not yet been experimentally tested. Phenotypic plasticity under exposure of naïve *D*. *melanogaster* to salty food has been described, including anal papillae size reduction in larvae and changes in expression levels of several genes involved in secretion. Otherwise, little is known about the mechanisms of adaptation of *D*. *melanogaster* to high NaCl concentrations in the course of evolution experiments. To our knowledge, there are no data on the possible contribution of microbiome (bacteria and yeasts, carried by the flies on the cuticle and in the gut) to such adaptation. On the other hand, data on *D*. *melanogaster* microbiome (which has become a popular object of research in recent years) indicate that such a contribution is conceivable. It has been shown that microbiome modulates host metabolic gene expression, metabolic response to diet, the efficiency of food resource exploitation, immune response, adult life expectancy, and larval growth rate, and that host genetic control of microbiome affects the nutritional status and other physiological characteristics of the flies. Microbes can rescue undernutrition in *Drosophila*. Moreover, microbiome species richness in *Drosophila* species correlates with the diet of the flies, while different species of yeasts present in the food substrate influence differently the survival of larvae and duration of larval development. Flies carry bacteria and yeasts in their guts and on the body surface, and progeny feeding on a substrate on which their parents had lived can ensure transgenerational transmission (‘inheritance’) of the microbiome. In the current paper, we focused on the possible impact of yeasts on the adaptation of *D*. *melanogaster* to high salt diet, because we noticed that fly lines reared on salty food usually carry more yeasts than the lines reared on standard food. Further research is needed to assess the role of symbiotic bacteria as well. The current study builds on two previous findings. First, we have shown previously that the two *Drosophila* lines (Fs1, Fs2) which have been reared on the salty substrate for 11 months became more tolerant to salty food and reproduced on this substrate more efficiently than the two control lines (Fn1, Fn2), which have been reared on the standard (favourable) substrate. The same four lines are used in the current study. Here we show that the lines Fs1, Fs2 are still more salt-tolerant than the lines Fn1, Fn2 after three years of the evolution experiment. Second, in order to assess the possible impact of the microbiome on adaptation of *Drosophila* to high NaCl concentration, we have previously compared the reproduction efficiency of the flies on the high-salt food medium pre-inoculated with homogenized flies from either salt-tolerant or control lines. We found that pre-inoculation with homogenized flies from salt-tolerant lines improved reproduction of the flies on a high-salt medium compared to pre-inoculation by homogenized flies from the control lines. We obtained this result in four laboratory lines (two salt-tolerant and two control ones) reared in population cages with overlapping generations. We also found remarkable differences in the abundance and taxonomic composition of yeasts between salt-tolerant and control *Drosophila* homogenates. These results agree with the assumption that the increased tolerance of *D*. *melanogaster* to salty food as observed in evolutionary experiments is probably due to quantitative or qualitative changes in the microbiome which enhance the fitness of the flies (and the entire assemblage—holobiont) on salty food. Here we test the robustness and replicability of this result using four other *D*. *melanogaster* lines (Fs1, Fs2, Fn1, Fn2), maintained under different conditions (in glass vials with non- overlapping generations). We also attempt, for the first time, to reveal specific components of microbiome that influence the fitness of *D*. *melanogaster* exposed to salty food. We show that yeast strains isolated from salt-tolerant fly lines (Fs1, Fs2) appear to enhance reproduction of naïve *D*. *melanogaster* on salty food more efficiently than yeast strains isolated from control lines (Fn1, Fn2). The study included four steps: 1. First, we ascertained that the two *Drosophila* lines (Fs1, Fs2) which have been reared on the salty substrate for three years in the course of our evolution experiment, are more tolerant to salty food and reproduce on this substrate more efficiently than the two control lines (Fn1, Fn2), which have been reared on the standard (favourable) substrate. We had tested the same four lines previously, 11 months after the start of the evolution experiment, and obtained positive result. In the current study, we show that this result is replicable after three years from the start of the experiment. 2. Next, we confirmed that pre-inoculation of the high salt medium with homogenized salt-tolerant flies (Fs1, Fs2) improved the reproduction of control (naïve) flies on this medium compared to pre-inoculation by homogenized flies from the control lines (Fn1, Fn2). Thus we show that the results obtained previously are replicable in four other fly lines, reared in different conditions (see above). 3. We evaluated the abundance and taxonomic composition of yeasts in the four experimental lines of *D*. *melanogaster*. Pure cultures were obtained from five yeast strains found in either control or salt-tolerant lines (three strains from Fs1 and two strains from Fn1). We show that there are robust differences between yeast communities from salt-tolerant and control fly lines. Most importantly, one yeast species, *Starmerella bacillaris*, is abundant in all salt-tolerant lines tested to date (Fs1, Fs2, and two other salt-tolerant lines tested in) and absent or very scarce in all control lines (Fn1, Fn2, and two other control lines tested in). 4. Finally, we examined for the first time the influence of each of the five yeast strains on the fitness of naïve flies exposed to salty food. We found that strains isolated from the salt-tolerant flies improved the reproduction of the flies on salty food better than strains extracted from the control flies, and better than *Saccharomyces cerevisae*, the common bakers’ yeast (which is absent in the homogenates of the experimental flies). *Starmerella bacillaris* yeast demonstrated the strongest positive effect. Thus, the results are compatible with the hypothesis that yeasts contribute to the adaptation of *D*. *melanogaster* to high salt substrates; we also identified components of the yeast microbiome which are likely to be responsible for this contribution. # Material and methods ## Design of the evolutionary experiment and experimental populations In the current study we used *D*. *melanogaster* lines which have been living for three years (about 80 generations) on one of the two food substrates: standard (favourable) or high salt (stressful). The initial population of *D*. *melanogaster* was derived from 30 wild flies caught in southwestern Moscow (Russia) in September 2014. The evolutionary experiment started in October 2014 with *D*. *melanogaster* lines derived from the initial population and cultivated on different food substrates. Each line was derived from 30 flies randomly chosen from the initial population; no isofemale lines were established. The flies were reared at 23–24°C and natural lighting. We used four lines in the current study: Fn1, Fn2: lines reared on the standard laboratory food medium (60 g inactivated yeast, 35 g semolina, 50 g sugar, 45 g crushed raisins, 8 g agar, 2 g propionic acid per 1 L of food). Hereafter, this food medium is denoted by the letter N. Fs1, Fs2: lines reared on a salty food medium (the medium N with addition of 40 g of NaCl per 1 L of food). Hereafter, this food medium is denoted by the letter S. Populations were maintained in cylindrical glass vials, 64 mm in diameter and 100 mm in height, cotton-plugged, with 66.7 mL of food per vial. Each vial also contained a drinking bowl: a 1 mL cylindrical plastic reservoir filled with moist cotton wool. Each population occupied one vial. Every two weeks, all adults in the vial were immobilized by carbon dioxide; subsequently, 10 males and 10 females were selected at random and placed in a new vial with fresh food; the remaining flies were discarded. ## Tests on the efficiency of reproduction on different food substrates In order to test the flies’ reproduction efficiency on different food substrates (with or without previous inoculation by yeasts or homogenized flies, see below), we placed 10 pairs of parents in a vial with food and kept them there for 7 days, after which the parents were discarded. Subsequently, we counted their offspring (adults and pupae) daily until the adult offspring ceased to emerge. We used the total number of adult offspring as a measure of reproduction efficiency. We used the same testing technique previously in the course of our experiment. We performed three tests on the reproductive efficiency (sections 1, 2 and 4 of the Results); 5 to 10 test vials were used for each fly line/substrate combination. **In the first experiment** (section 1 of the Results), we evaluated the fitness of the flies from the four lines on the substrates N and S. There were eight fly line/substrate combinations (parents from the line Fn1 tested on food N (Fn1/N), Fn2/N, Fs1/N, Fs2/N, Fn1/S, Fn2/S, Fs1/S, Fs2/S). Parents were selected at random from the experimental populations. **In the second experiment** (section 2 of the Results), we compared the reproduction efficiency of salt-naïve flies from the control line Fn1 on salty substrate pre-inoculated with homogenized flies from one of the four experimental lines. Inoculation was performed two days prior to placing the parents in the vial, so that microorganisms in the homogenate had time to reproduce. We used virgin males and females from the line Fn1 as parents. **In the third experiment** (section 4 of the Results), we evaluated the reproduction efficiency of Fn1 flies on the salty substrate which have been previously (two days prior to placing the parents) inoculated with one of the yeast strains. We used five yeast strains extracted from the flies and two controls: bakers’ yeast *S*. *cerevisiae* and no yeast (sterile water); seven to eight vials were used for each variant. Virgin males and females from the line Fn1 were used as parents. Virgin flies were immobilized by carbon dioxide within eight hours after eclosion, separated by sex and transferred to test tubes with food N. They were used in tests 8–12 days after eclosion. Tests were performed at 23–24°C and natural lighting in standard glass vials with 66.7 mL of food per vial. To prepare the homogenate, 110 randomly selected flies from one of the four lines (Fn1, Fn2, Fs1, or Fs2), aged 0 to 5 days after eclosion, were immobilized by placing in a freezer at −20°C for 3 minutes. Thereafter, the flies were transferred to sterile test tubes with 1100 μl of sterile water (10 μl per fly, which results in approximately 1:10 dilution, because the average mass of an adult *D*. *melanogaster* is about 1 mg), mashed with a sterile glass rod and mixed for 15 minutes using a Heidolph Multi Reax vortex, at 1600 rpm. Subsequently, we took 100 μl of the homogenate for the analysis of yeast abundance and species composition (see below). The rest of the homogenate (1000 μl) was put onto the food substrate with a pipette with a sterile tip, 150 μl of the homogenate per vial, and evenly distributed with a small sterile glass spatula. Thus the amount of the homogenate applied to the food substrate in each vial corresponds to about 15 homogenized flies. Two days after the homogenate was applied to each vial, we placed 10 pairs of parents (10 females and 10 males) and removed them 7 days later; after that, we monitored the development of the offspring. ## Analysis of yeast composition in *D*. *melanogaster* homogenates We studied yeast population of *D*. *melanogaster* lines Fn1, Fn2, Fs1 and Fs2 using the inoculation method on a dense growth medium. We added 900 μl of sterile water to 100 μl of the homogenate (described above) and mixed it using the vortex for three minutes. As a result, we had a dilution of 1 : 100 (one homogenised fly, weighing about 1 mg, per 1 mL of water). Using a pipette with a sterile tip, we placed the aliquot of the suspension (50 μl) on the surface of the dense growth medium GPYA (20 g/L glucose, 10 g/L peptone, 5 g/L yeast extract, 200 g/L agar, and 1 g/L levomycetin to prevent bacterial growth), spread in a sterile Petri dish. Subsequently, we evenly distributed the aliquot on the surface of the medium using a Drigalski spatula. We inoculated the fly suspension from each line (Fn1, Fn2, Fs1, Fs2) in tenfold replication (10 Petri dishes per line). Inoculated dishes were incubated for five days at a room temperature (20–22° C). By the end of this period, all yeast colonies were divided into types by morphology and counted. For each sample, we obtained the overall number of yeasts in colony-forming units (CFU) per fly. We extracted two to three strains of each colonial morphotype to obtain pure cultures, then grouped all the cultures by cultural and micromorphology traits. The final identification of the groups was based on analysis of the nucleotide sequences of the ITS1-5.8S-ITS2 region and D1/D2 of rDNA domain 26S (LSU) using the method described previously. Amplification of rDNA regions was performed using ITS1f (`5’-CTTGGTCATTTAGAGGAAGTA`) and NL4 (`5’-GGTCCGTGTTTCAAGACGG`) primers. The same primers were used for sequencing. We used an Applied Biosystems 3130xl Genetic Analyzer sequence by Syntol Scientific Production company (Moscow, Russia). Data analysis was performed using the NCBI (ncbi.nlm.nih.gov) or CBS database ([www.cbs.knaw.nl](http://www.cbs.knaw.nl/)). The obtained sequences were deposited in GenBank database (MK332454-MK332474). ## Use of yeast pure cultures for inoculation of the high salt medium We prepared suspensions of the five most abundant yeast strains (extracted from *D*. *melanogaster* lines Fn1 and Fs1) by taking 7 mm³ pure culture biomass and 1 mL sterile water and mixing them for three minutes in the Multi Reax vortex at 1980 rpm. The suspensions were spread on the surface of the high salt medium, 150 μl per vial, with a pipette with sterile tips and small Drigalski spatulas, under sterile conditions. The vials were then incubated for two days at a room temperature (20–22° C). Subsequently, we placed 10 pairs of parents (10 females and 10 males) into each vial, removed them 7 days later, and monitored the development of the offspring. # Results ## 1. Reproduction efficiency of *D*. *melanogaster* on the food substrates N and S The aim of the first experiment was to estimate the fitness of the four *D*. *melanogaster* lines (two salt-tolerant and two control ones) on the substrates N and S. We had performed the same experiment previously, in September 2015, 11 months after the start of the evolutionary experiment. We had previously found that the flies from the lines Fs1 and Fs2 produced more progeny on both foods (N and S), compared to lines Fn1 and Fn2. Re-testing was performed in August 2017, 34 months after the start of the evolutionary experiment. The results are shown in, raw data are available in. The figure shows that flies from the salt-tolerant lines (Fs1, Fs2) produced more offspring on salty food than flies from the control lines (Fn1, Fn2), which agrees with the earlier results. On standard food N the results are less conclusive: Fn2 produced significantly less offspring than both Fs1 and Fs2, but the difference between Fn1 and the two salt-tolerant lines was insignificant. The null hypothesis that all four lines belong to the same distribution is rejected (*p* = 0.000076, Kruskal–Wallis test) for the food N (left graph). Paired comparisons revealed significant differences between the line Fn2 and the other lines (for pairs Fn2/Fn1, Fn2/Fs1, Fn2/Fs2 *p-*value is 0.00500, 0.00062 and 0.00005, respectively; Dunn’s test). Other differences are insignificant. For the salty food S (right graph), the null hypothesis that there are no differences between the lines is also rejected (p = 0.000862, Kruskal–Wallis test). All paired differences between control and salt-tolerant lines are statistically significant (for pairs Fn1/Fs1, Fn1/Fs2, Fn2/Fs1, Fn2/Fs2 *p-*value is 0.01281, 0.01909, 0.00091, 0.00151, respectively; Dunn’s test). Thus, flies that were reared on salty food demonstrate a significantly higher reproductive success on the salty substrate compared to the control flies. The results are consistent with our previous conclusions that flies from the lines Fs1 and Fs2 have a higher fitness when exposed to both types of food; thus, their acquired tolerance to salt probably resulted in trophic niche expansion rather than specialization, although further experiments are required to test this possibility. ## 2. Influence of fly homogenates on the reproductive efficiency of *D*. *melanogaster* on salty substrate The aim of the second experiment was to evaluate the influence of the microbiome of different fly lines on the reproductive efficiency of flies on salty food. We compared the influence of pre-inoculation of food S with homogenized flies from the lines Fn1, Fn2, Fs1, Fs2 on the reproduction efficiency of the naïve (not salt-tolerant) flies from the control line Fn1. The experiment was performed in September–October 2017. The results are shown in, raw data are available in. The figure shows that flies produced more offspring in the vials where the homogenate of the salt-tolerant flies (Fs1, Fs2) was spread over the surface of the salty substrate, compared to the vials inoculated by the homogenized control flies (Fn1, Fn2). This agrees with the results obtained previously in the four other *D*. *melanogaster* lines. The Kruskal–Wallis test rejects the null hypothesis that all four lines belong to the same distribution (*p* = 0.01979). The paired comparisons showed significant differences between Fn1 homogenate and both “salt-tolerant” homogenates (Fs1, Fs2); *p*-value is 0.00748 and 0.01024, respectively (Dunn’s test). In the rest of the cases, the differences are not significant (for pairs Fn2/Fs1 and Fn2/Fs2 *p*-value is 0.10854 и 0.13419, respectively), which may be partly due to small sample size (only five vials per test). Thus we can conclude that the results previously obtained in four other *D*. *melanogaster* lines kept under different conditions (in population cages) have been successfully reproduced. We confirmed that the homogenate of salt-tolerant flies spread over the surface of the salty substrate improved the reproduction of *D*. *melanogaster* on such substrate compared to the homogenate of the control flies. These results, together with the ones obtained earlier, imply that some characteristics of the microbiome of salt-tolerant flies may contribute to the observed increase in salt tolerance. It is also possible that additional factors independent of the different microbiome between Fn and Fs homogenates could have affected fly performance on high salt food. Further experiments were performed to test the hypothesis that the increase in salt tolerance observed in Fs flies was at least partially due to changes in yeast microbiome, and to reveal specific characteristics of the microbiome that may improve salt tolerance in *Drosophila* lines reared on salty food. ## 3. Composition of the yeast microbiome We inoculated the *Drosophila* homogenates from each of the four lines in ten Petri dishes to assess quantitative and qualitative composition of the yeast microbiome. We used the same four homogenates as in the experiment described in the previous section. Five yeast species were detected in the homogenates: 1. *Pichia occidentalis* (in all four lines), 2. *Zygosaccharomyces bailii* (in Fn1 and Fn2), 3. *Starmerella bacillaris* (syn. *Candida zemplinina*; in Fs1 and Fs2), 4. *Candida californica* (Fs1), 5. *Pichia membranifaciens* (abundant in Fs2, a few cells in Fn1). The yeast microbiome composition is shown in, raw data are available in. As depicted by the figure, the four fly lines differ greatly in the total abundance of yeast: from 22,3 CFU per fly in line Fn2 to 15393 CFU per fly in line Fs2. Altogether, the homogenates of the salt-tolerant flies (Fs1, Fs2) contain more yeasts than the homogenates of the control flies (Fn1, Fn2). The species composition of yeasts also differ between the salt-tolerant and control flies. The most prominent differences concern *S*. *bacillaris* and *Z*. *bailii*. The former species is abundant in both salt-tolerant lines and absent in the control lines; the latter species, by contrast, is abundant in both control lines and absent in the salt-tolerant lines. ## 4. Influence of different yeast strains on reproductive efficiency of *D*. *melanogaster* on salty food The aim of the next experiment was to estimate the influence of the yeast strains present in the fly homogenates on the reproductive efficiency of the control (Fn1) flies on salty food. We chose three yeast strains extracted from the “salt-tolerant” Fs1 homogenate (*C*. *californica*, *S*. *bacillaris*, *P*. *occidentalis*) and two strains from the homogenate of the control line Fn1 (*P*. *occidentalis*, *Z*. *bailii*). Suspension of bakers’ yeast *Saccharomyces cerevisiae* and sterile water were used as controls. Thus, there were seven experimental conditions overall. The experiment was performed in November–December 2017. The results are shown in, raw data are available in. The Kruskal–Wallis test allows us to reject the null hypothesis that all seven samples belong to the same distribution (*p* = 0.00005); *p*-values in the paired comparisons (Dunn’s test) are shown in. The results shown in and agree with the assumption that yeast strains from the salt-tolerant Fs1 flies improve reproduction of salt-naïve flies on salty food compared to controls (bakers’ yeast and no yeast). At the same time, the positive effect of yeast strains from the control Fn1 flies is not significant. The strongest positive influence is shown by *S*. *bacillaris*, although differences between this strain and other yeast strains from the salt-tolerant flies are not statistically significant. # Discussion ## 1. Acquired tolerance of *D*. *melanogaster* to high salt food medium The food substrate with high (2% and more) NaCl concentration is an unfavourable (stressful) medium for *D*. *melanogaster*, because it induces high larval mortality and suppresses larval development. Typically, naïve adult *D*. *melanogaster* can endure up to 4% NaCl in the food medium. Above this concentration, flies continue to eat the food, but their lifespan shortens and fecundity decreases dramatically. Nevertheless, laboratory populations are able to adapt to NaCl concentrations up to 6–7% or even 7–8% over several dozen generations if the salt concentration increased gradually. Although there is some data on the mechanisms of short-term salt stress response in *Drosophila*, the nature of the flies’ long-term acquired tolerance to salt is not completely understood. The *D*. *melanogaster* lines Fs1 and Fs2, described in this study, have been reared on food with 4% NaCl since October 2014. Their reproductive efficiency on salty and normal food was tested for the first time 11 months after the start of the evolutionary experiment, in September 2015. We found that they produced more offspring on both types of food than the control flies reared on food N. In the current study, we replicated this result three years after the start of the evolutionary experiment. It is difficult to compare directly the results of the first and second tests because the former was performed in test tubes with 10 mL of food and two pairs of parents, while the latter was performed in glass vials with 66.7 mL of food and ten pairs of parents. In any case, we have confirmed that the flies from lines Fs1 and Fs2 are more tolerant to high salt diet than the flies from lines Fn1 and Fn2 three years after the start of the evolutionary experiment, which makes it reasonable to search for the mechanisms of this adaptation. ## 2. Microbiota contributes to the observed increase in salt tolerance We estimated the reproductive efficiency of the flies on salty food pre- inoculated with the homogenized salt-tolerant and control flies to assess the possible contribution of the microbiome to the adaptation of *D*. *melanogaster* to salty food. We have shown previously that pre-inoculation of the salty substrate with salt-tolerant homogenized flies reliably improved the fitness of *D*. *melanogaster* on food S. This result was obtained from the four *D*. *melanogaster* lines reared under different conditions (in population cages, with overlapping generations). This apparently indicates that some quantitative or qualitative characteristics of the microbiome of the salt-tolerant flies, along with other features of the salt-tolerant fly homogenate, improve the reproductive efficiency of the flies on salty food. One of the goals of the current study was to verify the reproducibility of this result in the other four lines maintained under different conditions (in vials, with 20 randomly chosen flies transferred into a new vial every 14 days). As shown in, in the current study the previous result was replicated in general. Although not all the paired comparisons between the control and salt-tolerant lines showed statistically significant levels of difference, the directions of differences in all cases agreed with the expected ones. Therefore, the results are consistent with our hypothesis that microbiome contributes to the adaptation of *Drosophila* to salt. This makes it reasonable to search for specific characteristics of microbiome responsible for this contribution. We acknowledge that additional factors independent of the different microbiome between Fn and Fs homogenate could have affected fly performance on high salt food pre-inoculated with different homogenates. Offspring number notably increased when the flies were given the Fn fly homogenate as compared to no homogenate (compare). This implies that flies likely fed on the homogenate for a while before fully relied on the fly food, and the nutritional content probably differed between the Fs and Fn homogenate. Moreover, the flies themselves could be producing factors that alleviate salt stress independently of their microbiome. This can confound the interpretation of the results. In order to reduce this uncertainty, several experimental approaches can be used. One possible approach is to compare the effects of Fn and Fs homogenates to those of axenic flies homogenates, or to add antibiotics and fungicides to Fn and Fs homogenates in order to remove the microbiome. However, such procedures would inevitably lead to new complications, e.g., rearing flies in axenic conditions would certainly change their physiology and nutritional value, while adding fungicides to homogenates would disrupt the normal growth of yeasts on the surface of fly food after inoculation. To this end, we have used two other approaches. First, we compared yeast microbiome composition in fly homogenates and found substantial differences between Fn and Fs homogenates. Second, we evaluated the effects of pure cultures of different yeast strains isolated from Fn and Fs flies on the reproduction efficiency of naïve flies on high salt food, and found that strains isolated from Fs flies tend to have the strongest positive effect (although their advantage over the strains from Fn flies is not statistically significant). The results are compatible with the hypothesis that yeasts do contribute to the observed increase in salt tolerance in Fs flies, as discussed in the next two sections. Importantly, these results do not provide a basis for a claim that microbiome is the only source of the increased salt tolerance in the Fs flies. Other factors, including genetic and epigenetic changes in the flies themselves, may also contribute to the observed adaptation; further research is needed to elucidate the relative importance of different factors. ## 3. Yeast microbiome composition differs between the salt-tolerant and control *Drosophila* lines The research of the host-microbe interactions on the *D*. *melanogaster* model is currently focused mostly on the gut bacteria. However, yeasts apparently form a very important part of the microbiome of *Drosophila*. Yeasts affect different aspects of physiology, immune response, and behavior of *Drosophila*. For instance, some yeast species influence the larval survival and growth rate, along with the adult body mass; in addition, larvae are selective to yeasts and prefer the species that enhance larval growth. *Drosophila* larvae and adults, on the other hand, can influence the species richness of yeast communities that develop on various food substrates. At least some yeast species survive the passage through larval and adult *D*. *melanogaster* guts, making it possible for the flies to be the effective vectors of yeasts in the wild. In the current study, we focused on the possible impact of the yeast component of the microbiome of *D*. *melanogaster* on the flies' adaptation to salty food. We compared the yeast composition in four *Drosophila* lines and found that the control and salt-tolerant lines differ in the total amount of yeasts (salt- tolerant flies carry much more yeast cells) and species composition. Yeast *S*. *bacillaris* is present in the homogenates of the Fs1 and Fs2 flies, but absent in the homogenates of the Fn1 and Fn2 flies; *Z*. *bailii*, by contrast, is abundant in both control lines, but absent in the salt-tolerant lines. It is interesting to compare these results with the ones from our previous study, where we performed a similar analysis in four other fly lines (two salt- adapted and two control ones) maintained under different conditions (in population cages). These lines have lower species diversity of yeasts than the ones kept in vials. We found only two abundant yeast species, *P*. *membranifaciens* and *S*. *bacillaris*, and only the distribution of the latter species was correlated with salt tolerance. *S*. *bacillaris* was abundant in both salt-tolerant lines and absent or scarce in both control lines. Thus, in all eight experimental *D*. *melanogaster* lines tested to date (four analyzed in this study and four analyzed earlier) we see a common pattern: *S*. *bacillaris* is always abundant in the homogenates of salt-tolerant flies but absent or scarce in the homogenates of the control flies. At the same time, other parameters of the yeast microbiome (the total amount of yeast cells and relative abundance of species other than *S*. *bacillaris*) vary a lot and do not show apparent relationship with either salt-tolerant or control fly lines. It makes *S*. *bacillaris* a likely candidate for the role of the microbiome component that positively contributes to the *D*. *melanogaster* adaptation to salty food. *S*. *bacillaris* is a genetically heterogeneous, psychrotolerant, osmotolerant, asporogenic species often found on grapes, in grape must and wine. It is sometimes recorded in soil, rotting watermelons and bananas; it was also found on *Drosophila* in the USA and Hungary, indicating that *Drosophila* can participate in the distribution of this yeast species. It has been shown that *S*. *bacillaris* yeast is edible for *D*. *melanogaster* larvae, and that larvae fed on bananas encourage the consistent development of this yeast species along with two others (*Candida californica* and *Pichia kluvyeri*), while simultaneously discouraging the growth of filamentous fungi. Moreover, some *S*. *bacillaris* cells (as well as some cells of *C*. *californica* and *P*. *kluvyeri*) survive passing through the larval gut. As far as we know, there are no data on specific relationship of *S*. *bacillaris* with salty substrates, or with the development of *Drosophila* on these substrates. It should be noted that the cells of *S*. *bacillaris* are generally smaller than those of the other yeast species found in our fly lines. This fact may be related to the positive effect of *S*. *bacillaris* on fly fecundity on high salt food (see next section); further research is needed to clarify this point. ## 4. Yeast strains from salt-tolerant flies enhance the reproduction of *D*. *melanogaster* on salty food depicts the results of the experiment in which we assessed the influence of five yeast strains (three strains extracted from the salt-tolerant Fs1 line and two strains from the control Fn1 line) on the reproductive efficiency of *D*. *melanogaster* on salty food. As controls, we used the baker’s yeast *S*. *cerevisiae* and sterile water. We found that yeast strains extracted from salt-tolerant flies significantly improved the reproductive efficiency of salt-naïve *D*. *melanogaster* on salty food compared to both controls. Yeast strains extracted from control (naïve) flies may also have some positive effect, but in our experiment it was not statistically significant. As we expected, the strongest positive effect was from the yeast S. *bacillaris*, although its difference from two other strains extracted from the salt-tolerant flies did not reach the level of statistical significance. *Drosophila* larvae can selectively improve the reproduction of some yeast species, including *S*. *bacillaris*, on their food substrate, at the same time preventing the reproduction of other fungi, which results in higher yeast community similarity in the presence of the larvae. Further research is needed to find out if the interactions between *D*. *melanogaster* and particular yeast species are robust and specific enough to consider them as an example of agricultural symbiosis, such as the symbioses described in ants, termites, and some other animals. Remarkably, the baker’s yeast *S*. *cerevisiae*, which we used as one of the controls, did not have any noticeable effect on the reproduction of *D*. *melanogaster* on salty food, unlike yeasts extracted from salt-tolerant flies. This seems to agree with the conclusion from that baker’s yeast, which is very rarely associated with *Drosophila* in the wild, is not the best model for studying the *Drosophila*-yeast interactions, despite the fact that this species is often used in such studies. It should be noted, however, that salt stress is not a natural condition for *D*. *melanogaster* (although it is for some other dipterans), and thus the failure of *S*. *cerevisiae* to alleviate this stress does not, in itself, provide enough evidence to consider *D*. *melanogaster*–*S*. *cerevisiae* interaction a poor experimental model. In our experiment, we did not measure the number of yeast cells used to inoculate food vials, but rather used the same volume of biomass for each yeast strain. As cell size of different yeast species vary substantially (e.g., *S*. *bacillaris* cells are generally smaller than those of the other species tested), we acknowledge that the different effects of yeast strains on fly reproduction may be partially due to different cell numbers in yeast inoculates. However, it is not evident that controling for cell number would be more appropriate than controlling for biomass. Whatever benefits the flies receive from the yeasts (e.g., nutrition, metabolites, or substrate transformation), these benefits may be more related to the biomass of the yeasts than to their cell numbers. For instance, yeasts are known to be a major food source for *Drosophila* in both adult and larval stages, yeast species with different cell sizes are palatable to *Drosophila*, and the nutritional value of ingested yeasts apparently depends more on biomass than on cell count. From the other hand, it is quite probable that the benefit received by the flies from *S*. *bacillaris* on salty food is related to the small size of the cells of this species. Further research is needed to clarify this issue. In general, our results agree with the hypothesis that some components of yeast microbiome contribute, along with other possible factors, to the adaptation of *D*. *melanogaster* to salty food observed in experimental evolution studies. It means that the possible role of microbiome should be taken into account when interpreting the results of such studies. The generally accepted (although usually implicit) presumption that the observed adaptation is exclusively due to genetic or epigenetic changes of the experimental population of the studied macroorganisms can be misleading. It is probably more reasonable to study the *Drosophila* adaptation to adverse conditions at the holobiont level, as proposed by the proponents of the “hologenome theory of evolution.” # Supporting information We are grateful to K.S. Perfilieva and M.B. Kornilova for their help in maintaining the *Drosophila* lines. We are also grateful to the anonymous reviewers for their insightful comments. [^1]: The authors have declared that no competing interests exist.

# Introduction Understanding the influence of a CEO’s cultural background on overinvestment decisions is essential, especially in the global economy’s diverse cultural contexts. China provides a particularly interesting case due to its cultural heterogeneity and the significant north-south cultural divide. In listed companies, the separation of ownership and management is a key feature. CEOs, appointed by shareholders, wield significant decision-making power, influencing the company’s operations and investment performance. Although previous research has examined the impact of CEOs’ personal traits on a company’s investment performance, there has been less focus on the influence of a CEO’s cultural background on overinvestment decisions. CEO characteristics, such as cultural background, have been found to systematically vary investment behaviors. Moreover, research has indicated that executives’ cultural backgrounds can influence company investment decisions. Despite these findings, the relationship between informal institutions, like a CEO’s cultural background, and investment decisions has been less explored. While existing literature provides a comprehensive understanding of the dominance of state-owned versus private enterprises in the capital market, the impact of a CEO’s cultural background on overinvestment decisions remains an under-explored territory. This study aims to address this research gap, providing an empirical investigation into the influence of a CEO’s cultural background on overinvestment, particularly within the unique context of Chinese businesses. Our in-depth examination of the regional cultural differences within China, specifically between the North and South, lends a distinct characteristic to this study. In this study, we aim to address this research gap by investigating the impact of a CEO’s cultural background on overinvestment in Chinese firms. Using data from listed firms between 2000 and 2018, we examine the influence of cultural background on corporate decision-making and its implications for firm value and investor wealth. In addition to cultural background, we also consider the role of the CEO’s age on overinvestment decisions, based on evidence that younger CEOs tend to take more risks than older ones. We also account for the agency problem and governance issues in state-owned enterprises, which can contribute to overinvestment. This study offers a unique contribution to existing literature by providing an in-depth analysis of the impact of a CEO’s cultural background on overinvestment decisions, a topic not sufficiently explored in previous studies. We focus on the regional cultural differences within China and their influence on overinvestment, adding a unique element to our study. By examining the complex interplay between CEO age, company ownership, and overinvestment behaviors, we hope to provide valuable insights that extend beyond China’s geographical context, enhancing the understanding of corporate overinvestment and its determinants. The practical implications of these insights are significant for companies, particularly those seeking to refine their investment strategies and governance mechanisms. The remainder of this paper is organized as follows. Section 2 discusses the hypothesis development, and Section 3 presents the data and the methodology used. Further, Section 4 discusses the empirical results of the findings. Finally, Section 5 concludes the study. # Literature review ## The correlation of CEO culture and firm investment The relationship between CEO culture and firm investment has been a topic of interest among scholars due to the potential impact of cultural background on investment decision-making behavior. Studies have shown that cultural preferences for risk can vary across different backgrounds, leading to differences in investment decisions. In recent years, evidence from Chinese sub- national institutional contingencies also showed how tournament incentives could spur CSR performance. Personal traits, such as values and risk tendencies, can also impact decision-making. The upper echelon theory suggests that top management teams’ characteristics can influence a company’s strategy and performance, with the CEO being a crucial figure in determining the company’s risk decisions. Further studies on the effect of CEO characteristics on firm performance across different cultural contexts, such as the study of Saudi Arabia Listed Firms, supplement this perspective. Understanding the relationship between CEO culture and firm investment can provide valuable insights into the role of culture in investment decision-making and potentially aid companies in improving their investment strategies. The influence of financial decisions, like capital structure, on corporate performance also plays a vital role. Traditional culture significantly impacts on an individual’s background, encompassing elements such as diet, religion, history, language, and personal habits. In China, a vast land with numerous ethnic groups, the main geographic boundary is the north-south division. The formation of different cultures in southern and northern China is influenced by factors such as climate, geography, and history, with geography being the most stable and influential factor in cultural formation. The north and south are divided geographically and culturally by the Qinling Mountains and the Huai River line, which corresponds with the zero-degree average temperature line in January, a boundary for the survival of winter crops. In terms of agriculture, the south mainly cultivates rice, while the north predominantly cultivates wheat. In comparison to Western countries with well-developed economic systems, traditional culture in China has a more profound impact on the economy than formal institutions. The difference between southern and northern cultures is shaped by the natural environment and production methods, with a character that is more adventurous, independent, and brave in the north and more commercially savvy, meticulous, and risk-averse in the south. Drawing on the cultural-attitude-behavior hypothesis and the upper echelon theory, this study proposes that there is a correlation between CEO cultural background and corporate investment decisions. Specifically, we hypothesize that firms led by CEOs with northern cultural backgrounds will exhibit a greater inclination towards overinvestment compared to those led by CEOs with southern cultural backgrounds. This is due to the fact that northern culture tends to exhibit less uncertainty avoidance and greater risk preference, leading to a higher propensity for overinvestment. Therefore, the following hypothesis is proposed: 1. Hypothesis 1: Firms led by CEOs with northern cultural backgrounds are more likely to engage in overinvestment than those led by CEOs with southern cultural backgrounds. ## The relationship of CEO age and firms investment The upper echelons theory proposes that top management teams’ characteristics significantly influence a company’s strategy and performance. Previous study, has found that executive traits are crucial factors affecting corporate investment decisions. Moreover, there has been growing interest in exploring the effect of CEO personal characteristics, including age, on company decision- making. Stduy has shown that CEO age can influence investment decisions and attitudes towards risk, as well as other company policies. Age is easily observable and has been shown to have different outcomes on corporate decisions, with studies suggesting that the average age of the top management team can affect organizational growth. Previous research has demonstrated that CEO age can significantly affect corporate investment decisions, with younger CEOs tending to take greater risks and be more aggressive in their investment decisions compared to older CEOs. In contrast, older CEOs tend to adopt lower-risk investment strategies and exhibit a more conservative ddecision-making style, reducing company risk through policies such as lower financial leverage, more diversified investments, and lower investment expenditures. The managerial signaling model developed predicts that younger CEOs will be more aggressive in investing. However, from the perspective of managerial signaling, there is a negative correlation between CEO age and risk-taking behavior, indicating that younger CEOs take on more risk compared to older managers. Prior studies found that older CEOs tend to avoid investing in high-risk projects and make more conservative decisions compared to younger CEOs. Older CEOs tend to seek more information, conduct more in-depth evaluations, spend more time making decisions, demonstrate greater commitment to the company, and put in more effort to achieve the company’s goals. Three reasons why younger CEOs are more aggressive in investing: First, older CEOs may be in a period of life where financial security and job security are more important. Second, older CEOs may have a greater commitment to the company’s current state. Finally, the mental and physical stamina of older CEOs may be weaker, or they may be unable to grasp new ideas and learn new behaviors. Therefore, this study, in line with the managerial signaling model and the reviewed literature, hypothesizes a negative correlation between CEO age and risk-taking behavior. It predicts that older CEOs tend to adopt lower-risk investment policies to reduce the firm’s risk exposure. Accordingly, it posits that as CEO age increases, risk avoidance and conservative decision-making tendencies increase. Furthermore, it anticipates that older CEOs can better resist the influence of a Northern cultural background on excessive investment behavior. Thus, the following hypothesis is proposed: 1. Hypothesis 2: There is a negative correlation between CEO age and risk- taking behavior. The older the CEO, the more likely they are to adopt a conservative investment policy and resist the influence of a Northern cultural background on investment behavior. ## The correlation of firm’s investment and ownership structure in China The ownership structure of Chinese state-owned enterprises (SOEs) has significantly influenced their investment decisions. Due to the lack of proper supervision and corporate governance in SOEs, the ownership structure is often overlooked, and executives prioritize political objectives over market-oriented ones, leading to severe agency problems. This issue is compounded by the lack of protection for minority shareholders in state-owned listed companies. On the other hand, privately-owned or township enterprises that have gone public and become listed companies have relatively market-oriented ownership structures, resulting in stricter management supervision and fewer agency problems. In Chinese state-owned enterprises, the ownership structure is highly interconnected, which leads to improper supervision and neglect of ownership. Additionally, state-owned enterprises suffer from imperfect corporate governance, with executive personnel often selected by their respective local governments, and their career transitions and promotions mainly controlled by their parent company or the state. As a result, executives of listed companies often prioritize political motivations over market-oriented ones, leading to serious agency problems. In China, the protection of minority shareholders in state-owned listed companies has become a serious issue, as they face significant hurdles in obtaining fair treatment and obtaining appropriate compensation in the event of mergers and acquisitions. Ownership structure plays a crucial role in the investment decisions of Chinese state-owned enterprises (SOEs). Due to inadequate supervision and corporate governance in SOEs, ownership is often overlooked, and executives prioritize political goals over market-oriented ones, leading to severe agency problems. Minority shareholder protection is also a significant issue in state-owned listed companies. Conversely, privately-owned or township enterprises that have gone public and become listed companies usually have market-oriented ownership structures, resulting in stricter management supervision and fewer agency problems. In Chinese state-owned enterprises, the ownership structure is highly interconnected, and executives pursue political motivations, leading to serious agency problems. Moreover, imperfect corporate governance in state-owned enterprises results in executive personnel selected by their respective local governments, with their career transitions and promotions mainly determined by their parent company or the state. Research indicates that state-owned holding listed companies have significantly higher agency costs than non-state-owned holding companies, and state-owned enterprises are more likely to engage in overinvestment due to overconfidence among management. Political relationships negatively affect investment efficiency in state-owned enterprises, with investment expenditures being less sensitive to growth opportunities than those of non-state-owned enterprises. State-owned enterprises also have higher corporate investment due to government intervention and lower financing constraints, increasing the likelihood of overinvestment. Non-state-owned enterprises have been found to have higher investment efficiency than state-owned enterprises. Therefore, based on past literature, this study posits that the ownership structure of Chinese SOEs has a significant impact on their investment decisions, with agency problems being more severe due to the lack of proper supervision and corporate governance. Compared to privately-owned or township enterprises, SOEs often prioritize political objectives, leading to overinvestment and lower investment efficiency. Hence, we hypothesize: 1. Hypothesis 3: State-owned enterprises are more likely to engage in overinvestment compared to privately-owned or township enterprises due to agency problems resulting from political objectives. # Methodology Our research methodology involves an empirical investigation using data from Chinese firms, allowing us to study the real-world impact of these factors. We employ regression analysis to understand the relationships and interactions among these variables. The findings from this study contribute to both theoretical understandings and practical applications of corporate investment decision-making, with implications for companies in China and beyond. ## Data This study leverages data from the CSMAR Guotai’an database, encompassing annual data from 2000 to 2018 for Chinese A-share listed companies, including CEO background, control variables, and data necessary to quantify overinvestment. The period from 2000 to 2018 is selected because it marks the maturation phase of China’s modern corporate governance structures and market reforms. This period also allows sufficient temporal breadth to observe long-term patterns and trends in CEO behavior and overinvestment. Financial institutions, given their unique investment and financing behaviors and their susceptibility to stringent government and regulatory supervision, are excluded from the study to eliminate any potential distortions or sector-specific anomalies in the analysis. Furthermore, this study does not consider B shares and H shares due to their different investor base and regulation, focusing solely on A-share listed companies to ensure data consistency and reliability. ## Variables definition ### Overinvesment This study uses the research framework proposed to calculate overinvestment. First, the total investment is calculated as the sum of the investment required to maintain current assets and the investment needed for new assets. The formula is expressed as $$\text{I}_{\text{TOTAL},\text{t}} = \text{CAPEX}_{\text{t}} + \text{Acquisitions}_{\text{t}} - \text{SalePPE}_{\text{t}}$$ where, total investment(I_total)is defined as the sum of capital expenditure(CAPEX)and acquisition expenditure(Acquisitions), minus the proceeds from the sale of property, plant, and equipment(SalePPE) Additionally, total investment(I_TOTAL) can be divided into two parts: investment for maintaining assets (I_MAINTENANCE) and investment for new projects (I_NEW). The equations are as follows: $$\text{I}_{\text{TOTAL},\text{t}} = \text{I}_{\text{NEW},\text{t}} + \text{I}_{\text{MAINTENANCE},\text{t}}$$ Thus, new project investment can be divided into two parts: expected investment with a positive net present value ($I_{NEW}^{*}$) and abnormal or unexpected investment ($I_{NEW}^{\varepsilon}$). The value of abnormal investment can be positive or negative, with the positive portion considered as overinvestment and the negative portion considered as underinvestment. Since this study focuses on overinvestment, we chose the positive portion of abnormal investment as the research target. The relationship equation is as follows:: $$\text{I}_{\text{NEW},\text{t}} = \text{I}_{\text{NEW},\text{t}}^{*} + \text{I}_{\text{NEW},\text{t}}^{\varepsilon}$$ Next, this study will use the investment expenditure formula to calculate expected investment($I_{NEW}^{*}$) and abnormal investment($I_{NEW}^{\varepsilon}$). The fitted values from correspond to expected investment ($I_{NEW}^{*}$), while the residual values correspond to abnormal investment ($I_{NEW}^{\varepsilon}$)The equation is as follows: $$\begin{array}{ll} \text{I}_{\text{NEW},\text{t}} & {= \beta_{0} + \beta_{1}\frac{\text{V}}{\text{P}}_{\text{t} - 1} + \beta_{2}\text{Leverage}_{\text{t} - 1} + \beta_{3}\text{Cash}_{\text{t} - 1} + \beta_{4}\text{Age}_{\text{t} - 1} + \beta_{5}\text{Size}_{\text{t} - 1} +} \\ & {\beta_{6}\text{Stockreturns}_{\text{t} - 1} + \beta_{7}\text{I}_{\text{New},\text{t} - 1} + \sum\text{Yeardummy} + \sum\text{Industrydummy} + \text{I}_{\text{New},\text{t}}^{\varepsilon}} \\ \end{array}$$ To measure V/P company growth, the calculation method is to divide the value of operating assets (V_AIP) by the equity market value, and the calculation formula for operating asset value V_AIP = (1 − αr) BV + α(1 + r) X-αrd is based on the abnormal earnings persistence parameter estimated from the Ohlson (1995) research framework. The calculation method of α is, α = (ω/(1 + r − ω)) for operating asset value is based on the abnormal earnings persistence parameter estimated from the research framework; r is the discount rate used in "r" is estimated to be 0.05 according to Chen et al. (2016), and "ω" is estimated to be 0.62 according to Ohlson (1995), and ω is estimated to be 0.62. BV is the book value of common stock, X is earnings after deducting depreciation expense, and d is dividends. is total liabilities divided by total assets, is short-term investments divided by total assets, is the natural logarithm of the year in which the company went public, is the natural logarithm of total assets, is the difference in market value between this year and last year, is a dummy variable controlling for calendar year fixed effects, and is a dummy variable controlling for fixed effects across different industries. ## Independent variables ### CEO cultural background The CEO cultural background is a dummy variable. This study defines CEO cultural background based on the "New Geographical Literature" in 1908 that the Qinling Mountains-Huaihe River line serves as the geographical north-south boundary in China. If the birthplace of the CEO of company i in year t belongs to a northern province, Culture = 1. If it belongs to a southern province, Culture = 0. ### CEO age CEO age data obtained.from the CSMAR database, This study defines CEO age as the age of the CEO of the i-th company in the t-th year. ### State-owned Enterprises (SOE) The relevant information of listed companies will be obtained from the CSMAR database. If the company is a state-owned enterprise, then SOE = 1, and if it is a non-state-owned enterprise, then SOE = 0. ## Control variables Based on past literature, this study will control for certain company characteristics that may affect overinvestment, including free cash flow (FCF), leverage, company size, and firm age. This study defines company size as the natural logarithm of total assets, and leverage as total liabilities divided by total assets. Free cash flow is defined as operating cash flow minus expected new investments, and is calculated using Eqs, and as per Richard’s research framework. Operating cash flow is the cash flow generated by a company’s operations, and is calculated by subtracting cash flows related to investments in maintaining assets from cash flows generated by operations, and then adding research and development (R&D) expenses. Expected new investments are calculated using depreciation and amortization expenses obtained from financial reports. In addition, this study includes corporate governance variables as control variables, including the average shareholding percentage of the top three shareholders (Herfi3), whether the largest shareholder holds more than 25% of the shares (D1), CEO duality (Isduality), board size (BoardSize), supervisor size (SupervisorSize), executive size (ExecutiveSize), outside directors (OutsideDirectors), non-paid directors (NonpaidDirectors), and non-paid supervisors (NonpaidSupervisor). Herfi3 is defined as the average shareholding percentage of the top three shareholders, D1 is defined as 1 when the largest shareholder holds more than 25% of the shares, and 0 otherwise. Isduality is defined as 1 when the chairman and CEO are the same person, and 0 otherwise. BoardSize is defined as the total number of directors on the board, SupervisorSize is defined as the total number of directors on the supervisory board, ExecutiveSize is defined as the total number of executives on the board, OutsideDirectors is defined as the proportion of outside directors to the total number of directors on the board, NonpaidDirectors is defined as the proportion of directors who do not receive compensation from the company to the total number of directors on the board, and NonpaidSupervisor is defined as the proportion of supervisors who do not receive compensation from the company to the total number of supervisors in the company. provides detailed definitions of these variables. ## Model development ### The relationship between overinvestment and CEO cultural background To test hypothesis 1: CEO with a northern cultural background will increase overinvestment in the company since CEO cultural background and corporate overinvestment are not causally related, CEO cultural background will not change due to corporate overinvestment. Therefore, there is no endogeneity issue between CEO cultural background and corporate overinvestment in this study. It is hypothesized that the two variables have a positive correlation (\>0), and therefore, this study establishes the following regression model for estimation $$\text{OI}_{\text{i},\text{t}} = \alpha_{0} + \alpha_{1}\text{Culture}_{\text{i},\text{t}} + \theta_{1}\text{CV}_{\text{i},\text{t}} + \varepsilon_{\text{i},\text{t}}$$ In this study, the variable for overinvestment is represented by "I", and the variable for CEO’s cultural background is represented by "Culture". When the CEO’s birthplace is in the northern region, it is considered to have a northern cultural background, and Culture is assigned a value of 1. If the birthplace is in the southern region, Culture is assigned a value of 0. The study also includes 14 control variables: free cash flow, debt, company size, company age, average shareholding ratio of top three shareholders, maximum shareholder holding more than 25%, CEO duality, board size, supervisor size, executive size, external directors, non-remunerated directors, non-remunerated supervisors, and fixed effects for year and industry. ### CEO age moderate the effect of CEO cultural background on overinvestment To test hypothesis two: whether CEO age moderates the relationship between overinvestment and CEO cultural background, we hypothesize a negative correlation between the two variables, α₃ \< 0. Therefore, we establish the following regression model for estimation in this study: $$\text{OI}_{\text{i},\text{t}} = \alpha_{0} + \alpha_{1}\text{Culture}_{\text{i},\text{t}} + \alpha_{2}\text{CEO}\mspace{360mu}\text{Age}_{\text{i},\text{t}} + \alpha_{3}\text{Culture}_{\text{i},\text{t}} \times CEO\mspace{360mu} Age_{i,t} + \theta_{1}\text{CV}_{\text{i},\text{t}} + \varepsilon_{\text{i},\text{t}}$$ The variables in the model are as follows: Overinvestment (a binary variable indicating whether the firm is overinvesting), Culture (a binary variable indicating whether the CEO has a northern cultural background, with 1 for a northern birthplace and 0 for a southern birthplace), Age (the age of the CEO), and 14 control variables including Free Cash Flow, Debt, Company Size, Company Age, Average Shareholding Ratio of the Top Three Shareholders, Maximum Shareholding of a Single Shareholder (\>25%), CEO Duality, Board Size, Supervisor Size, Executive Size, External Directors, Unpaid Directors, Unpaid Supervisors, Year Fixed Effects, and Industry Fixed Effects. ### Overinvestment and the impact of CEO cultural background To test hypothesis three: Holding other conditions constant, compared to non- state-owned enterprises, the age of state-owned enterprise CEOs will more significantly inhibit the impact of CEO’s northern cultural background on excessive investment in enterprises. It is estimated that the two are negatively correlated, α₃ and β₃ \< 0. Therefore, this study establishes the following regression model: $$\begin{array}{ll} \text{OI}_{\text{i},\text{t}} & {= \left( {\alpha_{0} + \alpha_{1}\text{Culture}_{\text{i},\text{t}} + \alpha_{2}\text{Age}_{\text{i},\text{t}} + \alpha_{3}\text{Culture}_{\text{i},\text{t}} \times Age_{i,t}} \right) \times \text{NONSOE}_{i,t}} \\ & {+ \theta_{1}\text{CV}_{\text{i},\text{t}} \times \text{NONSOE}_{i,t} + \varepsilon_{\text{i},\text{t}}} \\ \end{array}$$ $$\begin{array}{ll} \text{OI}_{\text{i},\text{t}} & {= \left( {\beta_{0} + \beta_{1}\text{Culture}_{\text{i},\text{t}} + \beta_{2}\text{Age}_{\text{i},\text{t}} + \beta_{3}\text{Culture}_{\text{i},\text{t}} \times Age_{i,t}} \right) \times \text{SOE}_{i,t}} \\ & {+ \theta_{2}\text{CV}_{\text{i},\text{t}} \times \text{SOE}_{i,t} + \varepsilon_{\text{i},\text{t}}} \\ \end{array}$$ In this study, overinvestment serves as the dependent variable, while the CEO’s cultural background is the primary independent variable of interest. CEOs born in the North are indicated by a value of 1, while those born in the South are assigned a value of 0. The control variables include non-state-owned and state- owned enterprises, as well as 14 additional variables: free cash flow, debt, company size, company age, average shareholding ratio of the top three shareholders, maximum shareholder holding \> 25%, CEO duality, board size, supervisor size, executive size, external directors, unpaid directors, unpaid supervisors, and fixed effects for year and industry. To test the hypothesis, this study employs Chow’s test, which assesses whether the linear regression coefficients of two distinct datasets are equal. Widely utilized in time series analysis, Chow’s test detects the presence of structural changes and was developed by economist Gregory Chow in 1960. The method involves estimating the same variable across two different periods or samples to identify any disparities. In this context, the samples are divided into state-owned enterprises and non- state-owned enterprises. Chow’s test examines whether a significant difference exists in the coefficients between Eqs and, assuming constant residuals. The null hypothesis is H0: β1 = β2 = 0. If a difference is detected, the null hypothesis is rejected, signifying a divergence between the two groups. Consequently, the study separates the sample into state-owned and non-state- owned enterprises, applying Chow’s test to evaluate the hypothesis. ## 1. Empirical result ### Descriptive statistic In, during the sample period, the average level of overinvestment in non- financial industries in China was 0.049, with a standard deviation of 0.066. The difference between the maximum value (3.610) and the minimum value (0.000) is 3.610, indicating that there is a large fluctuation in overinvestment by Chinese firms, which can lead to overinvestment or underinvestment under different corporate governance and business conditions. The average CEO cultural background is 0.102 with a standard deviation of 0.302. It can be observed that a larger number of CEOs are from southern China ( Distribution chart of CEO cultural backgrounds). The average CEO age is 48.015 years, with a difference of 56 years between the maximum value (80.00) and the minimum value (24.00). The average free cash flow is 0.009, indicating that Chinese companies generally have inadequate control over cash flow, resulting in insufficient cash flow. The difference between the maximum value (1.127) and the minimum value (-0.236) is 1.363, indicating that there is a significant difference in the free cash flow held by companies with different operating conditions. Other variables include the average and standard deviation of total liabilities, which are 0.589 and 6.324, respectively. The average company size is 21.954, with a median of 21.827. The average company age is 2.104. Corporate governance variables include the average and standard deviation of the top three shareholders’ ownership ratio, which are 0.171 and 0.126, respectively. The average ownership percentage of the largest shareholder is 0.279. The average CEO duality is 0.170. The average size and standard deviation of the board of directors are 9.109 and 1.906, respectively, while the average size and standard deviation of the board of supervisors are 3.917 and 1.296, respectively. The average size and standard deviation of the senior management team are 6.396 and 2.497, respectively. The average and standard deviation of external directors are 3.273 and 0.695, respectively, while the average and standard deviation of non-compensated directors are 2.366 and 2.002, respectively, and the average and standard deviation of non-compensated supervisors are 1.418 and 1.295, respectively. From a financial perspective, the article discusses the phenomenon of overinvestment among non-financial firms in China. It notes that the average level of overinvestment among these firms during the sample period was 0.049, with a standard deviation of 0.066. This indicates that there is a high degree of volatility in overinvestment levels among Chinese firms, which can lead to both overinvestment and underinvestment under different corporate governance and operational conditions. ## Correlation of cultural background and firms’ overinvesmt To investigate the relationship between CEO cultural background and overinvestment, this study used a least squares regression model for empirical analysis, and the results are shown in. Column 1 of shows that at the 1% significance level, the coefficient of CEOs with a northern cultural background is 0.00352, indicating a significant positive correlation between overinvestment and CEOs with a northern cultural background. This is consistent with the expected hypothesis, indicating that when the CEO has a northern cultural background, they may prefer risk, which leads to an increase in overinvestment. Column 3 of adds control variables and fixed effects. At the 1% significance level, the coefficient of CEOs with a northern cultural background is 0.00337, which is consistent with the expected hypothesis, indicating a significant positive correlation between overinvestment and CEOs with a northern cultural background. This explains "culture-attitude-behavior" hypothesis. Compared with southern Chinese culture, northern Chinese culture has a greater risk preference. Therefore, this study believes that when the CEO has a northern cultural background, overinvestment is more likely to occur. This study finds that over-investing is more sensitive to current free cash flow and consistent with the explanation of agency costs, as there is a positive correlation between over-investing and higher levels of free cash flow. The results indicate that companies with positive free cash flow are more likely to over-invest, which is consistent with prior literature. Among other corporate governance variables, this study finds that larger supervisor size can better suppress over-investing, which is consistent with. At the 1% significance level, the coefficient is -0.00320, indicating a negative correlation. It is found that both the average shareholding of the top three shareholders (Herfi3) and the largest shareholder holding \>25% (D1) are negatively related to over-investing, with coefficients of -0.0513 and -0.0117, respectively, indicating that more concentrated ownership can better suppress over-investing. The occurrence of over-investing significantly decreases as company age increases, and both are negatively correlated. The opportunity for over-investing increases with company size, and both are positively correlated. Flow, which is consistent with the explanation of agency costs. When free cash flow is high, overinvestment by firms tends to be positively correlated with it. This indicates that companies with positive free cash flow are more likely to overinvest. Among other corporate governance variables, this study found that larger board size can better restrain firms from overinvestment, which is consistent with. At the 1% significant level, the two variables show a negative correlation with a coefficient of -0.00320. It can also be found that the average shareholding of the top three shareholders (Herfi3) and the largest shareholder’s shareholding \>25% (D1) are both negatively correlated at the 1% significant level, with coefficients of -0.0513 and -0.0117, respectively. This indicates that the more concentrated ownership is, the better it can restrain firms from overinvesting. When a firm’s age increases, the likelihood of overinvestment decreases, and both variables show a negative correlation. When a firm’s size increases, the likelihood of overinvestment also increases, and both variables show a positive correlation. ## The correlation of CEO age among CEO cultural background, firms’ overinvestment To examine the relationship between CEO cultural background, overinvestment, and CEO age, this study includes CEO age as a control variable. In, column 1, without control variables and at a 5% significance level, the coefficient of CEOs with a northern cultural background is 0.0205, indicating a significant positive correlation between overinvestment and CEOs with a northern cultural background. After including CEO age as a control variable, the two variables show a negative correlation at a 5% significance level, with a coefficient of -0.000346. This means that as CEO age increases, it is more effective in suppressing the relationship between CEOs with a northern cultural background and overinvestment. In, column 3, with control variables and at a 1% significance level, the coefficient of CEOs with a northern cultural background is 0.00307, indicating a significant positive correlation between overinvestment and CEOs with a northern cultural background. This shows that when CEOs have a northern cultural background, it can lead to an increase in overinvestment in the company. Regarding CEO age, it is found that CEO age suppresses the impact of overinvestment. At a 1% significance level, the coefficient is -0.000208, indicating a negative correlation between CEO age and overinvestment. This means that as CEO age increases, it is more effective in suppressing overinvestment in the company. When multiplying CEO cultural background and CEO age, at a 1% significance level, the coefficient is -0.000556, indicating that CEO age can suppress the impact of CEOs with a northern cultural background on overinvestment. Therefore, this supports Hypothesis 2, indicating that as CEO age increases, it is more effective in suppressing the relationship between CEOs with a northern cultural background and overinvestment. This study finds that CEO age can effectively suppress overinvestment in the company, which is consistent with the manager signaling model. Therefore, this study concludes that as CEO age increases, it is more effective in suppressing the relationship between CEOs with a northern cultural background and overinvestment. ## Relationship between CEO cultural background, overinvestment, and CEO age grouped by ownership structure Due to agency problems, companies may experience overinvestment. In China, state-owned enterprises are the backbone of the economy. Previous research has shown a relationship between state-owned enterprises and overinvestment, which is believed to be caused by excessive confidence in management. State-owned enterprises in China have more severe agency problems and lower monitoring effectiveness, which makes them more prone to overinvestment. Therefore, this study divided the sample into state-owned and non-state-owned enterprises to investigate the differences between the two. distinguishes between state-owned and non-state-owned enterprises. At a 1% significance level, it was found that in state-owned enterprises, CEOs with a northern cultural background are positively correlated with overinvestment, with a coefficient of 0.0499, while in non-state-owned enterprises, the coefficient is not significant, at -0.00464. After adding CEO age, at a 1% significance level, it was found that CEO age can effectively suppress the positive correlation between CEO’s northern cultural background and overinvestment in state-owned enterprises, with a negative correlation coefficient of -0.000955, while in non-state-owned enterprises, it is not significant, with a coefficient of 0.0000686. This shows that the effects of CEO cultural background, overinvestment, and CEO age are more significant in state-owned enterprises. Therefore, this supports Hypothesis 3, indicating that the effect is more significant for state-owned enterprises than for non-state-owned enterprises. At the bottom of, this study used the Chow test. This test estimates whether two samples have differences based on the same sample data in two different periods or different samples. It can be seen that when the sample is divided into state- owned and non-state-owned enterprises, at a 1% significance level, the coefficient is 21.51, indicating that there is a difference between the coefficients of the two. ## Robustness To further expand this study, geographic clustering, period clustering, Tobit model, and alternative variables were used for robustness tests. Geographic clustering was done by dividing companies into North and South based on their registered addresses, with the Qinling-Huaihe line as the boundary between North and South China. Period clustering was done by dividing years into recession periods (2007, 2008, and 2015) and non-recession periods (other years). Alternative variables were used to construct an expected investment model, as a substitute variable for overinvestment. ### The effect of geographic location grouping on the company In, this study divides the sample into southern and northern companies based on their geographical location. At the 1% significance level, it can be observed that both CEO cultural background and overinvestment are positively and significantly correlated, regardless of whether the company is located in the south or north. When the company is located in the south, the coefficient for the influence of a CEO with a northern cultural background is larger at 0.0383, compared to 0.0248 when the company is located in the north. This suggests that when a CEO with a northern cultural background is in charge of a southern cultural company, they would bring in northern culture and influence the company’s investment decision, which is consistent with the upper echelon theory that emphasizes the impact of top management team characteristics on a company’s strategy and performance. When discussing CEO age, it is found that CEO age suppresses the relationship between CEO cultural background and overinvestment to a greater extent in southern companies, with a coefficient of -0.000727, compared to -0.000365 in northern companies. At the bottom of, the CHOW test is used, which estimates whether there is a difference between two identical samples in different time periods or under different sample data. It is found that there is a significant difference in both CEO cultural background and the interaction between CEO cultural background and age after they are grouped. ### Effectiveness of time period segmentation In, the sample is divided into recession and non-recession periods, with 2007, 2008, and 2015 being the recession periods, while the other years are non- recession periods. In, it can be found that the effect is significant in both the recession and non-recession periods, consistent with the results in. In column 1 of, at a 1% significance level, the CEO’s cultural background and the overinvestment of the company are positively correlated during the recession period, with a coefficient of 0.0914. In column 2, at a 1% significance level, the effect is positively correlated, with a coefficient of 0.0295. Regardless of the recession or non-recession periods, CEO age can effectively suppress the relationship between CEO cultural background and overinvestment of the company, indicating that overinvestment is more likely to occur in companies during a recession. When the CEO is older, they can better suppress the relationship between CEO cultural background and overinvestment of the company. ### Tobit model effect In this study, the Tobit model is used (note), and in column 2 of, it can be seen that the coefficient of CEO with a northern cultural background is 0.0457 at a 1% level of significance, indicating a significant positive correlation between the influence of CEO with a northern cultural background and corporate overinvestment, which is consistent with the hypothesis, suggesting that when a company’s CEO has a northern cultural background, they are more likely to prefer risk, resulting in an increase in overinvestment. Multiplying CEO cultural background and age, the coefficient is -0.000838 at a 1% level of significance, indicating that CEO age can inhibit the influence of CEO with a northern cultural background on corporate overinvestment, which is consistent with hypothesis 2, suggesting that the older the CEO, the more they can suppress the relationship between CEO with a northern cultural background and corporate overinvestment. These results are consistent with those in. ### Alternative variables So, we use the framework to construct the expected investment model. Non- discretionary investment is measured as the deviation between actual investment and expected investment given the company’s investment opportunities (measured by sales growth rate). The relationship is expressed as follows: $$Investment_{i,t + 1} = \mspace{360mu}\beta_{0} + \beta_{1}Sales\mspace{360mu} Growth_{i,t} + \varepsilon_{i,t + 1}$$ This study defines corporate investment as the sum of capital expenditures (Investment) and acquisition expenditures (Acquisitions), minus the resale income of sold properties, plants, and equipment (SalePPE). The sales growth rate(*Sales Growth*) is defined as the growth rate of sales between the current and the previous period, and using, the residual value (*ε*) is considered as the non-expected investment of the company. The non-expected investment (*ε*) is divided into four quartiles, with the lowest being under-investing and the highest being over-investing. Therefore, this study uses over-investing as a substitute variable in this context. In, column 2, at the 1% significance level, the coefficient of CEOs with a northern cultural background is 1.367, indicating a significant positive correlation between CEOs with a northern cultural background and over-investment in the company, consistent with Hypothesis 1. This study believes that in comparison to southern culture in China, the risk preference of northern culture is higher, so when the CEO of a company has a northern cultural background, it is more likely to lead to over-investment. After discussing the effect of CEO age, it can be seen that CEO age suppresses the impact of over-investment in the company. At the 1% significance level, the coefficient is -6.396, indicating a negative correlation between CEO age and over-investment in the company, indicating that the older the CEO is, the more effective it is to suppress over- investment in the company. When CEO cultural background and CEO age are multiplied together, the coefficient is -3.039 at the 1% significance level. It can be found that CEO age can effectively suppress the impact of CEOs with a northern cultural background on over-investment in the company, consistent with Hypothesis 2. This indicates that as the CEO of a company grows older, they can more effectively suppress the impact of a northern cultural background on over- investment in the company. # Conclusion Our research indicates a notable positive correlation between a CEO’s northern cultural background and corporate overinvestment within the context of China’s A-share listed companies from 2000 to 2018. This relationship can be attributed to the North’s distinct cultural characteristics, such as a reduced aversion to uncertainty and a higher preference for risk. We also examined the moderating effects of CEO age and company ownership on these dynamics. Consistent with established research, we found that older CEOs tend to employ more conservative strategies. This tendency reduces the incidence of overinvestment commonly associated with a northern cultural background. Furthermore, state-owned companies showed a pronounced influence of a CEO’s northern cultural background on corporate overinvestment. This observation aligns with recent studies suggesting state-owned enterprises have elevated agency costs and managerial overconfidence. Our study illuminates the complex role of a CEO’s cultural background, age, and the company’s ownership structure in influencing corporate overinvestment. It underscores the importance of these variables in shaping corporate governance and strategic investment decisions. Our findings add to the existing literature, providing valuable insights for researchers, investors, and policymakers. This is particularly beneficial in understanding the investment decision-making dynamics within fast-growing economies like China. However, our study also reveals the need for further exploration in this field. Future research could delve deeper into these relationships, considering other potential moderating variables and different cultural contexts. Longitudinal studies would also be beneficial in capturing these dynamics’ evolution over time. Importantly, our findings related to Chinese state-owned enterprises serve as a foundation for comparative studies in various cultural and institutional settings. This would help discern the universal and context-specific aspects of CEO decision-making. A deeper examination of the intersection between CEO age and gender can also contribute to a more comprehensive understanding of gender disparities in decision-making. The effects of leadership transitions, particularly across generations, on investment decisions warrant more investigation, as they could yield crucial insights into shifts in investment strategies and their impacts on firm performance. Moreover, widening the scope to include other CEO characteristics, such as educational background or personality traits, could enrich our understanding of leadership and strategic decision interplay. Additionally, the influence of board characteristics on CEO decisions, an often overlooked aspect, deserves consideration for its crucial role in guiding a firm’s trajectory. Undoubtedly, exploring these areas will significantly enrich our understanding of the intricate relationship between CEO characteristics and firm investment strategies. [^1]: The authors have declared that no competing interests exist. [^2]: ‡ YLH and WTC also contributed equally to this work.

# Introduction LPA is a potent signaling lipid molecule that is involved in numerous phenomena, such as cell migration, preventing cellular apoptosis, angiogenesis, and others. LPA modulates its biological functions through the activation of at least six G-protein-coupled receptors (LPA1-6). Liver regeneration is an important phenomenon after liver injury, and the reproducibility of a partial hepatectomy model has made it the preferred approach for studies on liver regeneration. Many studies have demonstrated that exogenous factors, such as pharmaceutical agents like acetaminophen, chemicals like CCl₄, and endogenous factors, such as angiotensinogen, IL-6, and interferon gamma receptors, are critically involved in liver regeneration. Ikeda et al. first demonstrated that LPA might affect the proliferation of hepatocytes and stellate cells in those liver diseases that disrupted platelet activation. Recently, by using a partial hepatectomy mouse model, Simo et al. found that liver regeneration after partial hepatectomy was associated with significant changes in circulating LPA levels (LPA increased significantly at 72 hours post- partial hepatectomy to 6.30 ± 0.67 μM as compared to 3.58 ± 0.37 μM in sham-operated mice) and that hepatic mRNA levels of LPAR1, LPAR3, and LPAR6 were expressed in a time- and cell-dependent manner. Additionally, their immunohistochemical staining results revealed that LPAR1 protein was expressed in non-parenchyma cells, and that LPAR3 and LPAR6 proteins were widely distributed in regenerating liver tissue. The study by Simo et al. clearly demonstrated the phenomena of increased LPA levels and expression levels of LPAR’s during liver regeneration. LPA receptors were originally defined as an endothelium differentiation gene (edg) subfamily of G-protein-coupled receptors. Liver sinusoidal endothelial cells have been found to play important roles during liver regeneration. In this study, we used endothelial cell-specific marker CD31-coated magnetic beads to isolate liver sinusoidal endothelial cells from mice, and confirmed their purity by determining CD45 positive rates. These liver sinusoidal endothelial cells were then used to investigate the effects of treating these cells with physiological LPA levels with regard to their expression of angiogenesis factors, cytokines, and chemokines by using proteome profile arrays. # Materials and Methods ## Isolation of mouse liver sinusoidal endothelial cells Our animal use protocols were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of National Taiwan University College of Medicine and College (IACUC Approval No: 20120247). For primary cultures, mouse liver sinusoidal endothelial cells were isolated from 20 normal 6–8 week-old male C57BL/6 mice per experiment. The yields of mouse liver sinusoidal endothelial cells were, on average, 5 x 10⁶ sinusoidal endothelial cells per 1 g of liver. A mouse was first anesthetized with 5% isoflurane inhalation and after sacrifice by C0₂ asphyxiation; the entire liver was carefully removed. Liver tissues were washed twice with 50 ml of Dulbecco's modified Eagle's medium (DMEM; Invitrogen Technologies), cut into small pieces, and then incubated with 2.5 ml of 0.1% type II collagenase in DMEM at 37°C for 1.5 hours. A cell suspension was passed through a 100-um mesh and washed twice with 5 ml of Hanks balanced salts solution that contained 10% fetal calf serum (Invitrogen). Isolated cells were incubated in 1 ml of cold M199 medium (Invitrogen) that contained 1μg/ml of rabbit, anti-mouse CD31 antibody (sc-1506-R; Santa Cruz Biotechnology) for 30 minutes with gentle agitation, followed by adding 50 μl of Dynabeads M280 sheep, anti-rabbit IgG (10 mg/ml), and then incubated for an additional 30 minutes. Endothelial cells that bound to the magnetic beads were removed from unbound non-endothelial cells by magnetic isolation using an MPC-1 magnet (Dynal, Oslo, Norway). The cells bound on beads were re-suspended and cultured in endothelial cell growth medium (Cell Applications, San Diego, CA, USA). ## Flow Cytometry Analysis The following antibodies/reagents were used for flow cytometry analyses: anti- mouse CD31 (sc-1506-R; Santa Cruz Biotechnology), anti-mouse CD45 (sc-25590; Santa Cruz Biotechnology), and normal rabbit IgG (sc-3888; Santa Cruz Biotechnology), used as a negative control. Stained cells were analyzed by flow cytometry (BD FACSCalibur). ## Preparation of conditioned medium (CM) Cultures of liver sinusoidal endothelial cells (2 × 10⁶/10-mm dish) were rinsed twice with PBS and then cultured in 5 ml of serum-free M199 with 5 μM LPA or vehicle for 24 hrs. Conditioned medium from liver sinusoidal endothelial cells was collected and clarified by centrifugation (10,000 rpm for 5 min at 4°C) to remove cellular debris prior to protein array analysis. ## LPA and chemical inhibitor LPA (Oleoyl-L-α-lysophosphatidic acid sodium salt, L-7260) and ki16425 (SML0971) were purchased from Sigma (St. Louis, MO, USA). ## Protein arrays Mouse angiogenesis, cytokine, and chemokine antibody arrays (Proteome Profiler, R&D Systems; Ary015, Ary006, and Ary017, respectively) were used to analyze angiogenesis factor, cytokine, and chemokine expression profiles according to the manufacturer’s instructions. Briefly, conditioned medium was first mixed with a biotinylated detection antibody cocktail at room temperature for 1 hour while the array membrane was blocked with blocking solution provided by the manufacturer. A digital imaging system (Bio Pioneer Tech Co., Ltd.) was used to detect chemiluminescent signals, which were further analyzed using ImageJ software. ## Enzyme Immunoassay (EIA) Conditioned medium from liver sinusoidal endothelial cells was used for determinations of Cyr61, TIMP-1, C5/C5a, M-CSF, MCP-5, SDF-1, gp130, CCL28, and CXCL16 protein levels. These were determined using specific EIA kits (R&D Systems), according to the manufacturer’s instructions. Each measurement was performed in duplicate. ## Q-RT-PCR Q-RT-PCR was performed as in a previous study (13). Briefly, total RNA was isolated from liver sinusoidal endothelial cells using RNAzol B reagent (Biotecx Laboratories, Houston, TX). Then, cDNA was prepared from 2 μg of total RNA using random hexamer primers (ImProm-II RT System; Promega, Southampton, UK). LPAR1-6, Cyr61, TIMP-1, C5/C5a, M-CSF, MCP-5, SDF-1, gp130, CCL28, and CXCL16 mRNA levels were determined with a quantitative real-time PCR detection system (Light Cycler DNA Master SYBR Green I; Roche Molecular Biochemicals, Indianapolis, IN). The primers used are shown in. The amplification program included an initial incubation at 61°C for 20 min, followed by 40 cycles of denaturation at 95°C for 10 s, annealing at 55–57°C for 10 s, and extension at 72°C for 10 s. The expression level of each mRNA was normalized to that of GAPDH mRNA. ## Statistical analysis Results are given as means ± SD’s. Two-tailed t-tests were used to compare the results between the indicated groups in Results. A p-value of \< 0.05 was considered significant. # Results ## LPAR1, LPAR3, and LPAR6 are expressed in liver sinusoidal endothelial cells Because lysophosphatidic acid receptor (LPAR) expression has been shown to be involved during liver regeneration, we focused on the effects of LPA on liver sinusoidal endothelial cells. We isolated liver sinusoidal endothelial cells by using Dynabeads to positively isolate CD31-positive endothelial cells from mouse liver tissue. The purity of isolated liver sinusoidal endothelial cells was verified by determining the CD31 positive rate by flow cytometry and by expression of the hematopoietic cell marker CD45 that has also been reported to be expressed on liver sinusoidal endothelial cells. The purities of liver sinusoidal endothelial cells during five serial passages were determined, which showed that from the first to the fifth passage, the CD31 positive rates were from 94.4 ± 2.3% to 80.3 ± 4% and the CD45 positive rates were from 82.5 ± 4.7% to 74.3 ± 3.9%. For this study, liver sinusoidal endothelial cells were only used after the fifth passage. We also determined the mRNA expressions of LPAR’s by qRT-PCR. This showed that LPAR1 and LPAR3 mRNA’s were strongly expressed and that LPAR6 mRNA was weakly expressed in liver sinusoidal endothelial cells. To determine the effects of LPA on liver sinusoidal endothelial cells, we screened the conditioned medium derived from LPA-treated liver sinusoidal endothelial cells for specific biological functions based on the results of proteome profile arrays. ## LPA treatment enhances angiogenesis related factors Cyr61 and TIMP-1 expression in liver sinusoidal endothelial cells Conditioned media derived from liver sinusoidal endothelial cells after vehicle treatment or after LPA treatment were used to determine angiogenesis-related factors’ expression. When using an angiogenesis related protein array, the spots for Cyr61 and TIMP-1 had different intensities between conditioned media after vehicle treatment and after LPA treatment. Quantitative results showed that Cyr61 and TIMP-1 expressions were 3.61 ± 0.2-fold and 2.53 ± 0.13-fold higher, respectively, with LPA treatment as compared to those with vehicle treatment. ## LPA treatment enhances cytokine C5/C5a, M-CSF, MCP-5, and SDF-1 expression in liver sinusoidal endothelial cells Conditioned media derived from vehicle treated and LPA treated liver sinusoidal endothelial cells were used to determine cytokines’ expression using a cytokine protein array. This showed that the spots for C5/C5a, M-CSF, MCP-5, and SDF-1 had different intensities between conditioned media after vehicle treatment and after LPA treatment. Quantitative results showed that LPA treatment had enhanced C5/C5a (3.56 ± 0.0.4-fold higher), M-CSF (2.17 ± 0.14-fold higher), MCP-5 (3.32 ± 0.21-fold higher), and SDF-1 (2.48 ± 0.13-foldhigher) expression relative to those after vehicle treatment. ## LPA treatment enhances chemokine MCP-5, gp130, CCL28, and CXCL16 expression in liver sinusoidal endothelial cells Different conditioned media were also used to determine chemokines’ expression using a chemokine protein array. This showed that the spots for MCP-5, gp130, CCL28, and CXCL16 had different intensities between conditioned media after vehicle treatment and after LPA treatment. Quantitative results showed that LPA enhanced MCP-5 (2.13 ± 0.13-fold higher), gp130 (2.12 ± 0.13-fold higher), CCL28 (3.33 ± 0.16-fold higher), and CXCL16 (2.53 ± 0.12-fold higher) expression relative to those after vehicle treatment. ## LPA induced angiogenesis factor, cytokine, and chemokine expression in liver sinusoidal endothelial cells is mediated primarily through LPAR1 and LPAR3 signaling LPA regulates several proteins based on the LPAR subtype. Our results showed that LPAR1 and LPAR3 mRNA’s were strongly expressed in liver sinusoidal endothelial cells after LPA treatment. Thus, we investigated LPAR1 and LPAR3 involvement in LPA-mediated angiogenesis factor, cytokine, and chemokine expression in liver sinusoidal endothelial cells by using the specific chemical inhibitor ki16425. This showed that pre-treating cells with ki16425 significantly inhibited LPA enhanced expressions of angiogenesis factors (Cyr61 and TIMP-1), cytokines (MCP-5 and SDF-1), and chemokines (MCP-5, gp130, CCL28, and CXCL16). However, LPA enhanced cytokine C5/C5a and M-CSF expressions were not inhibited by ki16425 pre-treatment. To determine if LPA had a direct or an indirect effect on the expressions of these LPA induced angiogenesis factors, cytokines, and chemokines, the time course of LPA effects on these genes’ mRNA expressions were determined. This showed that the mRNA expressions for angiogenesis factors (Cyr61 and TIMP-1), cytokines (MCP-5 and SDF-1), and chemokines (MCP-5, gp130, CCL28, and CXCL16) were significantly increased after 8 hours of LPA treatment. By comparison, LPA enhanced cytokine C5/C5a and M-CSF mRNA expressions were significantly increased after 16 hours of LPA treatment. Our results that ki16425 did not inhibit LPA enhanced C5/C5a and M-CSF protein expression and the late transcriptional regulation (16 hours) for LPA enhanced C5/C5a and M-CSF mRNA expression suggested that LPA might regulate C5/C5a and M-CSF through an indirect effect. Taken together, our results suggested that LPA might enhance several important angiogenesis factors, cytokines, and chemokines, including Cyr61, TIMP-1, C5/C5a, M-CSF, MCP-5, SDF-1, gp130, CCL28, and CXCL16, expression in liver sinusoidal endothelial cells that was mediated by LPAR1 and LPAR3 signaling. # Discussion Liver regeneration is an important phenomenon that reflects the reparative capacity of this vital organ. Several growth factors and cytokines, such as interleukin-6, tumor necrosis factor-α, and hepatocyte growth factor, have been found to be critically involved in liver regeneration, particularly in parenchymal cells. However, vascular endothelial growth factor (VEGF) and its receptors, flt-1 and KDR/flk-1, are expressed by non-parenchymal cells, including sinusoidal endothelial cells, in the liver after partial resection. Using a sFlt-1-expressing adenoviral vector to infect C57BL6 mice to express the dominant negative receptor for VEGF and after 70% partial hepatectomy, a liver regeneration model showed that this angiogenesis inhibitor significantly suppressed hepatic regeneration. By using VEGFR1 tyrosine kinase knockout mice, Ohkubo H et al. found that VEGFR1-expressing macrophages were recruited to the liver during hepatic ischemia/reperfusion and contribute to liver repair and sinusoidal reconstruction through regulating expression of pro-angiogenic factors. This study demonstrated that VEGFR1 activation is a potential therapeutic strategy for promoting liver repair and sinusoidal restoration after acute liver injury. Coulon S et al. demonstrated that the blockage of VEGFR2 could attenuate steatosis and inflammation in a diet-induced mouse model for nonalcoholic steatohepatitis. The role of angiogenesis in the pathophysiology in nonalcoholic steatohepatitis may be worthwhile for a preventive and therapeutic setting. By using an Innovative in vivo μCT methodology, Ehling J et al. found that CCL2-dependent infiltrating macrophages promote angiogenesis in progressive experimental liver fibrosis. Liver sinusoidal endothelial cells are known to contribute to liver regeneration after liver injury. In endothelial cell membranes, LPA is a well-known pleiotropic lipid molecule that has potent effects on cell migration and membrane permeability. The receptors for LPA that were first identified were designated the endothelium differentiation gene (edg) subfamily of G-protein- coupled receptors. LPA has been found to primarily act through the activation of at least six G-protein-coupled receptors (LPA1-6). In this study, we found that LPAR1 and LPAR3 mRNA’s were strongly expressed and that LPAR6 mRNA was weakly expressed in mouse liver sinusoidal endothelial cells. Based on these findings for LPA receptors, we used a physiological level of LPA (5 uM) to stimulate liver sinusoidal endothelial cells for 24 hours. The conditioned media that were derived from these cell cultures were used for angiogenesis factor, cytokine, and chemokine expression profile determinations. Our results showed that LPA treatment enhanced Cyr61, TIMP-1, C5/C5a, M-CSF, MCP-5, SDF-1, gp130, CCL28, and CXCL16 expression in liver sinusoidal endothelial cells. Cyr61 has been found to promote liver fibrosis regression through the induction of cellular senescence in hepatic myofibroblasts. TIMP-1 knockout mice had impaired liver function and histological preservation after hepatic ischemia and reperfusion injury. Further, TIMP-1 expression promotes the survival and proliferation of liver cells, regulates leukocyte recruitment, and reduces active caspase-3 levels and increases Bcl-2 expression and Akt phosphorylation. In C5-deficient mice, severely defective liver regeneration and persistent parenchymal necrosis were found after exposure to carbon tetrachloride. Additionally, murine C5 or C5a reconstitution in C5-deficient mice significantly restored hepatocyte regeneration after toxic injury, which results showed that C5/C5a contributed essentially to the early priming stages of hepatocyte regeneration. For osteopetrotic mice that genetically lack functional M-CSF, after these mice underwent 70% partial hepatectomy, the proliferation of hepatocytes was significantly impaired. However, when osteopetrotic mice were intraperitoneally administered mouse recombinant M-CSF before partial hepatectomy, the numbers of Kupffer cells were increased and liver regeneration was recovered. In massive liver injury models, it was found that oval cell repair was involved in up-regulating the expression of SDF-1 in hepatocytes. The major biological role of SDF-1 is as a potent chemoattractant for hematopoietic cells homing to the liver after hepatic resection. The interleukin (IL)-6 family of cytokines signal exclusively via the gp130 co- receptor, which subsequently dimerizes and initiates intracellular signaling. Activation of IL-6/gp130-mediated STAT3 signaling pathway is crucial for both acute phase genes’ regulation after partial hepatectomy and hepatic differentiation of adult bone marrow-derived mesenchymal stem cells. In this study, we also investigated the possible signaling pathway involved in LPA enhanced Cyr61, TIMP-1, C5/C5a, M-CSF, MCP-5, SDF-1, gp130, CCL28, and CXCL16 expression in liver sinusoidal endothelial cells. Based on our findings for LPAR1 and LPAR3 mRNA expression, we used ki16425 that selectively inhibits LPAR1 and LPAR3 mediated actions. These results showed that LPA enhanced C5/C5a and M-CSF expressions were not inhibited by ki16425. Combined with our mRNA level determination, we concluded that the regulation of LPA enhanced C5/C5a and M-CSF expression in liver sinusoidal endothelial cells may have been through LPAR1 and LPAR3 indirect regulation. Growth factors or cytokine could be regulated by autocrine effect, one factor may be first induced by LPA, following it may stimulate another factor to express through autocrine effect. Such as Lin CH et.al demonstrated that LPA-stimulated lymphangiogenesis in HUVECs is mediated through IL-1β-induced VEGF-C expression. The results of this study clarified the expression of LPA receptors in mouse liver sinusoidal endothelial cells and showed that important angiogenesis factors, cytokines, and chemokines were regulated by LPA in mouse liver sinusoidal endothelial cells. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HSL PHL CNC. Performed the experiments: CHC SLL. Analyzed the data: CHC WHL HSL. Contributed reagents/materials/analysis tools: CHC SLL CMH SHK FCP SYW. Wrote the paper: HSL CHC.

# Introduction Rhacophoridae (Old World tree frogs), are a monophyletic family with a high diversity, constituting ca. 6% of the world’s anuran species. Liem analyzed the skeletal morphology in 420 lineages, representative of 14 rhacophorid genera. However, despite their prevalence and existing data for adult morphology, the patterns and processes of skeletal ontogeny is largely unknown. Rhacophoridae has been subjected to considerable recent taxonomic revisions, resulting in the recognition of many new independent evolutionary lineages at genus level; 11 of the 18 recognized rhacophorid genera are recent descriptions. These new generic descriptions are largely based on molecular data and to a lesser extent on molecular data and morphology. One of the main impediments to new lineage identification, prior to the advent of molecular phylogenetic techniques, was the lack of consistent external morphological characters. In recent phylogenetic reconstructions of rhacophorid relationships, adult skeletal data have been sparsely used, but never skeletal ontogeny. The evolution of the spectacular diversity of reproductive modes in rhacophorids shows that both terrestrial direct-developing and foam-nesting species arise through gel-nesting ancestors, while basal rhacophorids are aquatic breeders. The terrestrial direct-developing forms, which spend their entire embryonic sequence within eggs, are placed in three well-supported clades―*Philautus*, *Pseudophilautus* and *Raorchestes*. Basal, fully aquatic-breeding, genera (*Buergeria* and *Liuixalus*) exhibit a biphasic lifecycle, i.e., eggs are deposited in water and a free-swimming larva metamorphoses into an adult. The gel-nesting species that lay terrestrial eggs with aquatic larvae are in several distinct clades (*Kurixalus*, *Mercurana*, *Gracixalus*, *Beddomixalus*, *Frankixalus*, *Feihyla*). Finally, foam-nesting genera are in two paraphyletic clades (*Rhacophorus*, *Polypedates*, *Taruga*, *Ghatixalus* in one clade and *Chiromantis* as another). These have terrestrial foam nests and postembryonic free-swimming tadpoles. Skeletal ontogeny across developmental stages of these forms is not known, except for a single study on *Pseudophilautus silus*. Molecular phylogenies are often used to map data such as morphology and life history traits. However, this has not been applied to many newly recognized taxa, including novel rhacophorids. While several classical studies concentrate on osteology and ontogeny to ascertain higher-level systematics, ontogeny is not used to resolve the Rhacophoridae. It is only now possible to analyze the osteology of Rhacophoridae in a phylogenetic context. Sri Lankan rhaocophorids belong to three independent evolutionary lineages representative of two major life history strategies: foam-nesting *Polypedates* and *Taruga*, and terrestrial direct-developing *Pseudophilautus*. Ontogenetic skeletal development of the newly recognized genus *Taruga*, which was previously referred to as a part of *Polypedates*, has never been studied. Currently *Taruga* and *Polypedates* are recognized as sister lineages. *Taruga* is an endemic genus, with adult morphological characters (prominent calcar at the distal end of the tibia, conical tubercles surrounding the cloaca) and tadpole morphologies (features of the buccal cavity and vent tube), distinguishing it from *Polypedates*. Using an almost complete series of differentially stained tadpoles, metamorphs and adults, we examine the postembryonic skeletal development and adult skeletal osteology of the four species belonging to the two sister lineages of foam nesters: *Taruga* (*T*. *eques* and *T*. *longinasus*) and *Polypedates* (*P*. *cruciger* and *P*. *maculatus*). Here, we compare the patterns and processes of ossification in these two genera to facilitate deeper level comparative analyses of morphological evolution within Rhacophoridae. # Materials and Methods ## Field collection and lab rearing Four freshly deposited foam nests from two species each of *Taruga* and *Polypedates* were collected from the field: *T*. *eques–*Agarapatana, 6.901820°N, 80.690482°E; *T*. *longinasus*–Kanneliya, 6.248910°N, 80.333669°E; *P*. *cruciger*–Kurunegala, 7.488060°N, 80.363873°E; *P*. *maculatus*–Kantale, 8.351803°N, 81.004516°E. The nests, together with the substrate in which they were laid, were carefully transferred to our lab (University of Peradeniya), in sealed polyethylene bags placed in a cooler. They were positioned above aquaria containing aged tap water so that hatching tadpoles would fall directly into the water to undergo further development. The tadpoles were raised in the lab under identical water quality, light (12 hr day/night) and feeding conditions. They were fed twice per day; food uneaten 10 min after feeding was siphoned off and the water levels adjusted using aged tap water. They were sampled periodically from hatching (Gosner stage 24) to a fully metamorphosed froglet (stage 46) to represent every developmental stage. Two adult males from each species were also collected from the field (same locations as above). Research was conducted under the permission of Department of Wildlife Conservation (permit no. WL/3/2/13/13) and Forest Department (permit no. R&E/RES/NFSRC/14) of Sri Lanka. Specific methods of collection, euthanasia, tissue sampling and fixation followed the guidelines for use of live amphibians were approved by the ethical committee of Postgraduate Institute of Science, University of Peradeniya at its 16th meeting held on 14th November 2014. ## Preservation and osteology Sampled tadpoles were euthanized using tricaine methanesulphonate (MS-222) and preserved in 10% neutral-buffered formalin. They were stored in 70% alcohol following a graded alcohol series of 30% and 55%; the specimens were kept overnight at each step. One to three larvae were taken from each stage between 25 and 46 (– Tables). Additionally, two adults from each representative species, *Polypedates cruciger* (average snout-vent length (SVL) = 55.30 mm), *P*. *maculatus* (average SVL = 54.78 mm), *Taruga eques* (average SVL = 38.76 mm) and *T*. *loginasus* (average SVL = 38.90 mm) were also differentially stained for bone and cartilage. Osteological preparations and descriptions are based on 150 specimens (– Tables). Specimens were cleared and differentially stained for bone and cartilage using alizarin red and Alcian blue, respectively. To minimize differences in cleared and stained specimens, all the specimens were processed at the same temperature and treated with same stock solutions. Each specimen was scored for presence of bone by using a stereomicroscope within 1–3 days of the staining process (– Tables). Ossification indices were calculated for each stage by dividing the number of bones observed at a specific stage by the total number of bones. Cartilage terminology follows. ## DNA barcoding and genetic analyses The 16S rRNA mitochondrial gene fragment from a single tadpole from each foam nest was amplified and sequenced to ascertain species identity. Tadpoles were euthanized in MS-222, preserved in absolute ethanol and stored at –20°C in the Department of Molecular Biology & Biotechnology (DZ), University of Peradeniya. DNA was extracted from ethanol-preserved tail muscle using a standard protocol. Portions of the mitochondrial 16S ribosomal RNA gene (600 bp) were amplified by PCR using primer sets 16Sar and 16Sbr. PCR products were sequenced directly with dye-termination cycle sequencing. Newly generated sequences were checked using 4peaks (v. 1.7.1). Published sequences from 37 closely related species of *Polypedates* and *Taruga* (following Li *et al*. Meegaskumbura *et al*.) were included in a dataset. Additionally, four mantellid species were used as the outgroup. The compiled 16S rRNA dataset was aligned using ClustalW as implemented in MEGA v. 6.0. Uncorrected pairwise distances were calculated using PAUP\* 4.0b10. Highly variable regions were manually removed from the dataset; the final dataset consisted of 465 bp. The best-fit model (GTR+I+G) was chosen using jModeltest v. 2.1.4. Maximum likelihood (ML) analysis was performed to infer relationships among the lineages and clades using the software GARLI (Zwickl 2006) on the Cipres Science Gateway. # Results ## Phylogenetic position Phylogenetic relationships among the rhacophorid taxa presented here, agree with previous analyses. Tadpole sequences of a given species cluster with their respective adult sequences, forming two strongly supported clades (*Taruga* and *Polypedates*); uncorrected pairwise distances between the tadpole sequence (GenBank accession numbers: KY111847–KY111850; given upon acceptance) and its respective adult sequence range between 0.001–0.002. ## Larval neurocranium and the first oropharyngeal arch (stage 35) A nearly complete ontogenetic series of tadpoles was examined for each of the four species, which represent two genera (– Tables). Description of the neurocranium and first oropharyngeal arch at stage 35 is followed by descriptions of cranial and postcranial bones. The neurocranium is slightly longer than wide in each species. The neurocranium is widest at the midpoint of the arcus subocularis in *Taruga eques* and *T*. *longinasus* but at the posterior end of the arcus subocularis in *Polypedates cruciger* and *P*. *maculatus*. The chondrocranium of each species is oval- shaped in dorsal view. ### Ethmoid region In all four species the trabecular horns diverge laterally from one another and form ligamentous attachments to the lateral alae of the suprarostral cartilages. Anterior ends of the trabecular horns are almost flat. Trabecular horns account for 32.2% of chondrocranial length in *T*. *eques*, 26.7% in *P*. *cruciger*, 31.5% in *T*. *longinasus* and 29.5% in *P*. *maculatus*. The lamina orbitonasalis is present anterior to the quadratocranial commissure. The nasal septum is absent. ### Braincase All four species have a single large opening (i.e. frontoparietal fontanelle) in the roof of the larval braincase. This opening is demarcated anteriorly by the lamina orbitonasalis, posteriorly by the tectum synoticum, and laterally by the otic capsules and taenia tecti marginalis. The taenia tecti medialis subdivides the posterior frontoparietal fontanelle into two lateral halves. The taenia tecti medialis is almost 50% as long as the frontoparietal fenestra except in *P*. *maculatus* (*T*. *eques*, 46%; *T*. *longinasus*, 41%; *P*. *cruciger*, 43%; and *P*. *maculatus*, 37%). Two pairs of craniopalatine and primary carotid foramina are present ventrally (not shown). ### Otooccipital region Each oval-shaped otic capsule possesses a large fenestra ovalis ventrolaterally. The prootic and oculomotor foramina are visible in the orbital cartilage; the latter opening is smaller and directed more ventrally. Laterally projecting crista parotica can be seen anterolateral to the otic capsules. The crista parotica anteriorly bears a well-developed anterolateral process, which is long, finger-like in *T*. *longinasus*, stout and triangular in *T*. *eques* and long and triangular in *P*. *cruciger* and *P*. *maculatus*. A laterally projecting, triangular posterolateral process is well developed in the two *Polypedates* species but reduced in the two *Taruga*. Paired occipital arches extend ventrally from the posteromedial margins of the otic capsules and form occipital condyles by fusing with the basal plate. ### Palatoquadrate cartilage The palatoquadrate lies lateral to the braincase; it is oriented anteroposteriorly and parallel to the longitudinal axis of the chondrocranium. The anterior quadratocranial commissure, the ascending process and the otic process connect the palatoquadrate to the neurocranium. The posterior curvature of the palatoquadrate is at the level of attachment to the ascending process of orbital cartilage. The quadratoethmoidal process is triangular extending from the anterior margin of the anterior quadratocranial commissure. The processus pseudopterygoideus is absent in all four species. The muscular and articular processes constitute the anterior processes of the palatoquadrate. The muscular process is dorsally expanded; it is 1.74 mm wide in *P*. *cruciger*, 1.50 mm in *P*. *macualtus*, 1.26 mm in *T*. *eques* and 0.94 mm in *T*. *longinasus*. ### Suprarostral cartilages Paired suprarostral cartilages are oriented perpendicular to the longitudinal axis of the chondrocranium and lie between the trabecular horns. They support the upper horny beaks. Each comprises a flat, rectangular ala laterally and a medial corpus. Adjacent corpora are fused ventromedially in *T*. *longinasus*, *T*. *eques* and *P*. *maculatus*, and dorsomedially in *P*. *cruciger*. The gap between posterior margins of the medial corpus is 6.8% of chondrocranial width in *P*. *cruciger*; the gap between anterior margins of the medial corpus is 11.5% in *T*. *longinasus*, 9.3% in *T*. *eques* and 6.0% in *P*. *maculatus*. Lateral alae curve rostrally, where they fuse with the medial corpus. The suprarostrals of *T*. *longinasus*, *T*. *eques* and *P*. *maculatus* appear U-shaped in anterior view but assume an inverted U-shape in *P*. *cruciger*. ### Lower jaw Paired Meckel’s and infrarostral cartilages form the lower jaw. Infrarostrals are wedge-shaped and triangular in cross section. Each transverse infrarostral cartilage is confluent with each other medially via a thin bar of cartilage (commissura intramandibularis). The posterior margins of each infrarostral are connected via a ligament with the anterior margin of the adjacent, sigmoid- shaped Meckel’s cartilage. Meckel’s cartilage articulates with the articular process of the palatoquadrate laterally. ## Ossification of the skull The cranial ossification sequence is depicted in. The following comparisons of adult bones among the four species emphasize phylogenetically informative characters (character matrix is given as) recognized by Liem and Scott. ### Parasphenoid The parasphenoid is a triradiate, medial and dorsoventrally flattened bone that invests the basal plate. It is the first bone to ossify in all four species. Ossification begins in the cultriform process but continues posteriorly along the midventral floor of the braincase to the exoccipital region, where it extends laterally to form the paired alae. The anterior, bifid tip of the cultriform process lies posterior to the planum antorbitale; it is serrated in adult *T*. *eques* and *T*. *longinasus* and sharply pointed in *P*. *cruciger* and *P*. *maculatus*. The alae extend laterally and are moderately long; they are 45.9% of cranial width in *P*. *cruciger*, 37% in *P*. *maculatus*, 46.2% in *Taruga eques* and 67.1% in *T*. *longinasus*. The caudal edge of the sphenethmoid articulates with the rostral edge of the parasphenoid seen at stage 46 in *P*. *maculatus* but only in adults in *P*. *longinasus*, *T*. *eques* and *P*. *cruciger*. Alae fuse with the otic capsules at stage 44 in *T*. *eques*, stage 45 in *T*. *longinasus*, stage 42 in *P*. *cruciger* and stage 46 in *P*. *maculatus*. ### Exoccipitals Ossification of the paired exoccipitals begins along the dorsal regions of the occipital condyles. It continues around the middle part of the occipital arch dorsally and ventrally, and also along the otic capsules and basal plate. Subsequently, ossification extends laterally into the posteromedial portions of the otic capsules, the occipital condyles and the margins of the jugular foramen. Exoccipitals and prootics fuse to form the posterolateral parts of the braincase and the anterolateral, anteroposterior and anteromedial margins of the otic capsules at stage 44 in *Taruga eques*, *T*. *longinasus* and *P*. *cruciger*, and at stage 43 in *P*. *maculatus*, ### Frontoparietals Ossification begins near the midpoint of the taenia tecti marginalis and extends rapidly along the longitudinal axis, growing anteriorly and posteriorly. The frontoparietal ossification also proceeds medially at a comparatively slower rate, thus increasing in length and breadth while flanking the frontoparietal fenestrae. In adults, frontoparietals are slender, long, paired bones, which are narrowly separated at the midline. The frontoparietal fontanelle is bordered by taenia tecti marginalis laterally, nasal cartilages anteriorly and sphenethmoid anterolaterally. The greatest width of the skull (in adults) is achieved posterior to the otic capsules in all four species. Only the two *Polypedates* species possess parieto-squamosal plates, which are placed laterally on the posterolateral ends of the frontoparietals. ### Prootics Osteogenesis of prootics is initiated as a small center along the anteromedial margin of the otic capsule. These bones are deposited gradually, laterally and posteriorly over the anterior and anterolateral margins of the otic capsules. Prootics articulate with the lateral margins of frontoparietals dorsally (at stage 43 in *T*. *eques*, stage 45 *T*. *longinasus*, stage 44 in *P*. *cruciger* and stage 42 in *P*. *maculatus*). In adults, prootics form the anterolateral and ventrolateral margins of the otic capsules, and posterolateral walls of the braincase. ### Septomaxilla Osteogenesis of the septomaxillae is initiated as tiny centers, located anterolaterally within the nasal capsules (see – Tables for ossification sequences of each individual species). In adults, paired, dermal, semilunar- shaped septomaxillae lie within the nasal capsules, below the nasal roof, supporting the external nares. Septomaxillae are clearly visible between the nasals and pars facialis of the maxilla in lateral view, and also can be observed between the oblique cartilage and nasals in the dorsal view. ### Premaxilla Initial ossification of the alary process is visible dorsal to the trabecular horns at stage 40 in *T*. *eques*, stage 41 in *T*. *longinasus* and stage 42 in both *P*. *cruciger* and *P*. *maculatus*. Dentary and palatine processes of the premaxillae appear next, respectively. Premaxillary teeth arise as short, pointed buds at stage 42 in *P*. *cruciger* (number of teeth, left/right: 6/5) and at stage 46 in *P*. *maculatus* (8/8) but after metamorphosis in *T*. *longinasus* and *T*. *eques*. After metamorphosis, premaxillae unite syndesmotically, completing the upper jaw anteriorly. The dentary process of the premaxilla bears a horizontally oriented dental ridge at stage 46 in all four species. The alary process reorients vertically during metamorphosis, expanding dorsally when the trabecular horns erode during metamorphosis. In adults, paired premaxillae complete the maxillary arcade anteriorly. The premaxillae are well separated from one another anteriorly and also from the laterally adjacent maxillae. The premaxillae are located anteromedially, and lay dorsal to the proximal trabecular horns. These bones are composed of dentary, alary and palatine processes. The alary process of the premaxilla is curved laterally and support the cartilage of the nasal capsules. The palatine process is posteromedially oriented, and serves as a site for attachment of the soft tissue lining the buccal cavity. ### Maxilla The maxillae are paired, dermal, dentigerous bones containing dentary, palatine and facial processes. Initial thin ossifications (just posterior to septomaxillae in dorsal view) are located on either side of the skull along the posterior margin of the suprarostral cartilages. These rapidly extend anteriorly and posteriorly to form the pars facialis. The maxillary ossification reaches the level of the posterior margin of the orbit by stage 43 in all species. The facial process articulates with the premaxilla, forming a pointed snout in *T*. *eques* and *T*. *longinasus* and a blunt snout in both *Polypedates*. Premaxillae and maxillae begin to overlap at stage 44 in *T*. *eques*, stage 45 in *T*. *longinasus*, stage 46 in *P*. *maculatus* and in adults in *P*. *cruciger*. In *T*. *eques*, the maxilla extends beyond the caudal margin of the eye at stage 45 and it articulates with the quadratojugal at stage 46. This articulation also occurs at stage 46 in *P*. *maculatus and T*. *longinasus*, but it is seen in adults in *P*. *cruciger*. The facial process of the maxilla articulates broadly with the quadratojugal. Maxillary teeth are first visible at stage 42 in *P*. *cruciger* (number of maxillary teeth: 9/8), at stage 46 in *P*. *maculatus* (12/13) and in adults in *T*. *longinasus* (12/12) and *T*. *eques* (11/13). ### Nasals The nasals are paired, crescent shaped, expansive bones roofing part of the nasal capsules. The long axes are arranged transversely, parallel to the maxillae in all four species. After the formation of cartilaginous tectum nasi, paired centers of ossifications are narrowly separated from one another. This separation occurs anteriorly to the frontoparietals on the dorsal borders of the nasal capsules. The long, tapering maxillary process of the nasal extends ventrolateral, and adjoins facial process of maxilla in adults in all four species. The crescent shape of the nasals is maintained by equal rates of ossification occurring anteromedially and anterolaterally, widening along the anteroposterior axes. Nasals do not articulate medially, or with the frontoparietals posteriorly or the sphenethmoid anteriorly. In later stages (stages 44, 45, 46 in all four species) the anteromedial tip of the nasal grows and stops abruptly. Nasals are not completely developed by stage 46 in all four species, but nasals overlap the palatines in all adult specimens and have a more lateral orientation. In adults, nasals do not overlap either the sphenethmoid or each other. ### Angulosplenials and dentaries The mandible comprises three paired bony elements: angulosplenial, dentary and mentomeckelian. The angulosplenials are dermal bones that occupy posterior and anterior regions of the lower jaw. They do not articulate with the dentaries and mentomeckelians. Osteogenesis is initiated along the center of the ventral side of transversely oriented Meckel’s cartilage. Ossification proceeds along the ventral and lingual sides. The angulosplenials approach the mandibular articulation posteriorly and invest the lingual surface anteriorly after completion of metamorphosis. In adults, the coronoid processes can be observed in the dorsomedial portion of the posterior end of angulosplenials. The dentary is a dermal, small, dentate, slender bone investing anterolateral and external surfaces of the lower jaw. The dentaries appear as long, thin ossifications associated with the straightening and fusion of the infrarostrals with Meckel’s cartilages. Initial ossifications appearing along the fused margin of the infrarostrals and Meckel’s cartilage proceeds posteriorly retaining a medial separation. The dentary fuses to the mentomeckelians after metamorphosis. ### Mentomeckelians The mentomeckelians are small, paired endochondral bones that occupy the anteromedial parts of the mandible. Their initial ossification appears on the ventral surface of the infrarostrals. In all four species, at stage 46, mentomeckelians remain cartilaginous at their medial tips, appearing neither fused nor articulated. ### Columella A long, laterally oriented columella appears between the fenestra ovalis of the exoccipital and the otic ramus of the squamosal. The proximal footplate and the basal part (pars media plectri) of the columella are faintly ossified (see – Tables for specific stages). However, ossification of the columella can be only seen in adults of *P*. *cruciger*. ### Palatine The palatines are paired, elongated, edentate, slim, and transversely oriented dermal bones that lie perpendicular to the longitudinal axis of the skull. They occupy the ventral side of the antorbital process, between the maxilla and the sphenethmoid, posterior to the paired vomers. Initial ossification is observed along the anterior margin of the planum antorbitale (see – Tables for specific stages). As growth proceeds, the palatines thicken and elongate. They are curved dorsally and extend while fusing with the sphenethmoid anteriorly (fusion of the palatines with the sphenethmoid is seen in adult specimens of *P*. *cruciger*, *T*. *eques* and *T*. *longinasus;* but observed at the stage 46 in *P*. *maculatus*) while terminating bluntly. ### Squamosal Dermal, paired squamosals invest the cartilaginous palatoquadrate. They are comprised of three rami: ventral, zygomatic and otic. The ventral ramus is the largest of the three. It appears as a small ossification on the anterior margin of the larval muscular process. During the metamorphic reorientation of the palatoquadrate, the squamosal moves from anterior to the orbit to a position lateral to the otic capsule. The final orientation is achieved by stage 44 in *T*. *eques*, stage 43 in *T*. *longinasus*, stage 42 in *P*. *cruciger* and stage 43 in *P*. *maculatus*. The squamosals articulate distally with the quadratojugal at stage 46 in both *Taruga* species, and postmetamorphically in both *Polypedates* species. The short, small zygomatic ramus, projects anterodorsally from the ventral shaft of the squamosal and articulates with facial process of maxilla. Otic ramus ossifies concomitantly with the zygomatic ramus at stage 46 in *T*. *eques* and *T*. *longinasus*, and in adults in *P*. *cruciger* and *P*. *maculatus*; the otic ramus has a slight posterodorsal orientation, lying laterally adjacent to the anterolateral corner of the cartilaginous crista parotica. The squamosal arms rapidly elongate to form triradiate T-shaped bones in adults. ### Pterygoid The pterygoids are one of the ventral components of the suspensorium. These robust, dermal, triradiate bones have anterior, posterior and medial rami. The initial ossifications are observed along the ventromedial surface of the articular process of the palatoquadrate. As larvae grow, ossifications continue along the anterior and posterior margins. Ossifications along the ventromedial surface of the cartilaginous pterygoid process give rise to the long anterior ramus, which terminates near the anterior margin of the orbit. The medial rami, which are much shorter in a dorsomedial direction, extend onto the ventrolateral margin of the otic capsule. The bones are triradiate after the three rami expand in both *Taruga* species at stage 46 and in adults in both *Polypedates*. ### Vomer The vomers form as tiny, paired ossification centers at the posteromedial corners of the internal nares. Prechoanal, postchoanal and anterior processes make up the lateral, posterior and anterior components of the vomers, respectively. The prechoanal process forms the anterior margin of the choana. The postchoanal processes appear after metamorphosis in all four species. They support the anteromedial margins of the choana. Initial ossifications underlying the nasal capsules expand rapidly lateral to the anterior tip of the parasphenoid. Vomerine teeth are seen in adults of *T*. *eques* and *P*. *cruciger*. However, vomerine teeth are not distinct in the stained adult specimens of *T*. *longinasus* and *P*. *maculatus*. ### Quadratojugal The quadratojugals are paired, slender, dermal bones. Laterally, the anterior ends of the quadratojugals overlap the posterior ends of maxillae, completing the maxillary arcade ventrally (at stage 46 in *T*. *eques* and in adults in the remaining three species). The posterior ends of the quadratojugal ossify on the ventrolateral surface of the articular process of the palatoquadrate. These ends articulate with the ventral ramus of the squamosals. Further bone development occurs by lengthening of the bones, where the quadratojugal, maxillae and squamosal are interconnected at metamorphosis. ### Sphenethemoid The sphenethmoid is an endochondral bone that contributes to the anterior part of the braincase, placed between the posterior margins of the nasal capsules and the anterior margins of the frontoparietals. The sphenethmoid originates as two centers of semi-lunar-shaped ossifications, in faint red color lateral to the anterior most part of the frontoparietals, at stage 46 in *P*. *maculatus* and in adults in *T*. *longinasus*, *P*. *cruciger and T*. *eques*. Subsequently, ossification proceeds dorsoventrally forming a deeply concave sphenethmoid anteriorly and posteriorly. Dorsally the bones are nearly covered by the frontoparietals and ventrally invested by the cultriform process of the parasphenoid (Figs). ## Development of the hyoid skeleton The four species possess well-developed hyobranchial skeletons by stage 44 with thin hyoid plates possessing pairs of hyales, anterolateral (“alary process”), posterolateral and posteromedial (“thyrohyal”) processes. Osteogensis of the posteromedial processes are initiated by stage 45, in all four species, along the center of the cartilaginous shaft and extends along the anteroposterior axis. Adults of *T*. *eques*, *T*. *longinasus*, *P*. *cruciger* and *P*. *maculatus* possess well-ossified posteromedial processes with cartilaginous epiphyses on the distal ends and blade-like anterolateral processes (Figs). ## Development of the axial skeleton The axial skeleton is composed of three regions: presacral, sacral and postsacral. There are eight presacral vertebrae (I–VII procoelous, VIII amphicoelous). The sacrum and sacral diapophyses make up the sacral region whereas the post-sacral region consists of the urochord and hypostyle. Each post-atlas vertebra consists of a cylindrical centrum (oval in a cross section), dorsal neural arch and three pairs of processes: prezygapophyses (at the anterior end), postzygapophyses (at the posterior end) and transverse processes (expanding laterally from the pedicels). The atlas lacks transverse processes and prezygapophyses in all four species and the articulation with the occipital condyles of the skull is via a pair of atlantal condyles. The ossification of neural arches begins in a single center (see – Tables for specific stages for the four species). This center enlarges into a vertical sheet of bone that later has three components including a lamina, pedical and a pair of lateral processes. The transverse processes appear as lateral ossifications in all species, later proceeding further laterally from the center of ossification while maintaining cartilaginous tips. The transverse processes of ΙΙΙ and ΙV are the most well developed of all four species. Completion of the ossification of transverse processes in all four species occurs at stage 41. The distal end of the sacral diapophyses articulates with the ilial shaft of the pelvic girdle. ## Development of the pectoral girdle Prior to ossification, each half of the pectoral girdle is composed of the cartilaginous primordia of the scapula, suprascapula and coracoid. The scapula ossifies along its longitudinal axis, expanding laterally and articulating with the suprascapula. Ossification of the suprascapula is limited to the anterior region; this gives rise to the cleithrum, which has a wider base adjacent to scapula, where it extends narrowly along the distal margin. The clavicle appears as a thin ossification along the anterior margin of the cartilaginous procoraccoid, whereas the coracoid, an endochondral bone, ossifies along the mid portion of the cartilaginous primordia of the coracoid. Ossification of these two bones is observed concurrently in all four species. The clavicle and coracoid extend laterally along their cartilaginous parts without merging or articulating with other bones. The epicoraccoid is a cartilaginous arch that adjoins the two halves of the pectoral girdle. It appears at stage 37 in *P*. *cruciger*, *P*. *maculatus* and *T*. *eques* and in stage 36 in *T*. *longinasus*. The cartilaginous bridge joins the two halves of the pectoral girdle at stage 41 in all four species. The omosternum and sternum form anteriorly and posteriorly to this cartilaginous bridge, respectively; these two bones are ossified at stage 46 in *T*. *eques*, *T*. *longinasus* and *P*. *macualtus*, whereas in *P*. *cruciger* ossification occurs at stage 45. In adults, the base of the omosternum is clearly forked in both species of *Polypedates* but not in *Taruga*. Furthermore, the cartilaginous distal ends differ considerably among the four species. The adult sternum (“metasternum”; Liem 1970) has a bony stylus and a cartilaginous distal end (forked in *P*. *cruciger*; Figs) ## Development of the forelimb The forelimb consists of the cartilaginous primordia of the humerus, radius, ulna, radiale and ulnare at stage 36 in *P*. *cruciger* and at stage 34 in *T*. *eques* and at stage 35 in *T*. *longinasus* and *P*. *cruciger*. Ossification begins in the center of the midline of the cartilaginous humerus primordium, and extends anteriorly and posteriorly. The radius and ulna ossify separately, later fusing proximally (stage 38 in all four species), distally and medially (stage 42 in all four species) to form the radioulna. Cartilaginous primordia of the proximal phalanges are present at stage 37 in *P*. *cruciger* and at stage 35 in *T*. *eques*, *T*. *longinasus* and *P*. *maculatus*. Ossifications of all the forelimb elements initiate along the mid portion and expand laterally. The phalangeal formula of all species is 2-2-3-3, where the proximal phalanges ossify first and the distal phalanges last. The ossification of the carpals can be observed only in adult specimens of *T*. *eques*. Distal phalanges are Y-shaped in all four species. Forelimb phalanges tend to ossify before those of hind limbs. ## Development of the pelvic girdle The ilium, ischium and pubis unite to form the pelvic girdle. The ilium articulates with the ventral surface of the well-expanded sacral diapophysis of the axial skeleton. As development progresses the iliac shaft begins to ossify along with the mid portion of the humerus in all four species. The ossifications proceed anteriorly and posteriorly along its longitudinal axis. ## Development of the hind limb In all four species, ossification of the hind and forelimbs start concurrently (– Tables). The femur, being the proximal element of the hind limb, articulates with the pelvic girdle. This bone along with tibia, fibula, fibulare, tibiale, metatarsals, tarsals and phalanges form the hind limbs. The tarsal region consists of the prehallux, element Y and tarsals 2–3. Proximal to the tarsal region, fibulare and tibiale can be seen, which are fused at its proximal and distal ends. The phalanges ossify also from proximal to distal. The final phalangeal formula of the hind limb is 2-2-3-4-3 in all species. The ossifications of the tarsals were not observed even at stage 46, and considered as postmetamorphic bones in all four species of *Taruga* and *Polypedates*. # Discussion This comparison of the skeletal morphology of the two closely related foam- nesting lineages highlights phylogenetically informative characters at two different levels: between species and between genera. Adult osteology and chondrocranial morphology of the rhacophorids have been examined in several studies; however, little information is available on the skeletal development. The four species differ in chondrocranial morphology. *Polypedates cruciger* and *P*. *maculatus* have larger chondrocrania than *Taruga eques* and *T*. *longinasus*. There are substantial differences between the genera in the shape of the anterolateral processes and presence/absence of posterolateral processes on the otic capsules. Furthermore, the inverted U shape of the suprarostral cartilage is considerably different in *P*. *cruciger* when compared with upright U-shaped suprarostrals in the other three species. This is due to a dorsomedial fusion of the central corpora *vs* ventromedial fusion of the suprarostrals. Ossification sequences do not vary significantly within a single species, but considerable interspecific variations are observed between the four foam nesters. Differences in the ossification sequence of the cranial bony elements in particular are seen towards the end of metamorphosis. For all four species, the parasphenoid, frontoparietals and exoccipitals are the first bones to ossify, coinciding with the other observations of metamorphic cranial ossification. Paired prootics are the fourth bony elements to form in *Taruga* and *Polypedates;* septomaxilla, maxilla, premaxilla, nasals, dentaries and angulosplenials follow next, showing slight variations in the order of their initial appearance (– Tables). Our results show that the major modifications of the chondrocrania occur early in *Taruga* (stages 40–41) when compared with *Polypedates* (stage 42). However, interpretation of metamorphic acceleration or delay is dependent on the ancestral ossification sequence, which cannot be inferred from the existing data. In anurans, it has been recorded that premaxilla, maxilla, septomaxilla and nasals ossify after the underlying nasal cartilages are formed, which is also true for these four species. Similarly, the squamosals appear before the metamorphic remodeling of the palatoquadrate in all four species; however, *T*. *eques* shows a comparatively slow ossification rate of this bone. However the order of the appearance of the bones, vomer, pterygoid, quadratojugal, palatine, sphenethmoid, and columella vary considerably among the studied four species (– Tables). Ossification of the maxillary arcade commences at the same time in all four species, where maxillae ossify first; however, premaxillae show a slow ossification rate in *P*. *cruciger* and *P*. *maculatus* compared to *T*. *eques* and *T*. *longinasus*. Regardless the differences in the ossification sequences, the four species studied are similar in the relative timing of the skeletal units; the very first bones are formed in the cranium, followed by the axial skeleton, and next, with a clear delay, the ossification of the forelimbs and hind limbs progress. Forelimb and hind limb start ossifying simultaneously in all species except in *T*. *longinasus*, where ossification of the forelimbs is initiated first. The osteogensis of the limbs begins only after the I–VIII neural arches are formed. In all four species, the postcranial skeleton is well developed prior to the completion of the ossification of the cranium. Interestingly, tarsals and carpals are only ossified in *Polypedates maculatus* by stage 46, the species with the widest distribution (even across India), whereas in the other three species, these bones were ossified only in adults. The adult osteology of the two genera also shows several conspicuous differences, helping outline the generic-level boundaries using osteological characters; some of these characters were also used by Liem to define family and generic-level boundaries of rhacophorids. Furthermore, Meegaskumbura *et al*. highlighted some cranial morphological characters (e.g., shape of the skull, frontoparietal, pterygoid, and orbit) to distinguish the two genera. In our study, we highlight two additional characters, which have not been discussed before, i.e., the posteriorly extending parieto-squamosal arch (present in *Polypedates vs*. absent in *Taruga*), and structures of the sternum (“metasternum”) and omosternum. The distinguishable, forked, distally dilated sternum of *P*. *cruciger* is conspicuous among the four; the forked base of the omosternum is broadly forked in *Polypedates*, whereas in *Taruga*, this is not evident. These results are comparable to published data of the direct developing *Pseudophilautus silus*―a related lineage of *Taruga* and *Polypedates*. The order of cranial bone formation in *Pseudophiluauts silus* is similar to the patterns found in metamorphic anurans rather than the unique sequences found in other well-studied direct developers (e.g., *Eleutherodactlyus coqui*). Direct development removes the need of larval specializations, which can permit developmental repatterning. However, in *P*. *silus* most of the larval specific characters are significantly reduced rather than entirely lost. Among the deviated characteristics seen in *P*. *silus*, the jaw suspension exhibits the greatest departure from the typical tadpole morphology in its modifications of the palatoquadrate cartilage; the palatoquadrate is present as an initial thin horizontal cartilage (later orienting in a vertical position), where the posterior end of the palatoquadrate is not connected to the neurocranium via ascending and otic processes. Both processes are present in *Taruga* and *Polypedates*. Initiation of bone formation occurs prior to hatching in this direct-developing species, unlike the foam nesters described here, where the ossification begins after hatching. The formation of the jaw bones (maxilla, premaxilla, dentary, angulosplenial) is accelerated in *P*. *silus*, possibly facilitating jaw-usage for active feeding in newly hatched froglets. Vomer, quadratojugal and palatine are absent in hatchlings of *P*. *silus* but are present in adults. However, these three bones are observed by stage 46 in *Taruga* and *Polypedates*. These conspicuous variations between the direct developers and foam nesters, which have been studied so far, indicate the extent of developmental changes associated with these life histories despite sharing a common gel-nesting (GN) ancestor. Characters such as the number of bones present at the end of metamorphosis, ossification sequence and adult cranial morphology are of systematic value, as they tend to consistently vary between species. Our study highlights the variation of these developmental features as they are analyzed in a phylogenetic context. # Supporting Information We wish to thank: James Hanken and Rohan Pethiyagoda for their comments to improve the manuscript; Carl Gans Award to GS for financial support to present this work at the World Herpetological Conference in China (2016); Gayan Bowatte for support during initial phase of the lab work; Nayana Wijayathilaka for field work; Department of Wildlife Conservation and Forest Department of Sri Lanka for research permits to MM. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** MM GS RK. **Data curation:** GS. **Formal analysis:** GS RK MM. **Funding acquisition:** GS MM. **Investigation:** GS. **Methodology:** GS RK MM. **Project administration:** MM GS. **Resources:** MM GS RK. **Validation:** GS. **Visualization:** GS MM. **Writing – original draft:** GS MM RK. **Writing – review & editing:** GS RK MM. [^3]: Current address: Department of Organismal Biology and Anatomy, University of Chicago, Chicago, Illinois, United States of America

# Introduction There is no doubt that there is rising enthusiasm among higher education institutions across nations of the globe to build the transnational higher education setting to meet developing global demands. As it is observed, the internationalization of higher education is experiencing a prominent changing scenario. Heterogeneous internationalization modes have resulted in altering workplace environments, various viewpoints, shifting rationales, theoretical and strategic management, as well as observational possibilities and challenges. Historically, the concept of internationalization of higher education began with academics travelling to academic centres around the world to study wherever they could. In recent decades, there has been a growing emphasis on the international components of higher education at institutional policies and international faculties in international higher education institutions. Furthermore, there is an early separation between market-driven interests in the recruitment of fee- paying overseas students and a cost recovery industry for a marginal activity aimed to support globalization in general. A rising number of practitioners perceive internationalization efforts as a strategy of boosting personal and professional growth as a means of changing the system. More subsequently, the word "internationalization" has developed and gained popularity as a fashionable worldwide phenomenon and has been linked to the United Nations’ Sustainable Development Goals. Transnational higher education has grown and attracted growing attention globally as globalisation has deepened in the second part of the twentieth century. China-foreign cooperative education, as a manifestation of transnational higher education in China, has grown significantly in recent years and is becoming a crucial subject in higher education research. King investigates China’s distinctive traits and particularities of internationalization challenges in international higher education, notably its fast expansion in international collaboration with African nations. From 71 in 1995 to 2238 in 2019, the number of China-Foreign Cooperative Education Programs between Chinese institutions and foreign universities has increased dramatically. Nevertheless, while the Chinese government supports internationalization of higher education as a means of improving domestic higher education’s global rankings, the surge of China-foreign cooperation programs has been accompanied by quality issues raised by the introduction of mediocre foreign institutions driven by profits, as well as a lack of well-established legislation and regulations, inner governance systems, and quality control systems. As just an outcome, in addition to the updating of legislation and regulation perspectives at the national level, initiatives undertaken by cooperation institution partners are the true engine for the actual running of cross-border education, ultimately promoting teaching quality and protecting the interests of all stakeholders and deserve more attention. Starting with a case study, this paper examines quality control in transnational cooperation from cultural and institutional governmentality perspectives, attempting to identify significant measures to ensure the long-term and potential promotion of teaching quality in transnationally cooperative education. The instance selected for this research is an applied tourism university in a world-renowned tourist location in China’s southwest (hereinafter GTUC). GTUC attempts to conduct a China-foreign education cooperation program to introduce the elite hospitality education resources of one University for hospitality education from Switzerland (hereinafter EHLS), which is one of the world’s first institutions of hospitality management and ranks top throughout this field worldwide, in order to meet the increasing needs of the local market for high- end international hospitality professionals. Since 2015, an independent teaching and training facility has been built as the foundation of the Faculty of GL (An International Hospitality Management School) (hence GTUC-GL, or GL for abbreviation), which is a GTUC school specifically designed to conduct this program and to be the centre of this research. GTUC has been approved by the Ministry of Education China to enrol bachelor students in this special hospitality management since 2017, after formally entering the certified collaboration with EHLS in 2016. The current study aims to demonstrate the importance of cultural appreciation and institutional governmentality in quality control in transnational education cooperation through this case study, and it proposes to expand the international influence and recognition of China-foreign education collaboration through quality international exchange and cooperation. Following a review of previous research, the paper examines the cultural appreciation reflected in GTUC-hospitality GL’s mindset, institutional governmentality, and quality control process in GL as a case study to discuss conceptual quality control in transnational higher education cooperation programs. # Literature review Before delving into the challenges posed by the globalization of higher education, we would want to clarify the definitions of important terminology used in this study. The international curriculum, as defined by the OECD’s Centre for Educational Research and Innovation (CERI), is "an international orientation in content, aimed at preparing students for performing (professionally/socially) in an international and multicultural context, and designed for both domestic and foreign students" (p9). In light of this, the following aspects will be investigated in this section of the literature as part of the present article: what are the theoretical foundations of internationalization and transnational higher education; what is currently known in the field; and what are the existing gaps throughout the concepts? In order to provide a more comprehensive response to these concerns, it has been proposed that we investigate topics such as the internationalization of higher education and the quality control challenges that arise throughout the process of internationalizing higher education. The dominant understanding of internationalization of higher education in China adheres to this definition, which, in a nutshell, comprises of the internalization of student sources as well as the globalization of curriculum and administration. ## The internationalization of higher education The importance of internationalization for higher education is self-evident, even though these themes have remained insufficiently underestimated in a Chinese context. Guided by several disciplines including anthropology, language and communication, business and marketing, futurist studies, strategic leadership and pedagogy, internationalization challenges have emerged as a top priority for foreign universities all around the globe. Chinese colleges and universities are not bystanders to these global trends. This is in part a reaction to the changing global environment and cultural surrounds, but it is also a response to the globalisation shift itself, which, pushed to its logical conclusion, is a bottom-line development plan for any ambitious international institution. Adopting a coupling coordination model, Geng and Zhao conduct a regional and temporal examination of the link between the characteristics of sustainable higher education development and the coupling coordination relationship. In addition, they present a number of concrete and actionable recommendations for ensuring the continued growth of the higher education sector. Many aspects of contemporary globalization make it necessary for institutions to modify and define the concept in accordance with their own goals; as a result, prevailing conceptions of the meta-discipline of globalization in discipline have become increasingly complex as a result of such development. To a certain extent, this is especially obvious in countries where institutional internationalization is still in its early phases and where traditional Western ideas of internationalization must be studied further for their relevance in local situations. In these countries, internationalization of institutions is still in its early phases. Our comprehension of the benefits and constraints associated with internationalization practice in China will be aided by the development and use of this concept in the context of such unique conditions. In addition, it is of equal significance to guarantee the high quality of such international educational practices concurrently with the process of globalization of education. Following this part, quality control in the internationalization process of higher education is then further explained in the next section, along with a review of relevant literature. ## Quality control in the internationalization process of higher education With the commitment to enhance the world ranking of domestic higher education, transnational higher education is favoured by the Chinese government. This denotes all types of higher education study where the learners are located in a country different from the one where the awarding institution is based. The term is recently interchangeably used with "cross-border education", "offshore education" and "borderless education" in related research. Transnational higher education may be conducted in different forms, such as franchise, twinning, double/joint degree, articulation, validation or virtual/distance. However, borderless education that neglects the existing borders in the delivery of transnational programs, are not included in this study. The employment of quality control in higher education has been a global practice under the background of globalisation and internationalization. The integration of global markets puts modern countries under huge pressure to maintain or promote national competitiveness facing overwhelming challenges. As the knowledge base for developing potential talents, higher education has been confronting the same situation. Initiated by the state, multiple measures are taken to elevate the world ranking of domestic universities, including the quality control system, to finally enhance the efficiency and effectiveness of higher education performance. Generated from the manufacturing sector, the concept of quality refers historically to consumer satisfaction, and synthesis of conformance, adaptability, innovation and continuous improvement. On this basis, higher education quality is usually examined via exquisite standards, consistency with standards, adequacy of purpose, effectiveness in achieving institutional goals, and meeting stakeholders’ explicit or implicit needs. Since the 1980s, a series of internal reforms have reshaped the development of higher education in China, introducing privatization and marketization to accelerate the massification of education programs, which grants universities more autonomy and flexibility in university-level governance. Quality control has become a critical issue for the orderly, healthy, and sustainable growth of China-foreign cooperative education as a result of the different disorderly circumstances and quality concerns that have developed since its emergence. The majority of research have concentrated on the macro- level of government regulation and specialized quality control methods, although attention has also been made to quality control specific challenges. Because transnational higher education transcends national boundaries and surpasses the realm of regulation in a single country, the problem of quality control is much more difficult than quality control in a single country. This research examines the quality control of Chinese-foreign cooperative education from institutional and cultural perspectives in order to identify the underlying reasons of quality control issues. As a consequence, in the endeavour to ensure education quality, underneath the macro efforts undertaken at the national or cross-national level, the establishment of a well-established quality control regime at the micro level of school partners demands much more academic attention and investigative effort. As the governing body in close touch with professors and students, whether a school has a solid governance structure in place directly influences the students’ community. The Chinese Ministry of Education’s revisions of the quality control system in higher education are primarily concerned with numbers pertaining to an educational institution’s physical foundation. A well-designed campus, on the other hand, provides more than just effective and efficient educational results. Quality control can be effectively achieved only when all stakeholders’ interests are considered and all stakeholders can actively participate in the governance system, and when the framework of institutional governmentality is well established and can function in promoting quality control in higher education. Using these notions, this research seeks to use GTUC as an example by performing a case study of the quality control system from the viewpoints of institutional governmentality creation and cultural appreciation in transnational higher education. Finally, this study emphasizes the significance of high-quality international exchange and cooperation in growing China’s worldwide impact and recognition of foreign education partnership with China. # Research site and research methodology The research methodology adopted in this study is taking a participatory action research (PAR) approach (See below) and a case study method, both of which are iconic approaches in qualitative research methodology. PAR is a rapidly growing research approach in education research, thus, a proper method for educational scholars to investigate a co-developing education program with stakeholders. As an important form of qualitative research, PAR through focus group interviews is an unstructured, direct, one-to-many group interview. Focus group interviews are conducted by investigators with advanced interviewing skills to reveal underlying motivations, attitudes and feelings about an issue, and are most often used in exploratory surveys, to build up subjectivity via gathering reliable pieces of evidence. By gaining a detailed understanding of complex behaviour and exploring sensitive topics, the authors conducted detailed interviews with the faculty management team, frontline staff, students and many other stakeholders of the faculty at GTUC. The main function of the interviews was to obtain rich and vivid qualitative information from which to generalise and draw conclusions through the researcher’s subjective and insightful analysis. As for the case study, this is a major methodological approach in sociological inquiry. A third approach combined in this research is documentary review. This research used documentary analysis to achieve its goals. The promise of this technique is built on the measurement and comprehensive evaluation of existing records from the institution. The methodology of this research will allow for a meaningful evaluation of the school’s current document on limited circumstances. The research technique is based on an assessment of current announcements, disclosures, notifications, and running profiles of this institution, as well as an investigation of several internal regulatory recordings. The many actions carried out throughout the study’s development will allow the formation of landscapes with nodes showing performance, implementation of conceptual subthemes, and development linked to its internal management and quality control approaches. The above-mentioned research methodologies are scientifically valid and applicable when applied to the case of Chinese-foreign collaborative education in China. The investigation was carried out voluntarily by interview participants during an initial period of about 12 months, from January 2020 to December 2021, with the agreement of the institution and the written consent of all interviewers. For the Ethics Committee of the author’s institution has accepted and fully supported the current examination of the research objects. All participants are encouraged to read the following statement in accordance with the qualitative inquiry: *I consent to participate in this qualitative understanding to transnational collaboration project in the current faculty*. *I concur that the response I made in the following interview to be used for academic research purposes by researchers of this scientific interpretation*. An international literature review, institutional document and policy review, and meetings with university administrators, program and course leaders, coordinators, and professional development lecturers to develop an interdisciplinary, cross- institutional case study of an internationalised curriculum are among the data resources for further analysis. The overall number of respondents in this study was 64, which included university officials, the dean of the Faculty of GL, faculty members, GL students, and a broader spectrum of Faculty of GL stakeholders. In this regard, a detailed technique or strategy for gaining an explanation of the objective and analysing the effect of modifications to that mechanism was offered by the stakeholder approach. This was achieved by determining who the major players or stakeholders in the system were and analysing the functions that they play in the mechanism from a strategic perspective. A definition of stakeholder that is widely recognized and accepted categorizes as stakeholder any group or individual that has the potential to influence or is affected by a development and/or the achievement of the goals of an organization. This definition applies to both the development itself and the goals themselves. The authors mediated the focus group interviews, and the interview procedures were recorded by research assistants. Based on current research on internationalization, cultural appreciation, and quality control in collaborative international higher education cooperation, several key open-ended issues were proposed. These prepared questions maintained the flow of discussion topics and regulated the focal scope of the major subject of this research endeavour, which focused on (1) How academics working in various institutional and disciplinary settings define the idea of curricular internationalization? (2) How does the school include faculty in the process of internationalizing the curriculum in academic practices? (3) How is the quality control system developed and monitored during the implementation process? Discussions were also encouraged throughout the question and answer segment to delve into the latent meanings behind the participants’ replies. The respondents were given the freedom to disclose their actual feelings, which was made possible by the researchers’ support. The researchers attempted to provide a comfortable and free interactive environment for the respondents, in order to encourage deeper and more free communicative conversations. Three rounds of focus group interviews were held in the same location to ensure dependability and credibility in qualitative educational research methods. Each interview lasted around 40 minutes to an hour. The researcher audio-recorded the interviews with the permission of the participants and subsequently converted them to text using technological techniques. The three interviews’ texts were combined. After acquiring the entire texts, the writers double-checked with the interviews in cases where they were dubious about the actual meaning of the interviewees. The interview inaccuracies were subsequently addressed. Text analysis of this research project followed a further systematic investigation. The original material was then transcribed as the first phase. Following each interview, the tapes were meticulously transcribed, and the source material was thoroughly examined and analysed. The second stage was to code. The current study’s information was reorganized and classified, using key themes or thematic emphasizing categories. The coding results were displayed as text information and analysed using theme categories. The language of the interviews was used to deliver the material, which helped the researchers comprehend the feelings and perspectives of the interviewees in the situation. Finally, the theme categories were compiled in text format and sent to the interviewers for a second round of validation. As a result, the authors created the graphic below to expound on the contributing elements linked to the internationalization problem in this particular situation in China. # Research findings Accompanied by the synthesizing of precedential literature, an overall investigation of the texts has shown the research framework for a transnational higher education collaboration in the case of GTUC-GL, as shown in previous figure. Three identifiers have been conceptualized: (1) mutual cultural appreciation; (2) institutional governmentality; and (3) quality control. Such triangle identifiers have been grounded in the focal case in this research, which is popularly cited in previous prestige pieces of literature of transnational education collaboration, for instance, culture issues, institutional structure, as well as the systematic process for high profile quality control. The research findings for each of the identifiers is then further shown and elaborated in detail in the following sections. Part of the verbatim quotes, which are greatly condensed into theoretical debates and discursive summaries, serve as testimony to corroborate the previously specified identifiers, as seen in above. ## Cultural appreciation in the transnational educational collaboration process of internationalization of curriculum This section analyses the hospitality mindsets from the perspective of cultural appreciation. The internationalization of the curriculum sits at the intersection of university policy and practice and is a source of fascination and achievement for students, academic staff and university administrators. Stakeholders in the process of internationalization must be equipped with strong cultural appreciation, accepting cultural diversity, equity and inclusion. Under the guidance of GTUC’s internationalized education planning, with the combination of the needs of China’s education and GTUC’s features, GTUC-GL integrates its mission, vision and distils them down to "GLers" as its core values, which consist of *Graceful Generalist*, *Life-Long Learner*, *Responsible Rudder and Sincere Socialist*. To integrate these core values into school life, GTUC-GL starts from the following aspects to develop students into qualified high-end hospitality talents in the future. To date, the GL teaching facility is well-formed in its lobby, guest room, banquet hall, and many other areas, and has produced a fully functioning hospitality teaching environment, complete with hotel service chain settings. The lobby is surrounded by a green planting space, a small sports field, and a sofa lounge area, providing the faculty teaching community with an open view and a pleasant ambiance in their spare time. Integrated cultures from both Switzerland and China have infiltrated the staff and students’ imaginations. Such impressions were instilled in teachers and students during their initial visits to the EHLS campus, and therefore from the time they were recruited to the institution. These accomplishments did not come easily or quickly. Interviewees who were present at the initial establishment of the combined higher education partnership revealed their opinions regarding the challenges of creating an extremely international-style teaching and learning environment. The same views have been instilled in bachelor students. The academic staff and students were to embrace the hospitality traditions of both Switzerland and China. GTUC-GL’s philosophy is represented in the software and hardware infrastructure, as well as the services given to its teachers and students. A genuine hotel was built to house a cooperative education program, giving a physical foundation for establishing the EHLS teaching paradigm and delivering a learning experience via practical training in actual hotel roles, particularly from the time freshmen enter this teaching structure. This authentic hotel was built to host this co-operative education program. It gives students the ability to gather learning experiences via practical training in genuine hotel positions and serves as the physical foundation for establishing the EHL teaching paradigm. This option is particularly open to students who are just entering this educational system. This sentiment was echoed over the course of the interviews, with interviewees expressing the perspective outlined below. > *I believe the most crucial aspect is the ladder concept of > professional training programme*. *The curriculum is organized in a > spiral progression of "practice → theory → practice → theory"*. *We > offer students with a totally realistic operational environment in > which they may study the essential professional courses*, *allowing > them to swiftly grasp the relevant information and abilities and fully > realize the synchronization between professional education and > industry demands*. (G1, Professor, Head of School, one of the main > directors of the School, interview in December 2021). Adopting an EHLS educational horizon, GTUC has built an independent hotel as an academic building, providing an infrastructure basis for students’ immersive learning. GTUC-GL has a total surface area of 21,858 square meters, which is made up of 11 multi-media classrooms and 13 practical training classrooms that enable students to engage directly in the operations of a real hotel, flawlessly blending professional experience with theoretical school learning. Students begin their first semester in the faculty by cycling through 14 mock-up roles in a genuine hotel environment. Further, to keep the core culture value deeply rooted in every staff and student, for instance, GTUC-GL designs its school badge, makes signage of its mission, vision and core values. The faculty incorporate these cultural symbols in the faculty teaching building. GTUC-GL absorbs EHLS’s talent for developing a routine of "Practice–Theory—Practice Again—Theory Again", to strengthen features of applied hospitality talents development, which are also termed as an experiential learning model. The GTUC- GL, which is based on hotel facilities, draws on EHLS’s successful experience in conducting practical training to divide must-have hospitality service skills into 14 positions and design related practical training courses, achieving a high fusion of teaching and operation, theory and practice. The GTUC-GL produces an official document to incorporate hospitality professional standards into university life. The GTUC-GL professors and students adhere to a formal dress code of hospitality professionals that corresponds to their curriculum, demonstrating the GLers’ positive professionalism. The School also displays dress code and international business procedure signs, tacitly inspiring its pupils to be industry talents. depicts cultural appreciation as a dynamic interaction process between EHLS and GTUC at the institutional level. Such mutual structural appreciation suggests that active and favourable interconnectivity between different university cultures, as Wu and Pullman note, enhance the credibility and are beneficial to the organization of joint international higher education collaboration. > *The special features for this unique school are these followings*: > *the school’s financial and policy support*, *the calibre of the > international institutions with which the school has partnered*, *the > employment of foreign specialists*, *and the educational experiences > of the school’s management and faculty members gained during their > time spent studying abroad*. *This function is contingent on the > applied talent development model that the GTUC uses as well as the > launch of the operations integration program that EHLS has > implemented*. *This characteristic is in line with the objective of > teaching practical skills*, *and it may be used to highlight the > school’s training of applied talent features*. (In December 2021, an > interview was conducted with G5, Professor, who was one of the > supervisors at the school who was responsible for the formulation of > the curriculum) To have a more in-depth discussion of the internationalization of higher education, researchers must include the internationalization of the curriculum, which includes international-style teaching and outcome-oriented student learning. Yet, the internationalization of the curriculum is little known as a concept and is evolving in practice. Planning to reach the goal of internationalised learning, it must be done in the context of integrated cultures and knowledge spheres, diversified behaviour change and practices of full engagement within a transdisciplinary approach. However, if academic staff do not have sufficient experience, skills or knowledge required to internationalise the curriculum, they are, thus, incapable of sufficiently engaging with fully adopting the internationalization of the curriculum. This is serving as competitive implications for universities’ international strategies and student learning. The intersections between disciplines, curricula, internationalization and student learning in higher education form a connected space and a glocalized curriculum that offers rich culture experiencing opportunities for students and faculty members. Participants in the interview offered the following examples to illustrate how the curriculum may be made more international: > *This course of study has the intestinal fortitude to break away from > the normal educational paradigm and makes an effort to integrate > Eastern and Western methods of instruction*. *Working with EHLS*, *an > institution that is recognized as a leader in the hotel management > industry*, *provides a one-of- a-kind opportunity to learn from one > another*. *For this reason*, *we are enthusiastic about the progress > that is being made with the program*. *Second*, *the educational > technique of the program is a good match with what I’ve learned from > my own experience studying in other countries*. *I have high > expectations for the project and look forward to acquiring the most > cutting- edge knowledge from both the Western and Chinese educational > systems as well as improving my general competence*. *I want to be > able to lead the program in such a way that it produces students who > have a more global perspective and a more holistic approach*, *in > addition to individuals who have professional skill and moral > integrity*. (Interview conducted in June 2021 with G4, Professor, who > is largely responsible for teaching courses at the school) Cultural factors also need to be taken into examination in the quality control system. Mutual cultural appreciation in social sciences usually refers to "the dependence of a phenomenon … in institutional, social, cognitive or cultural terms". The faculty members of hospitality management cultivate students’ comprehensive competencies through all kinds of cultural- and hospitality- related activities. Besides organizing competitions of professional skills such as the "GL Cup" Sommelier Services Competition and Competition of Creative Table Napkin Folding, GTUC-GL selects outstanding students to participate in professional competitions at the national and provincial level and Young Hoteliers Summit, to transfer students from inexperienced youngsters to potential professionals with strong cooperation social responsibility mindsets. Here, cultural appreciation is innovatively adopted, to refer to the integration of different cultures reflected in transnational higher education programs. The school culture where all stakeholders are surrounded is shaped through the combination of features of two national cultures and that of institution partners. The vision, mission and strategy are further adopted and applied to find the origin of the cultural appreciation. By absorbing the exceptional characteristics of each side, the entity operating transnational cooperation programs can share with the whole community the ideal framework of education outcomes, which will lead to the enhancement of education quality. For faculty members, they are both hotel operators and learning facilitators, passing on the spirit of dedication, professionalism and enthusiasm to students through their proper behaviours. These student contests are some of the methodologies for students to to acquire cultural perspectives. In hotel management education, the GTUC-GL faculty provides accessible services for workers’ job and life, as well as a focus on students’ physical practices and mental growth. ## Research finding 2: An institutional governmentality reflected in the faculty Institutional governmentality is a conceptual approach to public service administration study. In a similar vein, the next part employs the concept of institutional governmentality to further examine the faculty’s administrative organization. Using this method, this section explores the complexities of the GTUC-GL faculty’s internal institutional structure, similar to what Baxstrom notes, to exemplify the structural planning and operational multiplicity. To this extent, GTUC has set up a fully functional institution to implement the original aims and scopes of the school (See below). GTUC has formed a Worldwide Advisory Committee comprised of distinguished international hospitality education and industry specialists from Hong Kong Polytechnic University, the University of Surrey (UK), the University of Houston (USA), Sun Yatsen University (PRC), and the Banyan Tree Group. Each year, committee members convene to exchange ideas and provide constructive proposals in response to GTUC’s difficulties and current issues, in order to assist GTUC review its growth path and goals. The GTUC-GL management team, as a GTUC school, actively integrates these yearly proposals into the innovation of its talent training and operating direction. Respondents showed a high level of satisfaction and awareness when one respondent made comments about the concept of institutional governance. > *When I was teaching in such an environment*, *I was able to feel the > support of the institution*, *the support of the government*, *the > good participation from the foreign side*, *the nice atmosphere of > several languages being used*, *and so on*. *This endeavour has > garnered a great deal of support from a number of departments > throughout the institution*. *For instance*, *the department that is > in charge of international cooperation has assigned a member of their > staff to oversee the abroad component of the project*. *In terms of > the English language*, *the institution places a strong emphasis on > providing students with extensive active pre-entry English language > preparation*, *as well as expanded language classes and other related > activities*. *The standard living and learning environment of the > students contributes positively to the students’ use of the English > language*. *These initiatives have made a substantial contribution*, > *collectively*, *to the internationalization of the institution*. (E5, > teacher and teaching assistant, responsible for teaching courses in > second languages at the school who have experience studying abroad, > interview took place in June 2021) Furthermore, GTUC and EHLS have formed a China-foreign Cooperation Program Management Committee comprised of nine members from both sides, led by GTUC’s president. The Committee meets on a regular basis to summarize and report on the school’s operations, as well as to consider the school’s future direction and action plan. Student responses complimented the work of the Committee and the School’s teaching management team, with one student remarking, > *For student management*, *we have midterm and final exams*, *as well > as feedback on instructors’ teaching every weekend*. *Regarding > instructor management*, *the school employs stringent instructional > monitoring*, *biannual academic assessments*, *and periodic Swiss > visits*. *This curriculum deviates from standard higher education in > that it combines theory and practice in its instructional technique*. > *In accordance with this arrangement*, *we complete two industrial > internships over the course of two semesters in our second and fourth > years*, *and we have cycled through over a dozen break-related > responsibilities in our first year to provide culinary and lodging > services for students*, *teachers and real visitors*. *Through this > extensive practical industry training*, *we gained a comprehensive > understanding of our suitability for the business*, *as opposed to > entering our senior year knowing our professional goals*.(Interview > with T5 senior in this program as of December 2021) To guarantee teaching quality, the GTUC-GL has formed a teaching management team comprised of the vice president in charge of teaching, the director of each faculty team, and the teaching secretary. The school has a Party branch with a secretary, a deputy secretary, and an organizer to carry out student management in accordance with GTUC regulations, assist students in resolving study and life challenges, and collect and provide feedback on students’ views. GTUC-GL analyzes teaching and learning at the conclusion of each semester to ensure that the measures are fully implemented. The school monitors instruction via reciprocal assessment between instructors and students, as well as random review by professionals. The teaching secretary compiles the assessment data and communicates them to school authorities, detailing the faults that were identified. Next a meeting of all faculty members, the school adopts and implements appropriate actions to alter the curriculum for the following academic year. ## Research finding 3: Improvement and progress in teaching quality control This research recognized current signals of development as a quality control system, immersive learning environment, customized curriculum design, and hands- on learning chances as hotel intern managers. The Teaching Quality Monitoring and Assessment Centre at GTUC is in charge of monitoring teaching activities, analysing monitoring data, and offering comments. The key revision procedures for the curriculum and instructional materials at GTUC-GL are shown in the figure below (see below). The GTUC-GL offers students an immersive learning environment in the form of a real-life mock-up hotel. An independent hotel is built to provide a genuine operating environment in order to achieve integration of education and operation. To meet the learning goals throughout the teaching process, a four- in-one model has been developed: (1) the integration of Chinese and Swiss education in one faculty; (2) the educational school teaching and a training business in one location; (3) being instructors or students as well as workers in one location; and (4) employees and consumers in one school (for faculty members and students are consuming in the teaching building). 13 distinct experimental training courses are built up to integrate professional competences with the curriculum, establishing a high degree of integration of theory and practice, as well as teaching and operation in daily routine, based on the must- have hospitality services and operation capabilities. This characteristic is clear from the respondents’ comments from one of the students interviewed. > *This curriculum provides an advanced international teaching approach > with a high degree of practice and theory integration*, *with one year > of practice and three years of theory in four years*, *as well as two > genuine industrial internships alongside the theoretical courses*, > *according to my personal experience*. *At the same time*, *we are > obliged to enter the building in operational condition*. *In other > words*, *a formal professional appearance is required*. *We have four > years of experience and a very professional mindset with this > operation before we graduate*. *As students of the program*, *we think > we are extremely competitive in the business and possess the diverse > knowledge and professional skills necessary for success in the > hospitality sector*. (T7, a sophomore in the program who was still in > her first industrial internship at the time of the interview in June > 2021, was interviewed) Second, for training and hospitality professionals, a spiralling pattern for nurturing talent known as the Practice & Theory Progression model of curriculum design has been established. The curriculum is then well-structured based on a two-time spiral of Practice-Theory-Practice-Theory rotation. This innovative teaching method received a medal for the Guangxi Zhuang Autonomous Region’s educational excellence. As one of the primary student activities, the GTUC-GL creates a multidimensional and comprehensive training platform comprised of Innovative Business Start-ups. Innovative business activities and start-up contests have evolved into training opportunities for students to improve their overall competencies. To some degree, these kids establish a cross-border, multidisciplinary, and cross-grade multiple collaborative innovation and entrepreneurship. Business project training techniques, shown as teacher-student co-working styles, are designed to develop students’ potential for invention and entrepreneurship. Educators encourage and support students’ inventiveness and entrepreneurial potential when they participate in contests involving business ventures. Similar sentiments were expressed by students during interviews: > *We host several cultural activities to encourage cultural > integration*, *such as an annual food service etiquette competition > and a wine label design competition*. *In addition*, *we provide a > variety of student events*, *such as the school’s unique Friday Night > and our student-designed ’520 Special Food Service’ event*, *which > entails preparing special coffee and cocktails*. *These activities are > an excellent means of fostering cultural integration*, *motivation*, > *and an atmosphere for learning*. (T6, current university student in > the China-foreign collaboration program, third year, interviewed at > the beginning of the school year in September 2021) Finally, the GL institution selects students each year to participate in the "Internet+" Innovation and Entrepreneurship Competition and establishes various awards to encourage students to actively participate in such activities and improve their innovation and entrepreneurial skills, organizational and coordination skills, and so on. Students might apply for roles such as Intern Manager to work as a manager at a real hotel. Students are participants in hotel operations as well as learning actors, embracing the spirit of hospitality and enhancing their overall skill as hoteliers via immersive education. GTUC-GL has professional employees to provide all-around high-quality education and services. GTUC-GL presents EHLS’s education model, incorporates GTUC’s applied undergraduate program features, and offers a curriculum that blends theory and practice to prepare students to become professional hoteliers. The School also establishes scholarships and bursaries to recognize excellent students, assist needy students in overcoming financial difficulties, and give all-around services to enhance students’ campus life. # Discussion With the fast expansion of Chinese-foreign cooperative education, it is critical to establish and continually enhance a quality control system for Chinese- foreign cooperative education from institutional and cultural perspectives. Three identifiers are described in the following sections based on an examination of the interview data obtained. To get a more in-depth understanding of the data presented above, this study will continue to analyse these three identifiers in order to investigate strategies to enhance the worldwide recognition of Chinese education. Furthermore, this debate aims to represent the GTUC-GL scenario of the quality control system of this transnational higher education partnership between China and Switzerland. ## Quality control in transnational higher education collaboration benchmarked by institutional constructions Quality control must be one of the most important parts of the internationalization process to the cooperative international higher education partnership. Due to the unique characteristics of international cooperation, we should not simply apply the traditional mode of quality assessment of domestic higher education, but should actively learn from the common practice of international quality control of transnational higher education, actively investigate the effective mechanism of appropriate separation of management, administration, and evaluation, and strengthen cooperation with quality control institutions and organizations of transnational higher education. To improve and enhance the quality of China-foreign cooperative education, the government should strengthen cooperation with quality control agencies and organizations of transnational higher education exporters, as well as build an external quality assessment system with Chinese characteristics and in accordance with international standards. Accreditation is often recognized as the most prevalent instrument for managing access to and quality control in the transnational higher education market. The variety of Chinese and international relationships should be reflected in accreditation. Because programs and institutions are different and unique, it is not possible to adopt a single methodology to accredit and evaluate them all. It is critical to present the various types of Chinese-foreign cooperative education, as well as the various needs and standards of Chinese-foreign cooperative education, so that the interests of Chinese-foreign cooperative education providers, students, and teachers can be better reflected through industry cooperation. ## Quality control in transnational higher education collaboration benchmarked by an effective management system To fairly satisfy the criteria of the original purposes of its initial foundation, an efficient management system should be established for developing a co-operative education institution. Effective quality control and risk management methods may be properly built to govern their educational practices by effective communication and trust between the cooperating two parties, as well as honouring their separate worldwide reputations. Promoting institutional internationalization would therefore be accomplished by protecting their individual credentials and co-competed course planning quality, as well as attaining mental agreement based on structural disparities. In terms of quality control management models, the shift from management to governance should be accomplished gradually. It is critical to specify the responsibilities of the government, school operators, and social intermediate organizations throughout the quality control process. Each stakeholder in a China-foreign cooperation project should clarify the respective function by forming a synergy of corporate social responsibility and coordination by constructing a scientific, rational, and effective quality control system. The national governmental oversight authority should concentrate on the aims, motives, and goals of Chinese-foreign cooperative education. The government’s quality control obligations at the local level are mostly tied to the execution of national policies and regulations. The institution is the primary topic of quality control since it is the primary supplier of education, and its quality risk control is based on the degree of satisfaction of students with the educational materials they get and the educational delivery method. From a cultural standpoint, the necessary core components for a good transnational higher education partnership are the acknowledgment of heterogeneous knowledge and reciprocal acceptance of cultural appreciation. In accordance with this cultural viewpoint, the pursuit of complementary capabilities and interests should be emphasized in the provision of cultural integration to both local and foreign cultures, as expressed in policy decisions and the quality control process. In terms of academic quality, it is critical to stress the comparison of transnational education programs with national programs. In the course of the growth of China-foreign cooperative education, several quality criteria have been devised., such as the *Guidelines for the Evaluation of China-foreign Cooperative Education (for trial implementation)* and the *Guidelines for the Selection of Exemplary China- foreign Cooperative Education Projects (for trial implementation)*. Despite the fact that these standards are still in the prototype level, they have a significant impact on the cultural attitude of the faculties and students as a whole. To further enrich and strengthen quality control standards, a successful international exchange and cooperation project would be able to expand and secure the healthy growth of China-foreign cooperative education. # Conclusion The present study analyses concern of internationalization, cultural appreciation, and institutional governmentality for quality control in the instance of GTUC-GL as a transnational higher education partnership by employing participatory action research and a case study methodology. GTUC-GL presents its own applicable GL model of developing hospitality skills in collaboration with EHLS, in order to build a highlighted talent-cultivating culture. The GTUC-GL is devoted to be a leading institution in discovering and disseminating applied talent training methodologies, a cultivator of future high-end industrial talents, and a catalyst for school-enterprise collaboration. The program’s goal is to produce high-quality graduates while also offering a plethora of experience opportunities for the development of a distinguished undergraduate program. The findings of this work give valuable insights for tourism and hospitality researchers and faculty members in the institutional structure, as well as career counselling for scholars interested in Chinese international partnership for higher education. The study adds to current academics not only via its conceptual treatment of cultural concerns, governmentality, and quality control in this arena, but also through its application in practice. According to the findings, the quality control system should identify the relative functions and responsibilities of all stakeholders, transition from management to institutional governmentality, and progressively strengthen quality control methods. Countermeasures such as enhancing and increasing quality standards in its mutual cultural appreciation from both sides must also be incorporated. Because of the current constraints, the results and comments should be interpreted with future discretions. To begin, this research solely examines into one example through two collaborated institutions. Future study should investigate these research themes in broader, more substantial cases, using more diverse research approaches, such as a quantitative approach, or with a broader spectrum of stakeholders. Second, although this study adds to the literature by establishing a research framework and approach to the issue via three conceptual notions, it is advised that future studies compare these conceptualizations with other forms of philosophical thought Future research may help to enhance our knowledge of how to build a high-quality transnational higher education cooperation throughout the globe, as well as discover a more fundamental philosophic conceptualization on this education collaboration issue. 10.1371/journal.pone.0274989.r001 Decision Letter 0 Tarrósy István Academic Editor 2022 István Tarrósy This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 31 May 2022 PONE-D-22-11647Internationalization, cultural embeddedness and institutional governmentality for quality control in transnational higher education cooperation: An empirical assessmentPLOS ONE Dear Dr. Zhu, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Your choice of topic and research demonstrate a firm commitment to contribute to the better understanding of TNE in China. It is an important project, which may offer a fresh insight into the major items on your agenda, i.e. internationalization, cultural embeddedness and the question of quality control at institutional level. Although Reviewer 2 was suggesting a full green light, I agree more with Reviewer 1 and the detailed criticism provided over there. Please, revise your manuscript thoroughly in particular with regard to your theoretical framework, research questions and empirical work. As for the literatiure review, consult more recent pieces published in the relevant fields, also possibly with a wider look at internationalization and the geopolitics of higher education. Reviewer 2 raises several serious problems about your findings and how you interpret your data. Please submit your revised manuscript by Jul 15 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, István Tarrósy, PhD Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please provide additional details regarding participant consent. In the ethics statement in the Methods and online submission information, please ensure that you have specified what type you obtained (for instance, written or verbal, and if verbal, how it was documented and witnessed). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information. 3\. Please ensure that you include a title page within your main document. We do appreciate that you have a title page document uploaded as a separate file, however, as per our author guidelines (<http://journals.plos.org/plosone/s/submission-guidelines#loc-title-page>) we do require this to be part of the manuscript file itself and not uploaded separately. Could you therefore please include the title page into the beginning of your manuscript file itself, listing all authors and affiliations. 4\. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match.  When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section. 5\. Your ethics statement should only appear in the Methods section of your manuscript. If your ethics statement is written in any section besides the Methods, please delete it from any other section.  \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: N/A Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: There is no doubt that there is growing interest among higher education institutions across countries of the world to develop the transnational higher education setting to reflect evolving global needs. The current study – “Internationalization, cultural embeddedness and institutional governmentality for quality control in transnational higher education cooperation: An empirical assessment” is very important to the development of TNE in China. While I commend the authors for the time and effort invested in the study, I think that there are additional work that need to be done in all the sections in order to bring the research to the journal’s standards. First, there is a lack of a clear theoretical underpinning of the study. I do not think that the authors’ discussions on internationalization and transnational higher education provide adequate information on the theoretical foundation. Secondly, although the introduction sets a very good pace for the delivery of very informative discussion on TNE and what the authors term as cultural embeddedness and institutional governmentality for quality control, the literature review and empirical sections were short of fulfilling those expectations. The major problem identified with the research findings is the lack of “thick data” or “participants’ narrative” to inform the findings. Summarily, although the current study is very important to the field of transnational higher education, I do not think that the authors have demonstrated enough evidence to show what new knowledge emerges from their study. I have provided some additional comments below for their attention. Literature review I think that the literature review section needs to be improved substantially to provide readers with better understanding of internationalization and transnational higher education. In its current form, the literature review section reads like an extension of the background section. The authors should provide answers to some vet important questions. For example: What are the theoretical foundations of internationalization and transnational higher education? What is already known in the field? What are the existing gaps in the concepts? Etc. unfortunately, these very important information cannot be found in the current text. Research site and research methodology The authors have not provided adequate information concerning the population of their study, the sampling technique adopted to select participants for the study and how the questions were developed. I read the literature review section again to find out how the authors developed their research questions but I could not gather adequate information on that. Also information on the process used to analyze the data is very scanty. Seeing that that authors followed a systematic investigation process, readers should be informed about how the codes were developed into the three main themes. This further connects to my earlier comment concerning the gap between the questions and the theories/concepts used by the authors. Research findings I struggled to identify how the narratives of the focus group participants were captured in the findings section. Did the researchers decide to abbreviate the contributions of participants? I suggest that the authors take a second look at the content and structure of the research findings section and provide readers with the rich text that reveal the voices of study participants. The discussion section also relies heavily on the secondary data analysed rather than the findings from the narratives of participants. Reviewer \#2: The topic of this article is very relevant from the viewpoint of internationalization trends in higher education. The research methodology (PAR, case study) adopted in this study is proper. Conclusions are clear and well established. There are some typing mistakes (e.g. lowercase instead of capital letters) left in the text. Please correct them. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Dr. Yaw Owusu-Agyeman Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0274989.r002 Author response to Decision Letter 0 22 Jun 2022 Cover letter in R1 submission June 2022 Dear Editor-in-Chief: This article has been updated in response to the feedback provided by the editors and reviewers. Your encouraging email to request a major revision of the article with the code PONE-D-22-11647 "Internationalization, cultural appreciation, and institutional governmentality for quality control in transnational higher education cooperation: An empirical assessment" is very much appreciated. We are really appreciative of your editorial board's decision to provide us with the chance to resubmit our work. We have given the remarks made by the reviewers and your own ideas careful consideration. We have every reason to believe that this piece of writing will be effective in appealing to a wide audience of possible readers. We have also added, in accordance with your specific demand, the file of "point- to-point response to the reviewers" in the article submission system. This file includes a comprehensive explanation of all of the modifications that were made in this R1 version. We have high hopes that following this significant revision, you and your prestigious blind reviews will be able to once again provide us with favorable feedbacks and remarks that are encouraging. We also add one significant originator who began the research concept of the present work, Miss Shushu WANG, in this submission. She is included as the corresponding author in this resubmission procedure. In the interim, we have accompanied this submission with a Change to Authorship form that is formatted in accordance with PloS One. We really hope that this adjustment does not cause you undue stress. We hope that everything works out well for you in life! We are grateful to you for taking my concerns into account. Thank you for your kind consideration! With sincere regards, Jinsheng (Jason) ZHU, PhD / Shushu WANG Guilin Tourism University Regarding the response you provided in the email correspondence, we will address your remarks in the next section of the current part of the cover letter below with bold blue letters shown below. Dear Dr. Wang, We've checked your submission and before we can proceed, we need you to address the following issues: 1\. Thank you for updating your data availability statement. You note that your data are available within the Supporting Information files, but no such files have been included with your submission. At this time, we ask that you please upload your minimal data set as a Supporting Information file, or to a public repository such as Figshare or Dryad. Please also ensure that when you upload your file you include separate captions for your supplementary files at the end of your manuscript. As soon as you confirm the location of the data underlying your findings, we will be able to proceed with the review of your submission. Response to the editor: Thank you for responding so quickly to the data availability statement. We believe it is incorrect to add "supporting information file" in the statement section. Because the submission has no accompanying information file. Therefore, we removed "Supporting Information File" from the statement for this resubmission. We regret for causing you this confusion in this manner. 2\. Please amend your authorship list in your manuscript file to include all the authors. Response to the editor: We've provided complete author information in the manuscripts (clear version and the file with track changes). 3\. Please amend your list of authors on the manuscript to ensure that each all authors re linked to an affiliation. Response to the editor: In the updated manuscripts, we included all of the relevant affiliation information. 4\. Please provide additional details regarding participant consent. In the Methods section, please ensure that you have specified (1) whether consent was informed and (2) what type you obtained (for instance, written or verbal). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information. Response to the editor: On page 8, you'll find further information that we've supplied about participant consents. For example, on page 8, under the heading "Methods," we put the statements that are shown below. “I consent to participate in this qualitative understanding to transnational collaboration project in the current faculty. I concur that the response I made in the following interview to be used for academic research purposes by researchers of this scientific interpretation.” 5\. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match. When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section. Response to the editor: Thank you for your comment. We have made revisions to the funding information accordingly. Some of the funding do not have a grant number. 6\. Thank you for stating the following financial disclosure: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript." At this time, please address the following queries: a\) Please clarify the sources of funding (financial or material support) for your study. List the grants or organizations that supported your study, including funding received from your institution. b\) State what role the funders took in the study. If the funders had no role in your study, please state: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” c\) If any authors received a salary from any of your funders, please state which authors and which funders. d\) If you did not receive any funding for this study, please state: “The authors received no specific funding for this work.” Please include your amended statements within your cover letter; we will change the online submission form on your behalf. Response to the editor: The following phrases are included in the article's acknowledgment section. “In the article acknowledgement, we state the following words. “This article is part of academic achievements of first-class universities and disciplines in tourism management discipline (project) in Guangxi, China. The corresponding author has also been participating in research projects supported by Guilin Tourism University-China ASEAN Research Centre. This research project is financially supported by Guangxi Tourism Vocational Education Teaching Steering Committee - 2021 Tourism Vocational Education Research Project on Teaching Reform in Tourism Education (2021LYHZWZ001).” We will be explaining in detail here. The “first-class universities and disciplines in tourism management discipline (project) in Guangxi, China” was a project of the university that the authors are working in. The Guilin Tourism University-China ASEAN Research Centre is one of the research centers that the authors Dr. Jinsheng (Jason) Zhu are working with. The “Guangxi Tourism Vocational Education Teaching Steering Committee - 2021 Tourism Vocational Education Research Project on Teaching Reform in Tourism Education” is one of the research projects undertaken by Dr. Jinsheng (Jason) Zhu. Here we would like to restate that both the authors are working and taking salaries in Guilin Tourism University, the institution we disclosed in the author title page. However, “The university and the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript”. We've returned your manuscript to your account. Please resolve these issues and resubmit your manuscript within 21 days. If you need more time, please email the journal office at <[email protected]>. We are happy to grant extensions of up to one month past this due date. If we do not hear from you within 21 days, we will withdraw your manuscript. Please log on to PLOS Editorial Manager at <https://www.editorialmanager.com/pone/> to access your manuscript. You will find your manuscript in the 'Submissions Sent Back to Author' link under the New Submissions menu. Be sure to remove your previous manuscript file if you are uploading a new file in response to these requests. After you've made the changes requested above, please be sure to view and approve the revised PDF after rebuilding the PDF to complete the resubmission process. We are requesting these changes to comply with the PLOS ONE submission guidelines (<https://journals.plos.org/plosone/s/submission-guidelines>). Please note that we won't send your manuscript for review until you have resolved the above requests. Thank you for submitting your work to PLOS ONE and supporting our mission of Open Science. Kind regards, Richard Ibañez Dilla PLOS ONE Once again, thank you for your kind consideration! With sincere regards, Jinsheng (Jason) ZHU, PhD / Shushu WANG Guilin Tourism University 10.1371/journal.pone.0274989.r003 Decision Letter 1 Tarrósy István Academic Editor 2022 István Tarrósy This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 25 Jul 2022 PONE-D-22-11647R1Internationalization, cultural appreciation and institutional governmentality for quality control in transnational higher education cooperation: An empirical assessmentPLOS ONE Dear Dr. Wang, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ==============================I was happy to see that the original submission was improved, but still more thorough work needs to be done, especially, as Reviwer 1 also underscores, the data analysis and the results section must be improved. Please, also take sufficient time to compose your sound rebuttal to all the questions, critical remarks articulated by the reviewers. If these are then accepted, I can support the acceptance of your article for publication.============================== Please submit your revised manuscript by Sep 08 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, István Tarrósy, PhD Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: I must commend the authors for the hard work in revising the current manuscript. While I have seen that the authors have provided some additional information to the original version of their work, I think that there are some outstanding issues that needs to be resolved. First the authors should avoid abbreviations. For example on page 4 (literature review, line 1) the authors indicate that, “before delving into the challenges posed by the globalization of higher education, we'd want to….” The sentence should read as, “before delving into the challenges posed by the globalization of higher education, we would want to…..” Secondly, information on the process used to analyze the data is very scanty. Seeing that that authors followed a systematic investigation process, readers should be informed about how the codes were developed into the three main themes. This comment was contained in my earlier review report but they have not been addressed in the revised study. Lastly I still do not see an improvement in the results section where the narratives of the focus group participants are captured in the findings section. Reviewer \#2: The authors have successfully managed to revise their paper according to the recommendations. The revised sections (Literature Review, The internationalization of higher education, Research site and research methodology) have made this research article more suitable for publication from a theoretical and empirical point of view. It is also acceptable that the authors have only provided a summary of thoughts contributed by the interviewed participants. This does not reduce the academic quality of this paper. It is understandable to take into account the word limit. This paper will enrich the academic literature dealing with the internationalization of higher education with a special focus on quality control. I highly recommend the authors to continue their research and extend it to other Chinese universities which have a transnational cooperation in the field of tourism. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0274989.r004 Author response to Decision Letter 1 9 Aug 2022 I was happy to see that the original submission was improved, but still more thorough work needs to be done, especially, as Reviwer 1 also underscores, the data analysis and the results section must be improved. Please, also take sufficient time to compose your sound rebuttal to all the questions, critical remarks articulated by the reviewers. If these are then accepted, I can support the acceptance of your article for publication. Thank you so much, Dr. István Tarrósy, for your encouragement and consideration for the publication of the current paper. Your encouragement helps us a lot in the process of drafting, revising and final publication of this academic work, which we believe will be contributing to the topic of transnational higher education collaborations, quality guarantee, as well as cultural connectedness. We also believe that this paper will get high citations in the near future! In the following response to the reviewers, we started to response to the reviewer from page 5, starting from the comments of the first reviewer. Looking forward to hearing from you! Yours, Wang Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: I must commend the authors for the hard work in revising the current manuscript. While I have seen that the authors have provided some additional information to the original version of their work, I think that there are some outstanding issues that needs to be resolved. First the authors should avoid abbreviations. For example on page 4 (literature review, line 1) the authors indicate that, “before delving into the challenges posed by the globalization of higher education, we'd want to….” The sentence should read as, “before delving into the challenges posed by the globalization of higher education, we would want to…..” Secondly, information on the process used to analyze the data is very scanty. Seeing that that authors followed a systematic investigation process, readers should be informed about how the codes were developed into the three main themes. This comment was contained in my earlier review report but they have not been addressed in the revised study. Lastly I still do not see an improvement in the results section where the narratives of the focus group participants are captured in the findings section. Response to reviewer \#1：Thank you so much for your comments. We are strictly considering your comments and revised the manuscript according to your criticisms point-by-point here below. 1\. We have revised the unappropriated abbreviations according to your suggestion. 2\. With regard to the investigative material, we have thoughtfully selected some of the obtained original interview data for inclusion in the manuscript, which resulted in an increase of more than one thousand words to the total word count. This does not constitute a violation of the standards set out by the prestigious publication Plos One since they do not take into account the word count. 3\. In the beginning, as a result of our interview that took a full year to complete from the voice recording to the transcription to the words, we gathered a total of around fifty thousand words. As a result of this, we came to the conclusion that your responses are of the highest significance. As a consequence of this, we decided to use interview data to complement the key arguments presented in each area of the article, most notably the results section. Thus, the result section turned out to be the current form. We hope you appreciate its intellectual worth in this run of review. We are really grateful for your views and intellectual insights. Reviewer \#2: The authors have successfully managed to revise their paper according to the recommendations. The revised sections (Literature Review, The internationalization of higher education, Research site and research methodology) have made this research article more suitable for publication from a theoretical and empirical point of view. It is also acceptable that the authors have only provided a summary of thoughts contributed by the interviewed participants. This does not reduce the academic quality of this paper. It is understandable to take into account the word limit. This paper will enrich the academic literature dealing with the internationalization of higher education with a special focus on quality control. I highly recommend the authors to continue their research and extend it to other Chinese universities which have a transnational cooperation in the field of tourism. Response to reviewer \#2：Thank you so much for your complements! We will be very very willing to extend our research to a wilder range of scope in the near future. We hope to work with you in the near future, haha! 10.1371/journal.pone.0274989.r005 Decision Letter 2 Tarrósy István Academic Editor 2022 István Tarrósy This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 8 Sep 2022 Internationalization, cultural appreciation and institutional governmentality for quality control in transnational higher education cooperation: An empirical assessment PONE-D-22-11647R2 Dear Dr. Wang, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, István Tarrósy, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): As long as all the comments of the reviewers have been addressed neatly, which they (especially Reviewer 1) also accepted, and I can also see that the manuscript has improved a lot, I wish to support the publication of the paper in the journal. Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: I must congratulate the authors for the hard work. The revised manuscript looks good. I wish them all the best with their future research. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Yaw Owusu-Agyeman \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0274989.r006 Acceptance letter Tarrósy István Academic Editor 2022 István Tarrósy This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 Sep 2022 PONE-D-22-11647R2 Internationalization, cultural appreciation and institutional governmentality for quality control in transnational higher education cooperation: An empirical assessment Dear Dr. Wang: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. István Tarrósy Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction The mitochondrion produces most of the ATP in the cell, an energy source on which almost all physicochemical processes depend. Each cell contains dozens or hundreds of mtDNA genomes that are inherited as a single haplotypic block from the mother to the offspring. Germ-line mutations accumulate on top of existing haplotypes, and these haplotypes aggregate in human populations according to their demographic histories. Due to the particularities of the mtDNA molecule (i.e. matrilineal inheritance and lack of recombination), it is straightforward to reconstruct phylogenetic relationships between human haplotypes. Phylogenetically related haplotypes in the population are commonly grouped into clusters or haplogroups. Thus, haplogroups represent branches of the mtDNA phylogeny, and the set of diagnostic variants defining these clades are popularly known as the sequence motif. Screening for these variants in a given mtDNA molecule can provide sufficient information to allocate a particular mtDNA genome into a given haplogroup. In the last few years, a huge number of studies have been conducted addressing the presumable association of mtDNA haplogroups with different complex diseases, including cancer, Alzheimer, Parkinson –, schizophrenia, infectious diseases, diabetes, LHON, etc. Most of these disease studies are population-based, that means, the mtDNA variability is compared between cohorts of cases and representative healthy control (case-control studies), where the statistically significant over-representation of a given variant in cases regarding controls might point to a biological association of this variant with the disease. Statistical procedures are important in order to understand the presumable relationship between mtDNA haplogroups and diseases. Estimating *a priori* statistical power is fundamental in case-control association studies given that this is the way to evaluate to what extent a positive finding is likely to be not at random. However, case-control association studies targeting the mtDNA variation rarely compute power mainly due to the lack of the statistical procedures that are necessarily different to those employed using autosomal DNA markers. To the best of our knowledge, only Samuels et al. investigated the issue of statistical power in regards to cases-control mtDNA studies involving 2×*k*. These authors used a simulation-based permutation test (Monte-Carlo) in order to estimate power calculations for prospective case-control studies. According to these authors, very large cohorts are needed to reliably detect and association between mtDNA haplogroups and complex diseases. This study however only deals with the restricted scenario where the number of cases equals the number of controls. The particular biological application of Sánchez et al. on a mtDNA case has to do specifically with 2×3 tables, comparing a RFLP polymorphism (binary) and the three genotypes derived from a biallelic albumin marker. Several software packages and statistical procedures were designed for the calculation of statistical power and sample size. Most of the procedures developed to date can only deal with 2×2 tables (the great majority) or 2×3 tables. Thus, most of the software packages have been designed for the estimation of power or/and sample size in the most common scenario involving allele frequencies deriving from autosomal binary markers (SNPs), that is, involving allele or genotype association tests. Only two software packages, namely G\*Power 3 and Pass 12, are able to treat tables *r*×*k*; however, these two packages only deal with scenarios where the number of cases equals number of controls. Finally, osDesign is based on logistic regression, and although it can deal with *r*×*k* tables it does not allow estimating samples sizes. In the present study we consider more general case-control disease scenarios involving any number of cases and controls and 2×*k* tables. For instance, it is a common situation that only a limited number of patients can be recruited in a particular study; however, an increase in the number of controls could contribute to reach a reasonable statistical power. The model elaborated in the present study is based on simulations (Monte Carlo method) as a way to estimate the statistical power in case-control studies where there is interest in investigating the presumable relationship between a certain disease and a number of mtDNA haplogroups (or haplotypes or mtDNA SNPs \[mtSNPs\]). We consider the frequency of the risky allele or haplogroup in controls (*p₀*) and in cases (*p₁*), and the difference between these two parameters is proportionally distributed to the frequencies of the remaining allele or haplogroup categories in cases. In addition, a web-tool named mitPower has been also developed to implement all the statistical procedures developed in the present study. # Methods ## Data simulation We first build 2×*N_H* (in general, 2×*k*) tables (10,000 simulations), where denotes the number of haplogroups considered (but could also be any number of haplotypes or mtSNPs). Two multinomial samples are used to build the contingency tables, taking as frequencies the estimated frequencies, and as size, the number of controls and cases of our study. In the simulated tables the row variable represents the status of case or control, while the columns represent the allele variables or haplogroups into which individuals are sorted. For the sake of simplicity, *N_H* was set up to 11 (unless otherwise stated) but the method and mitPower has been designed to accept any number of haplogroups. As done in Samuels et al., the following 11 haplogroup frequencies were considered as example: H (41%), I (2%), J (11%), K (8%), M (1%), T (13%), U (15%), V (13%), W (2%), X (2%) and a residual haplogroup (2%). The power values obtained using MitPower have been validated with other tools in comparable scenarios that consider 2×2 tables and equal numbers of cases and controls. MitPower was additionally validated for 2×3 tables with the procedure shown by Sánchez et al.. ## Statistical analysis First, the computation of the probabilities of 2×2 contingency tables is the best option to test the homogeneity of control and case sample populations; however, the computational requirements increase with the dimension of the contingency tables. A way to overcome this problem is to implement a Fisher's exact test that estimates the probability of a contingency table using a Monte Carlo simulation approach. On the other hand, the Chi-squared statistic is computationally feasible for 2×*k* tables being *k* any entire positive number. Both tests yield very similar results and both are implemented in MitPower. From here onwards, we have used the Chi-square statistic, which compares the values obtained in our contingency tables against the values expected under the null hypothesis of homogeneity. First, we generate a number of tables 2×*N_H* under a given hypothesis. A random number between 0 and 1 is generated using the R function *runif*, and this number is used as the seed for simulations. Power value estimators are obtained as the percentage of simulated tables with *P*-value below a fixed significance level. In order to obtain the *P*-value for each simulated table, the distribution of our statistic has to be known. This distribution is approached here using two alternative procedures: the asymptotic and the permutation approach. The former approach is based on the asymptotic distribution of the Chi-square statistics. Note that some authors argued that an increase of the false-positive rate occurs when the Cochran's rule is not verified; so the asymptotic approximation should be considered acceptable when the Cochran's rule is verified ; that means that a contingency table cannot contain expected values below one, and that no more than 20% of the expected value can be below five. The permutation method aims to overcome this problem. First, a large number of permuted tables of our initial data (contingency tables) are generated, with the only restriction that total sums by rows and columns have to remain constant. For each of these permutations, the Chi-square statistic is computed, and the *P*-value is obtained as the proportion of permutations with a Chi-square statistics higher than the statistics in the original data set. There are several ways to obtained permuted tables, and we chosen the method provided by the function *chisq.test*. Theoretically, the asymptotic and the permutation approaches should have similar values as the sample size increases. Some experiments have been done in this direction (and see text below) in order to corroborate this expectation. Along the simulation experiments carried out in the present manuscript, the permutation method was preferred given that it generally performs better than the asymptotic one (see below). All the computations were carried out in R (<http://www.r-project.org/>), and using the functions *chisq.test*, *fisher.test*, and *pchisq* of package *stats*. ## mitPower: a web interface to estimate statistical power in 2×k tables mitPower is a web-based tool (<http://bioinformatics.cesga.es/mitpower/>) that allows estimating the statistical power in case-control association disease studies. Several other utilities are available in mitPower such as the estimation of: (i) the *a posteriori* statistical power, (ii) the sample size needed in order to reach a given statistical power, and (iii) the minimum deviation from the null hypothesis (of no association) detectable under a given statistical power (expressed as *OR* and haplogroup frequency in cases). The software mitPower allows using two calibration methods: the asymptotic and the permutation procedure. The permutation procedure can be computationally demanding (see below) so the asymptotic procedure might be more convenient for complex scenarios. The mitPower web interface is written in PHP, allowing users to enter their inputs through an HTML form. All the mitPower analyses are executed using R scripts (see above), which are called from the interface through *Rscript* and run at the web server. Their output is ultimately formatted for web display again by the PHP interface. Results links are kept for 24 hours in the server. The underlying R scripts in mitPower run on a web server hosted by the Supercomputing Center of Galicia (CESGA; <http://www.cesga.es>) located in Santiago de Compostela (Galicia, Spain). # Results and Discussion Two procedures were followed to calibrate the distribution of our statistic: the asymptotic and the permutation method. In order to test which of the two approaches performs better, a simulation experiment has been performed considering different sample sizes, number of simulations, and permutations. The experiments indicate that the permutation method performs better than the asymptotic one given that power estimates approach closer to the significant value under the null hypothesis for low sample sizes. However, both approaches yield good estimates when considering large sample sizes. This is in agreement with theoretical expectations given that for large samples, their statistical distribution should be equivalent, as the permutation distribution should converge to the tabulated distribution. The results indicate that (i) the permutation method tends to fit better to the significance level than the asymptotic approach when the null hypothesis is true (specially for low sample sizes), and (ii) computational requirements using permutation can be an issue when considering a large number of iterations (large sample sizes); in such situations, the asymptotic calibration method might be more convenient. As done in Samuels et al., we would assume that haplogroup mtDNA frequencies in controls are known (note that there exist hundred of human population studies carried out to a local, regional or continental scale where these frequencies are available, at least for the most common haplogroups). We then simulated increases in the frequency of a risky haplogroup in cases, with the differences distributed proportionally between the remaining haplogroups (therefore, assuming there is no *a priori* assumption of an association with any of the remaining haplogroups considered). In agreement with Samuels et al., we observed that power strongly depends on sample sizes, haplogroup population frequencies, and the deviation from the null hypothesis when using equal numbers of cases and controls (see solid lines). We next evaluate the situation where the number of cases differs from the number of controls. As shown in, statistical power strongly depends on the case:control ratio when the other parameters are fixed, but this dependence is not as simple. As expected, power can increase very substantially as more controls exist relative to the number of cases. For instance, the statistical power to detect an association of haplogroup J increases from 60% to 80% when doubling the number of controls respect to cases in the example provided in. Samuels et al. introduced the *N_scaled* parameter for the estimation of the power. This parameter considers the difference between haplogroup frequency for equal numbers of controls and cases: , being *p₀* the frequency of the risky haplogroup in controls, *p₁* the frequency of the risky haplogroup in cases, and *N* the number of cases and controls (the total sample size is 2*N*). We further consider the more general situation where the number of cases (N_ca) can differ from the number of controls (N_co), . *N_scaled* and *N_sc* measures the squared standardized difference between frequencies in cases and in controls for the risky haplogroup. For 2×2 tables and a sample size large enough, the *N_scaled* parameter follows a chi-square distribution with one degree of freedom due to the asymptotic normality of the standardized difference between frequencies. As shown in, there is a clear relationship between the parameter *N_sc* and the power values. These values follow the theoretical curve obtained for 2×2 tables and equal numbers of cases and controls using the *arcsin* transformation :Where *p₀* in the frequency in controls, *p₁*is the frequency in controls, *N* is the sample size for each arm and and are normal quantile for and.Simulations also indicate that the statistical power decreases as more haplotypes are tested. Samuels et al. introduced a parameter, *N_H* (number of different haplogroups), that raised to the power of 0.37 allows to fit the statistical power to a single theoretical curve. According to Samuels et al. *N_scaled* can be redefined as a function of the number of haplogroups analyzed: . Note that the value 0.37 seems to have been obtained empirically by Samuels et al, (no specific formulae or indications were given in this regard). In the analysis shown in we aimed to reproduce their findings. The simulations corroborate the fact that 0.37 is the value that allows to better fit the data to the theoretical curve for values of statistical power above 50%. Below 50% an exponent of 0.5 would perform better although it can assume that values of statistical power below 50% might be not relevant in association studies. Therefore, we observed that *N_H* raised to the power of 0.37 allows fitting the statistical power to those scenarios where the number of cases differs from the number of controls. Finally, *N_sc* allows to relate all the parameters involved in the computation of the statistical power: Samuels et al. (see A1 in their Appendix A) propose to use *N_scaled* to estimate directly the statistical power and to determine the minimum number of controls and disease cases (*N_Cmin*) required for a specific level of power (their formula (2)). However, this formula applies when number of cases equals the number of controls. The simulation method aims to overcome the limitation of this formula allowing for different sample sizes in cases and controls. We then adjusted the parameter *N_Cmin* for the more general scenario involving unequal numbers of cases and controls. Two options are possible: (i) estimation of the sample size given a control-case ratio; or (ii) estimation of the minimum number of controls (cases) when the number of cases (controls) is fixed. In the first situation, if *N* denotes the number of cases, the control-case ratio, and the number of controls, the minimum number of cases needed to reach a power with a significance level can be estimated from (5) as follows:where denotes the value providing a desired power and a significance level, while \[·\] denotes the integer part function. The number of controls can be estimated as. Note that a value of would reproduce the particular scenario considered by Samuels et al.. In the second situation, the minimum number of controls or the minimum number of cases given a significance level and a power, can also be estimated when the number of cases or the number of controls is fixed, respectively: where is the value providing the power desired for a significance level , and \[·\] denotes the integer part function. This allows estimating the number of controls (number of cases) needed to reach a required power given a number of cases (number of controls). Note that statistical power is limited by the restrictions) and (8); this is the reason of why power becomes stationary when the number of controls increases in regards to the number of cases. It is also worthwhile to estimate the minimum deviation of the null hypothesis that can be detected for a power value and a significance level, assuming we know haplogroup frequencies in controls and considering a given number of cases and controls. We can express being the deviation of null hypothesis. Note that it must verify. If we calculate from (1), it results (for a risky haplogroup): This difference can be expressed in terms of odds ratio. Thus If then,Where *OR_min* denotes the minimum OR that can be detected for a power value and a significance level. The applications above require knowledge of the value given a significance level and a power. This parameter can be obtained by way of simulations and nonparametric regression ( \[right\] shows the scenario where). Note that nonparametric regression seems to perform well for power values above 60%. The package *sm* implements local linear estimation, window-selector cross validation and Gaussian kernel, that allows to obtain values for different significance levels α and power values. These values can be used)–(10) in order to estimate the desired parameters (as done). The same simulation methods proposed to compute *a priori* statistical power can be applied for the estimation of the *a posteriori* power. Note however, that we treat *a posteriori* power in a different context as interpreted by others. Our procedure involves generating new data (tables) using the sample proportions and sample sizes obtained from a particular study. Therefore, the null hypothesis is tested by simulating new contingency tables. The procedures are analogs to the ones used to compute the *a priori* power. Computation of the statistical power is essential to anticipate if the positive findings obtained in case-control disease studies are reliable. The present study has been motivated by the fact that the only available procedures to date to compute statistical power in mtDNA association studies only allows to deal with scenarios involving 2×2 tables (or 2×3 tables), or if 2×*k* tables only study designs considering equal numbers of cases and controls (which does not represent the most common scenario in association studies). In the present study, we also provide with a web interface that implements the procedures developed in the present study (mitPower). # Conclusions During the last decade, a large number of mtDNA case-control studies have been published in the literature, most of them pointing to a number of haplogroups presumably associated with a complex disease. The validity of many of these conclusions might be questionable, given that most of them are underpowered. Most of these studies did not estimate the *a priori* statistical power because statistical tools were not available at the time. The procedures developed in the present study allow the computation of statistical power in common as well as complex case-control study designs involving 2×*k*. The results indicate that underpowered studies could reach reasonable power by increasing the number of controls and reducing the number of hypothesis testing (i.e. haplogroups). In order to reach a wide range of researchers, we provide a friendly web-based tool (mitPower) that implements all the statistical procedures developed in the present study; this software can be used in both retrospective and prospective case-control disease studies. Note that the term retrospective is considered here as done before: “*the prospective power that can be obtained ignoring the fact that data have been gathered and a hypothesis has been tested. In essence, it computes the prospective power of the test as if: (a) the study and analyses had not yet been conducted and (b) the sample effect size is the hypothesized population effect size*”. Further developments of mitPower could involve the implementation of multiple test corrections for the computation of the statistical power (in the sense as it was suggested before in 2×2 tables) and in two-stage case-control designs. Also challenging would be to explore the phylogenetic relationship existing between different haplogroups (the phylogenetic dependence) or mtSNPs and how this dependence could influence the estimation of power. Finally, other statistical/computational approaches could find their place in 2×*k* tables, such as the use of Markov Chain Monte Carlo methods (MCMC), already explored for 2×2×2 tables. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JPS WGM AS. Performed the experiments: JPS JA AS. Analyzed the data: JPS AS. Contributed reagents/materials/analysis tools: WM AS. Wrote the paper: JPS AS. Drafted the manuscript: JPS AS. Agreed on the final version of the manuscript: JPS JA WGM AS.

# Introduction The experimental identification of *cis*-regulatory sites based on transcription factor binding motifs (TFBMs) is a difficult and time-consuming task. In this regard, *in silico* analysis of TFBMs has recently attracted attention as a promising tool for discovering true *cis*-regulatory sites. Previous works attempt to find TFBMs to model the mechanisms underlying the control of gene expression levels. They assume that the gene expression levels are determined by the presence of certain motifs in the upstream regions of the genes. Based on this assumption, they find TFBM candidates which show a strong correlation with changes in the gene expression levels. Instead of modeling the expression levels, another solution is to model the binding affinities between a protein and its target genes based on the thermodynamics theory. However, the binding affinities are difficult to measure and related works use transcription factor occupancy to approximate binding affinity. In this article, we present PeakRegressor, a new tool for the identification of functional TFBMs from ChIP-Seq data. As far as we know, this is the first attempt at performing peak signal regression based on candidate motif models. Because PeakRegressor is computationally efficient and the models are easy to interpret, it is usable with large-scale datasets. We apply PeakRegressor to two ChIP-Seq datasets and show its ability to recover motifs involved in the binding of STAT1 and RNA Polymerase II. # Results and Discussion ## Results with PeakRegressor shows the correlation coefficients between the peak scores and their predicted values by PeakRegressor in the test dataset. We keep the highest correlation coefficient among various for each iteration of the 30-fold cross-validation, and those 30 correlation coefficients are averaged and shown here. Obviously, the filtering with peak existence probability, i.e., Q-value, over the control experiment enhances the regressions. The filtering with promoter region proximity improves the regressions of RNA Polymerase II but not of STAT1. In, we plot the STAT1 peak scores with two filtering methods such as Q-value and promoter proximity in the test dataset against their predictions by PeakRegressor. The correlation coefficient is as high as 0.65 between the peak and predicted values for the Q-value filtering, whilst it is as low as 0.41 for promoter proximity filtering. Interestingly, however, the data points that are selected by promoter proximity existed only in a biased region, leading to worse prediction. In and, we show the top ten motifs for STAT1 and RNA Polymerase II identified by PeakRegressor, respectively. The motifs are sorted according to the absolute values of their averaged regression coefficients. A motif with a positive (resp. negative) coefficient is thought to have a strengthening (resp. weakening) effect on the binding. In the case of STAT1, it is clear that our approach correctly identifies the classical GAS motif TTC\[TC\]N\[GA\]GAA as the main binding motif. Meanwhile, the RNA Polymerase II binding motifs also contain known Downstream Promoter Element \[AG\]G\[AT\]\[CT\]\[GAC\] and Initiator Site \[TC\]\[TC\]AN\[TA\]\[TC\]\[TC\]. ### STAT1 composite motifs As the most important feature of PeakRegressor, it can give us a list of putative composite motifs. Basically, it is difficult to evaluate whether a composite motif consists of the same motif or multiple (different) motifs. In order to identify the composite motifs, we proceed as follows. First, we consider the best set of motifs according to PeakRegressor (i.e., the set which corresponds to the best prediction accuracy). Among these, we select motifs which have a normalized coefficient higher than. We use these motifs to represent each peak sequence as a binary vector, indicating whether a motif is present or not in the peak sequence. Then we cluster the resulting peak vectors using the K-Means algorithm. Thus each cluster contains peak vectors which show similar motif patterns, i.e., sequences containing potential composite motifs. Here we show an example of a composite motif that are responsible for STAT1 binding signals: ## Comparison with other regression methods PeakRegressor identifies potential TFBMs by solving a regression problem. This regression problem is defined by a set of peak vectors and their corresponding peak scores. The goal is to predict the peak scores from the peak vectors. The fitted regression model is then used to infer the TFBM candidates. We expect the regression method to have three properties. First, it should identify the true binding motifs. Second, it should identify the strengthening and weakening motifs. Third, it should be computationally efficient in order to cope with large ChIP-Seq datasets. In PeakRegressor, we choose to use the L1-norm log linear regression to solve this problem. This approach favors sparse solutions (i.e., solutions with a small number of motifs) and therefore, we argue that it is more suitable for the TFBM identification problem. However, many other regression methods are available and can be used to solve the regression problem. How do these approaches compare with the L1-norm log linear regression with respect to the desired properties? In the following, we compare our L1-norm log linear regression based approach with other regression methods: linear least squares regression, ridge regression, partial least squares regression, and principal component regression. For each method, we evaluate its performance on the STAT1 and RNA Polymerase II datasets and discuss the results. ### Linear least squares regression In and, we show the top ten motifs identified by the linear least squares regression. In the case of STAT1, we can see that the true GAS motif appears within the top ten motifs. However, two problems appear. First, the regression coefficients of the GAS motif are very low compared to those of the top motifs (between). This means that according to the linear least squares regression, the true GAS motif has only a minor effect on the binding, which contradicts existing biological knowledge. Second, the most important motifs according to the linear least squares regression are CCCCTCCC and CCCCACCC. However, each of them is associated with opposite coefficients ( and for CCCCTCCC, and for CCCCACCC). Therefore, each of them is considered to have both a strengthening effect and a weakening effect on the binding, which is a contradictory result. With the RNA Polymerase II dataset, linear least squares regression is able to identify the initiator site and the downstream promoter element. However, the instances of the initiator site have opposite coefficients (\[CA\]CAGACT with, T\[CT\]\[TA\]T\[TG\]\[AC\]\[AT\] with , and TT\[TAC\]TTT\[CT\] with). As they are instances of the same motif, we expect them to have the same sign i.e., to have the same effect on the binding. In summary, for both STAT1 and RNA Polymerase II datasets, the results of the linear least squares regression are difficult to interpret biologically. This is a typical situation where we would like to reduce the number of motifs used by the regression model. Clearly, this is not possible with the linear least squares regression approach. ### Ridge regression In and, we show the top ten motifs identified by the ridge regression. In the case of STAT1, we can see that the ridge regression and the L1-norm log linear regression identify very similar motifs. In both cases, the classical GAS motif is clearly identified as the main binding motif. Both regression methods also identify CA\[TC\]GTGACT\[TG\]C as a strengthening motif and GGAGGGCG as a weakening motif. In the case of RNA Polymerase II, both methods are able to identify the initiator site (T\[CT\]\[TA\]T\[TG\]\[AC\]\[AT) and the downstream promoter element (A\[GC\]\[TAG\]CA). However, they differ greatly with respect to computational complexity. In, the authors present an algorithm for computing the L1-norm log linear regression solutions of many regularization parameters for the same computational cost as that of a single least squares fit. As a consequence, using the same STAT1 dataset, a 30-fold cross-validation takes approximately hours with the ridge regression, while it takes only hours with the L1-norm log linear regression (i.e., times faster). In summary, although both methods show very similar results with respect to binding motif identification, the ridge regression is slower and more difficult to use with large ChIP-Seq datasets than the L1-norm log linear regression. ### Partial least squares regression and principal component regression In and, we show the top ten motifs for STAT1 identified by the partial least squares regression and the principal component regression. We can see that both methods are able to identify the classical GAS motif. In, the partial least squares regression shows very similar results to the L1-norm log linear regression as both methods identify CA\[TC\]GTGACT\[TG\]C as a strengthening motif and GGAGGGCG as a weakening motif. In, the principal component regression identifies only the GAS motif and fails to identify any other motifs involved in the binding. In the case of RNA Polymerase II, both partial least squares regression and principal component regression are able to identify the initiator site and the downstream promoter element. However, the results of the partial least squares regression and the principal component regression are difficult to interpret. In the former , different instances of the downstream promoter element have positive or negative coefficients (T\[TG\]AACAC**AGTT\[TA\]** with, \[CT\]\[CG\]AG**A\[GA\]TCC**A\[GA\]\[CG\] with, and A\[AG\]\[GA\]\[AG\]**GGA\[GCA\]G**A\[GA\]A with). As they are instances of the same motif, we expect them to have the same sign, i.e., to have the same effect on the binding. In the latter, all the instances of the initiator site and the downstream promoter element have negative coefficients. However, these motifs should strengthen the binding and therefore, we expect their coefficients to be positive. The lack of interpretability of the partial least squares regression and the principal component regression lies in the fact that the regression is performed in a low-dimensional feature space. In the original motif space, the vector representation of the peak sequences has a meaning and each component of a vector measures how similar a motif is to a peak sequence. However, in the low- dimensional feature space computed by the partial least squares regression and the principal component regression, the vector components lose their biological meaning. From the computational complexity perspective, we also mention that both methods are very slow. Using the STAT1 dataset, a 30-fold cross-validation of the partial least squares regression with 10 components takes approximately hours. In summary, the partial least squares regression and the principal component regression are able to identify the classical GAS motif for STAT1 and the initiator site and the downstream promoter element for RNA Polymerase II. However, the results are difficult to interpret biologically and do not allow identification of strengthening or weakening motifs. In addition, they are too slow to be used with large ChIP-Seq datasets. ### Advantages of L1-norm log linear regression over other methods for TFBM identification We considered the following regression methods for TFBM identification: L1-norm log linear regression, linear least squares regression, ridge regression, partial least squares regression, and principal component regression. In, we summarize the correlation coefficients averaged on the test sets. As we can see, all regression methods demonstrate similar performance and are able to identify the classical GAS motif for STAT1 and the initiator site and the downstream promoter element for RNA Polymerase II. However, they exhibit marked differences with respect to biological interpretability and computational efficiency. The results of the linear least squares regression, the partial least squares regression, and the principal component regression do not allow identification of strengthening or weakening motifs. Therefore, they are difficult to use for binding motif identification. Both L1-norm log linear regression and ridge regression solve this problem by means of regularization. However, the ridge regression is very slow compared to the L1-norm log linear regression. Therefore, the ridge regression is difficult to use with large-scale ChIP-Seq datasets. In summary, the L1-norm log linear regression is the only method that can achieve all the desired goals for our task; it identifies the transcription factor binding motifs, the regression coefficients are easy to interpret biologically, and its implementation with the LASSO algorithm is fast and efficient. This justifies our choice of the L1-norm log linear regression in PeakRegressor. ## Parameter setting The performance of PeakRegressor depends on the choice of parameters that have to be set empirically. In this section, we explain how we choose two important parameters: the length of peak sequences and the number of motif candidates. ### Length of peak sequences In the dataset provided by, all the peaks correspond to various DNA sequences. These sequences have different lengths, ranging from 1 bp to several thousand bp. To conduct our analysis, we modify the peak sequences in the following way: - We shorten long peak sequences for two reasons. First, when using long DNA sequences, the computations of the motif finding algorithm MEME take too much time. Second, finding good quality motifs with MEME is easier with short DNA sequences than with long ones. - We widen short peak sequences. Due to the noisy nature of ChIP-Seq data, the motifs we are looking for may not be exactly on the provided peak sequence, but in the surrounding DNA neighborhood. Therefore, we decide to choose a uniform length for all the peak sequences. The choice of 200 bp is empirical; we try several values (100 bp, 200 bp, 400 bp, and 800 bp) and consider the one that achieves the best performance, i.e., the highest correlation coefficients (results not shown for other peak lengths). ### Number of motif candidates In the first step of PeakRegressor, we use MEME to find over-represented DNA motifs in the peak sequences. This step results in 800 motif candidates for STAT1 and 880 for RNA Polymerase II. Given the large number of motif candidates, we empirically observe the presence of similar motifs in the set of motif candidates. We may wonder if this redundancy could affect the prediction performance of PeakRegressor. However, we show that this is not the case. PeakRegressor uses a regression method called L1-norm log linear regression. In contrast with other regression methods, L1-norm log linear regression achieves its best prediction performance by removing redundant or uninformative motifs from the regression model. Therefore, the removal of redundant motifs is automatically performed when using L1-norm log linear regression. shows the set of motifs that achieve the best correlation coefficient for STAT1. We can see that some motifs are similar. For example, the motifs A\[CT\]**TTC\[TC\]\[TG\]GGAA**, **TT\[CA\]C\[TAG\]\[GA\]GAA**\[GA\]T, A\[TA\]**TTCC\[CT\]\[GA\]GAA**\[AC\]T\[CG\]\[AC\], and **TT\[CA\]\[TC\]\[GA\]GGAA**\[AG\] are short, similar motifs containing the STAT1 binding motif. In other experiments, we find that the prediction performance worsens when similar motifs are removed (results not shown). Hence, although the motifs appear similar and redundant, they actually contain complementary information for the prediction performance. Moreover, the motif weights computed by PeakRegressor are all different (resp.,). Hence, while other approaches, such as motif clustering, would consider all these motifs to be equally important, PeakRegressor is able to detect the relative importance of each motif and compute the corresponding weight. This is explained by the noisy nature of the DNA motifs found by MEME in step 1. For a given binding motif, PeakRegressor needs to use all the noisy PSSM approximations to achieve the best prediction performance. This is an important property of PeakRegressor, especially when the number of noisy motifs is very large. ## Candidate motifs and their potential rSNPs Single or composite motifs found in the PeakRegressor system may reflect actual transcription factor binding sites. If a single nucleotide polymorphism (SNP) occurs within the sites, regulatory control of neighboring gene transcription will be perturbed, thus leading to genetic diseases in some cases. Therefore, true binding sites may have SNPs less frequently than the non-binding sites. As an important verification, we check the number of known SNPs to be found within the STAT1 positions presented by PeakRegressor by using dbSNP database (<http://www.ncbi.nlm.nih.gov/SNP/>). We find that 0.36% (147 for 40,395 bp) of mapped positions with 10 STAT1 motifs in on the peak sequences contains SNPs, while as much as 0.53% (17,852 for 3,344,439 bp) of all positions contains SNPs on the peak sequences. The statistical difference between the above two ratios is highly significant such as according to the hypergeometric distribution. These sites are possible candidates of rSNPs because the slight change within the motif may affect the change of gene expression level and might cause diseases. # Materials and Methods ## PeakRegressor PeakRegressor is a system to find TFBMs that are statistically important for transcription factor binding signals, by taking ChIP-Seq data as input, and outputs a list of TFBM candidates. In contrast with previous approaches, PeakRegressor uses the peak scores (provided by) as a surrogate for the binding affinities. We argue that the peak scores provide more accurate approximations of the binding affinities than the methods based on transcription factor occupancy. Therefore, using the peak scores lead to better identification of functional TFBMs. In addition, PeakRegressor identifies not only primary TFBM candidates but also secondary motifs that may often synergistically strengthen or weaken the binding. The workflow is summarized in. ### Step 1 First, we define the peak sequences as the -bp genomic regions centered around the peaks. Then, we sort the peak sequences according to their ascending scores. We group the peak sequences into clusters such that each cluster contains 200 peaks of consecutive scores. Then, we apply MEME (<http://meme.sdsc.edu/>) to each peak sequence cluster. For each sequence cluster, MEME is parameterized in ZOOPS mode to find motifs of lengths. This strategy has two advantages. First, it allows us to identify motifs that may be associated with a given binding affinity level. If a cluster contains only low (resp. high) binding affinity peaks, the corresponding sequences may contain weak (resp. strong) binding motifs, i.e., motifs that are specific to low (resp. high) binding affinity. Second, it reduces computational time by parallelizing MEME computations. ### Step 2 In order to predict the binding affinity of the peaks, we need to represent each peak as a vector in the motif space. Let be the DNA sequence of peak. Let be the -length sub-sequence of, starting from position. Let be the PSSM of motif. Let be the length of and be the length of motif. We represent peak as vector, such thatfor. The quantity is a sum of log-odd scores, representing how well motif matches sub-sequence. Hence, the first term of the sum, corresponds to the best match when we slide motif along sequence. The term is the maximum score achievable by any sequence matching with the motif. Therefore, we always have, with for the best possible match. Next, we want all the to be positive for interpretability purpose. So we simply shift their values by substracting the lowest component:, where is the minimum value of the original. Finally, we normalize each data vector by dividing it with its euclidean norm:. ### Step 3 Quantities to be fitted are the log values of the peak enrichment scores, as given by PeakSeq. We can now solve the regression problem defined by pairs for. Linear regression is a simple and popular approach, but is prone to overfitting. Hence, we choose to regularize the model with L1-norm, i.e., we want to minimize the sum of squared errors and the L1-norm of the regression coefficient vector:where is a user-defined regularization coefficient. The L1-norm log linear regression is able to remove redundant or uninformative features, and to select a small number of features that best explain the fitted quantity. In our case, the features correspond to DNA motifs and hence, the result of this step is a set of motifs that best explain the binding signal values from ChIP-Seq dataset. We use Lasso, a popular algorithm for solving L1-norm log linear regression. Lasso is available as part of the LARS package for R (<http://www- stat.stanford.edu/~hastie/Papers/LARS/>). ## Other regression methods In this section, we present alternatives to the L1-norm log linear regression: linear least squares regression, ridge regression, partial least squares regression, and principal component regression. All these regression methods are used in the following way. Once a regression model is fitted to the peak dataset, we rank the regression coefficients with respect to their absolute values. Using this ranking, the top motifs are the potential TFBMs. ### Linear least squares regression The linear least squares regression is the simplest regression approach. It fits a linear model to the dataset by minimizing the sum of squared errors. Its difference with the L1-norm log linear regression (equation 1) is the absence of a regularization term. Therefore, the linear least squares regression is more prone to overfitting when the regression problem contains more dimensions than samples. ### Ridge regression The ridge regression minimizes, where the regularization term is, i.e., the Euclidean norm of. It is quite similar to the L1-norm log linear regression, and their main difference lies in the regularization term. The ridge regression seeks a solution with a low Euclidean norm. Although the Euclidean norm is a protection against overfitting, it does not favor sparse solutions (i.e., solutions with many motifs) as the L1-norm log linear regression does. ### Partial least squares regression and principal component regression The partial least squares regression and the principal component regression are two approaches of the same idea; they perform linear regression using the low- dimensional data matrix instead of the initial data matrix. This approach avoids overfitting problems. Therefore, the partial least squares regression and the principal component regression have been widely used in problems containing several dimensions (i.e., motifs) and few samples (i.e., peaks). In the principal component regression, the low-dimensional data matrix contains the most information about the initial data matrix (according to the singular value decomposition of). In the partial least squares regression, the low- dimensional data matrix is calculated using both the initial data matrix and the peak score vector. In both cases, linear regression is performed using instead of the initial data matrix. Both partial least squares regression and principal component regression are available as part of the PLS package for R (<http://mevik.net/work/software/pls.html>). Once the regression coefficients have been computed in the low-dimensional space, they are mapped back in the original motif space. Then, these coefficients can be used to identify potential binding motifs. ## Input ChIP-Seq datasets The ChIP-Seq dataset we used is provided by and is publicly available (<http://www.camda2009.org/>). The dataset provides various information about each peak, including the peak score, the peak center (for STAT1), and the Q-value that reflects the significance of the peak. The Q-values are derived from the P-values. First, they compute the P-values that reflect the significance of peak enrichment in the number of DNA tags, compared to control samples. These P-values are computed using the binomial distribution. Then, to account for multiple hypothesis testing, the Q-values are derived from the P-values. See for more details. For STAT1, we use -bp windows around the peak centers to define the peak sequences. For RNA Polymerase II, the peak centers are not available and thus, we use the peak start and peak end coordinates to define the peaks. When the length of the resulting sequence is less than bp, we enlarge it in both directions in order to reach bp length. When the length is more than bp, we trim it in both directions in order to reach bp length. As a result, all the RNA Polymerase II peak sequence lengths lie between and bp. ## Evaluation of prediction performance PeakRegressor predicts the peak scores and therefore, we have two different values for each peak. The “true” peak score is the score provided by, and is derived from the frequency of reads of ChIP-Seq data. The predicted score is computed by PeakRegressor using the peak sequence information. Ideally, the predicted score should be equal to the true score. We use correlation coefficients to evaluate the prediction quality of PeakRegressor. ## Experimental protocol For L1-norm log linear regression and ridge regression, we have to set the regularization parameter. First, we define for. Then for each value of, we perform a 30-fold cross-validation. In each fold, we split the dataset into a training set and a test set, with a ratio. The optimal value for is the one which corresponds to the lowest prediction error on the test set. All the results of L1-norm log linear regression and ridge regression are averaged over the 30-fold cross-validation. For partial least squares regression and principal component regression, the experiments were limited by the slowness of both methods. First we have to set the number of components used for regression. We tried, and performed a 30-fold cross-validation for each value of. In each fold, we split the dataset into for training and for testing. All the results of partial least squares regression and principal component regression are averaged over the 30-fold cross- validation. The authors thank the anonymous CAMDA reviewers for their helpful comments. [^1]: Conceived and designed the experiments: JFP WF. Performed the experiments: JFP HH TT. Analyzed the data: JFP HC WF. Wrote the paper: JFP HC WF. [^2]: Takeaki Taniguchi is employed by the Mitsubishi Research Institute, Inc. There are no patents, products in development, or marketed products related to this research, and his involvement does not alter the adherence to all the PLoS ONE policies on sharing data and materials.

# Introduction Worldwide 1.5 billion cattle (FAO, 2013) are negatively affected by extreme temperature changes with more frequent heat waves during summer periods. Heat stress leads to decreased milk production, reduced reproduction rate and growth of dairy cows. The economic losses caused by thermal stress was predicted to cost the US \$879 million annually, Australia 6,838–11,986 \$/per year for the cattle herd and milk production in Europe is expected to drop by 5.7–7%. Extended cooling costs and feeding systems for dairy cows conflict with the growing demand on milk and beef production. Metabolic heat production of dairy cows increases with the level of milk synthesis, making high-yielding dairy cows extremely susceptible towards environmental heat, whereas non-lactating cows being e.g. in the late pregnant period, produce less metabolic heat and are less susceptible to environmental heat. When ambient temperatures exceed the thermoneutral zone of lactating dairy cows, feed intake declines which contributes to reduction in milk production and loss of body weight. Also, exposure of non-lactating, late-gestating cows to heat adversely affects milk yield, milk constitute contents, metabolic health, embryonic development, and trigger carry-over effects in the subsequent lactation. Besides the reduction in energy expenditure, metabolic adaptation to thermal stress comprises also a shift in post-absorptive metabolism and nutrient partitioning aligned to reduce endogenous heat production. It has been proposed that lack of adipose tissue mobilization during heat stress reduces metabolic heat production from fatty acid oxidation and that metabolic fuel selection shifts towards glucose utilization in lactating dairy cows. We have recently shown using indirect calorimetry that whole body fat oxidation is blunted in cows experiencing heat stress, while fat oxidation increased in pair-fed cows at thermoneutrality. Metabolic responsiveness towards high ambient temperatures, however, depends on the physiological stage. In contrast to lactating cows, adipose tissue of non-lactating, late pregnant cows is not refractory to lipolytic, adrenergic stimuli, and the rate of amino acid degradation was lower than in the postpartal stage. However, the molecular mechanisms underlying the different adaptation processes during lactation and late pregnancy have not been thoroughly evaluated yet. The liver is the main organ involved in glucose homeostasis by producing glucose from ruminal propionate. During early lactation, gluconeogenesis and glycogenolysis are typically increased to provide glucose for milk lactose production. The increase in gluconeogenesis is much more pronounced in cows calving during a hot summer as compared to a cold spring period and presumably involves increased utilization of lactate and alanine for glucose synthesis. Furthermore, hepatic mRNA expression of genes encoding fatty acid oxidation increases during the transition period from late pregnancy into lactation. When cows, however, experience heat during the transition period, expression of genes associated with fatty acid oxidation is downregulated compared to control counterparts. A second major source of stored glycogen but also amino acids is the skeletal muscle. Heat stress seems to have global effect on the amino acid metabolism resulting in an increased mobilization from skeletal muscle protein, but the molecular mechanisms underlying muscle proteolysis have not been investigated yet. As the muscle has a reduced capacity to oxidize fatty acids it presumably relies on circulating and glycogen-stored glucose for energy needs. Regulation of pyruvate entry to the TCA cycle seems to play a major role to facilitates lactate and pyruvate-alanine flux to hepatic gluconeogenesis, although its contribution to cellular and system energetic homeostasis is unclear. Therefore, the main objective of the current study was to evaluate molecular adaption of major catabolic and anabolic pathways in late pregnant and early lactation dairy cows to thermal stress conditions. # Materials and Methods ## Animal experiments All procedures were approved by the ethics committee of the State Government in Mecklenburg-West Pomerania, Germany (LALLF M-V/TSD/7221.3–1.1-074/12). As described earlier, 14 German Holstein cows genotyped for HDP70.1 5´UTR 895 were grouped to heat-stressed (HS, n = 7) or pair-feeding (PF, n = 7) group. All cows were at the end of the 2^nd parity (ante partum, ap) and not milked within the 7 weeks prior to the expected calving date. Animals received a total mixed ration twice daily (at 0700 h and 1500 h). Both groups passed through a 13-day trial once in ante partum (HSap and PFap) and post-partum stage (HSpp and PFpp). Animals were halter-trained and well adapted to climate chambers with a light cycle ranging from 0600 to 1900 has described previously. Three weeks before and after parturition, the 13-day trial consisted of two 6 days periods P1 and P2 separated by one day of thermal transition. In the experimental period P1, both HS and PF groups were exposed to the same climate conditions (15°C, 63±1% relative humidity (RH) resulting in a temperature-humidity-index (THI) of 60) with *ad libitum* feeding. On the following transition day, the air temperature was continuously increased to permanent 28°C for HS, but remained at 15°C for PF animals (experimental period P2). RH adjusted within 24 h to 52±2% for HS animals (THI = 76). THI was calculated as described earlier. Feed intake was recorded daily. On the transition day and the following 6 days of period P2, reduction of daily ad libitum intake of HS cows was calculated as percentage of the mean daily intake in P1 to provide the same amount of feed to PF cows during P2. Cows had free access to water, which was tempered to 28°C for HS animals during P2. In the pp period, cows were milked at 0630 h and 1630 h and milk yield was determined daily. Due to severe sickness of individual cows, which were withdrawn from trial, groups amounted to: HSap n = 7, PFap n = 6, HSpp n = 6, PFpp n = 6. ## Blood sampling and Analyses At the first day of P1, cows were equipped with an indwelling jugular catheter (Certofix mono; B.Braun, Melsungen, Germany). Before morning feeding, daily blood samples were collected during period P1 and P2 into 9 ml-monovettes (Sarstedt, Nümbrecht, Germany) containing EDTA. Blood samples were centrifuged immediately after collection at 1,570 x g for 20 min at 4°C to obtain plasma which was stored at -80°C before analysis. Plasma insulin concentrations were determined by RIA (#1257; DRG Diagnostics, Marburg, Germany). Intra-assay variation was 3.7% and inter-assay variation 5.0–6.0%. Glucagon was detected by RIA (GL-32K; Linco Research, St. Charles, MO, USA) with an intra assay variation of 3.4% and an inter assay variation of 16.3–28.6%. Acyl ghrelin was measured as described previously using a RIA kit (GHRA-88HK; Linco Research, St Charles, Billerica, MO, USA). The intra-assay variation was 4.0% and the inter-assay variation 6.8–16.6%. Plasma urea, albumin, alanine aminotransferase (ALT), β-hydroxybutyric acid (BHBA) and lactate were analyzed photometrically by ABX Pentra 400 (Horiba Medical, Kyoto, Japan) using the following kits (urea LT- UR0010, Labor+Technik Lehmann, Berlin, Germany; albumin A11A01664, ALT A11A01627; BHBA A11A01667; lactate A11A011721, Axonlab, Stuttgart, Germany). Plasma amino acids concentrations were analyzed by high-performance liquid chromatography as described recently. ## Biopsies and RT-RCR *Semitendinosius* muscle and liver biopsies were taken before morning feeding on transition day (representing period P1) and on day 6 after PF or HS challenge (representing period P2), both ap and pp. Tissues were snap frozen in liquid nitrogen and stored at -80°C until analysis. Tissues were mortared under liquid nitrogen and total RNA was extracted from 50 mg tissue powder with TriFast Reagent (Peqlab, Erlangen, Germany). RNA isolation was performed by using an RNeasy kit (Qiagen, Valencia, CA) and RNA quality was assessed using an Agilent 2100 Bioanalyzer, yielding RIN factors for muscle and liver between 5.0 and 7.9 (median 6.95) and between 6.0 and 7.8 (median 7.3), respectively. First strand cDNA synthesis (750 ng total RNA) was completed using 2400 U RevertAid Reverse Transcriptase (Thermo Fisher Scientific, Dreieich, Germany) and 250 pmol random primers (Metabion International, Planegg/Steinkirchen, Germany). cDNA was purified with High Pure PCR Product Purification Kit (Roche, Basel, Switzerland) and stored at -80°C until use. Transcriptional expression was quantified by real-time PCR. Primers were designed using Primer3 Plus software or PrimerBLAST. For liver, one PCR reaction contained 2 μl diluted cDNA (10 ng/μl), 5 μl H₂O PCR grade, 400 nM of each primer, and 2 μl Light Cycler FAST DNA Master PLUS SYBR Green I Reaction Mix (Roche, Basel, Switzerland) and was carried out in duplicates using LightCycler 2.0 (Roche, Basel, Switzerland). For muscle tissue, qPCR reactions contained 2 μl diluted cDNA (10 ng/μl), 2 μl H2O PCR grade, 5 μl Luminaris Color HiGreen qPCR Master Mix (Roche, Basel, Switzerland) and 400 nM of each primer. Amplicons were sequenced on an ABI 3130 Genetic Analyzer (Life Technologies GmbH, Darmstadt, Germany) for quality control purpose. The obtained sequences were blasted using NCBI BLAST tool to confirm sequence identity. The efficiency of amplification was calculated using LinRegPCR software, version 2014.4 (Academic Medical Centre, Amsterdam, Netherlands;, yielding efficiency values between 1.84 and 1.91. Data were quantified by qbasePlus software (Biogazelle, Gent, Belgium). The gene expression stability of five candidate reference genes was determined in geNorm. For final analysis were used *lipoprotein receptor-related protein 10* (LRP10) and *hippocalcin-like1* (HPCAL1) for liver and LPR10, *ceroid- lipofuscinosis neuronal 3* (CLN3) and *emerin* (EMD) for muscle tissue as references genes. ## Western Blot Muscle tissue (50 mg) was homogenized in lysis buffer containing 50 mM Tris-HCl (pH 7.8), 1 mM EDTA, 10 mM NaF, 1% IGPEAL CA-630, 0.1% Triton X100, 0.5% deoxycholic acid (DCA) and 0.1% SDS. Protein concentrations were measured using Bradford method. Equal amounts of protein (50 μg) were separated by SDS-PAGE and transferred to a nitrocellulose membrane (Whatman Protran BA 83, Dassel, Germany). Membranes were blocked with 3% bovine serum albumin (BSA) and milk powder in Tris-buffered saline (TBS) containing 0.1% Tween 20 (TBST) for 1 h and then incubated overnight with the following primary antibodies: phosphorylated AMP-activated protein kinase Thr172 (pAMPK; \#2535, Cell Signaling Technology, Cambridge, UK), AMPK (#2603, Cell Signaling Technology, Cambridge, UK), short/branched chain acyl-CoA dehydrogenase (ACADSB; LS-C81878/19265, Lifespan Bioscience Seattle, WA, USA) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH; PA1-988, Thermo Scientific, Waltham, MA, USA). After incubation, membranes were washed in TBST, incubated for 1 h at room temperature with the corresponding secondary antibody (goat anti rabbit IgG HRP, sc-2004, Santa Cruz Biotechnology, Dallas, USA or donkey anti sheep IgG HRP, ab97125, Abcam Cambridge, USA) and washed again in TBST. Chemiluminescent reagents were applied and blots were exposed to hyperfilms (GE healthcare, Chalfont St Giles, Buckinghamshire, UK). Hyperfilms were scanned and quantified using ImageJ (version 1.49). The ratio of pAMPK/AMPK and ACADSB/GAPDH, respectively, were calculated. ## Statistical Analysis Repeated measures data were analyzed for the effects of group, day, and their interaction during period 2 as a completely randomized design using PROC MIXED with repeated measurements analysis with day as the repeated effect. Data were tested for parametric or non-parametric distribution. For the comparison of group differences on a daily basis, a Tukey-Kramer test was applied. For PCR and Western Blot analyses, differences between periods P1 and P2 in the same group were analyzed using the Wilcoxon signed rank sum test included in the UNIVARIATE procedure of SAS (Version 9.4, SAS Institute Inc., Cary, NC, USA). Analysis of differences between HS and PF in P2 at the same productive stage was performed using the exact Wilcoxon-Mann-Whitney test of the NPAR1WAY procedure. We did not test for difference between ap and pp, thereby ignoring carry over effects. Repeated measurements are given as mean ± standard error (SEM). Results were considered as statistical significant at P\<0.05 and trends between 0.05\<P\<0.07. # Results ## Environmental heat downregulated PDK2 mRNA in skeletal muscle After heat exposure, expression of pyruvate dehydrogenase kinase isozyme 2 (*PDK2*) mRNA tended to be reduced (P\<0.063) in late gestation and was significant decreased (P\<0.01) in early lactation in comparison to PF. Lactate dehydrogenase A and B (*LDHA*, *LDHB*) and *PDK4* remained unaltered between HS and PF in both stages. However, *PDK4* mRNA increased from P1 to P2 in late- gestating cows after exposure to HS (P\<0.05), whereas *PDK2* mRNA did not. In the pp period, only *PDK2*, but not *PDK4* mRNA expression decreased in HSpp, but not PFpp cows (P\<0.05). A significant reduction of *LDHA* and *LDHB* mRNA expression was observed in late gestation, but not in early lactation of HS cows. Furthermore, *LDHA* mRNA expression was significant reduced in PFpp cows (P\<0.05;). Activation (phosphorylation) of AMP-activated protein kinase (AMPK) did not show differences between HS and PF, but comparing P1 to P2 showed an increased phosphorylation in HSap cows (P\<0.05). ## Heat stress prevented pair-feeding induced increased transcription regulating fatty acid oxidation in skeletal muscle While mRNA expression of mitochondrial acyl-CoA dehydrogenase, very long chain (*ACADVL*) tended to increase after PF (P\<0.063), cows exposed to HS had significantly reduced *ACADVL* expression as compared to late gestation PF cows (P\<0.05). This effect could not be observed during early lactation. Also, protein expression of short/branched chain acyl-CoA dehydrogenase (ACADSB) was significantly lower in HS compared to PF cows during late gestation (P\<0.01), while it was unaffected in early lactation. However, mRNA abundances of mitochondrial 3-hydroxyacyl-CoA dehydrogenase *(HADH*) peroxisomal acyl-coenzyme A oxidase 1 (*ACOX1*), acetyl-coenzyme A acyltransferase 1 (*ACAA1*) and remained unaltered in late-gestating HS cows. In early lactation, HS cows showed a significant increase in mRNA expression of the fatty acid activating enzyme *ACAA1* (P\<0.05), while mRNA abundances of the oxidative enzymes *ACADVL*, *ACOX1* and *HADH* were not affected in early lactation HS or PF cows. The mRNA expression of peroxisome proliferator- activated receptor gamma coactivator 1-alpha (*PPARGC1A*), a master regulator of mitochondrial biogenesis, declined from P1 to P2 in late-gestating HS cows (P\<0.05), whereas in early lactation mRNA abundance did not change. ## Heat stress induced mRNA expression of FOXO3 controlling proteolysis in skeletal muscle The mRNA abundance of forkhead box O3 (*FOXO3*), a transcriptional regulator of protein degradation tended to be elevated in late-gestating HS (P\<0.07) and was significantly upregulated in early lactation HS cows (P\<0.05). Pair-feeding did not affect *FOXO3* abundances, resulting in significantly higher FOXO3 expression in HS compared to PF cows in both late pregnancy and early lactation. The mRNA expressions of the proteolytic enzymes alanine aminotransferase 2 (*GPT2*) and calpain 1 (*CAPN1*) decreased after PF but was not affected by HS during early lactation, however, differences between HSpp and PFpp groups did not reach significance. *CAPN1* mRNA decreased from P1 to P2 in HSap and PFpp cows (P\<0.05) and *UBA52* mRNA expression declined after HS exposure in the pp stage (P\<0.05). However, mRNA expressions of the proteolytic enzymes ubiquitin B (*UBB*), and calpain 2 (*CAPN2*) were not affected by HS or PF conditions. ## Heat stress influenced hepatic PCK1 and PC mRNA expression Pyruvate carboxylase (*PC*) mRNA expression increased or tended to increase from P1 to P2 in HS and PF cows, but only in early lactation PC abundance was significantly lower under HS compared to PF conditions (P\<0.05). The expression of cytosolic phosphoenolpyruvate carboxykinase 1 (*PCK1*) increased from P1 to P2 in HS but tended to decrease in PF cows before parturition, however, there was no significant difference between HSap and PFap cows. In early lactation, PCK1 abundance tended to decrease after HS (P\<0.063) but not after PF, without reaching significant group differences. Mitochondrial phosphoenolpyruvate carboxykinase 2 (*PCK2*) did not differ between HS and PF in both stages. Furthermore, *LDHA* mRNA expression tended to increase in P2 relative to P1 in HS in late gestation (P\<0.063), whereas in early lactation no changes could be detected. ## Heat stress did not alter mRNA expression of peroxisomal and mitochondrial fatty acid oxidation in liver Comparison between HS and PF did not reveal differences in peroxisomal *ACOX1*, acyl-CoA oxidase 2 (*ACOX2*), *ACAA1*, catalase (*CAT*), *ACADVL*, *HADH*, 3-ketoacyl-CoA thiolase (*ACAA2*) and peroxisome proliferator-activated receptor alpha (*PPARA*) mRNA abundances in late-gestating and early lactation animals. By contrast, early lactation HS but not PF cows showed a reduction in peroxisomal *ACOX1* (P\<0.05) and *CAT* (P\<0.063) from P1 to P2, whereas CAT abundance also tended to decrease in late pregnancy after PF (P\<0.063). ## No transcriptional adaption of the urea and TCA cycle in liver during heat stress HS and PF had no effect on the mRNA abundance of enzymes of the urea and the TCA cycle such as argininosuccinate lyase (*ASL*), mitochondrial carbamoyl-phosphate synthase (*CPS*) and citrate synthase (*CS*) in late gestation and early lactation. The mRNA expression of NADH dehydrogenase subunit 2 (*ND2*) and ATP synthase H⁺ transporting, mitochondrial F1 complex, beta polypeptide (*ATP5B*), did not differ between HS and PF, but *Atp5b* mRNA abundance was lower in P2 compared to P1 in HSpp cows (P\<0.05), without reaching significant difference between groups. ## Heat stress increases plasma alanine in late gestation but decreased plasma glycine and proline in early lactation Plasma alanine concentrations were 1.3-fold higher after HS relative to PF in late gestation (P\<0.05), but did not differ in early lactation. Glycine and proline concentrations were significantly lower whereas phenylalanine was higher in HS relative to PF cows in early lactation (P\<0.05). Plasma amino acids involved in the urea cycle, namely citrulline, arginine and ornithine declined from P1 to P2 in HSap cows (P\<0.031). With the exception of arginine, these decreases could also be observed in PFap cows (P\<0.031). Furthermore, glutamate, glutamine, asparagine, tyrosine, histidine, threonine, tyrosine, valine, methionine and lysine declined from P1 to P2 in PF cows during late gestation (all P\<0.031), but not in early lactation. However, the sum of ketogenic amino acids were lower in P2 compared to P1 in HSap (P\<0.031), whilst in PFap cows there was only a trend (P\<0.063). Furthermore, threonine, tryptophan and leucine concentrations declined, whereas glycine concentration increased in HSap cows from P1 to P2 (P\<0.031). In the early lactation period, significant reductions in plasma amino acid concentrations were only detected for glutamate and glycine plasma in HS cows (P\<0.031). ## Increased plasma urea concentrations due to thermal stress in late gestation but not in early lactation Plasma urea concentrations increased from P1 to P2 in early lactation HS cows and were significantly higher as compared to PF cows (P\<0.05), whose concentrations only marginally increased (P\<0.063;). On the other hand, plasma urea was unaffected in the ap period. Albumin, alanine amino transferase (ALT), and beta-hydroxybutyrate (BHBA) did not differ between HS and PF but plasma albumin tended to be lower in P2 compared to P1 in HSpp cows (P\<0.063). Analysis of plasma BHBA revealed an increase from P1 to P2 in HSap cows (P\<0.05), which was not observed in PF animals. Furthermore, PFpp cows showed greater plasma BHBA concentrations in P2 (P\<0.05). In the ap period, plasma ALT concentrations declined from P1 to P2 in HS and PF (P\<0.05). HSpp but not PFpp cows showed lower ALT concentrations in P2 relative to P1 (P\<0.05). ## Decreasing insulin concentration results in the reduction of the insulin/glucagon ratio in HSap cows In late gestation, plasma insulin concentrations differed between HS and PF cows during P1, however, only in HS exposure led to a significant decrease during P2 (P\<0.001) while it remained constant in PF cows (ΔHS 15.4 vs. ΔPF 7.59 μU/ml; P\<0.001). In early lactation, insulin concentrations did not change between groups in P1 or P2. Glucagon concentrations did also not differ between HS and PF cows during both stages. The insulin/glucagon ratio declined in HSap cows during P2 (P\<0.001). Furthermore, HS cows showed higher lactate concentrations compared to PF cows in the ap (P = 0.05) but not in the pp period. In addition, plasma ghrelin was not affected by HS or PF and remained unaltered in ante partum and pp stage. # Discussion Periods with longer heat waves are predicted to occur more often leading to tremendous changes in animal farming in the next decades. Due to their enormous metabolic rate, but also due to negative energy balance during early lactation, dairy cows are particularly sensitive to heat during the transition period from pregnancy to lactation. The perspective of this study was to determine the effect of heat stress on metabolism and expression of key genes involved in energy metabolism of liver and muscle tissue of dairy cows during late gestation and early lactation. ## Heat stress in late gestation Feed intake of late-gestating and early lactation cows was shown to decline by \~50% during heat periods. Under the conditions of reduced energy intake, muscle metabolism enters a glucose-sparing state at thermoneutrality, while it appears to increase aerobic glycolysis during heat stress. Muscle glucose metabolism is among others regulated by the activation of AMPK upon increase in cellular AMP/ATP ratio. Muscle pAMPK is not significantly different between HS and PF cows in late gestation despite AMPK phosphorylation increased after one week of heat stress, but not after pair-feeding. This result indicates that heat stress might regulate intracellular energy balance via AMPK activation to shut down energy consuming anabolic processes and induce catabolic pathways. The activity of the pyruvate dehydrogenase complex (PDC) is regulated by various PDKs while their individual functions are still not entirely resolved. *PDK2* mRNA expression tended to be greater in PF compared to HS cows, indicating decreased PDC activity and reduced conversion of pyruvate to acetyl-CoA in PF cows. This regulation allows PF but not HS late-gestating cows to utilize acetyl-CoA deriving from fatty acid oxidation, an assumption supported by the greater muscle *ACADVL* mRNA and ACADSB protein abundance as well as greater whole-body fat oxidation in PF compared to HS cows. Conclusively, HS cows limit heat production by preventing increase in fatty acid oxidation in skeletal muscle tissue. Phosphorylation of AMPK signals to activate *FOXO3* gene expression. Our data show that HS cows express more *FOXO3* mRNA compared to PF animals. FOXO3 is involved in the regulation of muscle proteolysis, which is reflected by a significant increase in plasma 1-/3-methylhistidine after heat exposure but not PF of late-gestating animals. Although *FOXO3* mRNA expression increased under thermal stress, higher mRNA abundance of calpains (calcium- dependent, non- lysosomal proteases) and ubiquitin proteasome system could not be detected. However, lack of altered mRNA expression does not exclude the involvement of the lysosomal pathway degrading muscle proteins, but this assumption needs further investigation. Total plasma amino acid concentrations were not altered after HS or PF challenge, but interestingly, plasma alanine concentration did not decline after HS in pregnant cows as they do in PF counterparts. This result might indicate a continuously proceeding Cahill cycle during HS, presumably to meet the high amino acid requirement of the fast-growing near term-fetus. Activation of Cahill cycling requires sufficient glucose supply and the shut-down of pathways competing for the use of muscle pyruvate. Accordingly, we found reduced LDHB mRNA expression in HS but not in PF cows, indicating less conversion of pyruvate to lactate and back. Also, plasma lactate concentrations were greater in late gestating HS than PF cows, further pointing to an alleviation of Cori in favor of Cahill cycling. Furthermore, hepatic PCK1 mRNA expression increased after HS but declined after PF, yet without reaching significance level between groups. Despite of this, opposed regulation of *PCK1* expression after HS and PF challenge may indicate that HS late-pregnant cows favor the cytosolic phosphoenolpyruvate (PEP) pathway, probably to export NADH to the cytosol necessary for gluconeogenesis. Favoring cytosolic over mitochondrial PEP production would allow greater mitochondrial NADH oxidation and thereby reducing mitochondrial heat production, an effect which does not occur in metabolically challenged early lactation cows. In addition, hepatic *PC* mRNA expression increased from P1 to P2 in both HS and PF animals and was not different between groups indicating that the level of feed intake but not heat stress *per see* regulates PC. In line with this, Shahzad et al. (2015) reported that greater *PC* mRNA abundance in transition cows with summer compared to winter calving and concluded that the expression responses could be related to lower DMI. ## Heat stress in early lactation Homeorhetic adaptation to early lactation involves reduced glucose utilization and increased lactate efflux from skeletal muscle, thereby decoupling the Cori cycle to contribute to the sudden enormous glucose requirements of the mammary gland. We found that HS in early lactation cows decreased skeletal muscle *PDK2* expression. Although the precise function and induction of *PDK2* has not been fully elucidated, recent findings indicate that PDK isoenzyme is physiologically important to activate skeletal muscle PDC and pyruvate oxidation. Expressions of enzymes coding for mitochondrial β-oxidation and its transcriptional coactivator PGC1α were not affected by HS or PF, whereas *ACAA1*, a β-ketothiolase involved in the final step of peroxisomal fatty acid degradation was upregulated in HS while tending to be downregulated in PF cows, pointing to increased long-chain fatty acid degradation in muscle peroxisomes. The degradation of muscle protein may start in late gestation or early lactation. With the exception of *UBA52*, cows in early lactation exposed to heat did not change the expression of proteasomal genes despite increased *FOXO3* abundance. These data imply that HS in early lactation does not change skeletal muscle proteolysis on mRNA abundance despite reduced plane of nutrition. In line with this, HS has been shown to exert a greater effect on protein synthesis than protein breakdown in chickens. Adaptation of hepatic gluconeogenesis in early lactation occurs in a different way than in late pregnancy (see above). In early lactation, *PC* mRNA abundance is greater in PF than HS cows. We conclude that when the supply of propionate for hepatic gluconeogenesis is limited, lactate and/or amino acid-derived pyruvate are used more readily as gluconeogenic substrates under thermoneutral PF conditions. However, *PCK1* expression tended to be reduced in early lactation but to be increased in late pregnancy after heat exposure, indicating that the transcriptional regulation of this gene under hot temperatures depends on the physiological status. However, downregulation of *PCK1* in HSpp but not PFpp cows suggests less utilization of amino acid-derived pyruvate in HS that PF early lactating cows. Furthermore, cows in early lactation, but not late-pregnant cows respond to HS by reducing the mRNA expression of *Atp5b*, a subunit of the ATP synthase. This result indicates that lactating cows reduce metabolic heat production by reducing ATP production, whilst oxidation through the respiratory chain complex I, as indicated by unchanged *ND2*, remained constant. Whether uncoupling of the electron transport chain leads to reduced ATP synthesis needs to be determined in future studies. In early lactation, plasma proline and glycine concentrations were lower after HS compared to PF challenge, suggesting a more intensive utilization of these amino acids as precursors for gluconeogenesis during high ambient temperatures, but this assumption needs further investigation. Our results also contrast the findings by Tian et al. (2015), who described greater plasma glycine and proline concentrations of mid-lactating cows sampled during the summer compared to spring season, but this effect may also be attributed conditions others than heat stress. Similar to the late gestation period, the mRNA abundances of urea cycle encoding enzymes are not affected by HS or PF in early lactation, but also in late gestation may be due to the slow responsiveness of these genes as described by Velez *et al*.. Yet, plasma urea concentrations are almost doubled in HS lactating cows. Wheelock et al. (2010) suggested that the quick increase in plasma urea nitrogen would be due to increased hepatic deamination of amino acids. Our present findings on unchanged urea cycle enzyme mRNA, however, argues against this assumption, but confirms that HS but not PF reduces kidney perfusion and enhances plasma water losses (also into milk), thus facilitating the accumulation of urea in plasma without altered amino acid deamination. Lactating cows are unable to mobilize adipose tissue and thus free fatty acids and therefore cannot increase fat oxidation during heat stress. Accordingly, mRNA abundance of enzymes involved in mitochondrial β-oxidation did not change after HS and PF. However, one week of HS showed lower *ACOX1* and *CAT* mRNA abundances of enzymes involved in peroxisomal β-oxidation, reducing peroxisomal heat production and oxidative stress. In cows that calved during hot summer compared to mild spring days, hepatic *ACOX1* was downregulated alike, but in addition to this, *CPT1A* and *PPARA* were also reduced, indicating a lower rate of total fatty acid oxidation in these HS cows. In summary, metabolic adaptation to heat stress and reduced feed intake differ between late pregnancy and early lactation of dairy cows. In liver and skeletal muscle, a shift in substrate utilization is required to maintain gluconeogenesis for fetus development in late gestation and milk production in early lactation, through the reduction of endogenous heat production. This knowledge is crucial for urgently needed strategies mitigating heat stress, risks for diseases and economic losses, while sustaining animal well-being and performance. # Supporting Information We thank S. Wiese, C. Arlt, U. Wiedemuth for technical support and the staff at the FBN ‘Tiertechnikum’ for assistance with the animal care and sample collection to perform biochemical analyses and protein/ gene expression studies. This study was supported by the core budget of the Leibniz Institute for Farm Animals (FBN). The publication of this article was funded by the Open Access fund of the Leibniz Association. The authors declare no competing interests exist. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** BK. **Formal analysis:** FK OL ME JW. **Investigation:** FK OL. **Project administration:** BK. **Resources:** JW. **Supervision:** BK. **Validation:** FK OL. **Visualization:** FK. **Writing - original draft:** FK BK OL. **Writing - review & editing:** FK BK.

# Introduction Breast cancer is the most common cancer types and the leading cause of cancer mortality in females in the world. It was estimated that approximately 278,800 new breast cancer cases with 64,600 deaths occurred in China in 2013. It is well known that cancer progression is driven by mutations in cancer genome. Somatic mutations in *AKT1*, *PIK3CA PTEN* and *TP53* genes were found at high frequency in breast cancer, with *PIK3CA* as 26.4%, *TP53* as 24.7%, *PTEN* as 3.8% and *AKT1* as 2.8% in the Catalogue of Somatic Mutations in Cancer (COSMIC) database. Recent large genomic landscape studies have showed that *TP53* and *PIK3CA* were the two most frequently mutated driver genes in primary breast cancer and the mutation spectrum of these four genes displayed subgroup specificity with great clinical significance in cancer classification and treatment. However, the spectrum of these four gene mutations in breast cancer is still largely unknown in Chinese population. Thus a comprehensive understanding of the prevalence and clinical characteristics of *AKT1*, *PIK3CA*, *PTEN* and *TP53* gene mutations in Chinese breast cancer patients is urgently needed. With the advance of next-generation sequencing (NGS) technologies, mutation analysis has become effective and feasible for routine clinical application in breast cancer. In this study, paired tumor and normal tissues from a cohort of 313 Chinese breast cancer patients were screened for *ATK1*, *PIK3CA*, *PTEN* and *TP53* mutations using microfluidic PCR-based target enrichment and NGS technology. Furthermore, clinicopathological characteristics of breast cancer associated with the mutations of these four genes were analyzed in parallel. # Material and methods ## Patients and tissue samples Fresh tumor and paired adjacent normal tissues (located at least 2 cm away from the site of tumor tissue) from 313 primary breast cancer patients were collected at Xiangya Hospital, Central South University from year 2013 to 2015. The clinicopathological characteristics of the 313 patients were summarized in. All breast specimens were reviewed by experienced pathologists. The breast cancer molecular subtypes were characterized based on the guideline of St Gallen International Expert Consensus (2013). All of the 313 patients have been tested for *BRCA1* and *BRCA2* mutations by NGS and validated using Sanger sequencing in our previous study. All the patients in this study were females of Chinese Han population without selection for family history or onset age. We declared that the experiments performed in this study comply with the current laws of the People's Republic of China. This study was approved by the Ethics Committee of Central South University, Changsha, China, and all participants had given written informed consent. ## Library preparation and NGS Genomic DNA of all samples were extracted using the TIANamp Genomic DNA Kit (TianGen Biotech, Beijing, China), and quantified using a Nanodrop 2000 spectrophotometer (Thermo Scientific, Wilmington, DE, USA). Totally, 60 pairs of primers were designed to amplify the complete coding regions and exon–intron boundaries of target genes, and the primer sequences were displayed in. The primers were designed using the on-line design tool, Primer3 (<http://bioinfo.ut.ee/primer3/>), by following the User Guide of Access Array^TM System for Illumina Sequencing Systems (Fluidigm, South San Franciso, CA, USA). All primers were validated by single-plex PCR with assessment of PCR products for expected size on agarose gel. The high-throughput target enrichment was performed on the Fluidigm Access Array (Fluidigm, South San Franciso, CA, USA) according to established workflows. Then the target gene libraries were sequenced on an Illumina MiSeq sequencer (SanDiego, CA, USA) using MiSeq Reagent Kit v2 (500 cycles). ## Sequencing data analysis and variant annotation The raw sequencing data were base-called and demultiplexed using MiSeq Reporter v.1.8.1 (Illumina, SanDiego, CA, USA) with default parameters and FASTQ files were generated for downstream data analysis. The adapter sequences and low quality reads were trimmed away from the raw reads using Trimmomatic v.0.32. Cleaned reads were aligned to the UCSC human reference genome hg19 using the Burrows-Wheeler Alignment tool (BWA) v.0.7.10. After alignment, the SAMtools (v.1.1) software was applied to convert the alignment files to a sorted, indexed binary alignment map (BAM) format. Base recalibration and realignment around indels was done with the GATK v3.1.1. Germline genotypes were called by the GATK UnifiedGenotyper (with paired tumor and adjacent normal tissues sample), and somatic mutations were called by MuTect (v.1.1.4) under the High-Confidence mode with default parameter settings. Both tumor and matched normal tissue samples from the same patient were sequenced together in a NGS run. The variants present only in tumor tissue samples were thus classified as somatic mutations. And variants present in both tumor and paired normal tissue samples were classified as germline mutations. We defined the final list of somatic mutations with the following filters: number of reads with the altered base in the tumor ≥10; frequency of the reads with the altered base in the tumor ≥ 5% except for variants that are also reported in COSMIC database; minor allele frequency \<0.1% in each of the following publicly available databases: 1000 Genomes (<http://www.1000genomes.org/>) and Exome Aggregation Consortium (<http://exac.broadinstitute.org/>). Variants were annotated using ANNOVAR (February, 2016) with the annotate_variation.pl script. This tool mapped variants to RefSeq genes, known variations from dbSNP 138 and annotated the predicted functional consequences of missense variants using two *silico* tools (SIFT and PolyPhen-2). Additional clinical variant annotations were obtained from NCBI ClinVar (<http://www.ncbi.nlm.nih.gov/clinvar>), and COSMIC database (<http://cancer.sanger.ac.uk/cosmic>). The reference sequences for numbering were based on the NCBI GenBank Database for *AKT1* (NM_005163.2 and NP_005154.2), *PIK3CA* (NM_006218.2 and NP_006209.2), *PTEN* (NM_000314.4 and NP_000305.3) and *TP53* (NM_000546.5 and NP_000537.3). Novel mutations were defined as variants that have neither been previously recorded in dbSNP (<http://www.ncbi.nlm.nih.gov/SNP>), ClinVar (<http://www.ncbi.nlm.nih.gov/clinvar/>), 1000 Genomes (<http://www.1000genomes.org/>), Exome Aggregation Consortium (<http://exac.broadinstitute.org/>) or COSMIC (<http://cancer.sanger.ac.uk/cosmic>), nor reported in literatures. In this study, all variants were classified according to the American College of Medical Genetics and Genomics recommendations. Variants resulted in non-functional or truncating-proteins were classified as pathogenic mutations (including stop-gain mutations, frameshift mutations and splice site mutations). In addition, we also considered variants as pathogenic mutations if they were annotated as “pathogenic” in NCBI ClinVar. The annotation and classification of the protein domains of these 4 genes was based on the NCBI’s Conserved Domain Database (CDD). ## Statistical analysis Continuous data were summarized using mean and standard deviation. The difference of age among patients with different gene mutation status was determined by the Wilcoxon Rank Sum test. And χ² test was used to compare categorical variables between groups across clinicopathological characteristics. Alternatively, Fisher’s exact test was used when χ² test was violated. The obtained *P* values were considered statistically significant if the *P* value is \< 0.05. The Holm’s procedure was used to adjust *P* values for multiple testing. All the computations were performed using the R software (version 3.1.0, [http://www.cran.r-project.org](http://www.cran.r-project.org/)). # Results ## Detection of mutations by NGS Microfluidic PCR-based target enrichment and NGS were performed to sequence the entire coding regions and exon-intron boundaries of *AKT1*, *PIK3CA*, *PTEN* and *TP53* genes in the cohort of 313 Chinese breast cancer patients. Total 120 somatic mutations were detected in 190 patients (190/313, 60.7%). The somatic mutation frequency of *AKT1*, *PIK3CA*, *PTEN* and *TP53* in this cohort was 3.2% (10/313), 36.4% (114/313), 4.8% (15/313) and 33.9% (106/313), respectively. Similarly, the somatic mutation frequency of these genes reported by TCGA was 2.4%, 35.5%, 3.2% and 35.3%, respectively. Notably, one synonymous variant (p.T125T in *TP53*) was included in this study, because it can lead to alternative splicing as previously reported. In addition, 6 germline mutations were also found (1 in *PIK3CA*, 2 in *PTEN* and 3 in *TP53*) in 6 of the 313 patients. Among these 126 mutations, 53 were considered as pathogenic (42.9%), including 52 somatic mutations and 2 germline mutations. All the somatic mutations detected in this study were confirmed in two different NGS runs. In addition, all germline mutations and the somatic mutations with allele fraction ≧20% in tumor tissues were confirmed using Sanger sequencing ( and Figs). ## Frequency and spectrum of *AKT1*, *PIK3CA*, *PTEN* and *TP53* mutations In *AKT1* gene, total 2 somatic mutations were detected in 10 of 313 patients (3.2%), both of which were missense mutations located in exon 3 within pleckstrin homology (PH) domain of the AKT1 protein. The mutation p.E17K, which occurred in 9 patients (9/10, 90%), and dominated the mutation spectrum of *AKT1*. No germline mutation was found in *AKT1*. In *PIK3CA* gene, total 17 somatic mutations were detected in 114 of the 313 patients (36.4%), which located across 7 different exons (exon 2, 5, 8, 9, 10, 14 and 21). Notably, 9 patients harbored two mutations of *PIK3CA*. Exon 10 and 21 were the two hotspot regions within PI3Ka and PI3Kc domain, mutations of which presented in 34 (34/114, 29.8%) and 70 (70/114, 61.4%) of cases, respectively. Among them, 26 patients had p.E545K mutation and 7 patients had p.E542K mutation in PI3Ka domain. Total 52 patients had p.H1047R mutation in PI3Kc domain, and 15 patients had a different p.H1047L mutation at the same spot. One novel somatic mutation p.I69N was found in the PI3K_p85B domain. In addition, one germline mutation (p.K733R) was detected in *PIK3CA*. By *in silico* analysis, it was predicted to be deleterious by SIFT and benign by PolyPhen-2. In *PTEN* gene, total 17 somatic mutations, which located across 7 different exons (exon 1, 2, 3, 5, 6, 7 and 8), were identified in 15 of the 313 patients (4.8%), and no recurrent mutations were found. Of these 17 mutations, 8 located in the phosphatase domain and 9 located in the conserved domain 2 (C2), including 8 nonsense mutations, 5 missense mutations, 3 frameshift indels and 1 splice site mutation. Notably, two patients harbored two mutations in *PTEN* (Patient 124 harboring p.Q17X and p.C211X, Patient 258 harboring p.Y27N and p.G165R). Three novel somatic mutations (p.K62X, p.N340IfsTer4 and c.635-12_636delTTAACCATGCAGAT) were detected in *PTEN*, all of which may lead to a truncated or non-functional PTEN protein. In addition, two germline mutations were found in *PTEN*. The mutation p.C136R was recorded in NCBI ClinVar database as pathogenic. Another germline mutation p.Q110E was novel and predicted to be tolerated by SIFT and benign by PolyPhen-2. In *TP53* gene, total 84 somatic mutations were identified in 106 of the 313 patients (33.9%), with 5 patients harboring two mutations. All the somatic coding mutations of *TP53* located in exon 4, 5, 6, 7, 8, 9 and 11, and additional 3 splicing variants located in intron 4 and 9 ( and). A large proportion of somatic mutations found in *TP53* clustered in the region from exon 4 to exon 8 within the DNA-binding domain, mutations of which presented in 97 cases (97/106, 91.5%). The somatic mutation of *TP53* included 62 missense mutations, 10 indels (5 inframe and 5 frameshift), 9 nonsense mutations and 3 splicing variants. Notably, 5 novel somatic mutations (p.Q136AfsTer5, p.K139_P142del, p.Y234dup, p.V274LfsTer31 and p.N310TfsTer35) were detected in *TP53*. All of these were frameshift mutations which may lead to deleterious effect on TP53 protein function. Additionally, 3 germline mutations were detected in *TP53*. The two missense mutations, p.G244S and p.P295L, were recorded in NCBI ClinVar as likely pathogenic and uncertain significance, respectively. The remaining one splicing variant c.559+1G\>A was classified as pathogenic mutations. ## Multiple-gene and recurrent mutations in *AKT1*, *PIK3CA*, *PTEN* and *TP53* Among the 190 somatic mutation carriers, 137 (137/190, 72.1%) harbored mutation in single gene. Total 51 patients (51/190, 26.8%) harbored co-mutation in two genes and 2 patients (2/190, 1.1%) harbored co-mutation in three genes. These included 2 patients (2/313, 0.6%) with co-mutations in *AKT1*-*PIK3CA*, 2 patients (2/313, 0.6%) with co-mutations in *AKT1-TP53*, 4 patients (4/313, 1.3%) with co-mutations in *PIK3CA-PTEN*, 40 patients (40/313, 12.8%) with co- mutations in *PIK3CA-TP53*, 3 patients (3/313, 1.0%) with co-mutations in *PTEN- TP53* and 2 patients (2/313, 0.6%) with co-mutations in *PIK3CA*-*PTEN*-*TP53*. No concurred mutation was observed in *AKT1*-*PTEN* and *AKT1*-*PIK3CA*-*TP53* genes. Total 25 recurrent somatic mutations were found in this study. Among them, 10 mutations each recurred in \>1% cases of this cohort of 313 patients, including 1 mutation in *AKT1* (p.E17K), 5 mutations in *PIK3CA* (p.N345K, p.E542K, p.E545K, p.H1047R and p.H1047L) and 4 mutations in *TP53* (p.A159V, p.R175H, p.R213X and p.R248W). Overall, 125 of the 313 (39.9%) patients harbored at least one of these 10 mutations accounting for 55.7% of all mutations found in *AKT1*, *PIK3CA* and *TP53*. We did not observe any recurrent mutation in *PTEN* gene in this study. All of the *PTEN* mutations only presented in one patient each. ## Association of somatic mutations with clinicopathological characteristics We analyzed correlations between somatic mutation status of the 4 genes and patient clinicopathological characteristics. Comparing mutation carriers and non-carriers, *PIK3CA* mutation carriers were significantly more likely to be ER-positive (*P* = 0.041), PR-positive (*P* = 0.004) and invasive ductal carcinoma (IDC) (*P* = 0.002). *TP53* mutation carriers had a significant higher proportion of patients to be ER-negative (*P*\<0.001), PR-negative (*P*\<0.001), HER2-positive (*P* = 0.002), IHC p53 mutation positive (*P* = 0.018) and with high Ki67 expression (*P*\<0.001) than non-carriers. No significant difference of clinicopathological characteristics was identified between mutation carriers and non-carriers of *AKT1* or *PTEN*. Furthermore, we assessed whether these somatic mutations were associated with deleterious germline *BRCA1*/*2* mutations. All of the 313 patients have been tested for *BRCA1*/*2* mutations by NGS in our previous study. As shown in, almost all of the *AKT1*, *PIK3CA* and *PTEN* somatic mutation carriers did not harbor *BRCA1/2* mutation, except that one *PIK3CA* somatic mutation carrier had a *BRCA1* mutation. Five *TP53* somatic mutation carriers co-harbored *BRCA1* mutation and one *TP53* somatic mutation carrier co-harbored *BRCA2* mutation. Notably, all of the five *BRCA1* mutation positive patients harbored *TP53* somatic mutations (*P* = 0.001). ## Somatic mutations distribution across different molecular subtypes The distribution of the somatic mutations of the 4 genes varied in different breast cancer molecular subtypes. *PIK3CA* mutations occurred at high frequency in luminal A (40.7%) and luminal B (38.3%) tumors, while relatively low in basal-like (32.5%) and HER2-enriched (18.9%) tumors. In contrast, *TP53* mutations were more common in basal-like (62.5%) and HER2-enriched (54.1%) tumors than in luminal A (11.6%) and luminal B (37.5%) tumors. *AKT1* mutations only occurred in luminal A (5.8%), luminal B (3.1%) and HER2-enriched (2.7%) tumors. *PTEN* mutations only occurred in luminal A (5.8%), luminal B (5.5%) and basal-like (7.5%) tumors. The associations between somatic mutation of the 4 genes and breast cancer molecular subtypes were analyzed. The *TP53* mutations showed significant association with breast cancer subtypes (*P*\<0.001) and had higher proportion of patients with basal-like (23.6% vs. 7.3%) and HER2-enriched (18.9% vs. 8.2%) tumors, comparing with non-*TP53* mutations. # Discussion In this study, by integrating microfluidic PCR-based target enrichment and NGS technologies, we sequenced the entire coding regions and exon-intron boundaries of *TP53* and three PI3K pathway genes (*AKT1*, *PIK3CA*, *PTEN*) in paired tumor and normal tissue samples from 313 Chinese breast cancer patients. Our results showed that somatic mutations of these genes occurred at high frequency among Chinese breast cancer patients. Previously, several studies have conducted mutational analysis including *AKT1*, *PIK3CA*, *PTEN* and/or *TP53* genes in breast cancer worldwide. However, few studies have focused on the comprehensive study of *AKT1*, *PIK3CA*, *PTEN* and *TP53* mutations altogether in Chinese breast cancer patients. Most of these studies focused on either selected hotspot sites or selected exons of these four genes. As shown in, due to the differences of detection methods and studied regions, the reported mutation frequency of these four genes varied a lot among different studies and different populations. The mutation frequencies of *AKT1*, *PIK3CA*, *PTEN*, and *TP53* in Chinese population were reported to range 0–4.4%, 7.5–38.8%, 0–4.8% and 10.0–33.9% respectively. In other populations, these frequencies were reported to range 1.4–6.0%, 7.1–45.0%, 1.0–5% and 27.2–38.8% respectively. In all studies, *PIK3CA* and *TP53* were consistently the top two frequently mutated genes, which confirmed their important role in breast carcinogenesis. In addition to high frequency of *PIK3CA* and *TP53* single-gene somatic mutation, *TP53-PIK3CA* co-mutations were detected as high as 12.8% in our cohort, compared that as 8.7% in a TCGA cohort. This co-occurrence pattern was also discovered by prior studies with frequency as 5.3% in 120 breast cancer patients and as 5.9% in 1766 breast cancer patients. Previous *in vivo* study has confirmed that *TP53* and *PIK3CA* mutations show cooperation in mammary tumor formation in mice. It have been reported that *TP53*-*PIK3CA* co-mutation carriers had worst disease-free survival comparing with non-mutation carriers, *PIK3CA*-mutation-only or *TP53*-mutation-only carriers. Since a high frequency of *TP53*-*PIK3CA* co-mutations was detected in our cohort, this mutation pattern need to be evaluated closely in clinical settings for Chinese breast cancer patient in the future. Cancer hotspot mutations carry valuable information for diagnosis, prognosis and treatment. In this cohort, total 10 mutations were found to be recurrently mutated in \>1% patients accounting for 55.7% somatic mutations in *AKT1*, *PIK3CA* and *TP53*. Of these 10 mutations, *AKT1* p.E17K, three *PIK3CA* mutations within the PI3Ka (E542K and E545K) and PI3Kc (H1047R) domains and two *TP53* mutations (p.R175H and p.R248W) within the DNA binding domain were well established hotspots in breast cancer. Additional 3 mutations (p.N345K and p.H1047L in *PIK3CA*, p.R213X in *TP53*) were also reported as hotspots by a recent study on a large number of tumors by a novel statistical algorithm. The p.H1047L mutation occurred at the same location as p.H1047R, which were also detected by a study on Chinese breast cancer patients. The mutation *TP53* p.A159V was also detected by another study on breast cancer with the frequency as 0.9% (5/560). These hotspot mutations may be important candidate target for clinical applications in cancer treatment and screening. Previous studies suggested that somatic mutation was one of the mechanisms leading to PTEN loss. In this study, the frequency of somatic mutations of *PTEN* was reported as 4.8%, while loss of *PTEN* in protein expression was reported as high as 48% in breast cancer. The reason was that other mechanisms such as promoter methylation, loss of heterozygosity, transcriptional or post- transcriptional regulation can also lead to PTEN loss. In this study, 11 out of the 17 somatic mutations found in *PTEN* were stopgain SNVs or frameshift indels which can cause truncated PTEN protein. And the other 6 *PTEN* somatic mutations were predicted to be deleterious or probably damaging. Taken together, all *PTEN* somatic mutations may lead to deleterious effect on protein function, which suggested that PTEN alteration play a critical role in breast tumorgenesis. AKT1 is a downstream mediator of phosphatidylinositol 3-kinase. In line with previous studies on Chinese breast cancer patients, we detected only one hotspot mutation (p.E17K) in the pleckstrin homology domain. Recently, it has been demonstrated that mutation *AKT1* p.E17K is a therapeutic target which is sensitive to AKT inhibitors in breast cancer patients. Thus 9 out of the 10 (90%) *AKT1* somatic mutation carriers with p.E17K mutations in this study may be good candidates for AKT inhibitors treatment. In conclusion, our results showed that somatic mutations in *AKT1*, *PIK3CA*, *PTEN* and *TP53* genes were common events in Chinese breast cancer patients and had distinct spectrum across different breast cancer subtypes. Total 60.7% of the patients harbored at least 1 somatic mutation. *PIK3CA* somatic mutations were significantly associated with ER-positive or PR-positive tumors. *TP53* somatic mutations were significantly associated with ER-negative, PR-negative, HER2-positive, *BRCA1* mutation, Ki67 high expression and basal-like tumors. These findings provided a comprehensive mutational characterization of *AKT1*, *PIK3CA*, *PTEN* and *TP53* genes in Chinese breast cancer patients with valuable implications for clinical management and optimal design of clinical trials in the future. # Supporting information [^1]: Xinwu Guo, Ming Chen, Limin Peng, Xunxun Xu, Jinliang Li, Lizhong Dai and Jun Wang are employees of Sanway Gene Technology Inc. Julia Xiaoning Day is a student from La Jolla Country Day School (La Jolla, CA, USA), who worked as an intern at Sanway Gene Technology Inc. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Bioturbation refers to the physical effects of animals on their substratum. It typically takes two forms in marine ecosystems: particle reworking and ventilation. These two processes, which primarily result from the activities of benthic (usually burrowing) macroinvertebrates, may have significant ecological implications. While sediment mixing by particle reworking occurs through faunal feeding, defecation and burrowing activities, gallery flushing with water by ventilation is primarily conducted for respiration and feeding purposes. Ventilation has ecological importance both below and above the sediment surface. This is particularly true when it occurs in blind-ended burrows because water pumped by the animal is forced to percolate through the porous sediment around the burrow. This advective bioirrigation and its oxidative transport have important consequences for biogeochemical cycles. Descriptive bioturbation studies are challenging because they attempt to describe biologically driven dynamic 3D processes that occur in an opaque sediment matrix. Although various techniques have been developed for this purpose, no single experimental approach has yet provided dynamic 3D images of bioturbation. Facing the lack of experimental options, the approach is currently to extract and combine the results obtained from a selected number of existing methods depending on the question addressed. Although this approach has been used with success, it has obvious limitations. To date the most complete spatial and temporal description of sediment bioirrigation has been achieved using dynamic 3D models but these cannot stand alone as they lack proper experimental validation. We present the hybrid medical imaging technique, positron emission tomography/computed tomography (PET/CT) to simultaneously measure ventilation and visualize biologically driven porewater advection in 4D. PET/CT is a radiologic procedure that merges two imaging techniques: (CT) computed tomography with transmission of x-rays and (PET) positron emission tomography with emission of gamma rays. These techniques have separately proven useful in marine science, but they have never been combined. While CT scan provides 3D structural information of biogenic structures, the PET component offers 3D information of porewater flow dynamics. A major advantage of these techniques is their non-destructive and non-invasive nature. To prove the applicability of PET/CT, we obtained a dynamic 3D visualization of porewater advection induced by the burrowing polychaete *Arenicola marina* (or lugworm). *A*. *marina* typically lives in deep (\>15 cm) blind-ended burrows, which are ventilated at a rate of about 1 ml water/min (;). The pumping activity and the associated bioirrigation generated by this species have been extensively studied because of its omnipresence in European coastal waters; making it a perfect study object for a comparative methodological study. ## Basic principles of PET/CT A PET/CT scanner combines two medical imaging techniques: computed tomography (CT) and positron emission tomography (PET). CT transmits x-rays through the studied object, whereas PET detects photons generated when positrons emitted by a radionuclide tracer inside the object is annihilated with electrons. Both techniques are tomographic *i*.*e*. the scanner generates a series of 2D cross- sections (thin slices), that are superposed by a computer algorithm to create a full 3D image of the object. Series of 3D images are acquired repeatedly over time to create a dynamic 3D image *i*.*e*. 4D. The CT scanner provides morphological information of the studied object, whereas the PET scanner yields information on the accumulation or transport of a radioactive tracer inside the object. Noticeably, the tracer is only used in very small quantities enabling the study of objects under physiologically normal conditions. PET/CT was commercialized in 2001, and has been increasingly used in various fields of medicine, but especially in oncology for detection, staging and response evaluation of a wide array of cancers. The following descriptions introduce the general terminology and basic principles of CT and PET. ## Computed tomography (CT) CT (computed tomography) is based on X-ray radiography and uses the X-ray attenuation properties of an object depending on its bulk density and its effective atomic number. Thus, each voxel (i.e. 3D pixel) on the resulting X-ray image has a distinct contrast corresponding to the average specific attenuation of the voxel through the object. When the X-ray tube rotates in a spiral around the studied object, it allows the reconstruction of an image with good contrast at sub-millimeter resolution for each slice of the object that is read in Hounsfield units (HU). ## Positron emission tomography (PET) Once the CT has been completed, the object is moved slowly and accurately by a motorized table to the PET where a radioactive tracer is (or has been) administered. Tracers may be a single radionuclide- for example fluoride ¹⁸F or an element that is incorporated into virtually any molecule—for instance H₂¹⁵O. One key feature of the radioactive isotopes is that the chemical characteristics of the element are not altered no matter which isotope is applied. The radionuclides used in PET must be positron (β⁺) emitting isotopes. The most frequently used radionuclide in medical imaging is fluoride ¹⁸F, but oxygen ¹⁵O, nitrogen ¹³N, and carbon ¹¹C are also commonly used. The half-life of most radioisotopes used in medicine ranges from minutes ¹⁵O (2 min) to hours ¹⁸F (110 min). The principle of PET is based on the detection of two collinear photons (γ-rays) generated when the positron emitted by the radionuclide is annihilated with an electron. The two opposing photons from each annihilation events are detected by a series of cylindrical detectors that surrounds the object. Then, each photon pair is assigned a set of spatial coordinates, from which 3D images can be generated by tomographic reconstruction. The transaxial field of view (FOV) is smaller than the detector ring diameter, and restricts the window over which images can be reconstructed for a given position. Image acquisition may be in *dynamic* or *static* mode depending to the desired outcome. A dynamic acquisition is a series of static images confined to a single position acquired over time with a temporal resolution of a few seconds. However, spatially displaced successive static scans can cover larger objects. For example, static scanning of a human body usually requires 20 to 30 min. A benefit of PET is its quantitative capacity, which means that the amount of tracer can be measured exactly in any region of the object after correction for attenuation and photon scatter. Corrections are based on the density of the material as derived from the CT scan acquisition. On the other hand, inaccuracy may occur in the localization of the tracer because the scanner detects the opposing photons from the annihilation and not the decay itself. However, this inaccuracy remains typically less than a few mm, and depends on the energy of the positron and the density of the medium in which the positron travels. # Materials and Methods ## Sample collection and preparation Lugworms (*Arenicola marina*) and sediment were collected in spring 2011 from Bregnør Bight on the southeastern coast of Odense Fjord on the island of Fyn, Denmark (N: 55.4790 E: 10.6129;). The sediment was passed through a 1 mm sieve to remove animals, gravel and shell debris before it was directly transferred to 10 plastic buckets (diameter: 25 cm). The buckets were filled with 9 liters of sediment to a height of 20 cm and covered by 1.5–2 liters of aerated seawater. The buckets were stored in 200 liter laboratory tanks containing well-aerated seawater at 21 ± 2°C and at a salinity of 20 ± 3 under a 12h/12h day/night cycle. The sediment was left to compact for 8–10 days until scanning. Lugworms (6 to 8 cm long) were collected from the same site 3 days before scanning. Intact animals were placed in Petri dishes containing aerated seawater to void their guts for at least 12 h before being weighed after brief blotting on a dry tissue. Worms were introduced to the buckets after careful transport to Odense University Hospital and left to establish burrows for 1 to 2 days before scanning. ## Scanning configuration and setup All scans were acquired from a GE Discovery RX PET/CT scanner (General Electric Healthcare, Waukesha, WI, USA). A helical CT-scan was acquired using a standard CT protocol with a scan field of view of 70 cm. Data was reconstructed with a standard filter into transaxial slices with a field of view of 50 cm, matrix size of 512x512 (pixel size 0.98 mm) and a slice thickness of 3.75 mm. CT-scan was followed immediately by a dynamic PET-scan with field view of 70 cm. The CT- scan was used for attenuation correction and the scatter was modelled using single scatter simulation algorithm. The PET data was reconstructed into transaxial slices with a matrix size of 128x128 (pixel size 5.47 mm) and a slice thickness of 3.75 mm using iterative 3D OS-EM (2 iterations, 21 subsets). Five buckets with active worms were selected for scanning. The activity and condition of the lugworms were assessed by two visual criterions: (1) formation of fecal casts at the sediment surface confirmed that lugworms were alive and feeding and (2) formation of white *Beggiatoa*-like bacterial mats at the sediment water interface confirmed that lugworm irrigation moved sulfide-rich water upwards in the sediment. One single CT scan was taken of each bucket for the structural description, whereas PET conditions were registered dynamically with an acquisition time of 15 seconds per frame. It is assumed that there was little change in the burrow geometry over the total time of the experiment (1 h) and that the slow and smooth movement of buckets on the sliding table during CT scan only had slight influence on worm position and activity. We conducted image selection based on the geometry of the burrow obtained from the CT scans. It was important for the experiment that burrows were contained within the FOV of the PET scanner (here, 15 cm in width) to obtain a dynamic acquisition. Approximately 80 MBq of Na¹⁸F dissolved in 1 ml water were introduced directly to the continuously aerated overlying seawater. This corresponded to an initial concentration of less than 0.40 nM Fluoride. The mixture of the tracer and the overlying water was done by continuous aeration. Fluoride ¹⁸F was chosen because it had the longest half-life of the radionuclides readily available, and has been shown to be applicable in similar studies. Fluoride is known for its strong affinity to CaCO₃, Fe and clay (R.C. Aller, pers. comm.) suggesting that its use as a conservative tracer in sediments containing high levels of these elements should be avoided. The Bregnør Bight sediment consisted of quartz sand and contained only little carbonate (3.5% DW), iron (0.3 mmol.gDW^-1) and clay (\<5% DW). However, the extent of fluoride precipitation on the obtained results is unknown, but assumed to be relatively small. The toxicity of ¹⁸F to lugworms was limited since it was introduced at very low levels and all lugworms were alive and in good condition at the end of experiment. Similarly, Roskosch et al. exposed the larvae of *Chironomus plumosus* to the same fluoride levels and did not report any toxicity. Finally, the 110 min half-life of ¹⁸F seems to be optimal for determination of lugworm activity as it allows a recording time of several hours. Once the scanning was finished, the buckets were closed and the ¹⁸F was left to decay for more than 48 hours before the sediment characteristics were determined. ## Sediment characterization Sediment characteristics were measured on two cores obtained simultaneously by introducing two core liners (i.d. 5 cm) into the buckets. The cores were sliced at 1 cm intervals to 20 cm depth and the following measurements were done for each layer. Density (g·cm^-3) was measured as the weight of a known volume of sediment; water content (%) as the weight loss after drying for 24 h at 105°C; and organic content as the weight loss after combustion for 5 h at 520°C. Porosity was calculated from of water content and density. Grain size was measured by passing about 20 g of wet sediment through a Wentworth series of sieves of decreasing size. The median grain size and sediment type were calculated using Gradistat 4.0. ## CT and PET data extraction ### Burrow dimensions from computed tomography Visualization of CT images was done using Advantage Workstation v. 4.4. (General Electric Healthcare, Waukesha, WI, USA). The 3D images of the water-filled burrow were created as the difference between attenuation of seawater and sand. The burrow dimensions were quantified from 10–30 sections for which both diameter and length were measured to calculate the average diameter and the total length of the burrow. More accurate measurements can be obtained, but it requires a more tedious procedure. ## 4D visualization of bio-irrigation The ¹⁸F was rapidly mixed homogeneously in the overlying water after addition. The initial images from PET obtained 1–2 min after tracer addition should therefore provide a valid starting point for a 4D visualization of water transport. Subsequent 3D images were then acquired dynamically with a temporal resolution of 1 minute and used as time lapse frame to create a dynamic 3D representation of the bioirrigation pattern. We could achieve much higher frequency, but the chosen time interval was sufficiently to visualize the ventilation pattern of the lugworm in a sensitive manner. A higher temporal resolution would result in more noisy images. The 3D PET images and CT images were combined on Advantage Workstation v. 4.4. (General Electric Healthcare, Waukesha, WI, USA). ## Quantification of ventilation and bioirrigation Corrections for e.g. tracer attenuation were done automatically when PET and CT images were combined. However, it was unknown whether the correction method used in clinical studies could be transferred to measurements in sediment. The radioactivity measurements were therefore validated by placing 6 closed spheres with a known concentration of Na¹⁸F (4 kBq/ml) in the water above the sediment and inside the sediment (volumes ranging between 0.5 ml and 27 ml). Based on the expected radioactivity in the various spheres, it was confirmed that PET quantification of radioactivity in the sediment was valid within an accuracy of ca. 5%. The volume of water transported by the lugworm into the sediment was estimated from the accumulation of ¹⁸F activity in the sediment. Two 3D regions of interest (ROI) were identified from the PET images and the total radioactivity in each of these ROI was recorded for each time frame (i.e. every 15 seconds). The reference ROI (ROI_ref) was placed in the overlying water to determine radioactivity of the source water pumped by the lugworm into the sediment. Sediment ROI (ROI_sed) was defined to encompass the volume of sediment affected by the bioirrigated zone as seen from the PET images. The volume of water pumped into the sediment was calculated as (*V* = *A_sediment* / *c*), where *A_sediment* is the accumulated ¹⁸F radioactivity (MBq) in the sediment ROI and c is the radioactivity (MBq/ml) in the overlying water. The underlying assumptions are that the background radioactivity of the water contained inside burrows is negligible and that the radioactivity in the overlying water remains constant (after decay correction) during the experiment. The latter assumption was supported by the low lugworm pumping (\<1 ml/min), which corresponded to less than 5% of the overlying water during the 1 hour experiment. This assumption was further verified a posteriori. To determine the volume of sediment affected by bioirrigation, the volume of water injected by the worm was divided by the sediment porosity. Finally, the porewater flow rate was computed by dividing the pumping rate of the lugworm by the cross section of the burrow calculated from the CT images. # Results ## Sediment characteristics The sediment consisted of organic poor (\<1%) moderately well sorted fine sand with median grain size of about 0.22 ± 0.02 mm. The average density of the sediment was about 1.78 g·cm^-3 and the water content was 16–20%. The average porosity of the sediment was constant below 1 cm at a level of about 0.27. ## Burrow dimensions The morphology of burrows was clearly visible on the CT images, but the lugworms were not identifiable because of too small difference in attenuation between the lugworm and water. Burrows were blind-ended and connected to the overlying water by one opening at the sediment surface. The architecture of 5 burrows ranged from a vertically (I-shape) or about a 45° angle (J-shape) descending from the burrow opening to more irregular and complex three-dimensional galleries. Overall, the average length of the burrows was 21 cm and the feeding pocket was located at an average depth of 14 cm. Larger worms had proportionally wider burrows (R²: 0.7, p = 0.055). ## 4D visualization of bioirrigation Three out of the 5 worms were actively pumping during the 1 hour scanning time. The combined 3D pictures obtained from PET and CT provided dynamic and detailed information of the spatial distribution of ¹⁸F relative to the burrow position as outlined by the CT images (Figs and ; and Movies). Water transport within the sediment radiated from the feeding pocket of the burrow, but with heterogeneous size and shape of the generated plume within the sediment. In some cases, the plume was localized only around the feeding pocket, and in other cases it was also evident 1–2 cm upward along the burrow shaft. ## Ventilation pattern The high frequency of our recordings enabled a detailed description of lugworm ventilation patterns (Figs and ;). Distinct periods of downward ventilation typically alternated with reverse ejections and periods of quiescence. The duration of these activity phases varied among lugworms, but the ventilation time appeared inversely related to ventilation intensity (ρ_spearman = 0.43). On average, the animals ventilated during 33 to 80% of the elapsed time at a rate of 0.17 to 1.33 ml min^-1. Overall the total amount of water pumped into the sediment by the lugworms ranged from 8 to 26 ml h^-1 and was not clearly related to worm body mass. Based on the obtained rates and burrow dimensions it was estimated that water in the burrow was renewed 1 to 4 times per minute during active ventilation. # Discussion ## Validation of the method The CT approach used in this study provides for the first time a 3D image of lugworm burrows. The results confirm that *Arenicola marina* live in blind-ended burrows, but with a more complex morphology than the simple J-shape traditionally described for this species. Image analysis provided additional information on the dimensions of lugworm burrows (length, diameter), confirming that CT is not only an exciting visualizing approach, but also an excellent measuring tool. The burrow dimensions observed in our study were in the low range of those measured directly by others probably owing to the relatively small size of the lugworms in our experiment. CT has been increasingly used in marine science within the last decades to provide 3D visualization and measurement of biogenic structures created by various macroinvertebrates in muddy sediment. Measurements of biogenic structures using CT image analyses currently represent the only non-destructive method that can provide the required high resolution (\<1 mm) of burrow structures. Perez et al. (1999) noted that it is difficult to obtain valid CT images across a 15 cm wide core of sand. This problem was also to some extent apparent in our study, but image quality remained sufficient to visualize burrows through a 25 cm sand column. The combination of PET and CT revealed for the first time the 3D and dynamic nature of bioirrigation induced by *A*. *marina*, which otherwise may require several invasive techniques. The pumping rates obtained during active ventilation periods are in the low range of the rate measured by others—likely due to the small size of the animals used in our experiment. PET images also revealed that the behavior of lugworms consisted of a succession of ventilation (5–20 min) and reverse ejection periods (1–4 min) that is typical for lugworms (\[–\]). Upward ejection of water in the tail shaft occurs either directly as backward pumping or through backward movement when the animal defecates at the surface. The quiescent periods (4–8 min) observed after reverse ejection is likely the result of feeding and burrow maintenance. These results confirm and extend the application of the PET techniques described by Roskosch et al. in muddy sediment. ## Potential applications The large applicability of PET/CT scan opens a new window of opportunities for the understanding of bioturbation by benthic macrofauna and its ecological implications. PET images provide valuable information on the irrigation pattern and capacity of hidden animals that lives in blind-end burrows such as the lugworm, *A*. *marina*, but it is also applicable for animals living in open- ended burrows as shown by Roskosch et al.. The ventilation of the blind-ended burrow of the lugworm was estimated based on the tracer transport and accumulation in the sediment. However, a technique based on ¹⁸F cannot be applied for open-ended burrows because most of the tracer passes rapidly through the burrow without significant decay. We suggest that this problem can be solved using short-lived radionuclides such as ¹⁵O labeled water (t_½ \< 2 min) that are supplied continuously to the overlying water until steady state is reached. The radionuclide decay should then be equal the transport of water into the burrow, and the volume of water pumped at any given time can be calculated. PET/CT scan also constitutes a valuable tool for experimental validation of bioirrigation modeling. One implication of the results obtained here is that non-localized bioirrigation should be considered in future modeling studies where homogeneous bioirrigation often constitutes one of the basic assumptions. Enhanced flow might occur through cracks or inhomogeneities in the sediment surrounding burrows. Such fine scale heterogeneity can be precisely elucidated by the images obtained from CT scan. Sediment reworking can potentially be traced by PET using particle bound radioisotopes. However, we recommend using CT instead because of its greater spatial resolution. 3D quantification and visualization of sediment reworking has been achieved following aluminum oxide particle transport. But, we hypothesized that the complementary capacities of PET/CT provide a complete 4D assessment of bioturbation, where sediment reworking can be visualized/quantified by CT and ventilation and irrigation by PET. Furthermore, Injection of radionuclides directly into burrowing animals opens the possibilities to trace their behavior when hidden inside the substratum. PET can record animal position and movements within sediments at high temporal frequency. Intra- or interspecific interactions among animals can be studied in great detail by injecting different tracers or concentrations into different individuals or species. The stress effect of injection and toxicity of tracer on the animals should be taken into account, but a pilot study revealed that there is no behavioral and survival effect from injection of 50–150 μl of Na¹⁸F (80 MBq diluted in 1 ml) into the coelom of lugworms. This technique could substitute the use of transparent and artificial substratum, like gelatin, to observe the behavior of burrow-dwelling animals. Other radiographic solutions have been used to visualize animal position within the sediment. Charbonneau et al. were able to visualize the location of mayfly larvae (*Hexagenia* spp.) on X-ray images by placing a mixture of glue and metal (Mo₂C) on the animals. This rather invasive approach could be applied with a CT and provide 3D visualization. Dufour et al. instead suggested the direct use of CT to observe the position of animals in the sediment. In this way, Mermillod-Blondin could actually distinguish between shells of *Mya arenaria* and *Macoma balthica*, but soft tissue animals, like lugworms, are not visible with CT. Another promising tool in the 3D exploration of biological processes in sediments is MRT (magnetic resonance tomography) or also called MRI (magnetic resonance imaging). Although it is based on completely different principles than CT, it can provide structural information at similar scale as CT. MRI is better than CT at distinguishing between soft tissues and Kleinhans et al. suggested that it should be possible to distinguish between animals and the water inside burrows. This postulate remains to be tested, but Bauman et al. used MRI to measure advection and diffusion processes in sediment without any tracers. A more advanced PET/MRI combination is advantageous compared with PET/CT because MRI and PET images can be recorded simultaneously without moving the object. Application of MRI or PET/MRI has to our knowledge not been proven for marine bioturbated sediment. The major drawback of the tomographic techniques is the high economic costs. Only few institutions have the capacity to acquire and maintain such expensive and advanced equipment. The cost of one clinical examination using PET/CT scan in Denmark is currently about 1500 €. This means tomographic-based studies will only be possible if ecologists collaborate with nuclear medicine specialists or radiologists. Since its first use in marine ecology by Perez et al. in 1999, tomography based-studies have tackled challenging issues that could not be elucidated otherwise. # Supporting Information We thank Emma Hammarlund for inspiring discussions, Peter Mygind Leth for CT- scan trial experimentation with iodine and Kim Brixen for his support in establishing collaboration between the different institutions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MD EK DC PEB JHD HT AT PFHC. Performed the experiments: MD DC PEB JHD HT AT. Analyzed the data: MD DC PEB AT. Contributed reagents/materials/analysis tools: PFHC EK PEB JHD HT AT. Wrote the paper: MD EK AT PEB DC PFHC JHD HT.

# Introduction The target of rapamycin (TOR) is a structurally and functionally conserved protein kinase that plays a key role in integrating nutrient and energy signaling to promote cell proliferation and growth. TOR proteins are found at the core of two evolutionarily conserved complexes, known as TORC1 and TORC2. TORC1 regulates cell growth in response to nutrient availability, growth factors and energy status by inhibiting catabolic processes and promoting anabolic processes at multiple levels including protein translation, ribosome biogenesis, gene transcription, protein degradation and autophagy. In fission yeast *Schizosaccharomyces pombe*, Tor2, a core kinase in TORC1, links nitrogen signals to cell proliferation. Nitrogen depletion suppresses Tor2 activity, and nitrogen starvation-responsive genes are largely controlled by Tor2. Several amino acid permeases are included in these genes. Amino acid permeases are a family of proteins that mediate amino acid flux with distinct substrate specificity and subcellular localization. Tor2 regulates transcription of respective amino acid permeases in distinct manners, either positively or negatively. In addition, Tor1, a core kinase in TORC2, is shown to be involved in transcription of several amino acid permeases in a manner distinct from Tor2. However, the underlying mechanism remains to be determined. To address this issue, it is critical to identify *cis*-regulatory elements and transcription factors which mediate Tor signaling for transcriptional regulation of amino acid permeases. In the last few years, the cap analysis of gene expression (CAGE) has been developed to study the transcriptional regulation on a genome-wide scale. This analysis allows high-throughput identification of sequence tags corresponding to 5’ ends of mRNA at the cap sites, thereby facilitating the prediction of core promoters. Using this method, here we have found that Tor1 and Tor2 oppositely regulate transcription of several amino acid permease genes, namely *isp5*^*+*, *per1*^*+*, *put4*^*+*, *SPBPB2B2*.*01*, and *cat1*^*+*, and that these opposite regulations are evenly distributed across multiple TSSs of the respective genes. The motif discovery analysis based on the CAGE results revealed the GATA motif as a potential *cis*-regulatory element for Tor-mediated transcription, and multiple lines of evidence suggest a critical role of a GATA transcription factor Gaf1 in Tor signaling for regulating transcription of amino acid permeases. # Materials and Methods ## Yeast strains, Growth Media, Drugs and General Methods The *Schizosaccharomyces pombe* strains used in this study are listed in. The media (EMM and nitrogen-depleted EMM), denotation and genetic methods have been described previously. Gene disruptions are indicated by the gene symbol preceded by Δ (for example, Δ*tor1*). Proteins are denoted by Roman letters with only the first letter capitalized (for example, Tor2). Drugs were obtained from the following sources: DMSO (Nacalai), rapamycin (Toronto Research Chemicals) and Torin-1 (Tocris). Database searches were performed using Pombe community database PomBase (<http://www.pombase.org>). ## Generation of Δ*tor1* Strain The Δ*tor1* strain was generated using a one-step gene disruption by homologous recombination. The *tor1*::*ura4*⁺ disruption was constructed as follows: The BamHI fragment containing *tor1*⁺ was subcloned into the BamHI site of pGEM-7Zf (Promega). Then, a PstI fragment containing *ura4*⁺ was inserted into the PstI site of the previous construct. The resulting plasmid pKB6751 was digested with BamHI, and the fragment containing the disrupted *tor1*⁺ gene was transformed into KP456 (*h*^*-* *leu1-32 ura4-D18*). Stable integrants were selected on media lacking uracil, and disruption of the gene was checked by genomic Southern hybridization (data not shown). ## CAGE Library Preparation, Sequencing and Data Analysis The prototrophic wild-type, *tor2-287* mutant and Δ*tor1* cells were cultured in EMM media to log-phase. After the cells were collected, the total RNA was prepared using the RNeasy Mini Kit (Qiagen) with on-column deoxyribonuclease digestion (RNase-free DNase Set, Qiagen). The no-amplification non-tagging CAGE library was prepared as described previously using 5 μg total RNA with RIN value of more than 9. The library was subjected to 50-base single read sequencing on a HiSeq 2000 sequencer. The sequencing generated 10 to 20 million reads from each sample. The reads with low-quality and ambiguous bases and corresponding to ribosomal RNA were removed using fastX tool kit \[<http://hannonlab.cshl.edu/fastx_toolkit>\] and rRNAdust \[<http://fantom.gsc.riken.jp/5/sstar/Protocols:rRNAdust>\], respectively. The remaining reads were mapped to *S*. *pombe* reference genome using bwa. CAGE read clustering, differential gene expression analysis and motif discovery were performed by RECLU ver3.1 pipeline (<http://cell- innovation.nig.ac.jp/wiki2/tiki-index.php?page=P000001286>), integrated into Maser (National Institute of Genetics; <https://cell- innovation.nig.ac.jp/index_en.html>) using default settings except with \> 1.0 absolute log(2) fold-change for differential expression analysis. The distribution of TSSs at 1bp resolution is visualized by Integrative Genomics Viewer (IGV) (Broad Institute). Reproducible results were obtained from two independent samples for respective groups. ## RNA Extraction and Semi-quantitative RT-PCR Total RNA was extracted from yeast cells using the RNeasy Mini kit (Qiagen) with on-column deoxyribonuclease digestion (RNase-Free DNase Set; Qiagen). cDNA was synthesized from the resultant total RNA using the High Capacity cDNA Reverse Transcription Kit (ABI) and subjected to semi-quantitative PCR with the SYBR Green PCR Master Mix (ABI). The primers for RT-PCR were summarized in. Signals were detected and analyzed with an Applied Biosystems 7500 Real-Time PCR System (ABI). The mRNA levels of amino acid permeases were normalized to those of *act1* according to the comparative CT method, and were statistically analyzed. ## Luciferase Reporter Assays A Renilla luciferase reporter plasmid for the 1,038bp region from the translation initiation site of *isp5*⁺ promoter (pKB8527) was previously constructed and reported. The CAGE data in this study revealed that the actual transcriptional start site (TSS) of *isp5*⁺ (chr1:5466753) is much closer to the translation initiation site (chr1:5466777) than that in the current PomBase data (chr1:5465377). Based on this new TSS, the 1038bp region from the translation initiation site contains 1013bp region from the TSS, and this promoter fragment was designated *isp5*^-1013 in this study rather than our previous nomenclature (i.e. *isp5*^-1038). A novel reporter plasmid for the 513-bp region from the TSS of *isp5*⁺ promoter (*isp5*^-513) was generated using the following primers: sense primer (4017) `5’-AAC TGC AGC TCC GGA TTT ATA AGA GCG-3’` and antisense primer (3941) `5’-CCG CTC GAG TTT AAT TTT TTG TTT GAT GG-3’`. The resultant fragment was subcloned into the PstI/XhoI-digested pKB5878, a phRG(R2.2)-basic multicopy vector (Promega) that contains Renilla luciferase reporter gene. The resulting plasmid was registered as pKB8632 (*isp5*^-513). A reporter plasmid with the 513-bp region of *isp5*⁺ promoter lacking the GATAAG motif (*isp5*^-513Δ) was constructed from pKB8632 by introducing the deletion of this motif by PCR using a sense primer (4983) `5’-GCT TGG CAA AAA TTG TTT TGT AAA TTG AAT GAA AAA ATG-3’` and an antisense primer (4982) `5’-CGA ACC GTT TTT AAC AAA ACA TTT AAC TTA CTT TTT TAC-3’`. The resulting plasmid was registered as pKB9112 (*isp5*^-513Δ). A Renilla reporter plasmid for the GATAAG motif containing three tandem copies of `GATAAGATAAG` was constructed as described previously, except that a *cis*-regulatory element was generated by annealing the following oligonucleotides: sense primer (4548) `5’-GGC TTT GAT AAG ATA AGA TAC ATG ATA AGA TAA GAT ACA CAT GAT AAG ATA AGA TGC AC-3’` and antisense primer (4549) `5’-TCG AGT GGA TCT TAT CTT ATC ATG TGT ATC TTA TCT TAT CAT GTA TCT TAT CTT ATC AAA GCC TGC A-3’`. The resulting plasmid was registered as pKB8742 (3xGATAAG Renilla reporter). A Renilla reporter plasmid for the CDRE (calcineurin-dependent response element) motif containing three tandem copies of AGCCTC was constructed as follows. The open reading frame of firefly luciferase of pKB5723 (3xCDRE firefly reporter) was replaced by the NcoI/NotI fragment from pKB5878 containing the open reading frame encoding Renilla luciferase. The resulting plasmid was registered as pKB9132 (3xCDRE Renilla reporter). To measure the Renilla luciferase activity, the cells transformed with a reporter plasmid were added with the substrate, coelenterazine (Promega), and the bioluminescence was real-time detected as relative light units (RLU) at 27°C every minute using a microplate luminometer (AB-2350, ATTO Co.), as described previously. Since the bioluminescence signals were peaked at about one hour in our conditions, we measured RLU for at least 2 hours. The peak RLU value during this period was determined and used for statistical analyses. ## Construction of the Cells Expressing Gaf1-YFP under its Native Promoter and Fluorescent Imaging Analyses of Gaf1-YFP Localization The yeast strain expressing Gaf1-YFP under its native promoter (KP6488) was generated as follows. A plasmid containing the open reading frame of Gaf1-YFP- FLAG-His₆ (Orfeome clone 37/H01) was purchased from RIKEN BioResource Center. This plasmid registered as pKB8931 was digested with KpnI to obtain the DNA fragment containing the open reading frame encoding Gaf1-YFP-FLAG- His₆ and *ura4*⁺ marker, and this DNA fragment was integrated into the chromosome at the *gaf1*⁺ gene locus of KP5079 (*h*^*-* *ura4-D18*) by homologous recombination. Fluorescent images were acquired using a microscope (Axioskop 2 Plus; Carl Zeiss, Germany) equipped with an alpha Plan-Fluor 100x/N.A.1.45 oil objective lens (Carl Zeiss) and a SPOT 2 digital camera in combination with the Spot32 software version 2.1.2 (Diagnostic Instruments, Sterling Heights, MI). Fluorescent images were processed using Adobe Photoshop CS6 only for illustrative purposes. ## Immunoblot Analyses A plasmid to express Gaf1-GFP under the *nmt41* promoter was generated as described below. The *gaf1*⁺ gene was amplified by PCR with the genomic DNA of wild-type cells as a template. The sense primer was (4582) `5’-CCG CTC GAG ATG GAT CTA AAG TTT TCC-3'`, and the antisense primer was (4583) `5’-CCG CTC GAG GCG GCC GCC CAT AAC GCT ATA CCA ATC C-3’`. The amplified product containing the *gaf1*⁺ gene was digested with XhoI, and the resulting fragment was subcloned into the XhoI site of BlueScriptSK (+) (Stratagene). Then, a XhoI/NotI fragment containing *gaf1*⁺ was ligated to the XhoI/NotI site of the C terminus of the GFP carrying the S65T mutation. The cells were transformed with the resultant plasmid (pKB9124) and cultured on EMM with thiamine (50 μM). The transformants were grown in fresh EMM without thiamine for 20 hours to induce expression of Gaf1-GFP. To examine Gaf1 phosphorylation, immunoblot analyses of Gaf1-GFP were performed, as previously described. λ protein phosphatase assay was performed, as previously described with minor modifications. Briefly, cells were resuspended in the lysis buffer (50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 5 mM EDTA, 10% glycerol, 0.2% NP-40, 20 mM β-glycerophosphate, 0.1 mM Na₃VO₄, 10 mM p-nitrophenyl phosphate, 10 mM NaF, 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, and Halt Protease Inhibitor Cocktail, EDTA-Free (Thermo scientific)) and disrupted with microbeads. The resultant lysate was diluted by 4-fold into 1 x PMP buffer supplied by the kit with 2 mM MnCl₂ and then incubated with λ protein phosphatase (NEB) at 30°C for 40 min without or with its inhibitor cocktails (50 mM EDTA, 10 mM Na₃VO₄, and 50 mM NaF). For other immunoblot analyses, the cell lysate was prepared using a 1.85M NaOH containing β-mercaptoethanol (7.5% v/v), and was neutralized with 50% trichloroacetic acid (TCA). The resultant lysates were subjected to SDS-PAGE with 10% precast polyacrylamide gels (Nacalai). Proteins were then transferred onto PVDF membrane. After blocking, the membrane was incubated with rabbit anti-GFP antibody (1:5000 dilution) or mouse anti-α-tubulin antibody (1:10000 dilution, clone B5-1-2, Sigma-Aldrich). Rabbit anti-phospho-(Ser/Thr) Akt antibody (#9611, Cell Signaling Technology) was used to detect Rps6 phosphorylation as a readout for Tor2 activity, as previously reported. The resultant membrane was further incubated with HRP- conjugated anti-rabbit or anti-mouse antibody (CST). Signals were detected using Clarity Western ECL Substrate (BioRad) and autoradiography films (FUJIFILM). ## Statistical Analyses Data are shown as means ± SEM. Comparison between two groups was statistically analyzed with unpaired *t*-test. Comparison of more than two groups was statistically analyzed with one-way ANOVA followed by Tukey’s multiple comparison test to evaluate pairwise group differences. Planned comparisons were performed for selected comparison groups, if it is necessary to assess statistical significance for the difference which could not be detected with multiple comparison tests across all the comparison groups. The *P* values less than 0.05 are considered to be significant. Statistical analyses were performed with Prism 6 (GraphPad). # Results ## Tor1 and Tor2 Oppositely Regulate mRNA Expression of Several Amino Acid Permeases Using CAGE technology, we obtained and compared the capped mRNA expression profiles of amino acid permeases in Δ*tor1* cells and *tor2-287* temperature- sensitive mutant cells with those in wild-type cells. In immunoblot analysis using Rps6 phosphorylation as a readout, *tor2-287* mutant cells showed reduced Tor2 activity at 27°C to avoid heat inducible gene expression, and this residual Tor2 activity was abolished at 34°C. In this analysis, we used the mRNA samples from the cells cultured at 27°C, and identified up-regulated and down-regulated transcriptional start sites (TSSs) in *tor2-287* cells or Δ*tor1* cells, compared with wild-type cells ( and Tables). Twenty-two amino acid permeases were significantly expressed in any one of the analyzed cells, and each amino acid permease showed distinct regulations by Tor2 and Tor1. In *tor2-287* cells, 5 genes (*isp5*^*+*, *SPBPB2B2*.*01*, *put4*^*+*, *per1*^*+*, *SPCPB1C11*.*02*) were up-regulated, and 2 genes (*cat1*^*+*, *SPCC74*.*04*) were down-regulated, by at least 2 fold. In Δ*tor1* cells, 2 genes (*cat1*^*+*, *SPCPB1C11*.*02*) were up- regulated, and 7 genes (*isp5*^*+*, *SPBPB2B2*.*01*, *put4*^*+*, *per1*^*+*, *SPBPB10D8*.*01*, *SPCPB1C11*.*03*, *SPCC74*.*04*) were down-regulated, by at least 2 fold. Notably, Tor1 and Tor2 oppositely regulate the expression of several amino acid permeases: Tor1 increases, and Tor2 decreases, the expression levels of 4 amino acid permeases *(isp5*^*+*, *SPBPB2B2*.*01*, *put4*^*+*, and *per1*^*+*), whereas Tor1 decreases, and Tor2 increases, the expression level of *cat1*^*+*. Then we analyzed the distribution of TSSs differentially used in Δ*tor1* cells and *tor2-287* mutant cells, compared with wild-type cells. The opposite regulations by the loss of Tor1 and Tor2 inhibition appear to be evenly distributed across multiple TSSs in each of the genes we selected (*isp5*, *per1*, *put4*, *spbpb2b2*.*01* and *cat1*;). Since the small sample size of our CAGE data precludes statistical analyses, all the above changes by the loss of Tor1 and Tor2 inhibition found by CAGE were confirmed to be statistically significant in semi-quantitative RT-PCR analysis with the larger sample size. These results indicate that Tor1 and Tor2 oppositely regulate transcription from the same TSSs in the gene of each amino acid permease. ### Tor1 and Tor2 oppositely regulate transcriptional activity of *isp5*⁺ promoter To confirm that Tor signaling regulates the promoter activity of amino acid permeases, we performed the Renilla luciferase reporter assay with the promoter of *isp5*⁺ as a representative amino acid permease regulated by Tor signaling. The *isp5*⁺ promoter used in this study covers the promoter region from the TSS to its 1013bp upstream (*isp5*^-1013). The basal activity of *isp5*⁺ promoter was higher in *tor2-287* mutant cells, and was lower in Δ*tor1* cells, compared with wild-type cells at both 27°C and 34°C. Nitrogen depletion, which is known to inhibit Tor2, increased the activity of *isp5*⁺ promoter in wild-type cells. Under nitrogen depletion, *tor2-287* mutant cells showed higher activity than wild- type cells at 27°C and 34°C. However, since the basal transcriptional activity was elevated in *tor2-287* mutant cells, nitrogen depletion-induced increase became smaller than that in wild-type cells at 27°C, and was abolished at 34°C, suggesting a critical role of Tor2 in nitrogen depletion-induced *isp5*^*+* transcription. These results suggest that basal Tor2 activity inhibits the activity of *isp5*⁺ promoter, and that nitrogen depletion disinhibits the *isp5*⁺ promoter through inhibiting Tor2. By contrast, in Δ*tor1* cells, the basal activity and the nitrogen depletion- induced activation of the *isp5*⁺ promoter were abolished, suggesting a critical role of Tor1 in the activity of the *isp5*⁺ promoter. The activity of the *isp5*⁺ promoter still requires Tor1 even when TORC1 activity is reduced, because the loss of Tor1 abolished both the basal transcriptional activity and nitrogen depletion-induced transcriptional activation in *tor2-287* cells. We also examined the effects of representative Tor inhibitors on the *isp5*⁺ promoter activity. Rapamycin, a Tor2-selective inhibitor, did not affect the *isp5*⁺ promoter activity in wild-type cells. However, in *tor2-287* mutant cells at 27°C with reduced Tor2 activity, rapamycin increased the *isp5*⁺ promoter activity. This rapamycin-induced transcriptional activation was abolished after shift to 34°C for 2 hours. Therefore, rapamycin appears to induce *isp5*^*+* transcription in a Tor2-dependent manner, although Tor2 activity in wild-type cells appears to be resistant to rapamycin treatment. Unlike rapamycin, Torin-1, an inhibitor for both Tor1 and Tor2, increased the transcriptional activity of *isp5*⁺ promoter in wild-type cells. By contrast, in *tor2-287* cells, Torin-1 treatment reduced the *isp5*⁺ promoter activity from the level with vehicle treatment at both 27°C and 34°C, suggesting that this inhibitory effect of Torin-1 in *isp5*⁺ promoter activity is independent from Tor2. Since Tor1 is critical for the *isp5*⁺ promoter activity, as described above, the inhibitory effect of Torin-1 could be due to Tor1 inhibition. ### The GATAAG motif is required for *isp5*⁺ promoter activity To identify potential transcription factors involved in Tor-mediated regulation of gene expression, we performed the motif discovery analysis using our CAGE results. The TSSs with at least 2-fold changes in *tor2-287* or Δ*tor1* cells were classified into 4 groups; up-regulated at top peaks, down-regulated at top peaks, up-regulated at bottom peaks, and down-regulated at bottom peaks. The top 100 TSSs with the highest concentration in each group were separately analyzed. We identified that GATA motifs are overrepresented in the promoter regions associated with the TSSs up-regulated at top and bottom peaks as well as those down-regulated at top peaks in *tor2-287* cells. GATA transcription factors are a family of transcription factors characterized by their ability to bind to the DNA sequence “GATA”. The GATAAG motif is one of the sequence motifs to which GATA transcription factors are known to bind. In the *isp5*⁺ promoter, there are two GATAAG motifs located at the -645\~-640bp region (distal) and the -124\~-119bp region (proximal) upstream of the TSS. The distal GATAAG motif was not critical for *isp5*⁺ promoter activity, since the shorter *isp5*⁺ promoter (*isp5*^-513) lacking this motif showed the normal basal activity of *isp5*⁺ promoter and its activation induced by nitrogen depletion and Torin-1 treatment. To determine the importance of the proximal GATAAG motif, we selectively removed this motif from the *isp5*^-513 promoter. The resulting promoter lacking the GATAAG motif (*isp5*^-513Δ) showed negligible basal activity and the lack of activation induced by nitrogen depletion, Torin-1 treatment or Tor2 inhibition. These results suggest that the proximal GATAAG motif at the -124\~-119bp region from the TSS is critical for the basal activity of *isp5*⁺ promoter and its activation induced by Tor2 inhibition and nitrogen depletion. ### Tor1 and Tor2 oppositely regulate GATAAG-mediated transcription We then examined whether Tor1 and Tor2 regulate transcription mediated by the GATAAG motif, using its Renilla luciferase reporter with the promoter containing three tandem copies of this motif (3xGATAAG). This GATAAG reporter recapitulated the patterns of Tor-mediated regulations of the *isp5*⁺ promoter, as follows. In wild-type cells, nitrogen depletion increased the activity of GATAAG reporter. Under nitrogen depletion, *tor2-287* mutant cells showed higher activity of GATAAG reporter than wild-type cells at 27°C and 34°C. However, since the basal transcriptional activity was elevated in *tor2-287* mutant cells, a relative increase induced by nitrogen depletion became smaller than in wild-type cells at 27°C, and was abolished at 34°C, in which Tor2 activity was fully suppressed. These results suggest that basal Tor2 activity suppresses GATAAG-mediated transcription, and that nitrogen depletion disinhibits this transcription through inhibiting Tor2. By contrast, in Δ*tor1* cells, the basal activity of the GATAAG reporter and its activation induced by nitrogen depletion were abolished. It is noted that the loss of Tor1 did not affect the activity of a reporter for the CDRE motif. Therefore, Tor1 is selectively involved in GATAAG-mediated transcription. We also examined the effects of rapamycin and Torin-1 on GATAAG reporter activity. The GATAAG reporter also recapitulated the effects of these drugs observed with *isp5*⁺ promoter. Thus, rapamycin did not increase GATAAG reporter activity in wild-type cells. However, rapamycin significantly increased this reporter activity in *tor2-287* cells at 27°C. This rapamycin- induced activation was abolished at 34°C, in which Tor2 activity is fully suppressed. Therefore, rapamycin appears to induce GATAAG-mediated transcription in a manner dependent on Tor2, although rapamycin alone may not suppress Tor2 activity in wild-type cells. By contrast, Torin-1 treatment increased GATAAG reporter activity in wild-type cells, whereas it decreased the reporter activity in *tor2-287* mutant cells at both 27°C and 34°C, indicating that the inhibitory effect of Torin1 is independent from Tor2. Since Tor1 is critical for GATAAG- mediated transcription, Torin-1-induced suppression could be due to Tor1 inhibition. ### Gaf1 is critical for transcription driven by *isp5*⁺ promoter and the GATAAG motif In fission yeast, there are five GATA transcription factors named Sfh1/SPCC16A11.14, Fep1/SPAC23E2.01, Ams2/SPCC290.04, Gaf1/SPCC1902.01, and SPCC1393.08 (<http://www.pombase.org/spombe/related/PBO:0000371>). We performed transcriptional activity of the *isp5*⁺ promoter and the GATAAG reporter in Δ*gaf1*, Δ*fep1* and Δ*SPCC1393*.*08* cells, three strains available from the whole-genome knockout library. In Δ*gaf1* cells, the basal activities of these reporters and their activation induced by nitrogen depletion and Torin-1 treatment were abolished. The other two strains appeared to show normal activities of these reporters (data not shown). The loss of Gaf1 also abolished the elevated basal activity of *isp5*⁺ promoter in *tor2-287* mutant cells. These findings show that Gaf1 is critical for the basal activity of *isp5*⁺ promoter as well as its activation induced by nitrogen depletion and Tor2 inhibition. Furthermore, the mRNA levels of *isp5*, *per1* and *put4*, but not *cat1*, were significantly decreased in Δ*gaf1* cells, suggesting a universal role of Gaf1 in transcription of multiple amino acid permeases. ### Tor2 inhibition induces nuclear localization of Gaf1 and its dephosphorylation To investigate how Gaf1 is regulated for transcription of amino acid permeases, we observed the localization of Gaf1-YFP expressed under its native promoter. Since expression of Gaf1-YFP rescued defective *isp5*^*+* expression in Δ*gaf1* cells, Gaf1-YFP is functional. In wild-type cells cultured in the EMM medium, Gaf1-YFP was mainly localized to the cytosol. Nitrogen depletion and Torin-1 treatment induced robust nuclear localization of Gaf1-YFP from the cytosol as early as 5 min. Although rapamycin treatment alone did not induce transcription of amino acid permeases, as previously reported and shown in this study, it also induced nuclear localization of Gaf1-YFP, albeit to the much lesser extent. We then investigated the involvement of Tor1 and Tor2 in nuclear localization of Gaf1. In Δ*tor1* cells, Gaf1-YFP was localized to the cytosol in the EMM medium, and was localized to the nucleus upon nitrogen depletion. However, nuclear signals of Gaf1-YFP appear to be weaker in Δ*tor1* cells, compared with wild-type cells. The *tor2-287* cells showed abnormal localization of Gaf1. Gaf1 was slightly enriched in the nucleus at 27°C, and was completely localized to the nucleus at 34°C. Therefore, basal Tor2 activity suppresses, and Tor2 inhibition and nitrogen depletion induces, nuclear localization of Gaf1. It has been reported that dephosphorylation of Gln3, a GATA transcription factor in budding yeast, occurs simultaneously with its nuclear localization. We examined whether Gaf1 phosphorylation is also regulated by nitrogen depletion and Tor2 inhibition in fission yeast. We performed immunoblot analysis using Gaf1-GFP expressed under the *nmt41* promoter to examine Gaf1 phosphorylation. Since expression of Gaf1-GFP rescued defective *isp5*^*+* expression in Δ*gaf1* cells, Gaf1-GFP is functional. In this analysis, Gaf1 signals were broadly distributed around the expected molecular weight. λ-phosphatase treatment increased the mobility of Gaf1 signals, so that the signals formed a sharper band. This effect of λ-phosphatase was not observed in the presence of its inhibitors. These results suggest that the mobility shift of Gaf1 signals in immunoblot analysis is due to Gaf1 phosphorylation. Torin-1 treatment and nitrogen depletion increased the mobility of Gaf1, whereas rapamycin treatment appeared to have no effect on Gaf1 mobility. In *tor2-287* mutant cells, Gaf1 mobility was also faster than in wild-type cells at both 27°C and 34°C. In Δ*tor1* cells, nitrogen depletion induced faster Gaf1 mobility similarly to wild-type cell. These findings suggest that nitrogen depletion and Tor2 inhibition, but not Tor1, induces Gaf1 dephosphorylation along with its nuclear localization. # Discussion Tor signaling is critical for transcriptional regulation of amino acid permeases according to nitrogen availability. However, the underlying mechanism remains largely unknown. Here we have found that the loss of Tor1 decreases, and Tor2 inhibition increases, transcription of *isp5*^*+*, *per1*^*+*, *put4*^*+* and *SPBPB2B2*.*01*, whereas the loss of Tor1 increases, and Tor2 inhibition decreases, transcription of *cat1*^*+*. These opposite regulations by Tor1 and Tor2 occur evenly across multiple TSSs identified by CAGE for respective genes of amino acid permeases. In luciferase reporter assay, nitrogen depletion and Tor2 inhibition induce the activation of *isp5*⁺ promoter, and Tor1 is critical for the basal activity of *isp5*⁺ promoter. The basal activity and Tor- mediated regulation of *isp5*⁺ promoter depend on one of its GATAAG motifs, and the GATAAG reporter recapitulates Tor-mediated regulations of *isp5*⁺ promoter. Furthermore, a GATA transcription factor Gaf1 is critical for the basal transcription of *isp5*^*+* and its activation upon nitrogen depletion and Tor2 inhibition. Therefore, this study substantiates a critical role of Tor-Gaf1 signaling in transcriptional regulation of several amino acid permeases. Since the GATAAG motif in *isp5*⁺ promoter is critical for its transcriptional activity, it is plausible that Gaf1 binds to this motif for *isp5*^*+* transcription, although the possibility that Gaf1 might indirectly regulate *isp5*^*+* transcription cannot be excluded. Consistent with the role of Gaf1, nitrogen depletion and Tor2 inhibition induce nuclear localization of Gaf1 and its dephosphorylation. These regulations of Gaf1 have also been reported by another group recently. Similar to these findings, in budding yeast, inhibition of TOR1, a core kinase of TORC1 in this organism, also leads to nuclear localization and dephosphorylation of Gln3, a GATA transcription factor. Nuclear localization of Gaf1 induced by nitrogen depletion and Tor2 inhibition could increase the concentration of Gaf1 in the nucleus, thereby promoting its binding to the GATAAG motif in *isp5*⁺ promoter. Indeed, nitrogen starvation with proline as a poor quality nitrogen source also induces nuclear localization of Gaf1, and increases Gaf1 binding to the *isp7*⁺ promoter containing three potential GATA motifs. However, whether and how Gaf1 dephosphorylation contributes to its nuclear localization and Gaf1-mediated regulation of *isp5*^*+* transcription remains to be elucidated. By contrast, since the loss of Tor1 abolishes *isp5*^*+* transcription as well as GATAAG-mediated transcription, it is plausible that Tor1 promotes *isp5*^*+* transcription through Gaf1 binding to the GATAAG motif in *isp5*⁺ promoter in a manner opposite to Tor2. If that is the case, Tor1 regulates Gaf1 in a manner different from Tor2, since the loss of Tor1 affected neither nuclear localization nor phosphorylation of Gaf1. The mechanism about how Tor1 regulates Gaf1-mediated transcription remains to be investigated. As discussed above, our findings suggest that nitrogen depletion suppresses Tor2 activity, thereby increasing *isp5*⁺ transcriptional activity through Gaf1. However, *tor2-287* cells showed much higher activity of *isp5*⁺ promoter than that in wild-type cells under nitrogen depletion. This difference cannot be explained by the magnitude of inhibition of Tor2 activity, because nitrogen depletion for 15 min abolishes Tor2 activity, whereas *tor2-287* cells showed reduced, but significantly Tor2 activity at the permissive temperature, as measured by Rps6 phosphorylation. Since long-term partial suppression in Tor2 activity in *tor2-287* cells altered expression profiles of various genes, Tor2 may indirectly regulate *isp5*⁺ transcriptional activity through some of these genes. In addition to *isp5*^*+* transcription, Tor1 and Tor2 also oppositely regulate mRNA expression of other amino acid permease genes, namely *per1*^*+*, *put4*^*+* and *SPBPB2B2*.*01*. Gaf1 is also critical for mRNA expression of *isp5*^*+*, *per1*^*+* and *put4*^*+*. These results suggest that Gaf1 coordinates transcription of multiple amino acid permeases upon Tor2 inhibition and nitrogen depletion. Although the GATAAG motif was not found in the promoters of *per1*⁺ and *put4*⁺, GATA-containing sequences similar to this motif exist. Whether these sequences mediate Gaf1-mediated transcription remains to be investigated. In contrast, Tor1 suppresses, and Tor2 promotes, the transcription of *cat1*^*+* in a direction opposite to *isp5*^*+*, *per1*^*+* and *put4*^*+*. Since the loss of Gaf1 did not affect the mRNA expression of *cat1*, another unidentified transcription factor must be involved. Therefore, Tor1 and Tor2 may regulate transcription of amino acid permeases in multiple ways through distinct transcription factors. Since each amino acid permease has its own substrate specificity and subcellular localization, this study also paves the way for elucidating the pattern of amino acid flux to promote cell proliferation and growth upon nutrient challenge. # Supporting Information We thank Dr. Takayoshi Kuno for generous supports, invaluable advice and encouragement to complete this study. We thank Dr. Mitsuhiro Yanagida and National BioResource Project for providing *tor2-287* strains. We thank Ms. Misako Takizawa for secretarial assistance. We thank Genome Network Analysis Support Facility in RIKEN CLST and Platform Project for Supporting in Drug Discovery and Life Science Research from Japan Agency for Medical Research and Development (AMED) for technical supports for the CAGE data analysis. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YM NM TF. Performed the experiments: YM NM QL YQ. Analyzed the data: YM NM TF. Contributed reagents/materials/analysis tools: YM TF RM. Wrote the paper: YM NM TF RM.

# Background Syncope and transient loss of consciousness (TLoC) account for a substantial proportion of healthcare resource use. Data from the United Kingdom (UK) show that in 2010–11, syncope and collapse were among the top three cardiology- related admissions and that \>90% of syncopal admissions are to emergency departments (ED). In Germany, around 150,000 syncope-related hospitalisations are reported yearly. Syncope presents a diagnostic conundrum to clinics. Patients are usually recovered by the time of presentation and the event itself was rarely captured by qualified bystanders. As a consequence, a large number of tests are typically employed to detect underlying aetiologies and the patient journey from presentation to diagnosis is often unstructured with low diagnostic yields, often \<50%. An analysis from the multicentre PICTURE registry of syncope care in Europe reported an average of 3 different specialists evaluating the typical patient and a median of 13 diagnostic tests performed, including many non-essential and expensive evaluations such as MRI/CT scan, neurological or psychiatric evaluation in almost 50% of patients. Other studies have reported rates of 40%. This long and resource-intense time from presentation to diagnosis results in costs and in lengthy delays in delivery of the appropriate treatment. In some patients, notably those with cardiac syncope, such delays mean the patient may remain at increased risk for cardiac events over substantial time periods instead of receiving appropriate therapies. An unstructured diagnostic strategy, while adhering to guidelines and satisfying clinical necessities, can lead to major insufficiencies in the system. The use of standardised protocols improves diagnostic yields and reduces hospital stay and in consequence, the establishment of syncope units in hospitals has been strongly recommended. A clinical care pathway defines goals and key elements based on guidelines, best practice and patient expectations; it sets the sequence and co-ordinates activities of the multidisciplinary care team, patients and their relatives. A hurdle is that the analysis and implementation of structural changes is often outside the clinical core competence of hospitals. We here describe the preliminary results of as pilot study on the application of principles from process improvement in commercial businesses, with a focus on quality control and lean management, to care pathways for syncope management in hospitals. Lean Six Sigma methodology was used to analyse and evaluate current patient journeys and to design and implement improved care pathways in close collaboration between the hospital teams and the Lean Six Sigma specialists. The project was designed to be compliant with current guidelines and to be process-focused, with no interference with clinicians' expertise, judgement and treatment of patients. # Methods Four hospitals took part in the project: The Northern General/Royal Hallamshire Hospital Sheffield and Victoria Hospital Blackpool in the UK, St. Marien Hospital Lünen, Germany, and CHU de Nîmes, France. The hospitals were included in the study based on a willingness to change and try out alternative approaches to their management pathways. There were up to four-fold variations in the sizes of hospitals, catchment areas and annual budgets between the participating centres but none of them specialised in syncope. In addition, the methodological approach was implemented at the Medical Center Alkmaar in The Netherlands. This hospital has a large geriatric patient population, for which additional tests (e.g. cognitive tests and nutrient tests) and a comprehensive geriatric assessment were performed. This made it an outlier and the data were analysed separately. The patients included in the analyses were selected randomly from those presenting at the hospital with syncope of undiagnosed cause. However, the nature of the study precluded a formal randomisation of patients to the pre- and post- intervention groups. Multidisciplinary project teams were formed at each centre with members representing a variety of specialties: Lean Six Sigma experts from the sponsor, consultant physicians, nurses, technicians and planners. The project studied processes, not treatments or patients; nor did the team evaluate or interfere with physicians' clinical judgement and treatments. Hence, no approvals were necessary by the hospital ethical review boards. However, the analysis phase sometimes required access to retrospective patient records to chart their journey through the system. Hence, all participating centres and consultants signed confidentiality agreements including clauses safeguarding patient privacy. All patient records/information was anonymised and de-identified prior to analysis. The objective was to use Lean Six Sigma methodology to identify scope for efficiency gains and to design and implement improved care pathways for patients presenting with recent syncope/TLoC. The analysis focused on two main improvements: the reduction of non-essential tests in a standardisation of steps in diagnostic pathways and a reduction in the time to patient diagnosis. Physicians were cardiologists and neurologists with special interest in syncope and willingness to improve cross speciality interactions in their hospitals. There was no interference with physicians' clinical judgements. Lean Six Sigma is a highly structured methodology that uses a set of management and statistical methods to optimise business processes. Initiatives are planned and implemented on a project-by-project basis. The typical sequence consists of five steps: Define, Measure, Analyse, Improve and Control (DMAIC). For the current application, these steps were condensed into 3 specific phases, each relying on the previous one for success. Each step had specific, quantified targets. The first, assessment phase, was mainly driven by the Lean Six Sigma experts who conducted a detailed analysis of baseline conditions at the participating hospitals. Performance was mapped at each centre during a time period of 1–3 months during with patients were followed in their pathways through the diagnostic processes. Assessments included detailed observations of activities, use of time and the resources drawn upon at each step of the pathway. A consolidated map of patients' journeys through the system from admission to diagnosis or discharge was produced in a standardised format. The charts provided a graphic overview of the similarities and disparities in pathways between different patients for further evaluation. Optimised diagnostic pathways for the average syncope patient were generated based on guidelines for syncope management: NICE guidelines for the UK hospitals; ESC guidelines for the hospitals in continental Europe. A gap analysis was performed on the current processes and patient flows against the benchmark pathways to identify bottlenecks and obstacles to a smooth process. A report with the analysis and key points for improvement was presented to the interdisciplinary team and used to start the second, improvement phase. The improvement phase was a collaborative effort between all specialties represented on the teams, and lasted for 3–6 months. Optimisation pathways and standardised best practice tools were developed. A new operational model was designed and an “optimal care pathway action plan” was drafted that included best practices and standardised decision trees. Decisions on prioritisation of new measures and processes were taken by the team after considerations of the impact on employees and ease of implementation. Documentation materials and forms were developed around the new pathways to ensure consistent implementation. Specific metrics were identified to measure the performance and the effects of new processes. To smooth the implementation of the new processes, the action plan included specific staff-training activities on the new processes, as well as support with change-management activities and counselling. The final, sustain phase, included continuing support with implementing the new pathways, a critical evaluation of the effects of the changes and the refinement of best practice tools. Specific multidisciplinary teams were established in the hospitals, with responsibility for the monitoring and fine tuning of new processes. The results of the new care pathways were assessed 3–9 months after the end of the improvement phase, at the point in time where the implemented changes were judged by the implementation teams as stable and sustained. A number of key indications were specified. Hospital resources were calculated as the overall cost per patient treated; number of tests taken; time from presentation to diagnosis and the time spent on each patient by physician, nurse or technician. Patient outcomes were evaluated on the diagnostic yield, which was defined as a diagnosis of the cause of syncope or a conclusion leading to a treatment. The patient experience was captured in the overall duration of patient journeys and diagnostic yield. Compliance with guidelines was monitored. A cost analysis is currently in progress that will include costs for time to diagnosis and associated hospital stay. The results of this analysis will be published separately. # Results The different pilot-projects were carried out between 2007 and 2012 over an average time period of 1 year (min. 8 months – max. 14 months). The average size of the project teams in the participating hospitals was 4 individuals (1 doctor, 2 nurses, 1 consultant). Some additional resources were drawn upon as necessary such as finance and IT departments. ## Assessment phase As illustrated by a typical chart of patients' journeys in one representative centre, care pathways at the beginning of the project were highly unstructured and there was little consistency in the number and kinds of tests performed, or in process durations between different patients. Although the diagnostic pathways were found to be compliant with current guidelines for the diagnosis and management of syncope, the inconsistencies and differences between centres and patients showed significant scope for improvements. The kinds of tests used in the centres are shown in. ## Overall results The assessment of the overall results were based on a total of 455 patients, 223 in the assessment phase and 232 admitted after the implementation of the new processes. Data on the number of patients in the post-intervention phase were lacking from Northern General & Royal Hallamshire. Only 20 patients were included at the CHU de Nîmes. Hence, p values could not be calculated for all data from these centres. The new processes led to statistically significant, sometimes dramatic improvements in the performance indicators at all participating hospitals. Overall the new processes led to a more efficient use of diagnostic tests and an improved prioritisation of the nature and order of tests. The number of tests was reduced by on average 24%; p = 0.046. Median time from admission to diagnosis was shortened by on average 59% (p\<0.048). The efficiency improvements came together with an increased diagnostic yield: from 42% before the implementation of new processes to 73% with the new systems (p = 0.007). All use of the assessed newly implemented processes (100%) was compliant with guidelines. The implantation of an insertable loop recorder (ILR) is often relevant to hospitals, as this decision is based on a well-founded suspicion of cardiac causes of syncope obtained by other tests. ILRs provide data over several years of follow-up and thus for a hospital, an ILR implant represents discharge and the end of a patient journey. When the definition of diagnostic yield was expanded to include ILR implant, yields were highly similar to those without including the ILR implant decision: 60% before the implementation of the procedures and 86% with the new procedures in place. ## MCA Alkmaar The MCA Alkmaar was unique in having a large geriatric population. The number of tests increased at this centre with the new processes as more tests such as nutritional status were introduced to improve diagnostic yield. Despite the increased number of tests, the time to diagnosis was shortened significantly at MCA Alkmaar, to similar extent as in the other centres. ## Case study: Victoria Hospital Blackpool As the individual hospitals varied in their needs and in the nature of the specific changes implemented to their processes, we describe the experiences in one specific centre, Victoria Hospital in Blackpool (UK). The characteristics of this hospital are shown in and. At baseline, the median time per patient from presentation at the hospital to diagnosis was 13.5 weeks and varied widely, up to 39 weeks. The objectives of the project were defined at the outset: the lead time from presentation to treatment should not exceed 18 weeks and the proportion of patients recommencing the diagnosis-to-treatment loop should be reduced to \<5%. In addition, unnecessary admissions of low to medium risk patients to the accident & emergency (A&E) department should be reduced. It was imperative to maintain compliance with NICE guidelines. The project team had 11 members, including managers, nurses and consultants from cardiology, ED and neurology. The assessment phase identified several points of potential efficiency increases. As there was no multidisciplinary approach to patient management, standardised referral criteria, definitions of blackout and standardised pathways were all missing and the content and quality of patient histories varied widely. Patients came into the diagnostic pathways at a multiplicity of entry points. Around half of patients presented after referrals from general practitioners and many of these referrals were of low quality or inappropriate. In consequence, a large proportion of patients (37%) were caught in a loop of repeated diagnosis, which slowed their progress through the pathway. A third of patients were referred by cardiologists. The remainder presented at the A&E department, were referred by neurologists or by other physicians. The most important change implemented was the creation of a “one stop shop” – The Lancashire Blackout Clinic – with the aim to provide a diagnosis in all patients admitted with clear referral criteria. The clinic is led by 2 nurses and overseen by the Consultant Cardiologist of the Week. Staff handle patients as they arrive and guide them through the pathways, to ensure a smoother process, clearer focus and optimised patient flow. The design and implementation of standardised, clear pathways were key to reducing redundancies and waiting times in the patient flow. Clear definitions, rules for risk stratification and timings for the different diagnostic steps were included in the pathway and checks were introduced to ensure these were kept. To reduce the number of inappropriate referrals, clear referral criteria were communicated and embedded in a mandatory referral form for all physicians. The multidisciplinary approach was abolished and replaced by regular interdisciplinary meetings where complex cases are discussed. As a result of the changes, the overall median lead time from referral to diagnosis was reduced from 161 days to 54 days, that from referral to first appointment from 20.5 to 14.2 days and the time from first visit to diagnosis from 17 to 1 day. The volume of referrals coming from outside the hospital's catchment area was reduced significantly. On average, 2.2 test were conducted with the new procedures v 4.0 in the old system, and the average number of visits per patient was reduced from 4 to 1.75. lists the test performed before and after project implementation. A comparison of representative patient journeys before and after implementation of the new syncope care pathways, including the tests performed, is shown in. The improved pathways were associated with increased diagnostic yield: from 46% to 83%. # Discussion The results presented from this pilot study, from four hospitals in three European countries, show a promising potential from applying a structured Six Lean Sigma process-optimisation methodology to the identification of efficiency bottlenecks and to the design and implementation of improved care pathways for patients presenting with syncope/TLoC. Over the 9–15 months assessment, improvement and sustain periods, there were significant improvements in lead times and diagnostic yield. In all hospitals the number of tests was significantly reduced. Even at the centre MCA Alkmaar, with a large geriatric population led to a higher number of relevant tests after the implementation of the new pathways, the new system reduced the time to diagnosis, as well as increasing the diagnostic yield. Compliance with the respective guidelines in each country was maintained. There is an acknowledged potential for the productive use of Lean or Six Sigma methodologies in health care. To our knowledge, this is the first study to show that a partnership between healthcare providers and experts on process improvement from commercial businesses, with a focus on quality control and lean management, can be used to structure, standardise and optimise care pathways for syncope/TLoC management in hospitals. Other successful attempts to improve syncope management have focused on establishing dedicated syncope units and implementing standardised protocols and sequences of tests. However, there has not previously been a focus on process management and on organising tests in the most effective fashion. Lean Six Sigma takes a highly structured approach to process improvement. The methodology has two focus areas, efficiency improvement (lean) and improved process quality (six sigma). Our baseline mapping revealed inefficiencies in the processes as well as a high variation in the diagnostic pathways between patients. Both factors were improved by the new pathways. While the Lean Six Sigma methodology is highly structured, there is flexibility within its application. Each situation is analysed individually and the initiatives are planned and implemented according to the needs defined for each specific organisation. This is relevant to hospital environments, which can vary widely in organisation, patient pool and regulation. One of the most efficient changes was the introduction of standardised referral criteria and mandatory referral forms, as done, e.g., at the Blackpool Victoria Hospital. The standardisation greatly reduced the number of patients sent back because of inadequate referrals or going into a loop of repeated testing. This is a common situation: an observational study in an Irish hospital has reported that 58% of admissions for syncope were unnecessary when compared with ESC guidelines. Even greater improvements than those at the Victoria Hospital may thus be achieved by, e.g., greater use of automatic methods to improve tracking and documentation of patients and visits. Such systems also have the potential to reduce errors due to multiple manual entries of the same data. In achieving an average of 1 day lead time to diagnosis, Victoria Hospital may be seen as a best case scenario, but improvements were achieved at all centres, indicating that Lean Six Sigma driven processes may be beneficial across a range of local conditions. Even in the case of MCA Alkmaar, where more tests were performed with the new system, the time to diagnosis was reduced. This further shows that the number of tests in itself is not a good measure of quality of care. Efficiency improvements in healthcare must not come at the costs of reduced quality of care or service to patients. Our results are reassuring on these counts. As untreated cardiac syncope is associated with increased morbidity and mortality, a shorter time to diagnosis and appropriate treatment can be expected to reduce mortality. A large part of the improvements can be attributed to reductions in tests not mandated by guidelines. The reasons for the increase in diagnostic yield are probably multifactorial. One factor was the reduction in the number of patients who were sent back because of inadequate referrals, which increased the rate of relevant patients entering the process. A number of factors can be identified that helped in successful implementation of changes and not all may be transferrable to non-syncope care pathways. The hospitals showed a strong engagement and willingness to commit resources and to implement change management. Without readiness to change, it is very difficult to overcome inertia in any organisation. Also, in the multinational environment of Europe, local knowledge and language skills are necessary to identify and evaluate information specific to each individual hospital. The allocation of appropriate resources is important at all stages of the process. The assessment phase is labour-intense, but to the hospital the additional personnel resources needed are minor, as most of the work is carried out by the consultants. The changes themselves require dedicated healthcare personnel and structural reorganisation of hospital procedures. The successful outcomes indicate that associated savings would compensate for the associated costs and possibly save overall costs. An analysis of these effects is in preparation. The short time, typically \<6 months, until benefits were seen with the new processes increases participant motivation. The availability of specific guidelines for the diagnosis and management of syncope enabled the generation of optimised diagnostic pathways to use as benchmarks. The complexity of Syncope/TLoC provides greater scope for process optimisation than diagnoses with simple pathways, fewer components and decision points. Several factors will affect the sustainability of the improvements. Dedicated project owners are necessary and the continuing aid of independent consultants may help. Some of the changes, such as a specially staffed blackout clinic, need resources and dedicated staff which may be susceptible to strategic decisions from hospital management or to political decisions beyond the control of the respective clinics. It is also important to maintain continued support from staff at all levels. Among weaknesses in this work should be mentioned the lack of formal randomisation of patients. We included patients randomly, but there was no randomised allocation of patients to the procedures before and after the management interventions. As the projects established individual new processes in each hospital it was not possible to run a parallel-group study. There is a theoretical possibility of differences in characteristics between the groups of patients admitted in the pre- and post phases, respectively. The random inclusion of patients, together with the large and consistent benefits with the new processes observed across participating hospitals make it highly unlikely that such systemic differences would exist in the other hospitals. Other weaknesses are the lack of data on the number of patients in the post- intervention phase at Northern General & Royal Hallamshire and the low number of patients included at the CHU de Nîmes. We did not systematically analyse the kinds of tests performed in the pre- and post- phases, not do we have data on the diagnostic yield of individual tests. Patient satisfaction was not captured in our analysis. It is known from other studies that diagnostic yield and waiting time are two important factors for patients – (S. Tsintzos, unpublished). Both these factors were reduced with the Lean Six Sigma guided pathways. Recently, the department of Neurosurgery at the UCLA Medical Center applied lean methods to streamline processes and co- ordinating tests, with resulting improvement in patient satisfaction. In summary, adapting a structured Lean Six Sigma approach to process optimisation in a hospital setting improved the efficiency and diagnostic yield of diagnostic pathways for the diagnosis of syncope. With appropriate modifications the strategy should be applicable to a variety of hospital organisations. As syncope represents 5–10% of admissions to an average European hospital and 50%–80% of patients presenting with syncope at an A&E department are admitted for inpatient care, wider adaptation of the methods may have beneficial implications on patient outcomes and clinic resources. The authors are indebted to the following individuals for their work with implementing the processes in the respective centres: Jaap H Ruiter MD and Rene W.M.M Jansen MD, MCA Alkmaar; Noémie Thomas, Alison Peace, Stijn J. Schretlen, Bryony Cresswell and Laura Verspui, all affiliated with Medtronic. The authors are also grateful to Stelios Tsintzos MD for helpful comments on the manuscript. [^1]: Leon Martens and Lucas Baggerman are employees of Medtronic. The other authors have declared that no competing interests exist. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: LM LB. Performed the experiments: LM GG JFHW LB GM CP. Analyzed the data: LM. Wrote the paper: PS. Reviewed the manuscript: LM GG JFHW LB GM CP PS LB. Contributed analytic insights and interpretations: LM GG JFHW LB GM CP PS.

# Introduction Mammary gland development during childhood does little more than keep pace with the general growth of the body until puberty. During puberty, the mammary gland exhibits a substantially dynamic process through which it gives rise to highly organized ductal branches from earliest rudimentary ducts. The biophysiological property of these ductal branches is characterized by continuous proliferation, migration, and differentiation of ductal epithelial cells and their adjacent myoepithelial cells, thus creating epical-basal luminal buds also referred to as acini, a basic functional unit of the mammary gland. During pregnancy and lactation, the mammary glands undergo vigorous proliferation and differentiation into a fully branched ductal network that orchestrates a secreted duct system capable of producing and collecting milk protein. It is well established that proper organization, maintenance, and function of the mammary ducts are largely ascribed to cell-cell adhesion and polarization of ductal epithelium and its interaction with extracellular matrix (ECM). These epithelial cells coordinate together to generate and maintain a polarized cellular layer that is surrounded by myoepithelial cells and ECM, contrary to inner acinar cells that lack attachment to ECM and rapidly undergo apoptosis, a death program analogous to anoikis. The ECM-rich basement membrane is able to interact with epithelial cells through binding different integrins accumulated at the abluminal membrane and to induce activation of FAK, PI3K, and Bcl2; thus enhancing the epical-basal polarization and survival of epithelial cells. YKL-40, also known as human cartilage glycoprotein-39 or chitinase-3-like-1, is a secreted glycoprotein originally identified from culture medium of a human osteosarcoma cell line MG-63. Structural analyses of YKL-40 have demonstrated that YKL-40 is highly conserved in human, porcine, cow, mouse, rabbit, and goat. Putative YKL-40-like proteins were also found in *Drosophila*, bacteria, and zebra fish. Human YKL-40 protein contains an open reading frame of 383 amino acids with a molecular mass of 40 kDa and it falls into a member of chitinase or 18-glycosyl-hydrolase gene family which can bind oligosaccharide. But it does not have chitinase/hydrolase activity because of the substitution of an essential glutamic acid with leucine in the chitinase-3-like catalytic domain. YKL-40 is normally expressed by a number of different cell types that include chondrocytes, synoviocytes , vascular smooth muscle cells, macrophages, and neutrophils. However, its normal function in these cells is incompletely understood. In a broad spectrum of human inflammatory diseases, serum levels of YKL-40 are elevated, including bacterial infections, rheumatoid arthritis, osteoarthritis, hepatic fibrosis, and bowel lesion. Thus the pathologic function of YKL-40 is implicated in the tissue remodeling and macrophage differentiation. Recently, YKL-40 null mice have exhibited markedly diminished antigen-induced Th2 inflammation and impaired macrophage activation and differentiation. Over the past decade, multiple independent studies have demonstrated that high serum levels of YKL-40 are associated with metastasis and reduced survival in a variety of human carcinomas such as breast cancer , colorectal cancer, ovarian cancer, leukemia, and glioblastoma. Consistent with these data, our recent reports unveiled an angiogenic signature of YKL-40 in the development of breast cancer and brain tumor. Although membrane receptors specific for YKL-40 binding have not yet been identified, the heparin-binding affinity of YKL-40 appears to be essential for its activity, resembling the heparin-binding property of other proteins such as extracellular matrix protein vitronectin and angiogenic factors bFGF and VEGF. Furthermore, YKL-40 was able to induce focal adhesion kinase (FAK) and MAP kinase Erk1/2 signaling cascades that mediate cell adhesion, spreading, survival, and migration in vascular endothelial cells. Likewise, YKL-40 displayed the ability to trigger phophoinositide 3-kinase (PI-3K)/AKT and MAPK signaling that regulates mitogenesis and survival of fibroblastic cells. While the expression levels of YKL-40 in normal mammary tissue remain to be investigated, there is compelling evidence showing that YKL-40 levels are rapidly increased in the initiation of mammary tissue involution as compared to the levels during pregnancy and lactation. For example, oligonucleotide microarray data analyzing a pregnancy-involution cycle of the mammary tissue demonstrated that YKL-40 was ranked as one of the top candidates in 145 up- regulated genes specific for involution. Consistent with these gene microarray data, YKL-40 protein levels were also detectable in milk secretion from weaning gland of bovine and goat, but were not detectable from lactating mammary tissue. These data suggest that elevated YKL-40 may be associated with mammary gland regression. To establish a functional role for YKL-40 in the normal mammary gland morphogenesis, here we tested the hypothesis that YKL-40 inhibits mammary epithelial differentiation and development during mammary gland development. # Results To investigate expression levels of YKL-40 during mammary tissue development, we utilized an immunohistochemical (IHC) approach in a series of different ages of parous and non-parous mice, including virgin mice (3-month old), mothers at the beginning of involution (7-month old), and parous animals (10-month old). Hematoxylin and eosin (H & E) analysis revealed that a few mammary ducts were scattered in fat-rich stroma from both virgin mice and parous mice, whereas extensive secreted ducts were observed in the tissue from early involution. In IHC staining, expression levels of YKL-40 were detectable exclusively in ductal epithelial cells but not in others (*e.g.* myoepithelial cells) in virgin mice. However, its levels were noticeably evaluated in the ductal epithelial cells from weaning tissue. After involution, the remaining ducts markedly decreased expression of YKL-40, suggestive of its function associated with ductal regression. The specificity of this anti-YKL-40 antibody (rAY) was validated by a test that pre-incubation of recombinant YKL-40 with rAY protected the interaction between tissue-derived YKL-40 and rAY, whereas collagen IV failed to resemble the inhibition of recombinant YKL-40. In addition, pre-immune serum rabbit IgG did not recognize YKL-40 in the tissue. Staining with an epithelial marker cytokeratin-8 (CK-8) showed that ductal epithelial cells of the virgin mammary tissue expressed CK-8, but its level was significantly reduced during involution. In the parous mice, CK-8 was resumed to the basal levels expressed in the virgin animals. In contrast, a myoepithelial marker smooth muscle alpha actin (SMa) remained at a strong expression level in the tissue from all different ages, indicating that myoepithelial cells do not play a major role in the mammary gland development. Interestingly, an epithelial surface marker E-cadherin (E-cad) responsible for cell-cell adhesion was significantly decreased during and after weaning compared with the ducts from virgin mice. These data suggest that ductal epithelial property is lost during involution and this physiological change may be associated with elevated expression of YKL-40. This *in vivo* data has encouraged us to test the likelihood that increased expression of YKL-40 inhibits differentiation of mammary epithelial cells. To explore this, we employed an *in vitro* three dimensional (3-D) matrix culture system in the presence of lactogenic hormones including prolactin (PRL), hydrocortisone (HDC) and insulin (INS), which represents the most common *in vitro* model capable of mimicking acinus-like secretion, differentiation, and polarization of ductal epithelial cells *in vivo*. A normal human mammary epithelial cell line 76N MECs were employed for the study. 76N MECs were initially established through transduction of a gene encoding human telomerase catalytic protein, and they possessed mammary stem cell/progenitor cell properties with ability to self-renew and differentiate into luminal epithelial and myoepithelial cells. 76N MECs were plated in growth factor-reduced Matrigel supplemented with PRL, HDC, and INS over 14 days. These cells aggregated to form an outer layer of a sphere, typically creating a lumen in the center. This phenotype demonstrated an apical-basal polarization of ductal epithelial cells. In contrast, lack of PRL, HDC and INS failed to stimulate this acinar polarization. In order to validate the differentiation and polarization capacity of 76N MECs, we tested for both a basal cell marker integrin α6 and an apical cell marker phosphoEzrin/Radixin/Moesin (pERM). As shown in, the sphere demonstrated the polarization with strong staining of integrin α6 on the basal side and pERM on the apical surface. Secreted milk protein was accumulated in the lumen. In addition, cytoskeleton protein actin and E-cad were also stained by the spheres. Altogether, these results suggest that 76N MECs retain differentiation properties of mammary epithelial cells capable of orchestrating a secreted luminal epithelium *in vitro*. Next, we treated these cells in the 3-D system with recombinant protein YKL-40 that was isolated and purified through a baculoviral infection system. YKL-40 was found to inhibit the acinar structure in 14-day culture by 60% of the controls in the absence or presence of YKL-40 small peptide (ySP) that contains 20 amino acids of the C-terminus of YKL-40. Immunocytochemical analyses of acinus-like structures unveiled that YKL-40 dramatically restrained the alveolus formation and growth, as pERM and integrin α6 did not correspondingly localize to an apical and basal side in the presence of YKL-40, in contrast to control cells which exhibited a larger, polarized acinus and produce milk protein in the lumen. YKL-40 did not influence 76N MEC proliferation as compared to the control cells or the cells treated with YSP in a monolayer culture dish. These data suggest that YKL-40 displays the ability to block 76N MEC secretion, differentiation, and polarization in the presence of lactogenic hormones. In an attempt to further determine the regulation of epithelial secretion by YKL-40, we measured expression of a milk protein β-casein in 76N MECs by immunoblotting and RT-PCR analysis. Consistent with the early immunocytochemistry results, β-casein was upregulated at both transcriptional and translational levels in response to PRL, HDC, and INS. However, treatment with YKL-40 abolished the induction of β-casein, confirming its inhibitory impact on epithelial secretion. Next, to address if YKL-40 has the ability to alter expression of E-cad and MMP-9, both of which play an important role in the cell polarity, motility, and extracellular matrix remodeling, the critical mechanisms that mediate mammary gland regression during involution. YKL-40 noticeably inhibited expression of E-cad but increased MMP-9 relative to controls treated with ySP. Accordingly, we evaluated cell motility using a cell migration assay. YKL-40 increased cell migration approximately 2-fold greater than did ySP. Collectively, these data suggest that treatment of mammary epithelial cells with YKL-40 leads to inhibition of cell secretion, differentiation and polarization, and increases in cell motility. To further validate this inhibitory signature for YKL-40 in the cells, we next determined if genetically acquired expression of YKL-40 can resemble the functions acted by recombinant YKL-40. Given that endogenous levels of YKL-40 in 76N MECs were not detectable (data not shown), we engineered a full length YKL-40 cDNA into the cells to express ectopic YKL-40. These cells resembled the activities of the parental 76N MECs exposed to recombinant YKL-40 in the Matrigel as studied earlier. Compared to vector control cells, 76N MECs expressing YKL-40 formed smaller acinus-like spheres, as a total of spheres were reduced by 45%. Consistent with this inhibitory phenotype, 76N MECs expressing YKL-40 also demonstrated disruption of the apical-basal polarity, when the spheres were stained with pERM and integrin α6. The results strengthen our hypothesis that over-expression of YKL-40 inhibits epithelial cell differentiation and polarity. We next sought to identify the impacts of YKL-40 on the expression of β-casein, E-cad, MMP-9, and corresponding invasive activity in these cells. Consistent with recombinant YKL-40 activity, ectopic expression of YKL-40 by 76N MECs resulted in suppression of β-casein and E-cad, but induction of MMP-9 and its activity. Accordingly, YKL-40-expressing 76N MECs also demonstrated increased motility as cell migration was elevated by 72% relative to the control. To further evaluate the invasive behavior acquired by the expression of YKL-40, we plated these cells on a diluted Matrigel that allows motile cells to spread and migrate. Following 7-day culture, 76N MECs expressing YKL-40 migrated and invaded the Matrigel, the phenotype contrary to the control cells that were restricted to grow as spheres. These data are highly in line with our earlier findings employing recombinant YKL-40, demonstrating that YKL-40 plays a key role in the inhibition of epithelial cell differentiation and the increase of cell motility *in vitro*. Furthermore, these findings supported the notion from *in vivo* studies that strong induction of YKL-40 in the early involution is associated with impaired mammary epithelial property and subsequent mammary gland regression. We also transplanted 76N MECs expressing YKL-40 into pre-cleared fat-pad tissue of SCID/Beige mice to monitor whether or not over-expression of YKL-40 has pathologic effects on normal mammary duct development. Mammary tissue developed from 76N MECs expressing ectopic YKL-40 exhibited well-organized epithelial ducts surrounded by fat-rich stroma, the phenotype identical to that found in mammary tissue derived from 76N MECs expressing an empty vector or host native cells in different ages of mice including virgin mice, mothers at the initiation of involution, parous and non-parous animals in H & E (data not shown) and IHC analysis of YKL-40. Although higher levels of YKL-40 were expressed by ductal epithelial cells in mammary tissue bearing YKL-40-expressing 76N MECs than that in counterparts or host mammary tissue in the different ages of mice except the involution period, none of pathogenic events was observed. In addition, IHC analyses for ductal differentiation and other activities in these ages did not show significant difference between mammary tissues containing YKL-40-producing 76N MECs, control 76N MECs, or host native cells, including: CK-8, E-cad, SMa, estrogen receptor (ER), progesterone receptor (PR), cell proliferation marker Ki67, and cell apoptosis staining TUNEL (data not shown). The data suggest that YKL-40 over-expression in normal mammary epithelial cells does not have the ability to induce ductal pathogenesis. # Discussion Our current study has utilized multidisciplinary approaches including genetic engineering, 3-D Matrigel cultures, and cleared fat-pad xenotransplantation to characterize a functional role for YKL-40 in the normal mammary gland development. To our knowledge, this is the first time characterizing YKL-40 in the mammary tissue morphogenesis, which will enhance our understanding of biological and physiological properties of YKL-40 in this field. We have found that YKL-40 is expressed merely by ductal epithelial cells in normal breast tissue at a low level throughout the life time except involution. This evidence suggests that the low expression of YKL-40 may be sufficient for the development of a limited numbers of mammary ducts in non-pregnant mammary tissue. In weaning, YKL-40 is markedly up-regulated in ductal epithelial cells, suggesting that YKL-40 may mediate mammary epithelial remodeling and regression. Consistent with our findings, milk levels of YKL-40 in goat and bovine were increasingly detectable during weaning; but were not detectable in lactation. Furthermore, in the mimicking of an *in vivo* environment for ductal morphogenesis, we used a 3-D *in vitro* culture system in the presence of lactogenic hormones and found inhibitory effects of YKL-40 on epithelial cell differentiation, secretion, and polarization, highlighting a physiological role of YKL-40 in the inhibition of mammary duct differentiation during involution. It would be quite interesting to understand molecular mechanisms of how YKL-40 suppresses epithelial cell secretion and differentiation induced by lactogenic hormones. During mammary gland development, epithelial morphogenesis and lateral ductal branching are controlled by a wide array of intra- and extra-cellular cues that are initiated from a variety of factors including: steroid hormones (*e.g.* estrogen and progesterone), polypeptide hormones (*e.g.* PRL, placental lactogenes), and growth factors (*e.g.* EGF, FGF, TGF-β, and insulin), ECM (*e.g.* laminin), and their binding receptors (*e.g.* integrins). In addition, a cell death process anoikis also plays an active role in the polarization of luminal epithelium, the event that normally contributes to the establishment of a hollow ductal phenotype due to deprivation of ECM in the core of acini. Aberrant expression or dysfunction of these factors could result in disruption of epithelial differentiation and acinar outgrowth, thus mediating or leading to a pathogenesis associated with mammary tumorigenesis. However, over-expression of YKL-40 alone by epithelial cells in the current study does not initiate pathogenesis towards epithelial dysplasia, hyperplasia, or carcinogenesis. These data suggest that YKL-40 is not tumorigenic. But, it may collaborate with other oncogenic factors in participation of tumorigenesis, as YKL-40-mediated inflammation is implicated in the development of ulcerative colitis-associated neoplasia in colorectal tissue. It is well established that elevated YKL-40 is associated with the malignancy of breast cancer. For example, there is accumulating evidence indicating that elevated serum levels of YKL-40 correlate with breast cancer progression and decreased disease-free survival. We recently identified that cancer tissue expression of YKL-40 was associated with tumor vasculature formation, demonstrating the angiogenic property for YKL-40 during breast cancer development. In context with current findings, all the data indicate that YKL-40 plays a pathological role in the late stages of breast cancer, rather than in the initial phase of the cancer. Normal mammary epithelial cells express strong E-cadherin, a membrane protein characteristic of cell-cell tight contacts. Dysfunction of E-cadherin is associated with loss of intercellular adhesion, the process of which is fundamental during mammary gland involution. For example, truncated E-cadherin in mammary epithelial cells led to cell-cell disassociation and thus loss of ductal epithelial polarization. In addition, increased expression of MMPs in mouse mammary epithelial cells was correlated with decreased E-cadherin, both of which mediate disruption of cell polarity and contribute to invasiveness. Our current study found that YKL-40 inhibited E-cadherin and induced MMP-9 expression in 76N MECs, concurrent with the changes in cell polarity and invasiveness in 3-D Matrigel culture, thus demonstrating an in important role played by YKL-40 during mammary tissue remodeling and involution. We also interestingly found that E-cadherin was decreased in parous mammary tissue. The data suggest that its regulation may also be dependent on other factors, not limited to YKL-40, such as hormones and their receptors, because E-cadherin expression in breast cancer is associated with increased levels of estrogen receptor. There is a wealth of evidence suggesting that elevated serum levels of YKL-40 in breast cancer patients serve as a cancer prognostic biomarker. However, a compelling study tested and compared YKL-40 levels between the blood and nipple aspirate fluid (NAF) from healthy women and patients with either breast precancer or cancer. Both pre-cancer and cancer patients contained higher concentrations of YKL-40 in NAF than did disease-free women. In addition, YKL-40 in NAF levels from health subjects was 600-fold higher than serum levels of YKL-40 in these normal women or cancer patients. Thus it suggests that NAF levels of YKL-40 may serve as a more sensitive marker than serum levels in the assessment of breast cancer progression. However, substantial epidemiological analyses are essential to firmly establish the relationship between epithelial expression of YKL-40 and pathogenesis of breast cancer from the same patients. In summary, the data presented here identified that YKL-40 expressed exclusively by ductal epithelia has the ability to inhibit mammary epithelial secretion and differentiation, impair epithelial polarization, and facilitate cell motility in the presence of lactogenic hormones, an essential mechanism that mediates mammary tissue remodeling during involution. These findings underscore the biophysiological activities for YKL-40 and also support the evidence that YKL-40 is significantly elevated in the involution. Therefore, elucidation of its inhibitory activity in mammary epithelial cell differentiation and mammary gland regression will help understand its pathological role in the breast cancer progression that is associated with poor differentiation of cancer cells. # Materials and Methods ## Culture of 76N MECs The cells (kindly provided by Drs. Vimla Band and Sallie Smith Schneider) were grown in RPMI 1640 medium with glutamine (Invitrogen) in the presence of 5 µg/ml insulin, 10 µg/ml hEGF, 5 µg/ml hydrocortisone (Sigma, MO, USA), and 10% FBS (Invitrogen). ## Generation of 76N cells stably expressing YKL-40 Full length of YKL-40 cDNA was subcloned into a retroviral pCMV-neo-vector. 293T retroviral packaging cells were transfected with YKL-40 or vector control DNA in the presence of pCL 10A1 vector using Fugene 6 (Roche, IN) as the delivery vehicle. Forty-eight hours after transfection, the supernatant was harvested and filtered through 0.45-µm pore size filters and the viral medium was used to infect 76N MECs. Selection with 800 µg/ml neomycin was started 48 hr after infection and the drug-resistant cell populations were used for subsequent studies. ## Purification of recombinant YKL-40 and polyclonal rAY Full-length human YKL-40 cDNA with a His-tag was subcloned into a pFastBac1 vector (Invitrogen, CA). Following transformation and amplification in DH10Bac E. coli, bacmid DNA containing YKL-40 was transfected into Sf9 insect cells by using CellFECTIN reagent (Invitrogen) and baculoviral medium was produced. A Ni- NTA column was used to purify recombinant YKL-40 according to manufacture’s instruction (Invitrogen) and YKL-40 pure protein was finally produced through a PD-10 desalting column (Millipore, CA). rAY was purified through an Econo-Pac serum IgG purification kit (BioRad, Hercules, CA) once it was generated from immunization of rabbits with a short peptide of YKL-40 encoding C-terminus of YKL-40. ## 3-D Matrigel assay 76N MECs (2×10³) were suspended with 50 µl of assay medium containing 4% growth factor-reduced Matrigel, 5 µg/ml PRL, 5 µg/ml HDC, 5 µg/ml INS and 1% heat-inactivated FBS, and then were transferred onto 96-well plates pre-coated with a layer of Matrigel and incubated for 10-14 days. In some conditions, the cells were suspended with the assay medium in the presence of 100 ng/ml YKL-40 or ySP. Top gel medium was replaced with assay medium every 4 days. To evaluate cell invasive behavior, Matrigel concentration in the assay medium was reduced to 2%. ## Migration assay 76N MECs (2×10⁵P) were transferred onto transwells (8 µm, 24-well plates) pre-coated with collagen IV (100 µg/ml). The lower chamber of transwells included RPMI medium with YKL-40 or ySP. After 4–6 hours of incubation, cells migrated to the membrane were quantified. For migration of 76N MECs expressing YKL-40 or vector, lactogenic hormones (PRL, HDC and INC) were added in the lower chamber. ## Gelatin Zymography Cell-conditioned serum-free media from 48 hr culture were collected for a zymograph analysis as described previously. ## Western Blot Analysis Cells were collected by scrapping and extracted in a lysis buffer (pH 7.4) containing 0.25 mM HEPES, 14.9 mM NaCl, 10 mM NaF, 2 mM MgCl₂, 0.5% NP-40, 0.1 mM PMSF, 20 µM pepstatin A and 20 µM leupeptin. The lysates were centrifuged at 10,000 g for 10 min at 4°C and the resulting supernatants were collected for 10% SDS-PAGE. Proteins were transferred to a nylon membrane (Invitrogen) and incubated with one of several antibodies: rabbit anti-YKL-40 (rAY, 1∶200), anti-β-casein (1∶200, Santa Cruz, CA), mouse anti-MMP-9 (1∶500, VWR, NJ), E-cad (1∶500, Invitrogen), and actin (1∶1000, Sigma, MO). Specific bands were detected using an ECL kit (Pierce, VWR). ## Immunofluorescence Matrigel samples were fixed with 2% para-formaldehyde for 15 min and permeabilized with 0.5% Trition X-100 for 10 min at 4°C. Then the samples were incubated with primary antibodies in PBS-base blocking solution containing 0.1% BSA, 0.2% Triton X-100, 0.05% Tween-20, 7.7 mM NaN₃, and 10% goat serum overnight at 4°C. Primary antibodies included rabbit anti-milk protein (1∶100; Nordic Immunological Lab, Neitherlands), anti-phospho- Ezrin/Radizin/Moesin (1∶400, Cell Signaling, MA), rat anti-integrin α6 (1∶200, Millipore, MA), and mouse anti-E-cad (1∶500, Invitrogen) and anti-actin (1∶1000, Sigma) antibody. Secondary antibodies including goat anti-mouse or rabbit Alexa Fluor 488 and Alexa Fluor 555 antibody (1∶1000, Invitrogen) were added for 1 hr at room temperature followed by nuclear staining with DAPI (300 nM, Invitrogen). ## RT-PCR Total RNA was extracted by a Tri-reagent (Molecular Research Inc., OH) from cell lysates. cDNA was synthesized through a reverse transcriptional reaction in the presence of oligo(dT) and a reverse transcriptase (Promega, WI). β-Casein cDNA was amplified through PCR reaction in the presence of 5’ (CAAGGGAGACCATAGAAAGCC) and 3’ (GACACTAATGGGGTTATGAACTGGGGC) DNA primers which covered 600 bp of β-casein gene. Samples (10 µl) were loaded into a 1% agrose gel to determine gene expression levels of β-casein and GAPDH as internal controls. ## 76N MEC transplantation into mammary fat-pad free tissue in mice All animal experiments were performed under the approval of Institutional Animal Care and Use Committee of the University of Massachusetts (IACUC ID 132669). Transplantation of 76N MECs into cleared fat-pad tissue of 4-week old SCID/Beige mice was described previously. In brief, the 4^th nipples and their epithelium-containing parts of the fat pad of mammary glands were excised. The incisions were closed with stitches and seven days later, 76N MECs (6×10⁶) expressing control vector or YKL-40 were injected into the right or left cleared fat-pad tissue, respectively. Host native fat pads and 76N MECs-injected fat pads were removed from 3-month old virgin mice, 7-month old mothers (twenty-one days after giving birth), 10-month old parous mice, and 10-month old non-parous mice. ## Histological analyses Paraffin-embedded tumor tissues were cut to 6 µm thickness and processed for immunohistochemical analysis. In brief, samples were incubated with 3% H₂O₂ for 30 min to block endogenous peroxidase activity, followed by incubation with blocking buffer containing 10% goat serum for 1 hr. The samples then were incubated at room temperature for 2 hr with mouse anti- Ki-67 (1∶100, BD Pharmingen, San Diego, CA), E-cad (1∶2000, Invitrogen), ER (1∶200, Santa Cruz), PR (1∶200, Dako Inc, Carpentaria, CA)`,` and SMa (1∶500, Dako) monoclonal antibodies, or rabbit anti-YKL-40 (1∶200) and CK8 (Abcam, Cambridge, MA) polyclonal antibodies. Goat anti-mouse or anti-rabbit secondary antibodies (1∶100) conjugated to HRP were added for one hr. Finally, DAB substrate (Dako Inc) was introduced for several minutes and after washing, methyl green was used for counterstaining. For detection of tissue apoptosis, we used the TUNEL assay that was described in the instruction of a TACS.XL DAB *in situ* apoptosis detection kit (Trevigen Inc., Gaithersburg, MD). # Supporting Information [^1]: Conceived and designed the experiments: RS SS. Performed the experiments: RS SS WY BB QJC. Analyzed the data: RS SS QJC. Contributed reagents/materials/analysis tools: RS. Wrote the paper: RS. [^2]: The authors have declared that no competing interests exist.

# Introduction Iron deficiency anemia in young children is recognized as a major public health issue and the most prevalent form of micronutrient deficiency worldwide. The global prevalence of anemia (defined as hemoglobin level of \<110 g/L) in children aged 6–59 months is 43% and half is attributable to iron deficiency anemia (IDA) which is defined as hemoglobin level of \<110g/L and ferritin level of \< 12 μg/L. IDA contributes substantially to childhood mortality and morbidity and is linked to impaired brain development and cognitive functions. IDA is also ranked as the third leading cause of disability worldwide and the 13^th leading risk factor for the global disability adjusted life years. Most of the burden of IDA is in the resource poor settings of Africa and Asia. In Pakistan the reported prevalence of IDA in children under five is between 40–70%. In Pakistani children IDA has been associated with growth retardation, impaired cognition, reduced physical activity and postulated as a contributor to the high national infant mortality rate. Widespread micronutrient deficiencies along with other clinical and social factors are believed to be the leading cause of IDA in Pakistan. However prevalence data is scarce with many studies more than a decade old, or based on small numbers and in small non- representative populations. Further, IDA is best defined as the combination of both low hemoglobin and low ferritin concentrations, whereas the published studies to date from Pakistan are mostly based solely on hemoglobin concentrations. As such, nationally representative robust data are still lacking to determine the socio-demographic factors associated with IDA that will enable the development of local strategies to treat and prevent IDA in Pakistani children. In this study we aimed to estimate the prevalence of IDA in Pakistani children and to evaluate factors associated with IDA by conducting a secondary analysis of the Pakistan National Nutrition Survey that was undertaken between 2011–2012. # Materials and Methods ## Data source The data used for this analysis were derived from the Pakistan National Nutrition Survey 2011–2012. The survey was conducted by the Aga Khan University (AKU) in collaboration with the Federal Ministry of Health in Pakistan and was funded by UNICEF. The survey was a stratified representative cross-sectional national survey with provincial specificity, with a two-stage stratified sampling design. The sampling frame in the form of Enumeration blocks (1500 in number) was provided by the Pakistan Federal Bureau of Statistics. Each enumeration block was demarcated, mapped and listed before the actual data collection and from each enumeration block 20 households were selected randomly through a computer generated program. A total of 30,000 households were included, 12,360 were urban households and 17,640 were rural, resulting in 27,963 respondents. Households with children between the ages of six months to five years were included in the survey. One married woman of reproductive age (15–49 years) with at least one child aged less than 5 years was chosen from each selected household. If there were multiple eligible women in a household, one woman was randomly selected. A pre structured and pre tested instrument was used for data collection. The instrument collected data on socio-economic status, maternal reproductive history, child medical history, food security, anthropometric measurements and biochemical measurements of various micronutrients. For anthropometric measurements weight was measured using lightweight SECA scales designed and manufactured under the authority of the United Nations Children’s Fund (UNICEF) and the height/lengths was measured by height boards made by Shorr Inc. Children under 2 years of age were measured lying down on the board and standing height was measured for older children. In children more than 2 years height-for-age z scores (HAZ), weight-for-age z scores (WAZ), weight-for-height z scores (WHZ) were calculated and for children less than 2 years length was used instead of height to calculate length-for-age z scores (LAZ) and weight-for-length z scores (WLZ). Body Mass Index (BMI) was calculated for mothers using height and weight. Children were reported as stunted, wasted or undernourished and mothers were reported as underweight, overweight or obese as per the WHO classification. Blood samples for hemoglobin measurement and micronutrient deficiencies were collected from one child in each household. Venous blood was collected and serum was separated within half hour of collection, using portable centrifuge machines with backup power. The hemoglobin was measured from the venous blood at field while the serum samples were transported to nutrition research laboratory at Aga Khan University under cold chain conditions through country wide network of AKU laboratories collection centres. Serum Ferritin concentrations were measured for iron deficiency because it has the highest sensitivity and specificity to detect iron deficiency in individuals. The ferritin levels were adjusted for alpha (1)-acid glycoprotein (AGP) and C-reactive protein (CRP) to take into account that ferritin is an acute phase protein and is raised in inflammatory conditions. Serum Ferritin was measured by electro-chemiluminescence immunoassay using Roche Cobas E41 Chemistry Analyzer and hemoglobin concentration was measured in whole blood using a Micro cuvette containing a dry reagent system and a dual wavelength HemoCue 201 Photometer. Hemoglobin concentration was also adjusted for altitudes of more than 1000 meters using the WHO recommended altitude formula. Blood samples were also analyzed for zinc deficiency and vitamin A deficiency. Serum was used to estimate the zinc and vitamin A deficiency and the samples were analyzed using Spectrophotometric method for zinc deficiency and Reversed Phase Chromatography method for vitamin A deficiency. All survey activities were monitored to ensure the data quality. The questionnaire was pre-tested prior to implementation in the field and a pilot survey was also conducted. Competency of field staff was also taken in account prior to hiring and pre-test and post-test was conducted for all field staff during trainings. The survey employed internal and external data monitors to ensure data quality with respect to the all the procedures and survey activities. The anthropometry and equipment HemoCue was calibrated daily for any possible errors. The team leaders analyzed the data (plausibility checks and digit preference) using ENA Smart software and provided regular feedback for improvement. Similar quality assurance steps were duly considered during data entry and cleaning. Ethical approval for the survey was obtained from the Ethical Review Committee of Aga Khan University and the National Bioethics committee of Pakistan. Written informed consent was obtained from the mothers of the selected child. In case of illiterate mother, consent was documented by a thumbprint on the consent form and a signature by a literate witness. All the names and personal information regarding the participants were kept confidential and data set was kept anonymous for analysis. ## Description of Variables Child IDA status was the main outcome variable and was defined as the having the combination of haemoglobin levels of \<110 g/L and ferritin levels of \< 12 μg/L. The selection of explanatory variables for analysis was informed by the literature and their availability in the dataset and is fully described in. Variables were grouped into three categories; household, maternal, and child. Under the household category area of residence which was categorized into rural and urban, socioeconomic status which was ascertained by computing the wealth quintiles (from being poorest to wealthiest) using the standard demographic and health survey tool and food security status using the standard household food insecurity access scale developed by Food and Nutrition Technical Assistance (FANTA) project and categorized as (Food Secure and Food insecure) were considered. In the maternal category maternal education which was defined as years of education completed (illiterate/years of education), maternal IDA estimated on the basis of low hemoglobin (\< 120 g/L) and low ferritin (\< 12 μg/L) concentrations and maternal BMI calculated on the basis of height and weight of mother and categorized as \<18.5 underweight, 18.5–24.99 normal weight, 25–34.99 overweight/obesity and ≥35 severe obesity, were examined. In the child variables sex, age, clinical anemia estimated through physical examination of conjunctiva, vitamin A deficiency defined as serum vitamin A concentration of \< = 0.70 μmol/L and categorized as deficient and non- deficient, zinc deficiency defined as serum zinc concentration of \<60 μg/dL and categorized as deficient and non-deficient, stunting defined as height/length- for-age z-score of \< −2, wasting defined as weight for height z-score of \< −2, underweight defined as weight for age z-score of \< −2 as per and history of worm infestation ascertained through recall of worm infestation in last six months. ## Statistical Analysis: All data analyses were conducted in IBM SPSS version 19 using a complex sample procedure to allow for adjustments of the sampling design implemented in the survey. The frequencies, along with weighted percentages, were reported for selected predictors and the mean with 95% confidence intervals (CIs) were calculated for the outcomes. The analysis started with simple univariate analysis followed by multivariate logistic regression. Unadjusted odds ratio with their 95% CIs were reported for the bivariate analysis. Variables significant at p\<0.25 were considered for inclusion in the multivariate model. Covariates that were insignificant at the multivariate level were dropped consecutively from the model after careful assessment of confounding. The final model was selected on the basis of theoretical and statistical significance of predictors. The Type 1 error rate was set at 0.05. The model estimates are presented with the adjusted odds ratios and 95% CI. # Results A total of 7138 children aged between 6–59 months were included in the analysis among them 2373 (33.2%) children had IDA based on low hemoglobin and ferritin levels. Analyses of the child blood samples found that overall (4264) 62.3% were anemic (hemoglobin concentrations \<110 g/L), of those who were anemic (283) 4.1% were severely anemic (\< 70 g/L) and (3981) 58.3% were moderately anemic (70 g/L – 109 g/L). Ferritin deficiency was identified in (3361) 47.1% children. Among those who had IDA the mean values for hemoglobin and ferritin concentrations were 70 g/L and 6μg/L respectively. The prevalence for vitamin A deficiency and zinc deficiency was 52.6% and 38.8% respectively. Clinical anemia was found in 30.1% children. The anthropometric data identified 44.5% of the children as being stunted, 15.5% as wasted and 33.6% as underweight. Among maternal factors the prevalence of IDA was found to be 20.0%, which was determined by using maternal hemoglobin and ferritin levels, and according to the maternal anthropometrics 16.0% of mothers were classified as underweight, 21.2% overweight and 11.1% obese. Among all children included in the analysis the majority of children came from rural areas (70.1%), 48.3% of children were female and 47.9% were under the age of 24 months. The two lowest wealth quintiles accounted for 43.3% of the study population, and the two highest was 36.0%. Of the sampled households 64.3%were food insecure. Only 8.3% of the mothers reported that their child had worm infestation in last six months. The univariate analysis showed that the IDA in children under five years of age in Pakistan was significantly associated with: age of less than 24 months (OR 1.45 1.29–1.62 p \<0.05), stunting (OR 1.44 1.27–1.64 p\<0.05), being underweight (OR 1.28 1.15–1.43 p\<0.05), presence of clinical anemia (OR 5.03 4.41–5.74 p\<0.05), history of worm infestation (OR 1.33 1.10–1.62 p\<0.05), having a mother with IDA (OR 1.91 1.65–2.21 p\<0.05) and household food insecurity (OR 1.23 1.08–1.41 p\<0.05). In the multivariate regression only being underweight was no longer a significantly increased risk with childhood IDA and of the variables that remained the adjusted odds ratios were slightly attenuated: age \< 24 months (AOR 1.40, 95% CI 1.18–1.55 p \<0.05), stunting (AOR 1.42 CI 1.23–1.63 p\<0.05), having a mother with IDA (AOR 1.72 CI 1.47–2.01 p\<0.05) and household food insecurity (AOR 1.20 CI 1.10–1.40 P\<0.05), except for presence of clinical anemia for which the adjusted OR increased (AOR 5.69 CI 4.93–6.56 p\<0.05). In contrast, living in a rural area (AOR 0.77 CI 0.65–0.90 p\<0.05) and being a female child (AOR 0.87 CI 0.76–0.98 p\<0.05) was associated with reduced odds of IDA. # Discussion In our study, the prevalence of IDA in children aged 6–59 months was 33.2% which according to the WHO criteria represents a ‘moderate burden’. We also found a substantial prevalence of low hemoglobin levels, vitamin A deficiency, zinc deficiency, stunting, wasting, underweight and food insecurity amongst children aged 6–59 months living in Pakistan. The prevalence of IDA in our study is lower than in previous studies in other low resource countries such as Palestine and Kenya and substantially below the estimate from a previous study in Pakistan. These differences could be attributed to variations in the study settings or factors such as the rate of parasitic infections and dietary habits. The previously published study from Pakistan, reported a prevalence of IDA of 63% from a smaller sample (n = 320) taken from a semi urban area, which is not comparable with this population based survey. Additionally it was conducted 18 years ago, before the introduction of the iron fortification in Pakistan. Our findings are consistent with IDA prevalence reported in Kazakhstan (32.4%), Yemen (34.2%) and the Pakistan National Nutrition Survey (36%) undertaken in 2001 but are higher than studies conducted in Morocco (20.4%), India (23.1%) and Iran (29.1%). Our study found that the food insecurity status of households is significantly associated with IDA in children, a finding that is consistent with the available literature examining this association \[, , and \]. Food insecurity is characterized by either unavailability of food or inability to procure and access food and has consequences in both macro and micronutrient deficiencies. In Pakistan the widespread food insecurity situation reflects the economic instability of many areas of the country. The study found a relationship between maternal iron deficiency anemia and IDA in children that confirms previous reports and highlights that IDA is common in pregnant and non-pregnant women of reproductive age in Pakistan. In our study clinical examination of children detected clinical anemia in 53.6% of IDA cases. This sign can therefore assist in diagnosis where facilities for biochemical testing are not available. We found the prevalence of IDA to be significantly associated with children’s age, with the youngest children having the highest odds of IDA. This finding is consistent with similar studies conducted in Iran, India and the Philippines. The first two years of life is a period of rapid growth with an increased iron requirement therefore risk of IDA is increased in this age. Moreover factors such as limited access to iron rich food, inadequate infant and young child feeding practices including lack of exclusive breast feeding, prolonged breast feeding and inappropriate weaning food and recurrent illnesses increase the chance of young children developing IDA \[, and \]. We also found that the odds of IDA increased when the child was stunted and food insecure, suggesting that malnutrition is a contributing factor for IDA. After adjusting for other factors, our study found that female children and those residing in rural areas were less likely to develop IDA. There is conflicting evidence with regard to the relationship between sex and IDA in children. Contrary to our study, data from Yemen and India found a higher prevalence of IDA in girls than boys; however studies from Western Kenya and Haiti found boys to be more at risk. In our setting children living in rural areas have more access to green leafy vegetables which may explain their reduced risk of IDA. The major strength of this study is the nationally representative sample using reliable methods to detect iron and adjustment for confounding factors. However this study also has some limitations. The cross-sectional nature of the study means that the temporal relationship between IDA and the associated factors cannot be established. The information on various indicators was collected using a structured questionnaire that may have been prone to recall bias and presence of some missing data is also a potential limitation. Moreover we did not measure serum vitamin B12 and folate levels and so cannot measure other forms of anemia, suggesting some misclassification bias. The data presented provides valuable insight into the prevalence of IDA in children which has remained fairly consistent in Pakistan over the last decade. Considering the burden of IDA in Pakistani children it is essential that interventions such as iron supplementation, food fortification and diet diversification should be used at scale, currently iron supplementation and wheat flour fortification are successful strategies being used worldwide. There is good evidence to suggest that iron supplementation improves hemoglobin level and reduces IDA prevalence in children. Studies conducted in Pakistan also reveal iron supplementation to children and mothers results in improved iron stores. Further a recent pooled analysis and studies in Central Asia, Venezuela and Iran suggest that wheat flour fortification can significantly improve iron status at the population level but there is a lack of such supplementation programs in Pakistan. To reduce the burden of IDA, Pakistan needs a holistic approach of short term vertical programs such as iron supplementation and long term horizontal programs including wheat flour fortification. Specifically, further efforts should be made to restore the national wheat flour fortification program as wheat is the staple diet of Pakistan. Additional investments are also required to improve; education, especially for females, food insecurity through food supplements and agricultural support. These may require novel methods such as behavioral change communications to improve dietary patterns and infant and young child complementary feeding (IYCF) practices. Finally for this generation of Pakistani children to benefit it is important that Pakistan acts swiftly to alleviate the prevalence of IDA. This manuscript is a part of MAH’s thesis to fulfill the requirement for a PhD at the University of Sydney. We are grateful to the Women and Child Health Division, Aga Khan University for providing the data of National Nutrition Survey 2011–12 for secondary analysis. We are also thankful to the University of Sydney for funding MAH's PhD scholarship (IPRS/APA) and CRG's funding through an NHMRC career development fellowship. We would like to thank all the participants who took part in the study, the interviewers, the laboratory technicians, the data editors, and the data operators. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ZAB SBS MAH KB CRG. Performed the experiments: MAH SBS IH. Analyzed the data: ZB MAH. Contributed reagents/materials/analysis tools: ZAB ZB SBS MAH KB CRG. Wrote the paper: MAH KB CRG SBS ZAB.

# Introduction Retinitis pigmentosa (RP) is one of the most common forms of inherited retinal degenerations, affecting about 1:3.000 individuals. It is characterized by a primary rod photoreceptor degeneration and consecutive cone photoreceptor death, leading to night blindness and subsequent progressive vision loss. Legal blindness at working age is common in RP patients, with devastating professional and social implications for the patients. Although typical characteristics of RP have been described, there is broad phenotypic variability, regarding both clinical presentation and age of onset. RP is genetically heterogeneous, with more than 60 disease-causing RP genes reported so far (RetNet, <https://sph.uth.edu/retnet/>). With novel upcoming therapeutic options such as gene therapy, the identification of the disease-causing mutations has gained importance. Diagnostic genotyping in RP has become very efficient with the implementation of next-generation sequencing (NGS),\[–\] enabling parallel sequencing of all known RP (and, if needed, many more) genes which improves our understanding of the disorder’s pathogenesis. Here, we report unexpected molecular findings and novel genotype-phenotype correlations in a cohort of 116 consecutive and unrelated patients seen in a retinal dystrophy clinic of a German tertiary referral center. In addition, the application of targeted NGS of all known RP genes provides insight into the mutational spectrum of this representative cohort. # Methods ## Patients This retrospective single-center cross-sectional study included 116 consecutive and unrelated patients who were investigated at the Department of Ophthalmology, University of Bonn, Germany. After clinical diagnosis of RP, molecular screening was performed. The study was in adherence with the declaration of Helsinki. Institutional review board approval (Ethics Committee, Medical Faculty, University of Bonn, Germany) and patients' informed consent were obtained. All patients underwent genetic counseling, which included family history assessment. Asymptomatic relatives of index patients received genetic counselling prior to mutation testing. ## Image acquisition and functional testing Clinical assessment included standardized anterior segment and dilated fundus examination, best corrected visual acuity (BCVA), and visual field testing (30–2 threshold program; HFA II; Carl Zeiss Meditec, Dublin, CA, USA), and, in selected cases, electroretinography (ERG). Retinal imaging consisted of spectral domain optical coherence tomography (OCT), fundus autofluorescence (AF) imaging (both, Spectralis HRA+OCT, Heidelberg Engineering, Heidelberg, Germany), fundus photography (Zeiss, Visucam, Oberkochen, Germany) and wide-field fundus imaging (Optos PLC, Dunfermline, United Kingdom). ## Molecular genetic analysis Genomic DNA was extracted from blood lymphocytes by a standard protocol. A two- tier procedure was implemented for patients suggestive for autosomal recessive or sporadic RP. A step-wise molecular screening was chosen, with initial Sanger sequencing of genes commonly involved in the pathogenesis of the suspected genetic RP subtype (*EYS* and *RP1* for autosomal recessive RP, *PRPF31* for autosomal dominant RP). If no mutation was identified in this initial sequencing of singular genes, targeted NGS on an Illumina Hiseq1500 system was carried out for the RP genes known at the respective time of analysis after enrichment using NimbleGen sequence capture technology (as described previously). Importantly, the NGS panels also contained the genes for clinically overlapping conditions such as cone/cone-rod dystrophies, Leber’s congenital amaurosis (LCA) and syndromes with retinal dystrophies, allowing for an extended genetic assessment if no mutation was found in genes previously associated with RP, without need for additional experimental efforts. To detect X-linked RP-causing mutations, we added NGS of amplicons comprising *RPGR*_*ORF15* to panel-NGS of the remaining exons of *RPGR* and *RP2*. Based on our step-wise molecular screening, Sanger sequencing and targeted NGS identified disease-causing mutations in 20 and 61 cases, respectively. Verification of mutations identified in NGS and segregation analyses were carried out by PCR and subsequent Sanger sequencing. To determine the most likely inheritance mode, pedigrees were generated based on the patients’ family history including at least three generations. Autosomal recessive inheritance was assumed in case of parental consanguinity and/or if only siblings were affected. Autosomal dominant inheritance was assumed if there was a positive family history for at least three successive generations, and likely autosomal dominant if different but not successive generations were affected in absence of known consanguinity (e.g., assuming reduced penetrance, or because a linking individual died early or lost contact). X-linked inheritance was assumed in families with only males being severely affected and at least considered if male-to-male transmission was lacking in families with RP patients in different generations. RP was categorized as sporadic in case of negative family history. If no family history was available, e.g. because the patient was adopted and without contact with his biological parents, no candidate inheritance mode was defined. Variants were filtered against dbNSFP v2.0, dbSNP v137, gnomAD (exomes) and the Human Gene Mutation Database (HGMD Professional 2017.3). The cut-off for the maximum minor allele frequency (MAF) was set to 1%. Nonsense, frameshift, large deletions and canonical splice site variants were regarded pathogenic. Rare non-synonymous single nucleotide variations were considered likely pathogenic when at least half of the algorithms of used in silico prediction software tools predicted that the variant is probably damaging and when it was predicted as conserved with conservation prediction algorithms. Functional predictions were carried out using SIFT, PolyPhen2, MutationTaster, MutationAssessor, FATHMM, LRT, VEST, CADD, PROVEAN and DANN. Splice sites were predicted with AdaBoost and RF. Assessment of conservation was done by PhyloP, GERP++, PhastCons, SiPhy, Grantham Distance and BLOSUM62. # Results and discussion Approximately 99% of all coding exons were covered at least 20-fold. Disease- causing mutations were identified in 81 (70%) out of the 116 analyzed patients, a high diagnostic yield compared to previous reports using targeted NGS in RP patients ( and Tables).\[–, –\]. The wide range of reported diagnostic yields in cohorts \>50 patients (25%-80%)\[, –\] may be due to differences in cohort sizes, populations, inclusion criteria, differences in clinical assessment, NGS platforms, and bioinformatic pipelines. Cases without mutations in the genes analysed herein can be due to causative variants in non-coding regions of these genes or in genes not yet known to underlie retinal degeneration, or–less likely–due to mutations in regions below optimal coverage. Overall, 110 different mutations distributed across 30 different genes were detected in this study, and 46 of these mutations were novel at the time of molecular diagnosis (and Tables). The clinical classification was revised in four patients who had variants in genes usually associated with an early macular degeneration (*CERKL*, *PROM1* and *ABCA4*). They presented with very late disease stages, preventing a clear classification of the initial phenotype as cone-rod or rod-cone dystrophy. However, re-evaluations of these patients’ histories were compatible with retinal dystrophies that initially affected the central retina and/or the cone system. Classification may be particularly difficult in patients with *CERKL* mutations, as their initial symptoms may manifest in dim light conditions, although their retinal phenotype may indicate a dystrophy primarily affecting the central retina. Thus, a combination of comprehensive clinical phenotyping and molecular testing improves the accuracy of diagnoses in genetically and clinically heterogeneous diseases. About half (48%) of the remaining 77 solved RP cases were explained by mutations in four genes: *RPGR* (n = 12, 16%), *EYS* (n = 10, 13%), *PRPF31* (n = 8, 10%) and *USH2A* (n = 7, 9%). Based on the genetic findings, inheritance turned out to be autosomal recessive in 56% (n = 45), autosomal dominant in 26% (n = 21), and X-linked in 19% (n = 15) of patients. If the family history was highly suggestive for a particular mode of inheritance (n = 45), a molecular diagnosis was achieved in 82–100%, then confirming the assumed inheritance. The mutation detection rate was lower (60%; 42 out of 70 patients) in cases of sporadic RP (see below), similar to a previous reported cohort of patients with macular and cone/cone-rod dystrophies. In the following, unexpected findings identified in 21 (27%) out of the 77 solved RP patients are described in detail (variants displayed). These include a high proportion of autosomal dominant or X-linked mutations in patients with sporadic RP, novel or uncommon genotype-phenotype correlations (with no mutations in genes knowingly associated with the respective phenotype), and X-linked mutations in females with apparently sporadic or dominant RP. ## High proportion of autosomal dominant and X-linked inheritance in sporadic RP In the majority of patients with sporadic RP, the retinopathy is autosomal recessively inherited, usually with a small recurrence risk in the offspring. Although this was confirmed by our findings, it is noteworthy that X-linked (n = 10) and autosomal dominant (n = 7) mutations accounted for disease in 24% of the sporadic cases. Similar rates of X-linked RP have been reported for overall sporadic (4–6%,\[, \]) or for male sporadic cases (15%-30%). Of note, two sporadic female patients with X-linked RP were identified in this study (see below). Most autosomal dominant mutations in sporadic RP patients were identified in *PRPF31* (n = 5, \#46–50). Only one patient carried a *CRX* mutation (see below, patient \#63), and one patient had a previously described assumingly pathogenic variant in *PRPH2* (c.422A\>G, p.Tyr141Cys; patient \#67; variant not carried by her mother, no further family members were available for segregation analysis). Previously it has been shown that mutations in *PRPH2* cause highly variable and mild RP phenotypes, and index patients may be erroneously classified as simplex cases because of undiagnosed family members with mild disease. Incomplete penetrance of *PRPF31*-associated RP is a well-documented phenomenon. Our study underlines the high rate of non-penetrance in *PRPF31* mutations and/or that the family history of relatively small pedigrees is less reliable in identifying such low- or non-penetrant dominant mutations. Segregation analysis for *PRPF31* mutations was possible in three families, revealing a *de novo* mutation in one patient (#47). Healthy individuals with *PRPF31* mutations were fully examined and showed no morphological fundus changes (including peripheral AF recordings). Of note, they had slightly reduced or borderline low responses on scotopic ERG testing, indicating that *PRPF31* mutation carriers may exhibit reduced retinal function at subclinical level. ## RP caused by a *CRX* mutation previously associated with cone-rod dystrophy An 83-year old patient (#63). without visual limitations throughout his professional life reported impaired dark adaption, nyctalopia and glare as first symptoms in his early 8^th decade of life. Funduscopy showed typical RP fundus changes. He also noted progressive loss of visual acuity over the past 7 years, and visual acuity was now 20/200 in the right eye and hand movements on the left eye. The visual field was severely constricted, and ERG responses were not detectable. Although there was no known affected family member, exclusion of a dominant inheritance of this late onset RP was not possible because most of his relatives died at relatively young age. Genetic testing showed a heterozygous missense mutation (c.122G\>A, p.Arg41Gln) in exon 3 of the *CRX* gene. The same mutation had previously been described in families with autosomal dominant late-onset cone-rod dystrophy, indicating a novel genotype- phenotype correlation for this particular mutation. ## Non-syndromic RP caused by mutations in *CEP290* A female patient (#36) with reduced night vision and dark adaption since childhood had graduated from university without visual aids. Starting in her mid to late twenties, she had noticed progressive vision loss and was diagnosed with RP at the age of 27. Now 70 years-old, visual acuity was reduced to light perception in both eyes. Clinical examination showed widespread atrophy of the outer retina and bone-spicule pigmentations. Targeted NGS revealed heterozygosity for a nonsense mutation (c.982C\>T, p.Gln328\*) in exon 12 and a deep-intronic mutation (c.2991+1655A\>G) in intron 26 that creates a strong splice-donor site in *CEP290* and a premature stop codon, p.Cys998\*. Compound- heterozygosity was deduced because the patient’s son carried the nonsense, but not the splice site mutation. Biallelic mutations in *CEP290* cause several syndromic ciliopathies and non- syndromic LCA.\[, –\] Of note, our patient had no abnormalities in motor or cognitive development, renal cysts or other morphological abnormalities indicating a *CEP290*-associated syndrome. Although an association of *CEP290* mutations with relatively milder non-syndromic retinal phenotypes has been described in a few previous reports, vision loss in those patients either occurred in the first decade of life or the history of vision loss was not described in detail. Only one recent comparable study reported a patient with an apparently similar disease course. The latter and our case also show that retinal disease due to *CEP290* mutations is not necessarily associated with poor cone function, since both patients had normal visual acuity apart from reduced night vision for at least the first two decades of life. It is likely that the phenotypic continuum of *CEP290*-related non-syndromic retinal dystrophies can be attributed to modifying retina-relevant variants elsewhere in the genome. The identification of *CEP290*-related RP, a second non-syndromic phenotype associated with mutations in this gene, further supports the categorization of Joubert syndrome (JBTS) genes as strong candidates for isolated retinopathies. *OFD1* and, very recently, *AHI1*, have also been shown to cause both JBTS and isolated retinal degeneration. ## Non-syndromic RP caused by *RPGRIP1* mutations A 42-year-old male patient (#39) was diagnosed with sporadic RP at the age of 13 years. He reported nyctalopia, dark adaption problems and glare since childhood. At the age of 37 years, he started using visual aids. He underwent cataract surgery, and visual acuity now was 20/50 in the right and 20/63 in the left eye. Visual fields were severely constricted and full-field ERG showed no detectable responses. Funduscopy showed alterations characteristic for RP, with widespread retinal pigment epithelial and photoreceptor atrophy, and bone spicule pigmentations. Targeted NGS revealed a novel homozygous deletion (c.3100_3238del139, p.Gln1034Thrfs\*23) of exon 19 of *RPGRIP1*, resulting in a frameshift predicted to result in unstable mRNA or protein truncation. Mutations in *RPGRIP* are a known cause of LCA and juvenile RP.\[–\] Besides its uncommon benign disease course, the presented case is exceptional because visual acuity was good in early life and remained relatively stable until an age beyond 40 years. This contrasts with previous observations which had suggested severe cone functional loss despite relative sparing of the fovea anatomy on OCT images. ## Non-syndromic RP caused by mutations in *MFSD8* A 35-year old patient (#37) who initially noticed nyctalopia, dark adaption problems and glare at the age of 31 years showed fundus changes typical for RP. Best corrected visual acuity was 20/40 and 20/25, and there was severe concentric constriction of the visual fields. ERG showed no detectable responses. Genetic testing identified a homozygous variant (c.1445G\>C, p.Arg482Pro) in exon 13 in the *MFSD8* gene. *In silico* assessment with various programs strongly supports the categorization as disease-causing mutation, as did the results from segregation analysis: The index patient’s likewise affected brother and a second brother who reported difficulties seeing in dim light (not examined) carried the *MFSD8* missense mutation in homozygous state, whereas a sister with normal vision carried the wild-type allele. *MFSD8* mutations are known to cause variant late-infantile neuronal ceroid lipofuscinosis (vLINCL; CLN7), an early-onset severe lysosomal storage disorder with intralysosomal accumulation of autofluorescent lipopigments, presenting with seizures, mental regression and retinopathy, which was either not further specified or suggestive for RP. Recently, *MFSD8* mutations have been described in families with non-syndromic autosomal recessive macular dystrophy with central cone involvement, isolated maculopathy and generalized retinopathy. Similarly, the patient in our study showed no neurologic features typical for vLINCL, confirming *MFSD8* as a non-syndromic retinopathy gene. The *MFSD8* genotype-phenotype correlation proposed earlier, with vLINCL resulting if both mutations are severe (in particular truncating mutations), whereas milder mutations (such as hypomorphic missense mutations) on at least one gene copy result in non-syndromic retinal degeneration, likely also applies in the family reported herein. ## *RP2* and *RPGR* mutations in female patients with apparently sporadic or dominant RP Patient A (RP2): In this 16-years-old young woman (#80), unilateral reduced vision was first recognized at the age of four years, but no further examinations had been initiated at that time. When examined at the age of 12 years, she reported impaired central vision, but no nyctalopia. Visual acuity in the right and left eye was 20/63 and 20/20, respectively. The right eye was emmetropic and the left eye was myopic (spherical equivalent -3,5 dpt). The visual fields were severely constricted in the right eye, and there was a nasal superior visual field loss in the left eye. The ERG showed extinct rod responses in her right eye, while responses in her left eye were severely reduced. When she was examined at the age of 16, visual acuity had deteriorated only in the right eye (now 20/100). Fundus examination revealed narrowed vessels, outer retinal atrophy and bone spicule pigmentations, all much more pronounced in the right eye. In addition, the left eye showed a tapetal-like reflex. Fundus AF confirmed the asymmetry and revealed a pattern of radial lines extending into the fundus periphery in the left eye, which is a characteristic finding in carriers of X-linked RP. NGS analysis identified a one base-pair duplication (c.829dupG, p.Ala277Glyfs\*11) in exon 3 in the *RP2* gene. No retinal disease was known in other family members, assessment of the parental retinal phenotype was not possible, and samples for segregation analysis were not available. Patient B (*RPGR*): This 39-years-old myopic (spherical equivalent -5 dpt in the right eye and -6,75 dpt in the left eye) woman (#69) with nyctalopia since childhood reported decreasing visual acuity and visual fields since her twenties. Visual acuity was 20/400 in the right eye and 20/63 in the left eye. ERG examination was not tolerated by the patient. Funduscopy revealed changes characteristic for RP including bone spicule pigmentation and attenuated retinal vessels. Fundus AF showed areas of increased and decreased AF in the right eye and a fine pattern of radial lines radiating peripherally from the fovea in the left eye. On OCT imaging, there was widespread thinning of the photoreceptor layer in both eyes with foveal sparing in the left eye. Furthermore, OCT imaging revealed thickening of the inner retina mainly around the optic disc, which was more obvious in the right than in the left eye. We identified a heterozygous four-base-pair deletion (c.2442_2445del, p.Gly817Lysfs\*2) in *ORF15* of *RPGR*. None of the patient’s parents had a history suggestive for retinal disease, and examination of the mother including AF-recordings showed no characteristics for a carrier state of X-linked RP. Genetic analysis of the mother revealed wild- type *RPGR* alleles, indicating a *de novo RPGR* mutation in the index patient. Patient C (*RPGR*): This 52-years-old myopic (spherical equivalent approximately -4 dpt in both eyes) woman (#76) with difficulty in seeing in the dark since childhood reported a progressive reduction of visual acuity over the past 10–20 years. Visual acuity was above 20/50 in adolescence, 20/400-20/200 around the age of 40 years and now 20/800. ERG examination showed no detectable responses. Funduscopy revealed changes characteristic for RP in both eyes. There was symmetric and widespread thinning of the photoreceptor layer on OCT imaging. Although fundus AF imaging showed no sign for an X-linked carrier state, we identified a heterozygous two-base-pair deletion (c.2405_2406delAG, p.Glu802Glyfs\*32) in *ORF15* of *RPGR*. The patient’s maternal great-uncle was visually impaired, and her maternal great-grandfather was blind, compatible with autosomal dominant inheritance with reduced penetrance. The mother of the patient died at the age of 50 years and had no visual problems. Female carriers of X-linked RP consistently have peripheral retinal pigment epithelial atrophy. Most carriers may experience mild or moderate reduction of visual function, with a minority becoming legally blind. Although rare, severe RP may occur in female carriers of X-linked RP and simulate autosomal dominant inheritance\[, –\], as in Patient C. Comprehensive genetic testing has been shown to detect mutations in *RPGR* or *RP2* in cohorts assumed to have autosomal dominant RP, leading to a genetic re-classification of those families. However, to the best of our knowledge, a sporadic female RP patient diagnosed with X-linked RP has only been reported once. Of note, parental testing for the *RPGR* mutation of patient B indicated that it occurred *de novo*. Herein, severe manifestations of X-linked RP were found in two sporadic female carriers and in one patient with a family history suggesting autosomal dominant inheritance with variable expressivity. Two of these three patients revealed very subtle signs for X-linked RP which may easily be missed by standard clinical examinations. This includes asymmetry between eyes which has been reported as a typical sign in X-linked RP-carriers, although marked asymmetry was specifically mentioned in only one out of 61 carriers. Another peculiar finding on fundus AF imaging in female carriers of X-linked RP is a radial pattern extending peripherally from the fovea,\[–\] as observed in a more or less subtle form in the less affected eye of Patients A and B. Moreover, peripapillary thickening of inner retinal layers as observed in patient 2 has been reported in males with X-linked RP. Thus, a subtle X-linked RP carrier phenotype was present in the less severely affected eye in two patients, but not in the third patient who presented with a progressed bilateral disease stage. Unbiased molecular genetic testing eventually revealed the correct diagnosis in these female patients with sporadic RP or with a family history suggestive for autosomal dominant inheritance with variable expressivity. # Conclusion In summary, this study demonstrates the enormous genetic heterogeneity of RP and a high detection rate of disease-causing mutations in RP patients using targeted NGS. Given the continuous decline of NGS costs, initial Sanger sequencing of likely disease-causing genes does not appear necessary anymore. Novel genotype- phenotype correlations (*CRX*, *RPGRIP1*) were uncovered, and recently reported novel correlations were confirmed (*CEP290*, *MFSD8*). *PRPF31* and *RPGR* mutations revealed unexpected inheritance modes in a subset of patients. Massively parallel sequencing of all known retinal dystrophy genes is a valuable diagnostic approach in RP patients. # Supporting information [^1]: I have read the journal's policy and the authors of this manuscript have the following competing interests: Dres. Neuhaus, Lenzner, Zahnleiter, Betz, and Eisenberger are employees of Bioscientia, a publicly traded diagnostic company. Dr. Bolz has been employee of Bioscientia until 2016. There are no patents, products in development or marketed products to declare. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. The other authors have declared that no competing interests exist. [^2]: ‡ These authors are joint senior authors on this work.

# Introduction Fibromyalgia syndrome (FMS) is a chronic disorder of unclear origin. Growing evidence suggests a combination of interacting neurophysiological, genetic, and psychosocial mechanisms as the cause of FMS. This syndrome is characterized by widespread musculoskeletal pain in association with fatigue, poor sleep quality, cognitive dysfunction, mood disturbances, and many other variable somatic symptoms. Prevalence of FMS in the general population varies from 1.0 to 4.9% in women and from 0 to 2.9% in men as demonstrated by studies from Europe, USA and Canada. There is currently no cure for FMS nor is there a “gold standard” of treatment. Management of this disorder is therefore aimed at reducing symptoms and maintaining optimal functioning. Interventions such as medication alone or the use of a single non-pharmacological treatment produce, at best, modest effects on patients' condition. Results of a meta-analysis of 49 studies published 15 years ago suggest that non-pharmacological treatments are more effective than drug interventions. A recent meta-analysis of 23 studies assessing the efficacy of psychological interventions for fibromyalgia showed small to medium positive effects on short and long-term pain, quality of sleep, functional status, depression, and tendency to catastrophize in the face of pain. Other recent literature reviews on the use of patient education, exercise activities, cognitive behavioural therapy (CBT), and multidisciplinary treatment suggest that a multimodal approach which combines at least one educational/psychological intervention with at least one exercise treatment can be effective for improving FMS symptoms including pain, fatigue, mood and/or quality of life (QOL). However, many of the reviewed studies suffer from methodological deficiencies (e.g., small sample size, single site study, unstandardized outcomes, short follow-up, etc), and well-designed trials are still needed. Based on the Interactional School of Low Back Pain, Barcellos de Souza et al. developed in 2007 a multimodal group intervention—the Interactional School of Fibromyalgia (ISF)—which combines exercise therapy and educational/psychological tools for self-management of FMS. Patient empowerment is an integral component of the intervention as is active patient participation. The authors conducted a randomized controlled trial (RCT) to assess the efficacy of their intervention and found positive effects on pain intensity and perceived overall capacity to manage FMS symptoms. Although promising, these results remain preliminary and need to be replicated in a RCT involving more than one site, and using a comprehensive set of well-validated outcome measures such as those recommended by the IMMPACT (Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials) Group. Furthermore, adding a qualitative research component to the study would be an asset to further capture the patients’ experience during the intervention. Finally, some aspects of the ISF needed to be updated and somewhat reorganized. We therefore adapted the ISF into a more structured intervention program entitled PASSAGE whose French acronym is <u>*P*</u>*rogramme d’*<u>*A*</u>*pprentissage de* <u>*S*</u>*tratégie*<u>*S*</u> *d’*<u>*A*</u>*uto-*<u>*G*</u>*estion* <u>*E*</u>*fficaces* (Training Program of Efficient Self-Management Strategies). The aim of the present study was thus to evaluate, quantitatively and qualitatively, the efficacy of the PASSAGE Program—a multicomponent interdisciplinary group intervention for the self-management of FMS. It was expected that the Program will lead to improvements in the clinical condition of patients suffering from this disorder. # Methods The French version protocol for this trial (as well as the English translation of the Methods section) and supporting CONSORT checklist are available as supporting information; see, and Protocols. ## Ethics Statement The research protocol of the present study along with the patient informed consent form were reviewed and approved by the *Comité d’éthique de la recherche sur l’humain du Centre hospitalier de l’Université Sherbrooke*, Sherbrooke, Quebec, Canada (May 26^th 2009, \#09–034) and by the *Comité d’éthique de la recherche avec des êtres humains de l’Université du Québec en Abitibi- Témiscamingue (CÉR-UQAT)*, Rouyn-Noranda, Quebec, Canada (May 15^th, 2009). The study was registered at the International Standard Randomized Controlled Trial Number Register \#ISRCTN14526380 (<http://www.controlled- trials.com/ISRCTN14526380/>). ## Protocol and Adjustments Six adjustments were made to the protocol prior to enrolment. First, the upper age limit (65 years) was withdrawn. Second, patients suffering from chronic pain disorders other than FMS (e.g., painful diabetic neuropathy) were not excluded from the study as long as the pain associated with FMS was their predominant complaint. Third, potential participants had to accept to not introduce new pain medications or other new therapeutic modalities for pain management during the 11 weeks of the intervention because such a change in treatment could have biased our estimation of the intervention efficacy and made difficult to isolate its effects). Fourth, an additional training session for the health care professionals acting as facilitators was conducted to clarify some issues, answer questions, review the procedures, and insure uniformity between study sites. To minimize costs, this session was conducted via video conference due to the large distance between the two study sites. Fifth, research assistants were instructed to calculate the participants' scores on the Beck Depression Inventory (BDI) upon reception of their questionnaire. If a score \> 30 was found and/or that participant reported suicidal ideas (question 9 of the BDI), the research assistant was instructed to contact the patient by phone and encouraged him/her to make an appointment with his/her treating physician (or psychologist) or to go the hospital emergency department. Finally, focus groups were added to the research protocol in order to document and further capture the patients’ experience. ## Design and Settings A mixed-methods, multicenter, open label, randomized, wait-list controlled trial with both a quantitative and a qualitative component was carried out in two university-affiliated settings between September 2009 and March 2011: 1) Sherbrooke, a suburban city located in the south of the province of Quebec (Canada), and 2) Rouyn-Noranda, a small city in the north of the province of Quebec (Canada). Study sites were chosen because of the clinical expertise of local teams with the ISF. describes the flow of participants through the study at each assessment point. ## Eligibility, Recruitment, and Randomization Subjects were eligible for participation in the study if they: a) were aged 18 years or older, b) were able to read, understand, and complete questionnaires in French, c) had a medical diagnosis of FMS based on the American College of Rheumatology (ACR) classification criteria for at least 6 months, d) reported FMS pain of at least moderate intensity (≥ 4/10) in the seven days prior to enrolment, the FMS pain being the chief complaint if the patient suffered from another chronic pain syndrome, e) were motivated to attend all group sessions and to integrate the proposed self-management strategies, and f) agreed to not introduce new pain medications or other new pain treatment modalities during the 11 weeks of the intervention. Exclusion criteria were the following: a) pregnant or lactating women, b) presence of an active cancer, uncontrolled metabolic disease and other major physical or psychiatric disorder that could compromise patient participation in the study, and d) outstanding litigation regarding patient’s claim for disability payments. Recruitment was conducted through announcements in local newspapers in both study sites between September 2009 and October 2009. Interested subjects were invited to call the research coordinator who explained the study, reviewed some of the eligibility criteria, and fixed a first appointment with the potential participants one month prior to the beginning of the intervention. At the time of the first appointment, a pain physician established the FMS diagnosis using the ACR criteria, and a physical/psychological evaluation was carried out to ensure the subjects met all the eligibility criteria including proper motivation to partake in the intervention. Written informed consents were obtained from all participants who were then randomly assigned to the Intervention (INT) Group (PASSAGE Program) or the Waitlist (WL) Group. Randomization was stratified by study site and gender, and was done by an independent third party using the Random Allocation Software—Version 1.0.0 (Isfahan, Iran). ## Description of the Group Conditions ### Intervention (INT) group As mentioned earlier, the PASSAGE Program is a structured multicomponent interdisciplinary group intervention aimed at reducing FMS symptoms and maintaining optimal function through the use of self-management strategies and patient education. The intervention consists of 9 group sessions with 8 participants lasting 2.5 hours each. As shown in, each session involved 3 major components—1) psycho-educational tools, 2) CBT-related techniques, and 3) patient-tailored exercise activities. Self-management of the main symptoms of FMS including pain, fatigue, poor sleep quality, and mood fluctuations were targeted during the course of the sessions as well as issues relating to stress management. An additional session was devoted to the pharmacological and non- pharmacological treatments of FMS. The first 8 sessions were held over a period of 11 weeks while the 9^th final session was carried out 6 months later to review progress and gain maintenance. The first two sessions were partly devoted to the establishment of a contract with the patient where she/he: 1) fixed three personal outcome goals to be met by the end of the intervention program, 2) determined the minimally acceptable changes to be expected, and 3) agreed to participate in all group sessions and to devote time during the week to the tasks prescribed at the end of each session—i.e., about 45 minutes/day, 6 times/week. Patients were informed that they will be excluded from the program if they missed 2 sessions. The sessions were always conducted in a well-equipped exercise room with mattresses, pillows, exercise balls, mirrors, sound system and computer equipment for Power Point presentations. These sessions were interactive and led by two health care professionals who both acted as facilitators, one being mainly responsible for the psychological aspect of the intervention and the other for its physical aspect. Patients were viewed as the “experts” of their condition, and were given a role of active partner in the management of their FMS. Except for Session 1, the others always started with customized exercise routines (15 min), including correction of posture and movements when needed. Participants were then asked to discuss their experiences with the prescribed tasks of the preceding week (including the practice of new self-management strategies) (15 min). Then, the two facilitators started the education part of the session during which various topics related to FMS symptoms and their management were covered. Participants were strongly encouraged to share their own experience and the tools/strategies they used to manage their condition which fostered patients' empowerment as well as their active participation as experts of their condition. This portion of the session, which lasted about 60 min, was followed by a 15-min break during which both participants and facilitators had the opportunity to socialize. In the second portion of the sessions, the facilitators proposed new self-management strategies, specific exercises, and respiration techniques. A strong emphasis was placed on the rationale behind the proposed strategies/techniques. Participants were invited to practice them during a 30-min period. Starting on Week 4, the exercise program ended with a relaxation session during which different techniques were taught and practiced (15 min). Finally, participants were prescribed tasks to be done during the following week(s). At any time during the sessions, participants were allowed to move, lie down, or use pillows to alleviate pain, if needed. In order to ensure uniformity and standardization of the intervention program in both study sites, all health care professionals acting as facilitators attended a one-day structured training session which was held in Rouyn-Noranda (Québec, Canada). The first part of the training was devoted to theory (e.g., rationale behind the intervention, key concepts, procedures to follow during each session, the “*to-do*” and “*not-to-do*”, etc), and the second part took the form of a practicum with presentations of scenarios, role playing, and video demonstrations. The comprehensive course manual used during the training and given to each facilitator also contained a detailed description of the content of the sessions along with an annotated paper copy of the slides to be used in each session. A second training session was held via videoconference to clarify some issues, answer questions, and review the procedures. To ensure facilitators’ adherence to the intervention protocol, all sessions were recorded and monitored (using back-surface mirror) by a health care professional with experience with the ISF and who participated in the development of PASSAGE Program. Conference calls involving the researchers and study coordinators from each study site were also made on a weekly basis to review the procedure, discuss issues raised during the sessions, and ensure uniformity. ### Waitlist (WL) group Participants randomized to the WL Group were instructed to continue their treatment(s) as usual until they could take part in the PASSAGE Program—i.e., 3 months after the INT Group had completed the program. Changes in pharmacological or non-pharmacological treatments were allowed during this period in the WL Group (usual care). ## Procedure ### Quantitative study Data were collected in both study groups at baseline (T₀), after the INT Group completed the 8 sessions of the PASSAGE Program (T₁), and 3 months later (T₂). Patients of the WL Group were then offered the Program and completed follow-up measures at the end of the intervention (T₁), and 3 months later (T₂), thereby providing efficacy data from another cohort of patients. Additional follow-up measures were also collected in the INT Group at 6 (T₃) and 12 (T₄) months after the completion of the PASSAGE Program so longitudinal data (T₀ to T₄) were available to assess the long-term benefits of the intervention in this group. Data were collected at each time point with a self-administered questionnaire which was mailed to the patients along with a stamped return envelope to be mailed back to the research team within the next 7 days. Reminder phone calls were made if the questionnaires were not returned on time. Upon reception, questionnaires were carefully verified, and a research assistant contacted the patients if any information was missing or if their depression scores on the BDI was \> 30 and/or they reported suicidal ideas (question 9 of the Beck Depression Inventory) (see Section Protocol and Adjustments). ### Qualitative study In order to document and further capture the patients’ experiences, face-to-face open-ended narrative qualitative group interviews were conducted in each study site. Interviews took place 6 to 9 months after completion of the PASSAGE Program, and were conducted by the same interviewer in both sites. The interviewer had an extensive experience in qualitative research interviews and was, until then, unknown to the study participants. Nine patients from the Sherbrooke site (Québec, Canada) and 7 from the Rouyn-Noranda site (Québec, Canada) volunteered to participate in the group interviews. The same interview guide was used in both study sites and it included open-ended questions aimed at covering three main topics related to the research objectives. Participants were asked to talk about 1) their experiences during the intervention, 2) its impact on their daily life, and 3) their general appraisal of the intervention. The group interviews lasted between 60 and 90 minutes, and were audio-taped, entirely typed-written (*verbatim)*, and annotated with the interviewer’s field notes. ## Outcomes ### Primary outcome Pain intensity was the primary outcome and was measured with a standardized numerical rating scale (NRS) where 0 indicated “no pain” and 10 “worst possible pain”. At each time point of the study, patients of both groups were asked to rate the average intensity of their pain as experienced in the past seven days. ### Secondary outcomes The choice of the secondary outcomes was based on the characteristics of the FMS symptomatology, the rational/objectives of the proposed intervention, and the IMMPACT Group recommendations as well as the 2012 Canadian Guidelines for the Diagnosis and Management of FMS. Two major sets of secondary outcomes, specific and global, were used to assess the effectiveness of the intervention. The selected measurement instruments are well-validated and widely used tools with documented psychometric qualities. <u>The first set of secondary outcomes</u> measured <u>specific</u> symptoms or dimensions of the patients’ condition prior to the beginning of the intervention (T₀) and at follow-up times—i.e., T₁ and T₂ in both groups, and T₃ and T₄ in the INT Group only. Severity of FMS was measured with one of the most widely used tool in this research field, the Fibromyalgia Impact Questionnaire (FIQ) which is a disease- specific instrument designed to evaluate the impact of FMS by providing a multidimensional assessment of the overall severity of FMS. The first 11 FIQ items ask about actual capacities regarding domestic activities and are answered to on a 4-point Likert scale ranging from 0 (always) to 3 (never). The last 9 FIQ items assess the presence and severity of various symptoms in the past seven days (pain, physical functioning, fatigue, morning tiredness, stiffness, depression, anxiety, job difficulty and overall well-being) using a numerical scale ranging from 0 (no symptoms) to 10 (major symptoms). The total FIQ score was calculated with a pre-determined algorithm ([*www*.*myalgia*.*com/FIQ*](http://www.myalgia.com/FIQ)) and ranges from 0 to 100, where a higher score indicates a greater impact of FMS. The extent to which patients’ pain interfered with various aspects of their daily living was assessed with the 10 interference items of the Modified Brief Pain Inventory (BPI). These items include general activity, mood, walking ability, normal work, relations with others, sleep, enjoyment of life, personal care, recreational activities, and social activities in the past seven days. Items are rated on a 0 (does not interfere) to 10 (completely interferes) scale. The global BPI interference is derived by averaging the 10 items. Considering the high frequency of sleep problems in FMS patients and the potential interrelations with pain, the Chronic Pain Sleep Inventory (CPSI) was also administered to all participants to assess the impact of pain on sleep quality during the past 4 weeks. The CPSI is composed of 4 items answered to on a scale ranging from 0 (never) to 10 (always). Items are: 1) trouble falling asleep, 2) needing sleep medication, 3) awakening due to pain in the night, and 4) awakening due to pain in the morning. The fifth item of the CPSI assesses overall quality of sleep using a 0 (very poor) to 10 (excellent) scale. A total Sleep Problem Index Score (SPIS) is calculated by taking the sum of the scores on items 1, 3 and 4. The SPIS can range from 0 to 30 and higher scores indicate greater sleep problems. The Coping Strategy Questionnaire (CSQ) was used to assess the type of coping strategies participants employed day-to-day to cope with their pain. The CSQ includes 21 items answered to on a 4-point Likert scale ranging from 1 (never) to 4 (always). Items assess 5 coping strategies: 1) Ignoring pain sensations, 2) Diverting attention, 3) Catastrophizing, 4) Reinterpreting pain sensations, and 5) Praying. A score is obtained for each subscale by summing the scores on each of its items. In addition, patients’ tendency to catastrophize while they are in pain, which is known to have a profound impact on the experience of pain (see critical review), was further investigated in the present study by administering the Pain Catastrophizing Scale (PCS). The PCS contains 13 items rated on a 5-point Likert scale ranging from 0 (not at all) to 4 (always). A total score is calculated by summing the score on each item. The Beck Depression Inventory (BDI) Version 1 was used to assess severity of depressive symptoms in the past seven days. This scale includes 21 items rated on a 4-point ordinal scale and a total score of the BDI (ranging from 0 to 63) can be obtained from the sum of all individual items. Higher scores indicate more severe depressive symptoms. Health-related QOL was assessed with a generic instrument—the Standard SF-12v2 (4-week recall). This questionnaire covers 8 domains (i.e., Physical Functioning, Role-Physical, Bodily Pain, General Health, Vitality, Social Functioning, Role-Emotional, Mental Health) and the scores on each of these domains are summarized into 2 scales, the Physical Summary Scale and the Mental Summary scale. Scores on each summary scale were calculated with standard scoring algorithms and normalized using the US general population values (mean = 50; SD = 10). <u>Our second set of secondary outcomes</u> was oriented towards patients’ <u>global</u> impression regarding changes in their condition and overall perception regarding their treatment responses in terms of pain relief. These measures were collected in both groups at T₁ and T₂, and at T₃ and T₄ in the INT Group only. Participants were asked about their global impression of change in the past 3 months regarding their 1) pain, 2) level of functioning, and 3) QOL, using a modified version of the Patient Global Impression of Change (PGIC) Scale. The scale ranged from 1 to 7 with “remained unchanged” as the mid-point, and “considerably deteriorated” and “considerably improved” as anchors. PGIC scores in each area were recoded into three categories: 1) Improved (slightly/ greatly/ considerably improved), 2) Stable (remained unchanged), and 3) Deteriorated (considerably/ greatly/ slightly/ deteriorated). Patients’ overall perceptions of their treatment responses in terms of pain relief was assessed on a 0 to 100% Pain Relief Scale where 0% represents no pain relief and 100% represents complete pain relief. Patients were asked to provide their ratings based on the preceding 3 months. Substantial improvement was defined with a cut-off point of pain relief ≥ 50%. ## Sample Size The sample size was calculated for the primary outcome (NRS = average pain intensity over the past 7 days; continuous scale ranging from 0–10) based on testing the inequality of two means in a repeated measures design. Previous reviews established that a 2-point reduction on the 0–10 NRS scale constitutes a clinically meaningful difference in pain intensity. Assuming the standard deviation of the NRS of 2.0 units and a study design with 3 repeated measurements having a compound symmetry covariance matrix, the sample size was determined based on the ability to detect, with a power of 80%, a change of 2 units or more in the average pain intensity score between the INT and WL groups at an two-sided alpha level of 0.05. Under these assumptions and conservatively assuming an autocorrelation coefficient (rho) of one, a group sample size of 16 patients, representing a total sample size of 32 patients was required. Given that the study was carried out in two sites and that each site was expected to have two groups, the sample size was doubled, and 64 patients in total were targeted (32 per study site). This strategy did not only increase our statistical power but also prevented reduced power because of patient loss to follow-up. Sample size estimation was performed using PASS 2008 and all statistical analyses were performed using SAS Version 9.2 (SAS Institute, NC, USA). The investigator in charge of the statistical analyses (A.L.) was blinded to group assignment. ## Quantitative Data Analysis Comparisons of the baseline (T₀) sociodemographic and pain characteristics between the INT and WL Groups were carried out with parametric and non-parametric tests depending on the type of variables and their distributions (i.e., t-test, Wilcoxon rank-sum test, Chi-square test, Fisher’s exact test). To detect if significant differences in continuous outcomes over time (T₀, T₁, and T₂) were due to the intervention (*Group* x *Time* interaction effect), linear mixed models for repeated measures were used: 1) unadjusted model, 2) gender and study site adjusted model, and 3) fully adjusted model (study site, gender, living arrangements, work status, pain duration and use of pain medication). When interaction effects were detected, post hoc pairwise comparisons using t-tests or Wilcoxon rank-sum tests were carried out with a Bonferroni correction. Effect sizes are presented as raw group differences (mean differences) and their 95% confidence intervals (95% CI). A negative effect size value indicates that the intervention was superior to the control group on negatively oriented outcome measures (i.e. pain intensity NRS, FIQ score, BPI interference score, CPSI sleep problem index score, catastrophizing CSQ score, PCS score, BDI score). A positive value indicates the intervention was superior to the control group on positively oriented outcome measures (i.e. CPSI overall sleep quality score, other CSQ subscales scores, SF-12v2 health-related QOL scores). Categorical outcomes such as the PGIC (proportion of patients reporting improvement) and pain relief (proportion of patients reporting ≥ 50% of pain relief) in the past 3 months were compared between the two study groups at the end of the intervention (T₁) and 3 months post-intervention (T₂) using Chi-square tests and Fisher exact tests where appropriate. Effect sizes were computed as odds ratios (OR) and their 95% CI. As previously mentioned, additional data were collected from the WL Group at the end of the trial once they had the opportunity to participate in the PASSAGE Program. These data were used to conduct sensitivity analyses to see if the pattern of results observed on the global outcome measures at T₁ and T₂ were similar to the one observed in the INT Group. With the 12 months follow-up data from the INT Group, additional analyses were carried out to assess the effect of the treatment over a longer period of time. One-way ANOVAs with repeated measures on one factor (within-subjects time effect between T₀, T₁, T₂, T₃ and T₄) were conducted. When significant differences were detected, post hoc comparisons were carried out using t-tests or Wilcoxon signed rank sum test. Raw effect sizes between T₀ and T₄ measures are presented as mean differences and their respective 95% CI. ## Qualitative Data Analysis A thematic analysis of the qualitative data was conducted using the methodology proposed by Mucchielli. All *verbatim* were reviewed line by line, summarized in words, and transformed into codes. MS Excel software was used to create a coding tree, and codes were combined to identify emerging themes which were then classified into main themes and associated themes. The analysis was done independently by two investigators (P.B., R.C-H) who then compared and reviewed their results until a consensus was reached. # Results ## Participants’ Recruitment As shown in, 24 subjects were excluded throughout the study selection process, leaving a total of 58 eligible patients who were randomly assigned to the INT Group (n = 29) and the WL Group (n = 29). Fifteen patients (31.0% in the INT Group (9/29) vs. 20.7% (6/29) in the WL Group) did not complete the 3-month trial: two out of fifty-eight were excluded from the program because of non compliance and the others withdrew either because they: 1) were no more able to attend the sessions due to a scheduling conflict (n = 3/58)), 2) developed a medical disorder unrelated to FMS (n = 3/58), 3) went through an episode of psychological instability (n = 2/58), or 4) for personal reasons (n = 1/58). Four out of fifty-eight participants failed to return their study questionnaires by mail at one time or another, and did not provided complete longitudinal data. Consequently, a total of 43 patients completed the T₂ measures: 20/43 received the intervention and 23/43 were on the waitlist. As mentioned earlier, this last group received the intervention at the end of the trial and were assessed up to 3 months post-intervention (17/23 patients completed follow-up). ## Participants’ Characteristics Socio-demographic and pain characteristics of the randomly assigned participants are presented in. Their mean age was 49.98 ± 9.23 years and 46.74 ± 11.42 years in the INT and WL Groups, respectively. As expected, there was a greater proportion of women in both study groups (\> 92%, 26/28 in the INT Group and 26/28 in WL Group). More than half of the subjects (18/28 in the INT Group and 15/28 in the WL Group) had completed a university education level. The mean duration of pain was \> 10 years in both groups (INT Group: 15.66 ± 11.12 years, WL Group: 11.94 ± 8.23 years) and the pain intensity levels on the NRS (average pain in the past 7 days) were comparable (INT Group: 6.57 ± 2.03, WL Group: 6.39 ± 1.83). The only statistically significant difference between the groups was the proportion of patients who were using prescribed pain medication; this proportion was lower in the INT Group (78.57%, 22/28) than it was in the WL Group (100%, 28/28). ## Efficacy of the PASSAGE Program up to 3 Months Post-Intervention ### Primary outcome As shown in, average pain intensity scores were comparable between the two study groups from baseline up to 3 months post-intervention. No significant *Group* x *Time* interaction was found (*P* \>.05). Effect sizes of -0.13 (95% CI: -1.37 to 1.11) and -0.55 (95% CI: -1.82 to 0.72) were found between study groups at T1 and T2 respectively. ### Specific secondary outcomes shows the variations in the mean scores (± SD), effect sizes and their 95% CI on the continuous secondary outcome (specific outcome measures) obtained in the INT and WL Groups at T₀, T₁, and T₂. The mixed models results revealed no significant interaction between time and group assignment on the majority of these outcome measures. In other words, when compared with the WL Group, patients in the INT Group did not show more or less improvements over time with regard to the severity of their FMS condition (Fibromyalgia Impact Questionnaire), the extent to which their pain interfered with their daily living, the quality of their sleep, the type of strategies they used to cope with their pain, their tendency to catastrophize in the face of pain, the physical component of their health-related QOL (Physical Summary Scale of the SF-12v2), and their psychological well-being (Depression BDI Scores, Mental Summary Scale of the SF-12v2). The only *Time* x *Group* interaction effect that reached statistical significance was on the Ignoring Pain Sensations Subscale (*P* = 0.010) of the Coping Strategy Questionnaire. Although these results were suggestive of positive impact of the intervention on this measure, the results of the post-hoc analyses were not statistically significant after application of the Bonferronni correction. ### Global secondary outcomes A different pattern of results emerged in the second set of secondary outcomes—i.e., measures of patients’ global impression of change (PGIC) regarding changes in their condition and overall perceived pain relief. Comparisons of the PGIC ratings revealed statistically significant differences between the two study groups at T₁ (end of intervention). Specifically, the likelihoods of reporting overall improvement in pain (OR: 30.67; 95% CI: 5.48 to 171.73; *P* \<.0001), level of functioning (OR: 8.80; 95% CI: 2.02 to 38.25; *P* =.002), and QOL (OR: 13.60; 95% CI: 3.36 to 55.04; *P* \<.0001) were higher in the INT Group than in the WL one. As shown in, 72.7% (16/22), 54.6% (12/22), and 77.3% (17/22) of the patients of the INT Group reported improvements on these indicators between T₀ and T₁ compared with 8.0% (2/25), 12.0% (3/25) and 20.0% (5/25) in the WL Group. Furthermore, patients in the INT Group continued to perceive significant improvements in the 3 months following the intervention (i.e., between T₁ and T₂). When compared to the patients of the WL group, those of the INT Group were significantly more likely to report improvement in pain (OR: 10.83; 95% CI: 2.42 to 48.52; *P* =.0008), level of functioning (OR: 5.00; 95% CI: 1.13 to 22.18; *P* =.0266), and QOL (OR: 6.46; 95% CI: 1.184 to 35.26; *P* =.0201). As for these three outcomes respectively, 61.9% (13/21), 42.9% (9/21), and 38.1% (8/21) of the patients in the INT Group continued to perceive improvements compared with 13.0% (3/23), 13.0% (3/23) and 8.7% (2/23) in the WL Group (one subject in the INT Group returned the questionnaire at T₂ but not at T₁). As previously mentioned, additional PGIC ratings were collected in the WL Group (17/23) at the end of the trial once they also completed the PASSAGE Program. Close to 60% of them reported improvements in their pain (58.8%; 10/17), functioning (58.8%; 10/17), and QOL (58.8%; 10/17) between T₀ and T₁, while the percentage of those who continued to improve on these indicators 3 months following the intervention was 35.29% (6/17), 23.53% (4/17), and 29.41% (5/17) respectively. Significant group differences were also found in the measure of patients’ overall perceptions of pain relief. The proportion of patients reporting ≥ 50% pain relief between T₀ and T₁ was significantly higher in the INT Group (36.4%; 8/22) than in the WL Group (12.0%; 3/25) (OR: 4.19; 95% CI: 0.95 to 18.53; *P* =.049). Three months later (T₂), one third (33.3%; 7/21) of the patients in the INT Group reported ≥ 50% pain relief compared to only 4.3% (1/23) in the WL Group (OR: 11.00; 95% CI: 1.22 to 99.25; *P* =.013). Once the WL Group completed the PASSAGE Program, 23.5% (4/17) of them reported ≥ 50% pain relief, and this percentage remained the same 3 months later. ## Efficacy of the PASSAGE Program up to 12 Months Post-Intervention A total of 18 patients (64.5%) assigned to the INT Group completed additional follow- up measures 6 (T₃) and 12 months (T₄) after the completion of the PASSAGE Program. One-way ANOVAs with repeated measures on one factor (time) revealed statistically significant mean differences across the different follow-up times (T₀ to T₄) regarding the NRS measure of average pain intensity in the past 7 days (*P* =.0263), the Fibromyalgia Impact Questionnaire (*P* =.0041), the Reinterpreting Pain Sensations subscale of the Coping Strategy Questionnaire (*P* =.0071), and the Pain Catastrophizing Scale (*P* =.0007). Although these results suggest that some clinical benefits of the intervention were maintained on the long-term on certain specific outcome measures, the results of the post-hoc analyses revealed statistically significant differences between the baseline scores (T₀) and those at 12 months post-intervention (T₄) only for the Reinterpreting Pain Sensations subscale of the Coping Strategy Questionnaire (Effect size: 1.83; CI:0.79 to 2.87) and the Pain Catastrophizing Scale (Effect size: -6.89; CI: -10.38 to -3.39). On the global outcome measures, a good proportion of patients continued to report improvements or remained stable as revealed by their scores on the PGIC scales (pain, functioning, and QOL) at 6 (T₃) and 12 months post- intervention (T₄). In fact, at T₃, 33.3 to 47.6% of the patients still reported improvements as experienced in the preceding 3 months while 23.8 to 38.1% reported being stable. Twelve months post-intervention (T₄), 11.1 to 16.7% of patients reported improvements as experienced in the preceding 3 months while 16.7 to 38.0% reported stable levels. When asked about the percentage of pain relief experienced in the past 3 months, more than one quarter (28.6%) of the participants reported ≥ 50% pain relief at follow-up 6 months (T₃) while 5.6% did so at follow-up 12 months (T₄). ## Qualitative Results Results of the thematic analysis of *verbatim* highlighted three major themes. The first one was UNCONDITIONAL ACCEPTANCE. All participants reported having greatly appreciated the intervention. Many of their comments were directed towards the facilitators—i.e., their unconditional acceptance, understanding, open-mindedness, kindness, availability and warm approach. For instance, some participants stated “*We felt understood; they know the disease*”, “*Just a look*, *and they knew what was wrong*” and “*I felt I was really important during the meetings*”. The majority of the participants stressed the difference between the PASSAGE Program and the care they usually receive where they do not feel supported by their healthcare providers: “*Our doctors don’t believe in our pain*, *we don’t know where to go and to whom to talk about FMS*”. All interviewed participants said they would partake in the intervention again and would recommend it to other FMS patients. The second emerging theme was GROUP cohesion. It appeared that being part of a group was more important than the participants expected it would be specifically in terms of the opportunity for: 1) sharing “*She may have a trick that you didn’t think of*. *It helps to find new strategies*”; 2) support “*I didn’t know that so many people suffered from FMS*, *now I feel less lonely*”; and 3) motivation “*Motivation of others helped maintain my own motivation*.”; “*The effect of the group stayed even when I was home*. *That is why I exercised even when I wasn’t in a mood to do so*.” and “*Even if I didn’t feel like going to the meeting*, *I went because I knew I would feel better*.” The last emerging theme was INCREASED EMPOWERMENT. Participants described that the impact of the intervention went beyond the symptoms themselves in that they acquired new knowledge and learned how to self-manage their condition. “*They gave me tools to gain control over my symptoms and to understand how to do it*.”; “*They helped me realize that I did too much exercising*. *I learned to manage my energy*.“; and “*I learned how to say no and to accept my limits*.”. The intervention also brought some behavioural changes among the participants: “*Now when I talk to my friends*, *I am not talking only about my disease*. *I ask them how they are feeling*.” and “*I now have leisure activities*, *I go out with friends*.”. Finally, the following verbatim provides a meaningful and insightful illustration of the global impact of the intervention: “*At the beginning of the intervention*, *I realized that my pain was like a budget*. *I understood that I will always have the same amount of money but I will now manage it differently*. *This is really different*. *When you manage your pain*, *it is less present*, *less intense*”. # Discussion The results of the present study suggest that the PASSAGE intervention had a positive short-term impact on patients’ overall perceptions of their condition as revealed by their scores on the PGIC (pain, functioning, QOL) and pain relief scales compared to the patients who were assigned to the waitlist during this same period. Additional follow-up measures in the INT Group also showed that patients continued to improve or remained stable at 6- and 12-months post- intervention. For instance, at 6 months post intervention, between 57% and 85% of the INT participants were either stable or still noticing improvements on the PGIC outcomes and the pain relief measure. Results of the qualitative component of the study further supported the quantitative findings by showing that the intervention was effective in helping FMS patients gain an impression of control over their symptoms. However, no improvements were found on the primary outcome (NRS pain intensity) and the secondary specific outcome measures. The present study has a number of implications which are discussed below. ## Short-Term Impact of the Intervention The hypothesis of the present study was that a multicomponent interdisciplinary self-management intervention would lead to a reduction in pain intensity and a lower impact of pain and FMS symptoms on various aspects of daily living including sleep quality and emotional well-being (depression), improved health- related QOL, less tendency to catastrophize in face of pain, and better pain coping strategies. However, no significant differences on these measures were found between the INT and the WL Groups. Nevertheless, significant improvements were observed on patients' global impression of change in pain symptoms, functioning, and QOL, as well as improvements in their perceived pain relief. It is thus possible that a measure assessing patients' global impression of change may be more reflective of the impact of an intervention like the PASSAGE Program than the primary outcomes generally used. Numeric scales, like the 0–10 pain intensity scale, are often considered as the primary outcome in pain treatments clinical trials, although these measures are more and more criticized. In the present study, no significant changes were found on the 0–10 pain intensity numeric scale which further suggests that this measure may not be the ideal primary outcome when it comes to the evaluation of an intervention designed for the FMS population. Similar conclusions have being reached by other researchers as PGICs are increasingly being used as a gold standard in chronic pain treatments clinical trials. For instance, improvements as measured by PGIC alone can be recognized as a response to a pharmaceutical treatment. In pharmacological trials among FMS patients, correlations have been reported between PGIC and clinical pain, physical functioning, fatigue and impact on daily living. Furthermore, because of the multidimensional nature of pain, QOL, improvements in functioning are increasingly used as markers of clinical significance when evaluating the efficacy of treatments. A meta-analysis of the efficacy of pain medications such as gabapentin and pregabalin concluded that, even with a moderate pain reduction, QOL could be very much improved in chronic pain patients and FMS patients. Thus, greater attention should be given to the selection of primary outcomes when studying the FMS population with a specific concern for assessing patient global impression of change and overall perceived pain relief. ## Teaching Self-Management of Symptoms The results of the qualitative and quantitative components of this study further suggest that the PASSAGE Program was highly successful in teaching patients to self-manage their illness and to take control over their pain management. One fascinating analogy emerged from the qualitative interviews where one participant described the management of his pain as one would describe the management of a budget. This participant mentioned that even though she will always have the same amount of money (—i.e., pain) what really matters is how she manages it. The better you manage your money (—i.e., pain), the less poor you feel (—i.e, the less intense and present the pain feels). The empowerment described by this patient and many others shows that the PASSAGE Program was successful in helping patients improve the self-management of their pain through the learning and implementation of new strategies. Furthermore, the quantitative results regarding the improvements of coping strategies and the increased ability to ignore pain sensations further demonstrate that the program was successful in this regard. Elements of the PASSAGE Program such as patient education and aerobic exercise could have had a positive influence as they are known to increase global well- being and physical functioning as well as to decrease pain (e.g.. For instance, Hävermark and Langius-Eklöf evaluated the impact of a physical therapy-based educational program (including information, exercise and relaxation) with groups of 15 FMS patients seen twice weekly for 2 hours over a period of 10 weeks, and found a positive short-term effect on symptoms as well as a long-term effect on well-being. However, it is important to note that the PASSAGE Program goes further than simply education and exercise. Above all, the principles of CBT where the role of thoughts, beliefs, and expectations are believed to have a major impact on symptoms were followed. The present results support the effectiveness of an intervention based on CBT principles. For instance, in the qualitative interviews, the unconditional acceptance of group facilitators, the group cohesion and the empowerment over the illness had a significant impact on patients. Other authors have also reported that facilitators, group membership and the sharing of strategies and life changes such as attitudes and healthy behaviours can significantly influence patients' pain recognition and their sense of control over their disease. Also, being part of an educational program appears to be helpful in reducing FMS patients' anxiety levels. The present results are further in line with the 2012 Canadian Guidelines for the Diagnosis and Management of FMS which specifically mention that multimodal management strategies must be used when dealing with FMS and that patients should play an active role in their care. ## Long-Term Impact of the Intervention Although the absence of a WL group at the T₃ and T₄ follow-up times limits our ability to draw conclusions regarding the long-term impacts of the program, some interesting results deserve to be mentioned. First, long-term improvements (up to 12 months post-intervention) on the PGIC scales and perceived pain relief were found among the patients of the INT Group. In addition, the average pain intensity in the last week, impact of FMS on daily life, pain catastrophizing and some coping strategies also appeared to improve over time. As in most quasi-experimental studies, we cannot determine if these changes were due to the intervention, to the effect of time or simply to the mere participation in a study. Time effects (independent of the intervention) were found on different follow-up outcome measures which suggests that the participation in the study, taken alone, can bring improvement in the medical condition. These preliminary results support the long-term effect of the PASSAGE intervention, but future randomized controlled studies are nonetheless needed. ## Strengths and Limitations To our knowledge, this study is the first which evaluates the clinical impact of a multicomponent interdisciplinary self-management intervention for FMS patients using a two-site mixed-methods design. Such a design (quantitative and qualitative) insured appropriate triangulation of the data, and allowed for a richer exploration of participants' experiences during the intervention. Other strengths to the present study which insure its internal validity deserve to be highlighted. A randomized controlled design was used, a rigorous training of intervention facilitators was carried out, standardized recruitment and data collection methods across study sites were used, validated and recommended measurement scales were utilized, and mixed models allowing the adjustment for confounders and the handling of missing data were used. Furthermore, this study was conducted on two different sites located in two different regions of the province of Québec (Canada) which ensures a better representativeness of FMS patients. Finally, the retention rate was good and in line with other intervention studies with this population. In the INT Group, 78.6% of the patients completed the intervention (18.8% of those in the WL Group dropped out during this period), 71.4% completed the measures at 3 months post-intervention, 67.9% did so 6 months post-intervention and finally 64.3% completed the 12-month follow-up questionnaire. Some limitations to the present study should nonetheless be addressed. The sample size remained relatively small and might have limited the detection of some group differences (Type II error). Although a Bonferroni correction was applied in some analyses, the possibility of type I error resulting from the high number of statistical tests that were conducted cannot be excluded. Also, the actual usage of the strategies taught to the patients during the intervention was not assessed which limits our ability to determine in which way the strategies were helpful or not. Another limitation is the self-report nature of all the measures used in the present study. Future research should include methods such as observation-based assessments of patients' functioning. Finally, although the baseline characteristics of the patients were similar to what is found in the general population of FMS patients, it should be remembered that the present RCT had strict selection criteria which may lead to the selection of patients that are not representative of the general population. For example, participants in the present study had to report moderate pain intensity (≥ 4/10) in the seven days prior to enrolment leading to the exclusion of a number of mild FMS patients. In fact, the recruitment of participants with pain ≥ 4/10 may explain the limited impact of the intervention on the 0–10 intensity of pain numeric scale considering the difficulty to improve outcomes in more severe FMS patients. ## Future Research and Clinical Implications Our next step will be to evaluate the clinical impact of our interdisciplinary self-management intervention in a pragmatic trial conducted in many clinical settings of the province of Quebec, thus providing further evidence for the effectiveness of the intervention in “real-life” settings. The PASSAGE Program could also be improved by involving family physicians in the intervention in order to provide the patients with optimal medication for their symptoms. It has been suggested that FMS patients should be treated rapidly in the primary sector of care before symptoms get worse and that a shared-decision making process between the patient and the physician can help in improving the quality of the interactions. Furthermore, in a recent report, Oldfield and colleagues discussed how primary care physicians' moral judgements resulting from their skepticism regarding the legitimacy of FMS greatly impair the patient-doctor relationship for women with FMS. It is thus essential to better educate family physicians and insure that the patients get the treatment they need. In this regard, Fitzcharles and colleagues published a set of recommendations in order to discuss the legitimacy of FMS and to guide its diagnosis and management. Thus, involving primary care physicians in the PASSAGE Program could be very beneficial for FMS patients in order to provide this often neglected clientele of patients with effective self-management of their symptoms. In conclusion, our new multicomponent interdisciplinary self-management intervention for FMS was found to be effective in improving the patients’ global impression of change in terms of pain, functioning and QOL as well as in increasing their perceived pain relief. We suggest including this type of measures in future clinical trials on FMS as they appear to capture an important aspect of the patient experience. Further research on the long-term efficacy of the PASSAGE Program is nonetheless needed as well as studies involving primary care physicians in the intervention. # Supporting Information We are grateful to all the patients who took part in the study, and to the members of the study team, Edith Bérubé-Quesnel, Josée Boucher, Éric Chaize, Christian Cloutier, Roxanne Daoust, Sophie Duhaime, Annie Lachapelle, Émilie Lagueux, Serge Lalonde, Sylvie Lamoureux, Marjolaine Landry, Niet Nguyen, Kathy Perrier, Vanessa Roy, Mario Siane from the Université de Sherbrooke and the Université du Québec en Abitibi-Témiscamingue. We also wish to thank Mr Louis Coupal (Impacts Inc.) for his helpful comments and suggestions for the statistical analyses. Finally, thanks are due to Geneviève Lavigne PhD who helped in the preparation and edition of the present manuscript. [^1]: The authors have read the journal's policy about competing interests. The present study was funded by a special grant program in which the Canadian Institutes of Health Research partnered with AstraZeneca Canada Inc. to put in place a research initiative entitled “Community Alliances for Health Research and Knowledge Exchange in Pain”. Additional funding was also obtained from Pfizer Canada Inc. (<http://www.pfizer.ca>) under the form of an unrestricted research grant. Except for MC, all the authors of the present manuscript have declared that no competing interests exist. At the time this study was carried out, MC was a member of the Scientific Review Committee of the Pfizer Neuropathic Pain Award and received honorarium to review grant proposals and assist to the meetings of the Scientific Committee. The authors also certify that the funding sources do not alter the authors’ adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: PB AL SM RCH JC IG JBS MC. Analyzed the data: AL PB RCH MC. Wrote the paper: PB AL SM RCH JC IG JBS MC.

# Introduction It is well known that the human immunodeficiency virus (HIV) enters the brain soon after the onset of infection (within 3 to 6 days). HIV crosses the blood- brain barrier through infected macrophages and microglia in a process called a ‘Trojan horse’ mechanism. Due to this process the involvement of the central nervous system (CNS) in the course of HIV infection is already often observed in the early stage of the disease. The hepatitis C virus (HCV) is a frequent co-infection with HIV as a result of similar routes of viral transmission. HIV/HCV co-infection is associated with worse clinical outcomes of both diseases in terms of systemic disease progression and mortality. It has been evidenced that HCV may invade the CNS as well, since that virus is not exclusively hepatotropic, as it can also replicate in leukocytes, including monocytes/macrophages. The infected monocytes may cross the blood-brain barrier and thus provide access of HCV into the CNS in a process similar to that described for HIV. Moreover, several studies have evidenced that HCV infection is associated with cognitive dysfunction, fatigue and depression. Since both viruses mentioned above can infect the brain and impair CNS function, we hypothesize that they may cause alterations in cerebral perfusion, as measured by in vivo magnetic resonance (MR) perfusion weighted imaging (PWI). PWI is a method that brings information on cerebral flow at the capillary level (microvasculature). Among a few PWI techniques, dynamic susceptibility contrast MR imaging (DSC-MRI) is the most often used one. DSC MRI enables noninvasive measurements of relative cerebral blood volume (rCBV), thus providing information similar to that obtained in PET and SPECT studies but with several advantages, such as lack of ionizing radiation, high spatial resolution and low relative cost. The aim of our study was to evaluate the early perfusion changes using PWI in the neurologically asymptomatic HIV-1-positive non-treated (naive), HIV-1-positive treated with combination antiretroviral therapy (cART), HIV-1/HCV-positive untreated with any antiviral drugs and HCV-positive patients before antiviral therapy, who presented a normal appearing brain on plain MR. The other purpose was also to evaluate correlation of the perfusion measurements with immunologic alterations in HIV-1-positive subjects as well as with the liver histology activity index (HAI) in HCV-infected patients. It should be stressed that in the available literature there are only a few studies on the use of PWI in the assessment of early cerebral perfusion alterations in asymptomatic HIV-1-infected and HCV-infected patients with only mild liver disease,. Furthermore, those previous studies did not evaluate the differences between HIV-1-positive non-treated, HIV-1-positive cART treated, HIV-1/HCV-positive and HCV-positive patients in one research. To the best of our knowledge this is the first report focusing on the analysis of early perfusion changes within normal appearing grey and white matter in a large group of various neurologically asymptomatic subjects. # Materials and Methods ## Patients Fifty-six neurologically asymptomatic patients: 17 HIV-1-positive who had no need of cART according to the recommendations of EACS \[European Aids Clinical Society (*EACS*) Guidelines 2013, Version 7.0, 6–7\] (4 women and 13 men; mean age 33 yrs), 18 HIV-1-positive receiving cART (8 women and 10 men; mean age 39.3 yrs), 7 HIV-1/HCV-positive (2 women and 5 men; mean age 36.6 yrs) before antiviral therapy for HIV and HCV infections, 14 HCV-positive naive patients (6 women and 8 men; mean age 38.2 yrs) and 18 normal control subjects (6 women and 12 men; mean age 34.69 yrs) were enrolled in the study. HIV-1-positive patients with HCV coinfection as well as HCV-positive subjects suffered from chronic viral hepatitis with HCV RNA positive results for longer than 6 months. In HIV-1-positive as well as HCV-positive patients plasma viral load (HIV RNA or HCV RNA) at the time of imaging was determined. The Cobas TaqMan test version 2.0 with High Pure system isolation, the Roche Diagnostics real-time PCR method for HIV or HCV was used. In all the HCV patients, a liver biopsy was performed. Histopathological examinations of the samples revealed HAI (histology activity index) inflammation score and HAI fibrosis score of 0–2, which means only mild liver disease. The clinical characteristics of the studied groups are shown in. ## Ethics statement The study was conducted in accordance with the guidelines of the local University Ethics Committee for conducting research involving humans. All participants provided their written informed consent to participate in this study, according to the Helsinki Declaration, and the study was approved by the Commission of Bioethics at Wroclaw Medical University (number of permission: KB-225/2008). ## Neurological evaluation All patients underwent physical and neurological examinations, including determining their Karnofsky score (max. 100). The inclusion criteria were lack of neurological abnormalities in neurological examination and a Karnofsky score above 80. The exclusion criteria were as follows: 1. History of neurological diseases (e.g. inflammatory changes, brain neoplasms); 2. Psychiatric disease history (e.g. depression, schizophrenia, dementia). 3. Patients with CNS AIDS related events were excluded from the study. ## Psychological evaluation Two cognitive tests: Wisconsin Card Sorting Test (WCST) as a measure of executive function and Brickenkamp’s d2 concentration endurance test as a measure of visual attention were used in order to assess possible deterioration of cognitive functions. ## The control group (CG) The healthy control group with negative anti-HIV and anti-HCV antibodies had no history of drug abuse or liver disease and consisted mainly of the hospital staff from the Department of Radiology and Department of Infectious Diseases. ## MR imaging protocols Imaging was performed with a 1.5T Signa Hdx scanner (GE Healthcare) using a 16-channel coil dedicated for head and spine imaging. Conventional sequences included: axial, sagittal and coronal T2-weighted images, axial T1-weighted and FLAIR (fluid-attenuated inversion recovery sequence) images, as well as diffusion-weighted imaging (DWI). Only those subjects who had normal signal intensity of gray and white matter in all sequences mentioned above, without evidence of cerebral atrophy were included in our study. ## Perfusion weighted imaging (PWI) PWI was performed with a DSC method, using a gradient-recalled T2\*-weighted echo-planar imaging (EPI) sequence. The parameters used were as follows: TR/TE = 1900/80 ms, flip angle = 80°, number of excitations = 1, matrix size = 192×128, and slice thickness = 8 mm (with no gap). Perfusion images were obtained in axial slices parallel to anterior commissure – posterior commissure (AC-PC) line. Image acquisition started 10 seconds before contrast agent injection to establish a pre-contrast baseline. Ten seconds after the start of image acquisition, 0.2 mmol/kg of body weight gadopentetate dimeglumine (MultiHance, Bracco, Germany) was injected with a power injector (Medrad) at a rate of 5 mL/s through an intravenous catheter placed in the antecubital vein. This was immediately followed by a bolus injection of saline (20 mL at 5 mL/s). Total duration of acquisition was 1 minute and 24 seconds. The dynamic images were post-processed using Functool software (GE Healthcare, ADW 4.4). CBV maps were computed on a pixel-wise basis from the first–pass data as described by Belliveau et al.. All rCBV values were normalised to the mean values in the cerebellar cortex, which is the region minimally affected in patients with neurocognitive disorders as reported in the literature. CBV values were obtained from several regions of interest (ROIs) and were placed manually in both cerebellar hemispheres (circular ROIs, size 525 mm²), temporoparietal cortices (8 mm above AP-CP), frontal cortices (16 mm above AP- CP), the posterior cingulate gyri (PCG) region (single circular ROI, size 525 mm², covering cortex of the right and left cingulate gyrus), frontoparietal white matter in the centrum semiovale (circular ROIs, size 525 mm²) and the basal ganglia regions (irregular hand-drawn ROIs outlining the putamen, globus pallidus and caudate nucleus, size 500–600 mm²). Temporoparietal and frontal ROIs were almost rectangular in shape (size 800–1000 mm²). ## Statistical analysis Analysis of variance (ANOVA) followed by the post hoc Tukey least significant difference (LSD) test was used for statistical evaluation of the results. Correlations between rCBV measurements and the immunologic data (CD4a, CD4n T cell count) in HIV-1-positive patients as well as the liver histology activity index (HAI) in HCV-positive subjects were estimated using Pearson’s correlation coefficients. Statistical computations were performed using the Statistica PL software package version 10.0, and p\<0.05 was set as a significant level. Additionally, corrections for multiple comparisons were performed. Applying the Bonferroni correction the significant p-values were established as p\<0.0055 (p\<0.05/n, where n = 9 variables/testing hypotheses: 9 rCBV measurements). # Results ## Cognitive tests There were no statistically significant differences in executive functions measured with all the WCST scales, but HCV patients performed the d2 attention task significantly slower than controls (variable TN, t-test, p = 0.03) and had lower values of speed-accuracy variable (TN-E, t-test, p = 0.02). ## Comparison of rCBV values between HIV-1 and HCV patients and control group (CG) Compared to controls, HIV-1 naive patients showed significantly reduced rCBV values in the right temporoparietal cortex (p = 0.009), left frontal cortex (p = 0.020), as well as in posterior cingulate gyrus (p = 0.012). HIV-1cART treated as well as HIV-1/HCV naive patients demonstrated a significant decrease of rCBV values in both temporoparietal cortices (right: p = 0.007; left: p = 0.036 and right: p = 0.007; left: p = 0.005, respectively), both frontal cortices (right: p = 0.009; left: p = 0.0009 and right: p = 0.032; left: p = 0.027, respectively), as well as in posterior cingulate gyrus (p = 0.033 and p = 0.045, respectively). In HCV-infected patients rCBV values were significantly reduced compared to controls in both temporoparietal cortices (right: p = 0.010; left: p = 0.003), in the left frontal cortex (p = 0.005), as well as in posterior cingulate gyrus (p = 0.045). In contrast, in both BG regions HCV subjects showed significantly elevated rCBV values compared to CG (right: p = 0.0002; left: p = \<0.0001), as well as to HIV-1 naive, HIV-1cART treated and HIV-1/HCV naive patients. There were no significant differences in rCBV values among the studied groups in both frontoparietal white matter regions. ## Correlations of PWI measurements with the immunologic data in HIV-1 subjects Correlations of rCBV values with the CD4a and CD4n T cell count were assessed for all HIV-1-positive and HIV-1/HCV-positive subjects enrolled in the study. We found no statistically significant correlations between rCBV values in all evaluated regions and CD4a as well as CD4n T cell count. ## Correlations of PWI measurements with the histological data in HCV subjects There were no statistically significant correlations between rCBV values and the inflammation and fibrosis HAI score in HCV infected patients. ## Comparison of rCBV values between HIV-1 and HCV patients and control group (CG) using the Bonferroni correction (significant p\<0.0055) According to the Bonferroni correction HCV subjects showed a significant decrease of rCBV values in left temporoparietal and left frontal cortices compared to CG. This group revealed a significant increase of rCBV value in both basal ganglia regions compared to CG, as well as to HIV-1 naive, HIV-1cART treated and HIV-1/HCV naive patients. We also found a significantly decreased rCBV value in HIV-1/HCV naive patients in left temporoparietal cortex. There were no statistically significant changes between the studied groups in rCBV values in other analyzed regions according to the Bonferroni correction. # Discussion In our study we assessed early cerebral perfusion alterations in HIV-1 naive, HIV-1cART treated, HIV-1/HCV and HCV-positive naive patients, who presented a normal appearing brain on plain MR. We found significantly lower rCBV values (p\<0.05) in several cortical regions in all HIV-1-positive and HCV-positive patients compared to controls, while there were no significant perfusion alterations in the white matter areas. The HIV-1-positive cART treated patients presented with more pronounced perfusion changes compared to non-treated HIV-1-positive subjects, as the cART treated patients revealed hypoperfusion in more cortical regions. Moreover, HCV-infected patients seemed to reveal the most significant cerebral perfusion changes, as they presented with the significant alterations even with the Bonferroni correction (p\<0.0055). Hyperperfusion in basal ganglia in HCV patients is an interesting finding, which we try to explain. The studied groups were neurologically asymptomatic. There were no significant differences between HIV-1 and HCV-infected patients and the control subjects in executive functions measured with all the WCST scales. We only found that HCV patients performed the d2 attention task significantly slower than healthy controls and had lower values of speed-accuracy variable. Diminished visual concentration endurance and intact executive function suggest that HCV infection is related to specific cognitive disturbances. These disturbances result from changes in brain functioning of a rather focal than generalized nature. We found the decreased rCBV value within the frontal and temporoparietal cortex bilaterally, as well as in the PCG region in HIV-1-positive and HCV-positive patients. This hypoperfusion of several cortical areas may confirm CNS injury due to HIV-1 as well as HCV infection. There have been only a few studies which have provided evidence of early cerebral perfusion alterations in neurologically asymptomatic patients with HIV infection – as well as in a macaque model of neuro-AIDS using PWI. Moreover, these studies assessed the perfusion changes in single brain regions, especially in basal ganglia (BG) or in cortical areas. In contrast, we evaluated rCBV values in 9 brain regions including posterior cingulate gyrus (PCG), bilaterally temporoparietal and frontal cortices, parietal white matter areas as well as both BG. Wenserski et al investigated 10 HIV-1-positive patients with normal motor function, and assessed the mean rCBF (regional cerebral blood flow) only in BG and showed no significant perfusion changes compared to the control group. These findings are partially consistent with our results, as we did not observe significant perfusion alterations within the BG regions in HIV-1-positive asymptomatic patients. On the other hand, Ances et al examined the impact of HIV on rCBF within BG and visual cortex in 33 HIV-1-positive neurologically unimpaired subjects and reported significantly reduced rCBF within BG and visual cortex for both acute/early as well as chronic HIV-infected patients. They concluded that perfusion disorders which occur soon after seroconversion may reflect neuronal or vascular injury of the brain among HIV-infected individuals before they even demonstrate neuropsychological impairment. Our results showing decreased rCBV values in the temporoparietal and frontal cortical regions as well as in PCG are in accordance with previous findings from PET and SPECT studies, which also evidenced hypoperfusion within cortical and subcortical areas in HIV-1 positive patients. Recently, Towgood et al published an interesting paper combining evidence from arterial spin labeling (ASL) and PET method in asymptomatic HIV-1-positive patients. They found significant age effects on both ASL and PET with reduced rCBF and regional cerebral metabolic rate of glucose uptake (rCMRglc) in related frontal brain regions, and consistent, although small, reductions in rCBF and rCMRglc in the anterior cingulate cortex (ACC) in HIV-1 infected patients. The mechanisms for cerebral perfusion disorders observed in the course of HIV infection remain unknown. However, it has been hypothesised that one possible mechanism could be the direct effect of HIV on platelet function. HIV-1-positive individuals demonstrate an increase in the number of activated platelets, which may present increased susceptibility to adhere to each other as well as to blood vessel walls. This process of platelets adhesion stimulates serotonin release and vasoconstriction resulting afterwards in perfusion disturbances. The other possible mechanism may occur due to direct interaction between the virus and endothelial cell migration. Within the brain HIV propagates an increase in cytokine and chemokine release, which could cause deleterious effects to neurons and subsequently may result in perfusion alterations observed in HIV-infected subjects. Another factor contributing to the disturbances in the cerebral microcirculation may be an increased atherosclerosis documented in HIV-1-positive individuals,. Furthermore, the studies mentioned above did not evaluate the differences in perfusion alterations between the non-treated and treated HIV-1-positive subjects. It should be stressed that in our study the HIV-1-positive cART treated patients exhibited more pronounced perfusion changes (decreased rCBV values in 5 brain regions) compared to non-treated subjects who revealed hypoperfusion only in 3 brain areas. We suggest that this could result from prior brain impairment (advanced HIV-1 infection and high viral load) before treatment. Secondly, these changes may be caused by possible neurotoxicity of antiretroviral drugs. The increased frequency of dyslipidemia and glucoregulatory disorders in HIV- positive treated patients has been reported, as well as increased rates of metabolic disorders leading to atherosclerotic disease and associated vascular pathology or increased β-amyloid deposition in the brain. These findings of significantly decreased rCBV values especially in treated subjects are consistent with our previous report showing the most pronounced metabolic disorders in HIV-1-positive cART treated patients. In contrast, as already published in our pilot study, we observed increased rCBV values in the BG region in HCV-infected patients. We hypothesize that the hyperperfusion of BG may be a sign of inflammation in the early stage of HCV brain injury in patients with mild liver disease. This hypothesis corresponds with the data from the literature. Several reports have demonstrated the occurrence of hyperperfusion in HSV encephalitis in PET studies, as well as in other viral CNS infections in SPECT examinations. Focal hyperperfusion as visualized by SPECT or PET appears to be an indicator of active inflammation of brain tissue. Moreover, it has been reported that the first stage of HIV-1-associated Minor Motor Disorders is characterized by hypermetabolism within the basal ganglia. This hypermetabolism, as assessed with PET, is correlated with hyperperfusion of the BG region in PWI. We would like to point out that HCV-infected patients, although remaining neurologically asymptomatic, were the only subgroup which revealed some subtle differences when performing the d2 attention task compared to controls. This fact may also suggest the brain injury reflected by an inflammation process within BG due to HCV infection similarly to the first stage of HIV-1-associated Minor Motor Disorders. On the other hand, it should be also considered that hyperperfusion in the BG region may indicate the local compensatory increase of perfusion due to hypoperfusion in several cortex areas. To make the statistical analysis more powerful, we added correction for multiple comparisons. According to the Bonferroni correction there were no statistically significant changes between HIV-1 naive and HIV-1cART treated patients and the control group in rCBV values. However, it should be stressed that we still found significant changes (p\<0.0055) in PWI measurements in HCV-infected subjects in the left temporoparietal cortex, the left frontal cortex and in both basal ganglia regions as well as in HIV-1/HCV patients in the left temporoparietal cortex. Our results indicate that HCV-positive and HIV-1/HCV patients with only mild liver disease demonstrate the most pronounced changes in cerebral microcirculation compared to controls as well as to HIV-1 naive and HIV-1cART treated patients. To the best of our knowledge such findings have never been reported before. Our results of significant cerebral perfusion disturbances in HCV-positive subjects coincide with the current knowledge concerning HCV-related CNS disorders. The spectrum of CNS complications encountered in HCV-positive patients encompasses not only cerebrovascular events but also autoimmune syndromes. The cerebrovascular disorders, which have been documented in HCV- infected individuals included occlusive vasculopathy and vasculitis. Isolated CNS vasculitis has been confirmed by angiographic evidence of multiple focal narrowing of cerebral vessels. Recently, it has been reported that HCV infection could be an independent risk factor for increased carotid wall thickness and plaque formation, which contribute to significant cerebrovascular mortality in HCV-positive subjects. However, the exact mechanism responsible for the CNS vasculitis process related to HCV infection remains poorly understood. It has been hypothesised that recurrent precipitation of cryoglobulins with complement fixation and direct complement pathway activation by HCV itself, might be involved in ischemic and inflammatory tissue damage. On the other hand, the vasculitic process in cryoglobulin-negative HCV individuals, as in our study, may be probably secondary not only to the direct complement activation but also to an interaction between the virus present within the CNS and the host immune system. Additionally, we investigated the correlations between rCBV values and the immunologic data (CD4a and Cd4n T cell count) in HIV-1-positive patients as well as the inflammation and fibrosis HAI score in HCV infected patients. However, we did not find any statistically significant correlations. In the other studies, mentioned above, concerning perfusion changes in asymptomatic HIV-1-positive individuals, the authors did not analyse the possible associations between PWI measurements and the immunologic data. Chang et al reported that rCBF abnormalities correlated significantly with clinical disease severity as measured by CD4 count, plasma viral load, Karnofsky score, and HIV dementia scale, but they evaluated patients with HIV-cognitive motor complex, who presented with the mean AIDS dementia complex (ADC) stage ranging from 0.5 to 2 (moderate dementia). In contrast, we enrolled in our study only HIV-1-positive asymptomatic patients with ADC stage 0. Furthermore, except for our previous pilot study concerning early cerebral changes in HCV-positive patients, there were no other papers reporting perfusion cerebral alterations in that group of subjects and no other possible correlations between PWI measurements and the inflammation and fibrosis HAI score in HCV infected patients have been published in the literature. # Conclusion Perfusion-weighted MR examination is capable of depicting cerebral perfusion alterations in neurologically asymptomatic HIV-1-positive patients in the early stage of disease before progressing to HIV-associated neurocognitive disorders (HAND) as well as in HCV-positive patients with only mild liver disease. The HIV-1-positive cART treated patients presented with more pronounced perfusion changes compared to non-treated HIV-1-positive subjects. Moreover, HCV-infected patients seemed to reveal the most significant cerebral perfusion changes. Hyperperfusion in basal ganglia may be an indicator of brain inflammation in HCV patients. In our opinion rCBV measurement could be a noninvasive neuroimaging biomarker for assessing early cerebral microcirculation impairment in HIV-1 as well as in HCV infections. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JB BK AZ PS KM MJS AK. Performed the experiments: JB AZ AK PS KM BK JG MF. Analyzed the data: JB AZ AK PS BK KM JG MF MJS. Contributed reagents/materials/analysis tools: JB AZ AK PS BK KM JG MF MJS. Contributed to the writing of the manuscript: JB BK KM MJS AZ. Manuscript editing and review: JB AZ AK PS BK KM JG MF MJS.

# Introduction Electronic health records (EHRs) are primarily used for routine medical care, but secondary use of EHR data for observational research is becoming increasingly popular especially in studying of drug effects postmarketing. In this era data is used to generate information on drug safety and effectiveness in a cost-efficient way and by exploiting actual care patterns, which differ largely from experimental settings. In an experimental setting such as in randomized clinical trials, the choice for a treatment is randomized, which would take care of potential confounding by indication. In actual care the treatment decision is usually influenced by measurable patient characteristics such as medical history, concomitant drug intake but also by personal prescriber preferences, which cannot be measured easily. This phenomenon of preferential prescribing is also known as channeling and may lead to confounding by indication. A well-known example of channeling is the preference of doctors to prescribe selective cyclo-oxygenase-2 inhibitors (COX-2 inhibitors) over nonselective (ns) non-steroidal anti-inflammatory drugs (NSAIDs) to patients at risk of developing upper gastrointestinal bleeding (UGIB), as the COX-2 inhibitors were developed on purpose to mitigate the GI effects of NSAIDs. Although clinical trials showed that COX-2 inhibitors are ‘safer’ than nsNSAIDs in relation to UGIB, observational studies showed no large differences between the rate of UGIB between COX-2 inhibitor and nsNSAIDs, possibly due to residual confounding by indications arising from channeling. In order to obtain unbiased estimates in observational studies this confounding must be dealt with adequately. However, it is challenging to capture all relevant channeling factors in the EHR databases because information is not primarily recorded for research purposes. Moreover, relevant information may also be recorded in EHRs in an unstructured way. Attempts to construct methods that deal with confounding have resulted in the propensity score method, the propensity score is an estimated conditional probability of receiving one particular treatment over another given a set of measured covariates, it can be regarded as a comprehensive way to look at channeling. Propensity score methods can be used to control for the unbalance between the treatment groups in order to estimate the comparative effectiveness of treatments. Four different methods of using the propensity to reduce confounding have been described: 1) matching on propensity score; 2) stratification on the propensity score; 3) inverse probability of treatment weighting using the propensity score; 4) and covariate adjustment using the propensity score. Typically, all variables related to either the outcome and/or exposure, are included in the propensity score model, sometimes these variables are not the exact confounding factors but proxies thereof. Yet, identifying appropriate proxies in large EHRs is challenging. Schneeweiss *et al*. proposed a high-dimensional propensity score (hd-PS) algorithm to empirically identify a large number of relevant covariates, with high prevalence, to control for confounding. In a case study on coxibs and NSAIDs using claims data in the USA, application of the hd-PS algorithm to control for confounding was found to produce an effect estimate for the risk of upper GI complications between coxibs vs. NSAIDs that was comparable to the one found in randomized trials. The hd-PS model is constructed by using many covariates of which some could serve as proxies for unobserved factors that otherwise may not be considered. Typically, only structured information such as diagnostic or procedure codes that is available in the claims databases, are included in the model. Rassen et al. evaluated whether adding two-word phrases, present in patients’ unstructured free-text data, to the propensity score model could improve validity of pharmacoepidemiology studies. Adjusting for two-word phrases resulted in an improvement in confounding adjustment. Electronic health records comprise much unstructured data and we propose that this information could also be used as proxies for potential confounding factors. The aim of this study was therefore to assess whether unstructured text in EHRs can be used to construct a propensity score model that would allow to properly deal with confounding. We assessed the performance of propensity score models in addressing confounding by indication using as an example the association between selective COX-2 inhibitors and nonselective NSAIDs in relation to upper gastrointestinal bleeding. # Methods ## Data source We used data from the Dutch Integrated Primary Care Information database (IPCI), a population-based general practice EHR database. This database contains prospectively collected routine care data representing real-life practice. In the Netherlands, all citizens are registered with a general practitioner (GP), who acts as a gatekeeper to secondary and tertiary medical care. IPCI contains information on more than 1.8 million patients from 340 GP practices. For each individual person, all relevant medical information from primary and secondary care is documented in the medical record. Apart from patient demographics, the recorded information in the EHRs contain medical notes (including symptoms, physical examination, assessments and diagnoses), drug prescriptions, laboratory results, referrals for hospitalization or specialist care, and hospital discharge summaries. In the IPCI database, drug prescriptions are recoded according to the Anatomical Therapeutical Chemical (ATC) classification for research purposes. Diagnoses are coded according to the International Classification for Primary Care (ICPC). Almost 60% of the medical record are clinical narratives, which do not contain coded information, but contain important information such as patient-reported symptoms and notes from the GP. ## Selection of NSAID cohort We created a cohort of all new adult (≥18 years) users of NSAIDs between 1996 and 2013. Patients had to be enrolled for at least one year in the database in order to be eligible for cohort entry. ATC codes used for NSAID exposure are presented in. Within the NSAID cohort we created episodes of ‘new’ NSAID use according to the following criteria: (a) at least six months of data available before NSAID exposure, (b) no prescription of any nonselective NSAID or selective COX-2 inhibitor in the previous six months (c) no mentioning of drug names, in the free-text, corresponding with NSAID-related ATC codes in the previous six months. The duration of a prescription was calculated by dividing the prescribed quantity by daily dose regimen. An NSAID episode continued when consecutive NSAID prescriptions started before or within 30 days of the end of the duration of the previous prescription. The end of the episode was defined as the end of the last NSAID prescription. Episodes were classified as an nsNSAID or COX-2 inhibitor episode based on the first prescription in that episode being an nsNSAID or a COX-2 inhibitor, respectively. If a patient switched between exposure (from COX-2 inhibitor to nsNSAID or vice versa), the duration of the NSAID episode was ended at the switch of the exposure. A patient could have multiple NSAID episodes, but only if the above-mentioned criteria were met. ## Selection of upper gastrointestinal bleeding patients Within the cohort of new NSAID users we identified all potential subjects who experienced an upper gastrointestinal bleeding (UGIB) via an automated search. UGIB was defined as all forms of ulcer complications such as bleeding, perforation, or obstruction. The entire medical record of all potential UGIB patients was extensively reviewed to ensure the diagnosis and the date of onset. Any other cause of UGIB (such as variceal bleeding or Mallory Weiss bleeding) was excluded. The date of UGIB was determined as the date of first mentioning of symptoms leading to the UGIB diagnosis or if this date was unknown, the date of diagnosis. ## Propensity score model A propensity model was fitted using all information (structured and unstructured) in the EHR. To reduce the number of potential variables we first converted all text to lowercase after which we removed special characters, words not starting with a letter or a digit, stop words (such as *de*, *het–*the article *the* in English), and punctuation. All unique words (also known as unigrams) in the 6 months prior to cohort entry were extracted and used as textual features (potential covariates). This approach is commonly known as bag- of-words (BoW) model. We tested two methods to limit the number of covariates that would be included in the regression. The first method generated models using covariates of which the frequency in the cohort was above a certain threshold, e.g., 1000 without any further selection. In the second method, we generated a model using covariates that were associated with the outcome. The chi-square test was used to select covariates that were statistically significantly associated with the outcome (p-value less than 0.05). Another propensity model (method 3) was added for comparison, where only the established confounders (i.e. age, sex, and the exposure to low-dose aspirin) were included in the propensity score model. We used patients’ prescription information to calculate exposure to low-dose aspirin. The selected features were subsequently used in a large-scale regularized regression using a LaPlace prior with the hyper-parameter of 0.01 to construct a propensity model for each method. We experimented with several hyper-parameters and results of each are presented in the. The advantage of using a regularized regression is that it can handle high-dimensional data. A flowchart depicting the process of propensity score model generation (for methods 1 and 2) is presented in. We used three-fold cross-validation to evaluate the predictive accuracy of the models. The data set was randomly divided in three equally-sized subsets or folds. In three cross-validation runs, each time, the model was successively trained on two folds and tested on the third fold. For each cross-validation run, an area under the receiver operating characteristic curve (AUC) was calculated. The averaged AUC was used as the overall performance measure. ### One-to-many propensity score matching The propensity score that was generated in each of the two models was used to account for the preferential prescribing of COX-2 inhibitors to patients at high-risk of developing an UGIB. In this study, we used the greedy one-to-many matching as described by Rassen et al.: 1. For each COX-2 inhibitor cohort member the difference in PS with each nsNSAID users was computed 2. Starting with the lowest difference, each COX-2 inhibitor cohort member was matched with one nsNSAID cohort member. Once an nsNSAID user was matched, he or she was precluded from further matching. A caliper of 0.01 was used, meaning no matches were made if the difference in PS was greater than 0.01. 3. After all COX-2 inhibitor cohort members were matched with one nsNSAID cohort member, the process was repeated until all nsNSAID users were matched or there was no match possible. The algorithm ensured that all COX-2 inhibitor cohort members were matched with at least one nsNSAID cohort member if such a match was available within the caliper. ## Statistical analysis To estimate the risk of UGIB among COX-2 inhibitor users compared to nsNSAID users we calculated hazard ratios with their corresponding 95% confidence intervals (CIs) using Cox proportional hazard regression. We conducted the analysis for four datasets: 1) a crude comparison (unmatched, no propensity score); 2) matched on age (± 2 years) and adjusted for sex and exposure to low- dose aspirin, no propensity score; 3) matched on PS with covariate frequency above 1000 and then adjusted for age, sex, and exposure to low-dose aspirin; and 4) matched on PS with covariates having an association with the outcome and then adjustment for age, sex, and exposure to low-dose aspirin. # Results ## NSAID cohort From the source population of more than 1.8 million patients we identified 518,768 new users of NSAIDs based on ATC codes. We then processed the unstructured free-text in the entries of the new users to identify mentioning of drug names corresponding with NSAID-related ATC codes. In total, 36,188 new users were removed because either an nsNSAID or COX-2 inhibitor drug was mentioned in the free-text in the six months preceding first NSAID exposure. This resulted in 482,580 new NSAID users in the study cohort. Out of these, 459,701 (95%) were nsNSAID users and 22,879 (5%) were COX-2 inhibitor users. Within the NSAID cohort we retrieved 11,994 potential UGIB patients. After reviewing the medical records we retained 1,048 UGIBs. The average duration of episodes for initiators of COX-2 inhibitors was 94 days and 66 days for initiators of nsNSAIDs. Baseline characteristics of initiators of COX-2 inhibitors and nsNSAIDs are shown in. Most of the episodes of COX-2 inhibitors and nsNSAIDs were started after the year 2004. ## Propensity model In total, we extracted 2,762,326 covariates (i.e., unique words, out of almost 96 million words) from approximately 2.4 million entries in the 6 months prior to NSAID episodes from the medical records of 482,580 new NSAID users. shows the performance of the propensity models built using different covariates selection methods. The first model used all covariates with a frequency of 100 or more in the cohort, which resulted in 95,078 unique covariates entered into the model. Increasing the frequency to 1,000 resulted in a reduction of the number of covariates to 27,619. The number of covariates further reduced when frequency was increased to 5,000. The performance of the models in terms of their predictive accuracy was comparable. The predictive performance of the propensity model that was built using 3,650 covariates that had an association with the outcome according to the chi-square test. This resulted in an AUC of 70.59. The performance of the propensity model that included only the established confounders resulted in an AUC of 66.27. The number of covariates in the models however were only 111. ## Risk of upper gastrointestinal bleeding The crude hazard ratio of UGIB for COX-2 inhibitors compared to nsNSAIDs was 0.50 (95% 0.18–1.36). When matched on age, the hazard ratio of COX-2 inhibitor use compared to nsNSAID use was 0.36 (95% CI: 0.11–1.16). Further adjusting for sex and exposure to low-dose aspirin resulted in HR of 0.35 and 0.36 respectively. Matching on PS only, using one-to-many matching with a covariate frequency above 1,000, reduced the hazard ratio to 0.42 (95% CI: 0.13–1.38). Subsequent adjustment for age resulted in a hazard ratio of 0.36 (95% CI: 0.10–1.22). Matching on PS limiting to covariates that were associated to the outcome also provided a hazard ratio of 0.42 (95% CI: 0.13–1.39). Adjusting for age reduced the hazard ratio to 0.32 (95%: 0.09–1.09). The top-25 covariates, in terms of their weights (beta values), from both propensity score models are presented in the and Tables. # Discussion In this study, we generated a propensity model using unstructured information from EHRs. We tested different methods to construct this and demonstrated the feasibility to do so as well as its performance. Since electronic health records are now widely available for secondary use, we need to develop methods and test performance of these methods for use in epidemiological evaluations such as drug effects. Our method to generate a propensity score model is substantially different from the high-dimensional propensity score (hd-PS) approach proposed by Schneeweiss et al. The hd-PS algorithm that was developed for claims data uses structured information such as diagnostic codes, in-patient procedure codes, and drugs dispensed. In each identified data dimension, the highest ranked codes are selected to enter in the hd-PS model. The use of two-word free-text phrases in addition to the structured information has also been positively evaluated in the context of hd-PS models. Our method is different since we used as the basis all unstructured text to generate propensity models, using a large-scale regularized regression, without pre-identified data dimensions. Several methods other than logistic regression such as data-adaptive and classification trees have been proposed for fitting a propensity model. To reduce the number of ‘meaningless’ features, we needed various textual data cleaning steps. We subsequently extracted all unigrams from the cleaned free-text, which served as potential covariates. Here we applied different approaches, to look at the impact of our choices. In the first method, the most-frequent covariates in the cohort were selected to enter the propensity score model. Since the covariates were selected merely on the basis of their frequency in the cohort, this method is prone to include covariates that may actually be instrumental variables. Instrumental variables have an association with the exposure but not with the outcome except through their effect on exposure. If covariates are included that are not true confounders, the variance increases and sometimes a small amount of bias may be introduced. In order to mitigate the potential to include covariates that are instrumental variables we included covariates with a significant association with the outcome to the propensity score model in the second method we applied. We used three-fold cross-validation to evaluate the predictive performance of exposure to nsNSAID or Coxib for each generated PS model. In the first method where covariates were selected based on their frequency, increasing the frequency threshold for covariate selection reduced the number of covariates that entered into the propensity score model but the performance of the models was still comparable. This suggests that the performance of the models was mostly based on a few covariates with high occurrence in the text. Reducing the number of covariates reduced the computation time needed to fit the model. By selecting covariates with an association with the outcome, we significantly reduced the total number of covariates without greatly affecting the performance. The propensity models generated using covariates with only high frequency in the cohort performed better than the one where association with the outcome was verified. This may be due to the presence of some instrumental variables which can result in an increase in predictive performance. We used another propensity model for the comparison purposes where only the established confounders age, sex, and exposure to low-dose aspirin were included. The predictive performance of this model was lower than the other two models which were generated from the free-text covariates. The second method, where covariate association with the outcome was verified, showed large decrease in the hazard ratios after further adjustments. Whereas previous studies have constructed the hd-PS with structured information, such as ICD and READ codes across different data dimensions in different sources, large proportions of information may be unstructured. We showed that this unstructured free-text can be used to construct propensity models. Initially, the new user cohort was created based on the prescription tables containing ATC codes. A high number of removals (7%) from the cohort based on the drug mentioned in the free-text indicates the importance of processing unstructured free-text instead of only relying on the structured information. Our study also has several limitations. First, by including covariates based on their frequencies we might have selected covariates that are not necessarily related to the outcome or the exposure, which could introduce bias. Second, since we only used unigrams, covariates like ‘*congestive heart failure’* cannot be recognized as such. Instead it will be recognized as individual words ‘*congestive’*, *‘heart’*, and ‘*failure’*, which might lead to over- and underestimation of some covariates. Like previous studies using hd-PS methods, we also used the known association between NSAIDs and UGIB as an example. It is unclear whether our findings regarding the PS generated from unstructured free- text apply to other treatment-outcome pairs. Since the PS algorithm in general relies on the information present in the cohort, a similar approach using a different data set might have different results even when using known example of NSAID-UGIB. The majority of COX-2 inhibitor episodes started after the year 2004, the period after the withdrawal of rofecoxib from the market because of cardiovascular risks. This may explain the strong protective effect of COX-2 inhibitors in the crude analysis which we would expect, but is different from previous observational studies that were done more closely to the introduction of coxibs. Since most of our patients started after the contra-indications were introduced, channeling towards high risk patients was less of an issue. In conclusion, our study showed that PS models can be created using unstructured information in electronic healthcare records. We also showed that the PS model where covariates were filtered on their association with the outcome provide an improvement in adjustment for confounding. This is useful for database studies using a large amount of unstructured free-text as in EHRs. Better methods for extracting meaningful covariates from the free-text may be required for effective proxy adjustment via propensity scores. # Supporting information [^1]: This study was supported by the VICI project 91896632 of the Netherlands Organization for Health Research and Development ZonMw. Janssen Research and Development provided support in the form of salaries for author M.J.S. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Technological advances made over the past few years have increased the digitalization of medicine, thus leading to a considerable growth of clinical, research and public health datasets. These data sources are increasingly related to the *big data* environment and they include, amongst others, genomics and other-omics related-data collections, electronic health records (EHRs), patient registries, medical imaging, administrative data and clinical trials data. However, a good part of such datasets are often kept and analysed in silos and not adequately shared. If properly integrated into the clinical life cycle, such collections of data stand to offer a unique opportunity to drive scientific discoveries and improve healthcare research. For example, they may allow a better understanding of the aetiology of illnesses and subsequently help in improving the management, prevention and treatment of diseases. This is even more promising in the framework of learning healthcare systems, where clear boundaries between research and care are dissolving and the same data are used both for improving scientific knowledge and providing better care. In this context, developing the harmonization of health data—described as the sum of all “efforts to combine data from different sources and provide users with a comparable view of data from different studies” —is crucial to improve clinical research and practice. Such standardized approach requires not only better quality, re-usability and interoperability of data, but also more open and collaborative communication between the different stakeholders active in the health data environment. The fact that a good percentage of healthcare spending are being wasted as a consequence of under-exploiting data potential in several healthcare systems around the world should be considered as one important factor urging for such changes to happen. Harmonized health datasets are also laying the foundation of a new era of biomedical research, where three concepts are currently converging, namely precision medicine, learning healthcare systems and implementation science. The harmonization of health data is a complex procedure which involves significant changes in how data are collected, shared and linked. Harmonization can be either prospective, when modifications occur in the study design to subsequently render the pooling of data more straightforward, or retrospective, when pooling is performed with data collected previously according to different study designs. In practical terms, harmonization can be achieved through two distinct but complementary approaches, namely a “stringent” and a “flexible one”. By means of a “stringent” approach, data are harmonized through the use of standard collection tools and standard operating procedures, implementable only in a prospective way. With the “flexible” approach, on the contrary, different data collection tools might be used, as long as operating procedures are standardized. In achieving the harmonization of health data, careful consideration needs to be given to already well-known as well as novel challenges related to the processes of collection, sharing and linkage. Such challenges are drastically intensified by the vastness and the hyper-connectedness of data at present time, which may result in unforeseen connections or cross-referencing between datasets, drastically increasing re-identification risks for data subjects. The presence of these challenges has resulted in the emerging of several barriers to the effective use and sharing of health-related data. Although these have been categorized as technical, motivational, economic, political, socio-cultural, ethical and legal, a more precise mapping of the exact content of such barriers, and of the solutions that have been elaborated to mitigate them, is lacking. Within this framework, the aim of this systematic review is to identify more precisely some of the barriers and facilitators encountered in the effort to achieve harmonization of health data—including the processes of data sharing and linkage—by a comparative analysis of studies conducted in two countries having different healthcare systems and data infrastructures, namely Denmark and Switzerland. These countries where chosen because, although they both offer high quality healthcare, they have two very diverse healthcare systems and two different data infrastructure models for healthcare. Denmark has a Beveridge- based national healthcare system and a long tradition of data linkage in health through its nationwide registries. On the contrary, Switzerland is based on a federalist Bismarckian organization of healthcare and started much later to develop strategies in the field of Health Information Exchange. In this perspective, this review seeks to identify past and current studies related to the field of harmonized health data collection, sharing and linkage in these two countries and list the barriers encountered and the facilitators that make these projects successful. Furthermore, the review aims to provide some insights on the complexities associated with the use of health data that can be of relevance also in the broader international context. # Methodology ## Search strategy and study selection This study conformed to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines, and its protocol was registered on January 3^rd 2018 on PROSPERO (CRD42018081424). A systematic literature search was performed on four search engines and electronic bibliographic databases namely PubMed, Web of Science (all databases), EMBASE (no Medline) and CINAHL for publications with dates ranging from 1^st January 2008 to 31^st December 2017. The time period is aligned with the adoption of the Swiss national eHealth Strategy in 2007, with the aim of introducing electronic patient records at national level. The search was repeated for the period of 1^st January 2018 to 31^st March 2019 to include additional publications and to ensure that our systematic review is up-to-date. Reference lists of included publications were screened to identify other potential harmonized health data collection, sharing or linkage projects. A search strategy was developed for each electronic database. The literature search included Medical Subject Headings (MeSH) terms and free applicable text to health data collection, sharing and linkage. The search strategy consisted of three components, namely (1) types of health data, (2) keywords for data collection, sharing and linkage and (3) country of interest. For instance, the search strategy for Switzerland on PubMed was: ("Administrative Claims, Healthcare"\[Mesh\] OR "Health Records, Personal"\[Mesh\] OR "Clinical Coding"\[Mesh\] OR "Patient Discharge Summaries"\[Mesh\] OR "Clinical Trials as Topic"\[Mesh\]) AND ("Databases as Topic"\[Mesh\] OR "Data Collection"\[Mesh\] OR "Medical Informatics"\[Mesh\] OR "Medical Record Linkage"\[Mesh\] OR "Information Dissemination"\[Mesh\] OR "Data Integration" OR "Data Sharing") AND ("Switzerland"\[Mesh\]) \[filters used are Articles types (Clinical Study, Clinical Trials (including controlled and Phases I to IV), Comparative Study, Evaluation Studies, Journal Article, Multicenter Study, Observational Study, Pragmatic Clinical Trial, Randomized Control Trial, Technical Report and Validation Studies), language (Danish, English, French and German) and species (Human Studies)\]. We did not include harmonization as an imperative component in our search strategy since the exact boundaries of this concept are still controversial and the addition of the term “harmonization” as an imperative component drastically reduced the number of publications for each country. Eligibility criteria for this study were: (i) publications based on health data collection, sharing or linkage projects. There was no restriction on study design and type, i.e. qualitative, quantitative or mixed method studies, and clinical or observational studies were included; systematic reviews were excluded; (ii) there were no restriction on age, gender, disease and ethnic group of participants involved in these studies; (iii) the studies had to involve some health data collection, sharing or linkage at cross-institutional, cross-national or cross-regional levels in at least one of the two countries; (iv) only English, French, German and Danish language articles were included, and (v) publication year of articles ranged from January 2008 to March 2019. ## Data extraction and quality assessment The literature search results were catalogued on EndNote^TM X8, a reference manager software. The titles and abstracts of all articles were screened independently by two authors (LDG and AM). The full-texts of the included publications were reviewed by LDG and AM to ensure that they met the eligibility criteria to be included in the systematic review. LDG and AM performed independently the data extraction from the included articles through a standard data extraction form developed progressively by the authors of this review. Additional publications gathered through reference screening went through title and abstract, and independent full-text screenings and data extraction by MCM. Another review author, TW, validated randomly twenty percent of the publications reviewed by LDG, AM and MCM, to assess the quality of the data extraction process. A disagreement level of less than 10% for the data entries was considered acceptable. The data extraction form included (i) study information (author(s) and publication year), (ii) sources of health data, (iii) cross-institutional or cross-national nature of the study, (iv) presence or absence of primary and secondary health data collection, analysis and sharing, and lastly (v) the categorization of barriers and facilitators to harmonized health data collection, sharing and linkage. The sources of health data were categorized as having three standard origins, namely health services, public health and research. Other sources of health data falling outside these three categories were classified in a residual category (“Other”). LDG and AM performed a categorization of the identified barriers and facilitators separately, and came to consensus on the final categorization of these elements for accuracy and inclusiveness. Disagreements were solved with the mediation of TW. The identified barriers and facilitators to harmonized health data collection, sharing and linkage were subsequently clustered into main categories, which were then sub-clustered into smaller categories to highlight the most common barriers and facilitators in these main categories (the full clustering/sub-clustering of barriers and facilitators is shown). For the purpose of this systematic review, we defined harmonization techniques as methods which would allow the coherent pooling of different data sources, involving health data collected either prospectively, retrospectively or both. Examples include the use of standard case report forms or data dictionaries, a central review of the collected data, training provided to researchers/stakeholders and leadership role by one of the partners for coordinating data collection, sharing or linkage activities. ## Analysis A narrative synthesis of included publications was carried out. This step involved the categorization of health data collection, sharing and linkage projects based on their national or cross-national dimension, their source of health data, and barriers and facilitators identified in these publications. This step was important to highlight similarities and differences between projects in Denmark and Switzerland. The statistical software, STATA ® version 15.0, was used for the different analyses. # Results A total of 1928 papers were initially retrieved from the search engines and electronic bibliographic databases for the period of January 2008 to December 2017. The search was repeated for the period of January 2018 to March 2019 (upon request of the journal) resulting in a total of 425 additional papers. The result of the two searches were combined for each stage of the PRISMA resulting in an overall total of 2353 papers retrieved for the period of January 2008 to March 2019. Duplicates (n = 170) were removed either automatically using ENDNOTE X8 or manually after reviewing abstracts and their titles. The remaining 2183 papers went through title and abstract screening, which resulted in the exclusion of 1789 papers. In-depth full-text screening was performed for 394 papers, and 115 more papers were excluded for not meeting the inclusion criteria (see for reasons). Reference screening of the 279 included papers, resulted in the identification and inclusion of 66 additional papers which met the eligibility criteria for this systematic review. The 345 included papers are summarized in, where they are categorized based on their national (n = 240) or cross-national (n = 105) dimension, their sources of health data and the total number of barriers and facilitators reported for each project. We identified 200 Danish and 40 Swiss national projects, and 105 cross- national projects. Among these cross-national projects, 14 projects involved the use of health data from both Denmark and Switzerland, 51 and 40 projects involved a Danish partner and a Swiss partner respectively. Overall, the number of projects which involved primary health data collection, sharing and analysis were 106 (30.7%), 92 (26.7%) and 106 (30.7%) respectively. Comparatively, the number of projects which involved secondary health data collection and analysis were 283 (82.0%; if a study collected both primary health data and secondary health data, it was counted for both). Of the 345 projects, 199 used health data from health services, 211 from public health sector, 94 from research and 62 from other health data sources. ## Overview of barriers Barriers of an ethical nature were reported 19 times in the included records and they concerned mainly issues related to privacy (n = 9) and respect for autonomy of study participants (n = 6). As to legal barriers, these were reported 17 times and they included issues associated with national data protection regulations (n = 4), differences in national legislations concerning data security and privacy (n = 4) and “Other” (n = 9) (e.g. legal uncertainty concerning health data collection or sharing, market restriction, etc.). Overall, the type of barriers that were more often reported, however, were those of a technical nature. In the records, 416 technical barriers were mentioned and they were classified as data quality issues (e.g. data incompleteness, potential misclassification of data, etc.) (n = 234), lack of data standards (data structure and semantics, e.g. ambiguous terminologies, temporal evolution of data standards, etc.) (n = 151), limited technical capabilities (e.g. no unique identifier, etc.) (n = 21) and “Other” (n = 10) (e.g. time constraints on physicians preventing the use of standard procedures for data collection). Financial barriers were also reported, but only a limited amount of times (n = 9), and they were principally referring to the unavailability or inadequacy of financial support (n = 8). Only 13 political barriers were found and they comprised institutional/constitutional organization issues (e.g. federalist system and different healthcare systems) (n = 6), mistrust between stakeholders (n = 3), data ownership issues (n = 2) and “Other” (n = 2) (e.g. no official guidelines for data sharing). Studies also reported some motivational barriers, including lack of research incentives (n = 17) (including additional workload imposed on physicians/researchers), data re-use prevented by stakeholders as they are deemed unfit for secondary use (n = 2), stakeholders’ competing interests (n = 2) and additional barriers of a diversified content, thus labelled as “Other” (n = 4) (e.g. study participants not showing up for part of the study). Finally, 6 socio-cultural barriers were reported in the included records, half of which were related to a “cultural clash” (n = 3), which we defined as issues resulting from different cultures in data collection, sharing and linkage of the partners involved in the project. ## Overview of facilitators Facilitators of an ethico-legal nature were reported 582 times in total, and they were classified as official/legal approval of study (e.g. Danish Data Protection Agency) (n = 148), ethical approval by a REC/IRB (n = 135), legislation permitting to proceed with health data collection, sharing and linkage without consent or REC/IRB approval (n = 79), obtaining informed consent from participants (n = 69), health data anonymization (n = 58), the presence of legislation requiring mandatory reporting (n = 41), confidentiality measures (n = 29; e.g. data security audits), project done according to international laws and regulations (n = 8), data access rights for patients (n = 4), clear legislation for data collection, sharing or linkage (n = 3) and “Other” (n = 8) (e.g. study data made available by researchers upon request). Facilitators of a technical nature were reported 981 times in total, which were grouped in three categories, namely techniques for data harmonization (n = 798), data linkage (n = 155) and “Other” (n = 28) (e.g. study allowed the creation of optional and mandatory datasets, whereby a minimum of data are classified as mandatory). Facilitators of a financial nature, especially explaining how funding was successfully secured, were mentioned 12 times. These referred, for example, to public-private partnerships, where both partners would gain some benefits from the collaboration, as a solution for funding issues (n = 3). 169 facilitators related to politics were reported. These referred to the structure of the health system as an advantage for harmonized health data collection, sharing and linkage (n = 139), data access control by the players (n = 11), the presence of a data sharing agreement between the stakeholders (n = 9), building and maintaining stakeholders’ trust for collaboration (n = 7) and “other” (n = 4). There were 14 motivational facilitators, which included monetary incentives to incite researchers/stakeholders to abide by standardized procedures for data handling and management (n = 7), improved data collection tool to ease the workload of researchers/stakeholders for data collection/sharing (n = 3), a memorandum of understanding between partners to ensure collaboration till end of study (n = 2) and “other” (n = 2). Lastly, there were 8 socio-cultural facilitators, which included data subjects controlling access to their data (n = 4) and “Other” (n = 4) (e.g. transparent policies for the participants). Country-wise distribution for all six facilitators categories are presented in. ## Barriers and facilitators identified in national Danish and Swiss projects When considering only national projects (n = 240) involving either Denmark (N = 200) or Switzerland (N = 40) alone, there were 323 identified barriers and 1234 facilitators. Technical barriers and facilitators were most frequently reported. For comparison purposes and compensation for the imbalances in the number of national projects identified in each country, the absolute numbers and the number of barriers and facilitators per 1,000 national projects for each country are illustrated in. Interestingly, the only identified category of barriers which was comparatively almost equally reported in Swiss and Danish single-country projects was that of technical barriers. Otherwise, ethical, legal, financial, motivational and socio-cultural barriers were reported 5.0, 3.3, 3.3, 5.7 and 10.0 times more in Swiss projects than in Danish projects respectively. On the contrary, a Swiss project reported on average more facilitators than a Danish one (only financial facilitators were reported equally in both countries). Ethico-legal, technical, motivational and socio-cultural facilitators were reported 1.2, 1.3, 4.2 and 1.3 times more in Swiss projects than in Danish projects respectively. Only facilitators related to politics were reported 4.5 times more in Danish projects than Swiss projects. ## Barriers and facilitators identified in cross-national Danish and Swiss projects With respect to cross-national projects (n = 105), there were 182 identified barriers and 532 identified facilitators. Technical barriers and facilitators were more frequently reported than those of another nature. For comparison purposes and compensation for the imbalances in the number of cross-national projects involving each country, the number of barriers and facilitators per 1,000 cross-national projects was calculated (excluding cross-national projects involving both countries) and illustrated in. Concerning cross-national projects, we observed a reverse tendency as compared to national projects. Studies involving a collaboration with a Swiss partner have, on average, reported 1.8 times less barriers than those involving a Danish partner. More in detail, projects including Switzerland reported 4.7, 1.9, 1.6, 3.1 times less barriers of an ethical, technical, financial and political nature respectively, than those with a Danish partner. However, cross-national projects involving a Swiss partner, reported 2.1 times more barriers of a motivational nature than those with a Danish partner. Comparatively, cross-national collaboration involving either a Swiss or Danish partner reported almost the same number of facilitators. Ethico-legal and political facilitators were identified 1.1 and 2.4 times more in cross-national projects with a Danish partner as opposed to cross-national projects involving a Swiss one. However, technical facilitators were identified 1.2 times more in cross-national projects with a Swiss partner than in those with a Danish one. # Discussion This systematic review provides a comprehensive overview of projects from either Denmark or Switzerland which involved the collection, linking or sharing of data and of the barriers and facilitators related to the usage of health data therein reported. Our study includes a broad range of projects relying on data from different sources and contexts (health services, public health, research and other) and it confirms that studies involving the harmonization, linking or sharing of health data still encounter a high number of obstacles, but also underscores that barriers have prompted the development of numerous solutions. We will here address and discuss the findings related to barriers and facilitators of each cluster that was identified. ## Ethico-legal barriers and facilitators Although ethico-legal factors are often described as some of the most problematic elements when it comes to linking and sharing health-related data, our results show that barriers of this nature are rarely reported. The small amount of ethico-legal barriers identified might either mean that such barriers were rarely present or that they were present but underreported. In our view, the latter option is more probable for at least two reasons. Firstly, as the records included in this review were all published articles, the explicit mentioning of ethico-legal complications might have been avoided to bypass problems related to publication. Secondly, ethico-legal factors are often less tangibile and transparent in comparison–for example–with technical ones and they are thus more likely to be superseded. Moreover, underreporting would confirm that ethico-legal aspects related to processing of health data are still underappreciated, which is a major obstacle to the final success of research projects. This also suggests that there is some resistency by authors to openly disclose and discuss ethico-legal problematics. For the future, a less cautious approach would be much more beneficial, since it would allow new research projects to build on the issues encountered by old ones. Ethico-legal facilitators were more widely mentioned. Results show that Swiss projects are still predominantly anchored to the “consent or anonymise” approach, according to which the solution to solve ethico-legal problematics concerning health data is to either anonymize information or to require explicit authorization by data subjects. Differently, Danish projects have made vaster use of alternative solutions, such as relying on specific confidentiality tools, and, more importantly, exploiting regulation that allows—upon certain conditions—to share and link health-related data without the need of obtaining consent by data subjects or REC approval. This demonstrates that the development of proper regulations to facilitate the harmonization and linking of health data offers practical solutions that projects developers are then willing to use. In this framework, another important finding concerns the role of the data protection authority. Whereas in Switzerland this public office–although existing–does not play a defined role with respect to research, results show that Danish studies have a more active interaction with the Data Protection Agency, as they need to apply for permission to use health data. The nature of the application to the national Data Protection Agency that Danish projects need to file is not explicitly described in the records reviewed, but it has been presented elsewhere as a less demanding procedure, resembling a simple duty of notification. Thus, many Danish projects dealing exclusively with health data–in accordance with national regulation–do not need to apply for full ethical review from a REC or IRB, an often demanding and lengthy process, but simply have to obtain clearance from the Data Protection Agency. This institutionalized interaction with the public authority responsible to ensure compliance with data processing rules can be an important factor helping project developers, since it incentivizes to proactively tackle privacy concerns. This interaction could thus be considered as a model to inspire changes in the regulatory framework in Switzerland. ## Technical barriers and facilitators In this systematic review, data quality issues were the most commonly reported barriers, followed by the lack of data standards and limited technical capabilities. Although Denmark has a developed health data infrastructure, numerous identified projects described that data quality problems still affect health services, public health and research datasets. This is confirmed by other studies, such as a review on the Danish National Patient Registry (DNRP) where the authors concluded that data incompleteness and heterogeneous validation methods of data limited the research potential of this registry. Although relevant, data quality issues can be mitigated in a system like the Danish one, since linkage between data from different registries can be easily performed using the personal identification number (CPR) provided to all Danish citizens at birth and to stable residents. Comparatively, Swiss projects and projects involving a Swiss partner also reported slightly more issues related to data quality than to data standards. However, in comparison to their Danish counterparts which reported almost twice more issues related to data quality than data standards, the difference in reporting of data standards and data quality issues was smaller in Swiss projects. This more equivalent reporting could imply that data standard issues are considered as important as data quality issues for the success of Swiss projects. Indeed, the high levels of data-heterogeneity in the Swiss healthcare context might stem from the fragmented nature of the healthcare system, where each of the 26 cantons \[federal states\] has a high degree of autonomy and where more than 55 health insurers are active. These findings underline how technical issues are interconnected with the context where projects are carried out, and that also external systemic factors–and not simply internal complications of the projects themselves—affect the emerging of these barriers. In Denmark, for example, the presence of nation- wide registries fosters the development of studies relying on secondary use of routinely collected data, where researchers are more likely faced with issues about the quality of data, since the latter was originally collected for a different purpose. On the contrary, in a country like Switzerland—where data are more often prospectively collected—issues about the absence of common standards because of fragmentation are also likely to be evident, on top of data quality issues. Our findings suggest therefore that even technical issues concerning data are strongly embedded in the surrounding where projects are conceived. This should induce project developers to communicate and learn from each others, since the barriers they will encounter and the solutions they will find are more likely to be dependent also on the context where they act, and not only on the specific features of their research. For example, since Switzerland’s healthcare sector does not use a universal personal identification number because of privacy concerns, linkage of data will almost certainly represent a technical challenge, regardless of the features the single project or the data that it aims at using. ## Motivational and financial barriers and facilitators With respect to motivational and financial factors, our findings are partly in line with the literature. Previous research had underscored that the key motivational and financial aspects concerned the lack of research incentives from resource-limited institutions, the fears of being ‘robbed’ of data before publication or of losing reputation because others might identify errors in the data, the reluctance to facilitate access due to potential inappropriateness of further uses, the need to secure resources for data sharing activities and the necessity to make arrangements between institutions for data management costs. Overall, national and cross-national Swiss projects combined reported more frequently motivational and financial facilitators than their Danish counterparts. This suggests that in a country with a less institutionalised system of data sharing and where studies often have a prospective design, more strategies are elaborated to deal with financial and motivational issues related to data, since–with a lower systemic support–single project developers have to make a greater effort. In contrary, in a context like Denmark—with the high prevalence of studies with retrospective design and the reliance on secondary uses of routinely collected health data–the need for financial and motivational facilitators might be lower. In fact, when health data harmonization is prevalently retrospective, a lower number of actors is involved–since primary data collectors are rarely included–thus reducing the urgency to create motivational or financial incentives for a large number of collaborators. Another important finding related to financial aspects is that the presence of economic constraints can be the source of additional barriers related to data harmonization, such as data quality issues. For instance, the Swiss project AMIS Plus—concerning a register for acute coronary syndrome—could not envisage systematic site visits to assess data quality or more in-depth questionnaires due to resource limitation. In Denmark, similarly, with the Copenhagen School Health Records Register—a health examination register for schoolchildren containing data on more than 350,000 individuals—financial constraints made it impossible for the authors to computerize the entire health card, thus limiting the understanding of potential confounding variables. This indicates even more that barriers of different natures are interconnected and that new projects need to acknowledge this interconnectedness of the barriers to successfully address them. ## Political barriers and facilitators Danish national projects did not report any barriers of a political nature, whereas cross-national collaborations mentioned a few, such as data ownership and organizational issues. This suggests that an institutionalization of data processing practices, similar to what occurs in Denmark, helps to remove political obstacles. Moreover, the presence of a centralized healthcare system structure also proves helpful, because it reduces the number of actors involved and thus the presence of competing interests. Political issues, however, might re-emerge when projects are cross-national and thus abandon the relatively *safe-haven* created at the national level. In a context like the Swiss one, on the contrary, political barriers seem to be more relevant for national projects, because these fuel internal conflicts related to the diversity of interests within healthcare and to the difficulty of implementing uniform and centralized policies. In fact, the two most mentioned political facilitators in Switzerland–building trust amongst stakeholders and stakeholders retaining control over data access–are both related to the attempt to coordinate the numerous different parties operating in the health data field and accommodate their competing interests. This might also explain why less political barriers are reported for Swiss cross-national projects. In fact, when projects from a context like the Swiss one go to a supra-national level, the chances of disputes related to in-country political antagonism to emerge is lower. Our results are thus in line with the literature, where mistrust between stakeholders, absence of comprehensive guidelines for data sharing and lack of legal accountability were identified as major political issues. However, our results further show that the incidence of political barriers seems quite different in single-country studies if compared to cross national ones. This finding is particularly important since it underlines that sometimes the choice of a national or cross-national design might have an impact on the number of political issues encountered. ## Socio-cultural barriers and facilitators Barriers and facilitators of a socio-cultural nature were rarely mentioned in the included records. Comparatively, the incidence of cultural barriers seems to be higher for Switzerland, where cultural clashes were mentioned more often than for Danish projects. Such difference could be due to the higher degree of fragmentation of the Swiss healthcare system in comparison to the Danish one, which is centralized and state-funded. In fact, one Swiss study reported that the choice for a distributed model in the managing of data was based on prior failures to implement centralized systems of health data and public mistrust towards the concept of centralization. Socio-cultural facilitators were mostly related to the involvement of data-subjects by allowing them to retain control of data access. For instance, the Swiss project reported that data subjects had the possibility to decide which part of their medical records could be considered “stigmatizing”, and thereafter blinded to healthcare professionals, other than their designated and trusted physician. The designated and trusted physician would have access to the full record. It is naturally impossible to determine whether socio-cultural barriers were actually overlooked or simply not reported. In either case, the limited mentioning of these factors signals an underappreciation of their importance. On the contrary, socio-cultural aspects should be carefully considered by project developers, since the harmonization of health data cannot ignore the cultural peculiarities of the single contexts from where data are pooled. Harmonization, linking and sharing do not happen in a vacuum and opening up the dialogue between data processors and society at large can be an important success factor for the harmonization of health data in the long run. ## Limitations The limitations of this systematic review include choices that we made regarding the number of databases used for our search, the fact that we did search using English key words, and that only 20 percent of included papers went through double checking for data extraction consistency. We could have thus missed valuable studies that were published only in Danish, French, and German which we could have found if key words had been in those languages. Given the high number of papers included and resources related constraints, we were unable to double check for all information recorded, but in light of low discrepancies found in the portion of records which were double-checked, we are confident in our output. A reporting bias of barriers and facilitators identified in the included papers cannot be excluded as published papers are focused mostly on the effectiveness of their interventions rather than on the implementation phase. It is possible that our results are thus biased towards barriers and facilitators more likely to be reported in the papers (e.g. those of a technical nature). Given the low numbers of certain types of reported barriers and facilitators, it is difficult to compare the situation in the two countries without under- or over-exaggerating their presence or absence in the two countries. However, the main objective of this systematic review was to identify barriers and facilitators to harmonized health data collection, sharing and linkage in Denmark and Switzerland. Causal inference was not part of this review’s primary objectives. # Conclusion This systematic review gathered evidence from Switzerland and Denmark to map and describe barriers and facilitators concerning data harmonization, sharing and linkage. Given the focus of this review on Switzerland and Denmark, part of the findings has specific relevance for these two countries. In particular, for Switzerland it has emerged that fragmentation in the health data environment is a key challenge for harmonizing, sharing and linking of data. Since the implementation of more centralized governance systems—which are of great use in Denmark—might not be a viable option for Switzerland because of the political structure of the country, a distributed governance model, which emphasises interoperability of health data, seems to be the preferable way forward. The introduction of Blockchain technology for patient records, which insures security and respects decentralization, is reportedly an auspicious technology as its use in the Estonian healthcare system described by Mettler suggests. This review outlined that the existing data infrastructure at the national-level in Denmark incentivizes the completion of retrospective registry-based studies relying on data reuse. Although barriers are still reported, the existence and comprehensiveness of this data infrastructure confirms that past efforts to improve the health data framework have proven successful. For the future, efforts should focus on easing projects involving cross-national collaborations. However, other findings are meaningful well beyond the borders of the two countries specifically considered. In particular, in this review it has emerged that, although a great number of barriers and facilitators are mentioned by the projects involving health data harmonization, sharing and linking, reporting focusses predominantly on specific aspects–above all technical ones. Whereas technical aspects are certainly important, the reluctancy to mention also issues of other natures is detrimental to the more general effort of the scientific community to favour the harmonization of health data. Referring more openly to the difficulties encountered at the ethico-legal level, for example, might be of help both for new projects to develop appropriate approaches and for policy makers to gather evidence on which regulatory interventions are needed. The under-appreciation of ethico-legal, socio-cultural and other context-specific complexities is a faux-pas, since the trust of both data-subjects and society at large is indispensable for the success a community in improving the health data context, like the experience of Iceland has demonstrated in the past. There, the project to build a national “health sector database” with health information of all citizens imported from their medical records failed also due to the underappreciation of ethico-legal issues (e.g. informed consent and privacy). Specifically, the population complained that inclusion of personal medical records into the database was supposed to happen without consent by individuals or the possibility to opt out. This was felt like a violation of privacy, because of the risk of re-identification and also due to the fact that the database was supposed to be run by a private company. A privacy complaint was brought in front of the national high court, who ruled against the project to build the database. For this reason, the project was definitely aborted. In summary, the success of current and future projects is likely to depend on a better understanding and appreciation of the complexities associated with harmonizing, sharing and linking health data. In the same line, proposed solutions to harmonization issues should not underestimate the contextual particularities of the country, in which such health data processes occur. # Supporting information The authors acknowledge the financial support received from the Swiss National Science Foundation (SNF NRP-74 Smarter Health Care, grant number 407440_167356). The SNSF is not involved in any other aspect of this review and the views expressed are those of the authors and not necessarily those of the funder. This systematic review is part of a broader project funded by the SNSF titled “*Advancing Smart solutions and eliminating barriers for the Acquisition*, *Analysis and Sharing of Health data in Switzerland* (SMAASH)”. MCM was an independent researcher when the work for this systematic review was carried out and she is currently affiliated with the Institute for Social and Preventive Medicine (ISPM) at the University of Bern. [^1]: I have read the journal's policy and the authors of this manuscript report the following non-financial competing interests: LDG and MCM are married and this does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Cell therapy is an exponentially growing field with \>2,500 clinical trials in the world over the last 10 years. Cell-based therapeutic products are positioned as a billion dollar per year industry with anticipated market growth. The method of using cells as drugs is particularly advantageous when a higher-order approach to treatment is required. A fundamental issue, particularly for transient cell therapies which are the largest of the drug class (50%), is the lack of predictive measures of the body’s response to the therapy and vice versa: traditionally known as a drug’s pharmacokinetics (PK) and pharmacodynamics (PD). Without a rigorous PK/PD model of the mechanism of action of transient cell therapies, processes cannot be optimized to formulate a drug and physicians will be unable to effectively monitor and communicate the benefits and risks of a cell therapy to a patient. Mesenchymal stem cells (MSCs) - an adult multipotent progenitor cell population initially derived from bone marrow – have over a decade of clinical testing in thousands of patients worldwide. These cells have appealing manufacturing properties such as the ease of isolation, expansion, and cryopreservation, as well as the tolerability of allogeneic cell therapy. MSCs are widely being evaluated for the combinatorial treatment of a variety of diseases including myocardial infarction, bone marrow transplantation, stroke, autoimmune disease , and wound healing. The predominant use of MSCs as an immunomodulatory agent is substantiated by the observation that MSCs can inhibit the activation and effector function of numerous immune cell types. The mechanism of action of immune cell inhibition by MSCs is primarily due to the release of secreted factors by cells. MSCs have proven effective in early stage clinical trials, yet, several Phase II and Phase III industry-led trials either have undergone early termination or have failed to meet primary endpoints. New strategies to optimize this therapeutic are needed to help usher this promising cell population to the clinic. A deep understanding of MSC pharmacology can provide insight on how to best deliver this therapeutic. A prevailing view of MSC trafficking is that, upon intravenous administration, cell homing, engraftment, proliferation, and/or differentiation are critical to induce a therapeutic effect. Sensitive monitoring techniques have demonstrated little, if any, long-term engraftment (*\>*1 week) of MSCs upon systemic administration with the majority of administered MSCs (*\>*90%) accumulating immediately in the lungs and then cleared with a half-life of 24 h. Therefore, MSCs can be viewed conceptually as a transient cell therapy that delivers a “payload” of secreted factors to alter acute disease progression. We put the concept of MSCs as a therapeutic delivery vehicle to the test with a focus on evaluating the systemic release of secreted factors. Investigators have studied the release of natural secreted proteins such as TSG-6 and TRAIL or engineered antibodies from MSCs in the systemic circulation after transplantation, although a thorough pharmacological analysis was not performed nor compared to purified proteins. In this study, conventional cell biodistribution data was combined with serum profiles of natural or engineered secreted factors released by the cells to understand rate-limiting processes that alter exposure to intravenous MSC therapy, the most widely used administration method. Based on our combined PK analysis, we observed that molecular monitoring of serum secreted factors revealed interesting phenomena regarding the delivery of natural or engineered proteins, an active cell secretion mechanism, and the host immune response to the graft. This study can aid in designing an optimal cell delivery regimen to maximize MSC therapy. # Results ## Comparison between the Bioavailability of Transplanted MSCs and Secreted Molecules Secreted factors released by MSCs after transplantation have been a general phenomena that has been observed in several pre-clinical therapeutic studies. It is reported that MSCs can secrete a wide range of cytokines and growth factors. We designed a pharmacokinetic study to monitor both the viability and distribution of MSC transplants and overlay the serum profiles of a specific and detectable level of interleukin (IL)-6, a pleiotropic molecule secreted by MSCs. Human MSCs were engineered with a luciferase reporter gene, injected via intravenous (IV) route (the convention used by the cell therapy community) and detected after injection using whole-body, bioluminescent imaging (BLI). By nearly 8 hours after IV cell injection the BLI signal was almost undetectable. At the same imaging time points, using human MSCs that were not engineered, we also measured the serum levels of human IL-6. We observed that the release of human IL-6 by MSCs was short-lived and tracked accordingly to the cellular BLI signal of the graft. Similar pharmacokinetics of two other protein secreted factors, monocyte chemoattrive protein (MCP)-1 and IL-8, were observed after IV injection in vivo further supporting our observation. Future experiments continued to explore and evaluate the impact of these pharmacokinetic profiles, with a focus on molecular delivery of IL-6 by IV administration. ## Cell Delivery of IL-6 Achieves Greater Exposure than Molecular Delivery of IL-6 Cells can be considered a “carrier” for their secreted factors. We used this perspective to understand the differences of delivering IL-6 in a purified form compared to a cell transplant. Conditioned medium from MSCs (referred to as MSC- CM) was generated in order to compare MSC-derived IL-6 without concerns of post- translational modifications that take place in other protein expression systems for recombinant forms of IL-6 that may affect bioavailability. Concentrated MSC- CM was prepared and a volume of 400 µL (containing 40 ng IL-6) was injected into mice. The content of soluble factors in this volume is equivalent to the secreted levels when 3×10⁶ cells are cultured for 2 days in vitro and was normalized to 1×10⁶ cells for comparison to cell transplants at the same cell dose. MSC-CM was administered by IV and the serum levels of IL-6 followed a classical bolus pharmacokinetic profile. Pharmacokinetic parameters for serum IL-6 levels were calculated with the assumption that the clearance of IL-6 itself was a constant 218 ml/hr based on the literature. The maximum serum IL-6 concentration was increased by ∼400% when delivered by cell transplants. This amounted to a greater than 7-fold increase in the exposure of the subject to IL-6 as measured by area under the curve (AUC) analysis. The temporal kinetics of IL-6 was also artificially extended by way of cell transplantation. The time to reach maximum serum concentration, the half-life, and the elimination constant of IL-6 were all significantly modified for prolonged duration of IL-6 by the use of cell transplantation. These data highlight the supplementary changes to molecular pharmacokinetic parameters by way of cellular delivery. ## MSCs Utilize an Active, Golgi-Dependent Secretion Mechnaism to Release IL-6 *In Vivo* The presence of human IL-6 in serum after cell transplantation could be due to passive mechanisms, such as cell rupture and cytokine release, or by active secretion pathways. Brefeldin A (BFA), a protein-transport inhibitor, was used to block IL-6 production by inhibiting Golgi apparatus-dependent vesicle secretion. A non-toxic concentration of 5 ug/ml BFA was chosen by evaluating a dose response of BFA to MSCs in vitro. After incubating MSCs with BFA for one day, the in vitro secretion of IL-6 in the supernatant was significantly inhibited compared to the untreated controls. The secretion of IL-6 was not restored after a day of incubation with BFA until 72 hrs later. We were satisfied with blockade of IL-6 for \>60 hours in vitro and advanced this cell formulation with impaired IL-6 secretion for in vivo PK studies. The PK profile of MSC-derived IL-6 was dramatically different, specifically a reduced maximal effective concentration (∼5×) and AUC (∼1000×) compared to the cells that were not treated with BFA. This study highlights a powerful active mechanism that MSCs employ to deliver IL-6 to the bloodstream that requires the Golgi apparatus. ## Exposure of MSC-derived IL-6 is a Function of Host Immune Cells To study the natural process of protein release from non-engineered cells, we designed our PK model using human cell transplants in mice. This model afforded us the ability to detect human proteins in mouse serum with high specificity and sensitivity and correlate that to *in situ* cellular production. We were also interested in studying the immune response, albeit a xenogenic rejection response, and its ability to alter in vivo protein release. We transplanted human MSCs by IV administration in mice strains that are immunocompetent (C57Bl/6), less immunocompetent (Foxn1−/−, thymic and peripheral loss of T cells), or severely immunodeficient (NOD-SCID-IL-2rg −/−, loss of B, T, and NK cells). The effect of the immune response on MSC-derived IL-6 serum kinetics was pronounced after IV administration. NOD-SCID-IL-2rg −/− mice treated with human MSCs had a delayed maximum effective concentration and a longer half-life of serum IL-6 than the same data gathered from Foxn1−/−, or C57Bl/6. The exposure of the subject to MSC-derived IL-6 was quantifiably different amongst recipients, suggesting that molecular monitoring of cell therapy can distinguish clearance mechanisms of the MSC transplant and/or secreted factors, in particular a form of immune clearance. ## Enhanced Systemic Exposure of Secreted Factors by Engineered MSCs MSC-derived IL-6 is a natural and highly expressed secreted factor that enabled our molecular monitoring approach, however IL-6 can be influenced by the *in vivo* microenvironment that MSCs encounter. In order to understand the maximal exposure of a subject to MSC secreted factor that was uninfluenced by host regulation, we developed genetically engineered MSCs with constitutive expression of the naturally secreted Gaussia luciferase (Gluc) reporter. Gluc has a circulation half-life of 5–10 minutes in mice, and has been used as a highly sensitive reporter (detection of ∼1000 cells) for quantitative assessment of cells *in vivo* by measuring its level in the blood ex vivo. GLuc activity can be easily quantified in blood by adding its substrate coelenterazine and measuring emitted photons using a luminometer. MSCs were first transduced with a lentivirus vector to stably express Gluc and GFP under the control of the constitutively active CMV promoter which yielded high transduction efficiency as monitored by fluorescent microscopy and flow cytometry for GFP. Gluc expression and secretion was also readily detected in MSCs conditioned medium. When infused into NOD-SCID mice, GLuc was detectable over a one week period with a time to peak concentration occurring at ∼8 hours post-cells transplantation. Engineering of MSCs with GLuc revealed a longer exposure to cell therapy and suggests engineered cell formulations for therapeutic studies may be useful for minimizing cell dose and/or frequency to achieve durable responses. # Discussion Pharmacology is a powerful discipline with many available methods to help understand and shape the characteristics of a drug to meet a therapeutic need. Monitoring a drug’s PK is a necessary study of a drug formulation that has been modeled extensively by balancing the infusion, absorption, metabolism, and elimination of a drug, assuming the body is defined by a steady-state of discrete compartments that are permeable to a drug. Drug formulations that contain living cells are an ever-increasing class of therapeutics that can benefit from such pharmacological analysis. The classical monitoring strategy for a cell therapy is to define the trafficking and viability of a cellular formulation after introduction to the body. In this study, we coupled this strategy with molecular monitoring of secreted factors released by a cell transplant that became detectable in the systemic circulation. We identified that these two PK profiles - that is, the PK of an administered MSC population and MSC-secreted IL-6 - were discordant. This was dependent on the imaging modality used in this study (BLI), which has known limitations in whole-animal reporting due to tissue diffusion distance of light emission. BLI showed a prominent lung expression of the MSC graft for a period of several hours, which engineered MSC-Gluc releasing a secreted reporter could be detected for days. Exploration of the dynamics and rate-limiting aspects of MSC-derived secreted factors in blood serum, rather than the cells themselves, may provide more predictive power in their use as a therapeutic. The PK of cell-derived molecules can be considered a lumped model of classical drug PK with additional complexity to consider due to cellular delivery. Cells as a delivery vehicle may be responsible for the difference in profiles compared to isolated molecules. Administration of a molecular drug bolus typically follows an exponentially decaying trajectory in terms of serum concentration of the drug. After cell administration, we observed that serum levels of secreted IL-6 followed an exponential decay independent of administration route. This may be attributed to limited engraftment MSCs entrapped in the lung. The diffusion of IL-6 from the cells would subsequently follow and be limited by tissue transport barriers before being detected in the bloodstream. In the systemic circulation secreted factors would likely follow their natural clearance pathways, which, in the case of IL-6, are very rapid (∼minutes in half-life). We propose that traditional PK models can be modified to account for an active, cellular production term as well as the diffusion of material from an engrafted tissue bed. The elimination, or clearance, of a drug compound can occur by metabolism or excretion of the drug in voided volume. In the case of a MSC transplant, we could detect the influence that the immune system has on the exposure of a subject to MSC secreted factors. The immune response we observed is likely a xenogenic rejection response that can be compartmentalized by arms of innate and adaptive immunity. Previous studies have shown that the immune competency of the host can have an impact on the cellular viability after transplantation. There are a number of reports that describe the role of NK cells and other mechanisms that affect the viability of allogeneic MSCs after transplantation including the generation of immunological memory to the transplanted material over time. The clearance of MSCs was substantial in IV transplanted MSCs, which were entrapped in lung tissue. NOD-SCID-IL-2rg −/− mice had a much greater exposure to MSC- derived factors compared to Foxn1−/− mice or C57Bl/6 mice. The difference between the two immunocompromised strains lies in the absence of B cells, NK cells, and defects in cytokine signaling in NOD-SCID-IL-2rg −/− mice suggesting that these cell populations and cytokines may be accountable for MSC clearance. This study, however, cannot distinguish immunologic elimination of the human cells versus elimination of the human secreted protein without further evaluation. We, and others, have observed protection of acute liver injury by molecules derived from MSCs. MSC molecules collectively had a dose-dependent effect on injured hepatocytes and stimulated a proliferation response in vitro and in vivo. In this study, we used IL-6 as a model secreted factor to study the cellular release profile of this cytokine in vivo. Although specific mediators of therapy still remain elusive, there is considerable evidence that IL-6, itself, may be liver-protective.. IL-6 is also shown to be important in improving acute inflammatory response. We envision that other cytokine and growth factors released by MSCs will help broaden the overall PK profile of this transient cell therapy and begin to identify associations with relevant diseases that can be combated through the combination of factors released by MSCs. Cognate receptors for specific secreted factors will help clarify the pharmacodynamics of MSC action and if dosing is enough to activate downstream signaling of MSCs are amenable to *ex vivo* engineering to express therapeutic secreted factors such as IL-2 for cancer immunotherapy. In this study, we first report the use of a non-specific GLuc reporter that was engineered into MSCs for sensitive blood detection. MSC-Gluc revealed a longer exposure of the subject to a secreted factor implying that MSCs were persistent in the body although undetected by BLI. MSCs were genetically engineered ex vivo with a self- inactivating lentivirus vector which integrates the Gluc cDNA within the genome of the cells, leading to stable expression. This study cannot rule out the possibility that MSCs could fuse with a host cell (e.g. myeloid cells) after transplantation and thereby maintain a longer serum level. These data also suggest that constitutive secreted factors are necessary to reveal the true bioavailability of MSCs, as IL-6 and presumably other natural secreted factors may be regulated at the gene expression level by the host. The initial release dynamics of IL-6 were contributed by an active Golgi-dependent mechanism, which is presumed to be necessary for continued secretion of other protein factors such as GLuc. Non-protein secreted factors, such as nitric oxide or prostaglandin E2, may not be rate-limited by Golgi-secretion pathways but instead follow enzymatic reaction kinetics that require stimulation to generate these mediators *in vivo*. MSC-Gluc can be an extremely useful tool for cell pharmacology studies that evaluate administration routes, initial dosage, and dosing frequency for optimizing the exposure of a cell therapy product. This work serves as the first application of a combined cell biodistribution and molecular PK modeling approach to MSC therapeutics. By combined *in silico* modeling and empirical analysis, a functional PK/PD model can now begin to be developed to predict the nature of MSC therapy given a particular formulation and administration route. Allometric scaling laws that help predict the conversion of parameters from animal to human models may be applicable to guide clinical trials using MSC therapeutics. Although we focus on a few key mediators, a more comprehensive view of all bioactive MSC secreted factors can lead to second-generation models that better capture potential non-linearity in the data. In addition, this theoretical framework may serve as the foundation for other clinically used cell variants such as hematopoietic and embryonic stem cells, or T cells. Such predictive, *in silico* analysis of cell-based therapies may reduce experimental costs due to a systematic minimization of required testing, increase throughput of discovery, and ultimately lead to more efficacious treatment regimens. # Materials and Methods ## Mice Athymic Foxn1−/− mice (nude, male, 6–8 weeks old), C57BL/6 mice (male, 8 weeks old), Balbc/J mice (female, 8 weeks old), and NOD.Cg-**Prkdc^scid Il2rg^tm1Wjl**/SzJ (NSG mice, male, 8–10 weeks old) were all purchased from Jackson Laboratories (Bar Harbor, ME) and housed at Massachusetts General Hospital Animal Facility following approved experimental protocols by the IACUC. ## Human MSC Isolation and Expansion Human MSCs were isolated and expanded following a previously established protocol. Briefly, fresh human bone marrow aspirates were purchased from Lonza. Mononuclear cells were separated by Ficoll density gradient centrifugation (GE Healthcare) and plated on a T-175 flask (1×10⁶ cells per flask). Mononuclear cells were cultured at 37°C with 10% CO₂ in MSC expansion medium. MSC expansion medium was composed of 15% fetal bovine serum, 2% penicillin and streptomycin, 0.2% gentamycin, 1 ng/L fibroblast growth factor, alpha-MEM with ribonucleosides and deoxyribonucleosides. Medium was changed 1 week later and unbound cells were washed away. The following week, colony- forming adherent cells were re-plated into a new flask for expansion. Medium was changed every 3–4 days. MSCs were subcultured when they reached 70–80% confluence. Only passage 2–5 MSCs were used for experiments. outlines the immunophenotype of MSCs using antibodies purchased from BD Biosciences. ## MSC Administration and Measurement of Human IL-6, MCP-1, and IL-8 Levels in Plasma MSCs (1×10⁶ MSCs in 200 µl FBS-free medium) were injected into mice by IV infusion. At 30 minutes, 3 hours, 8 hours, 24 hours and 72 hours, mice were anesthetized with 60 µl ketamine, 30 µl xylazine, and 60 µl saline per mouse and blood was withdrawn by cardiac puncture. After centrifugation at 14,000 rpm for 10 minutes at 4°C, plasma were collected and stored at −80°C before use. Human IL-6, MCP-1, and IL-8 levels were measured using an ELISA kit from BD Bioscience following the supplier’s recommended procedures. ## Preparation of Human MSCs Expressing a Firefly Luciferase Gene Reporter The lentiviral vector pHR’MND-LRT containing a firefly luciferase reporter was constructed as previously described. Infectious virus was produced by triple transient co-transduction of 293T/17 cells (ATCC) with pHR’MND-LRT, pCMVΔR8.91 i.e. packaging vector, and pMD.G i.e. VSVG pseudotyping vector. The titer of virus was determined by transduction of 293T cells followed by flow cytometry analysis of the mRFP reporter (Ex: 594 nm/Em: 620±15 nm). Cultures of 30–40% confluent human MSCs in a T-175 flask were incubated with the virus at a multiplicity of infection of 4 in a total of 20 ml expansion medium containing 8 µg/ml polybrene. This transduction protocol was repeated one more time. In each round, cells were incubated with the viral supernatant for 8 hours and then in MSC expansion medium for 16 hours. After the second round of infection, fresh medium was added to each flask and cultured for 3–4 days. Luciferase activity of transduced MSCs was confirmed with a luciferase activity assay before in vivo use. ## Bioluminescence Imaging A total of 1×10⁶ luciferase-engineered MSCs were given to C57Bl/6 mice either IM or IV. At specific time points after cell injection, mice received an intraperitoneal injection of 4.5 mg of luciferase substrate solution (Molecular Imaging Products) and were imaged thereafter. The bioluminescent signal was measured in anesthetized mice on an IVIS-100 imaging system (Caliper LifeSciences) until a peak signal was reached. Data are expressed as photons/second/cm², encompassing a region of interest over the implanted cells, including lung, leg and whole body. ## Preparation of Concentrated Conditioned Medium MSCs were cultured to 70–80% confluency in T-175 culture flasks before 15 ml DMEM media consisted of 0.05% BSA and 2% penicillin and streptomycin were added. Cells were further cultured for 1–2 days and then the supernatants were collected and filtered. Cell number was quantified using a hemacytometer after trypsinization. Culture media were concentrated 20–50 folds using an Amicon filter (MWCO: 3,000 Da) by centrifuging at 3500 rpm for 2–3 hours. The human IL-6 levels were measured by ELISA by appropriate serial dilution before injection into mice. Concentrated conditioned medium (400 µl) were injected into C57Bl/6 mice by IV or IM routes. ## Engineering and *In Vivo* Monitoring of MSCs with Secreted Gaussia Luciferase MSCs were allowed to grow up to about 70% confluency before viral transduction. A lentivirus vector carrying the expression cassette for Gluc and GFP, separated by an internal ribosomal entry site, under the control of the CMV promoter was previously described with a titer of 6.1×10⁷ IU/ml. Polybrene was added to each T-175 flask diluted down to a final concentration of 1×. Then 1 mL of virus was added to each flask. Cells were allowed to grow over night and then the virus-containing media was aspirated and replaced with fresh virus-free media. Transduction efficiency was confirmed by analyzing GFP expression using fluorescence microscopy and flow cytometery. Fully confluent cells were trypsinized with 1× Trypsin (Fisher) and re-suspended in conditioning media at a density of 1×10⁶ cells per 200 uL. 200 uL of cell suspension was injected into each mouse via tail vein. Blood was collected in Eppendorf tubes containing 4 µL of 20 mM EDTA via tail vein at 0.5 hour, 3 hour, 8 hour, 24 hour, 72 hour and 1 week post-MSCs injection. 10 uL blood was mixed with 100 uL 5 ug/ml coelenterazine substrate in a white, opaque 96-well plate and luminescence was detected using a BioTek microplate reader. ## Brefeldin A Treatment of MSCs and Proliferation Assay In a 6-well plate, MSCs were incubated with brefeldin A (Sigma-Aldrich) at a final concentration of 50, 10, 5, and 1 µg/ml for 24 hours. Supernatants were collected, and then MSCs were washed 3 times using PBS. Fresh medium was replaced, and cells continued to culture up to 3 days. Supernatants were collected at different time points for subsequent human IL-6 measurements. To measure the proliferation, BFA-treated cells were reseeded in a 96-well flat bottom plate and cultured with fresh medium for 72 hours. Cell proliferation was measured using a MTT assay kit (ATCC) following the supplier’s recommended procedures. ## Statistical Analysis In all studies batches of 3–8 mice from 2–3 independent experiments are reported. Raw pharmacokinetic data were analyzed using a 1-tailed Mann-Whitney U test for non-parametric data with the mean ± SEM shown or using a two way ANOVA with Tukey’s multiple comparison correction where pharmacokinetic parameters were calculated based on MATLAB software package models. # Supporting Information The authors are grateful for assistance from Ms. Jessica Sullivan and Mr. Peter Waterman in bioluminescence imaging. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FW MLY BT BP. Performed the experiments: JE RM FW KS SG KC BP. Analyzed the data: JE RM FW BT BP. Contributed reagents/materials/analysis tools: KC RW BT. Wrote the paper: JE RM BT BP.

# Introduction Globally, nonspecific low back pain (LBP) is one of the leading causes of years of disability and absenteeism from work. In 2017, an estimate of 577 million people around the world suffered from LBP. The lifetime prevalence of LBP is 70 to 80%. Most of those affected recover within the first six weeks. However, 10 to 40% remain in a state of nonspecific chronic low back pain (CLBP). CLBP is defined as pain below the costal margin and the gluteal region, with or without radiation to one or both legs, that persists for 12 weeks or more. CLBP, as a major health problem, is a long-lasting, extremely common musculoskeletal condition without a clear pathology. It limits activities and results in a high personal, social, and economic burden. Patients with CLBP have an increased presence of psychological factors such as kinesiophobia, maladaptive coping, anxiety, depression, catastrophising, or low self-efficacy. Consequently, CLBP is often conceptualised as a biopsychosocial condition, i.e. as a complex and dynamic interaction between physical, psychological, and social elements. Focusing on the different influencing factors of CLBP inpatient multidisciplinary biopsychosocial rehabilitation (MBR), an intensive, multicomponent inpatient intervention is frequently recommended in clinical treatment guidelines for CLBP. As the condition is already chronic, the primary objective of MBR is to restore the daily functioning of the participant, not to cure the pain. It is usually provided by multidisciplinary teams of healthcare professionals in rehabilitation centres or specialised pain clinics and has been shown to be very effective in reducing pain and disability. Outcomes are generally patient-reported physical functioning and pain intensity in the short, medium, and long term. However, MBR in secondary or tertiary care is limited in numbers, not widely available in healthcare settings, is associated with long waiting times, and requires a large number of treatment hours from different healthcare professionals. This leads to a high economic burden on individuals and the health system. According to habit theory, individuals changing their context are more likely to perform non-habitual behaviour. But as they return to their home environment, a relapse to old habits is more likely to occur. The best evidence for effective outpatient treatment of CLBP is exercise delivered in a high-dose of at least 20 treatment hours that is maintained continuously after the initial intervention. It should be noted that effective exercise therapy, which reduces pain and disability, often does not correlate with improved physical function of the musculoskeletal system. It may be that other exercise-induced changes such as emotional functions and improved cognitions (e.g., reduced anxiety, improved self-efficacy, less catastrophizing and less fear-avoidance behaviour) influence pain and disability more than changes in muscle strength, mobility, or muscular endurance. To increase the chance of long-term training maintenance and improved psychological outcomes, a focus could be placed on participants’ motivation, self-efficacy, and change in daily habits with a combined approach of exercise and psychological therapy in the form of a primary care biopsychosocial intervention (PCBI). Furthermore, outpatient PCBI could support people in establishing plans to cope with difficult situations, anticipate and practice high-risk situations in the social or physical environment. Another advantage of PCBI is its possible administration by trained physical therapists. Therefore, it seems to be an interesting and potentially cost-effective approach, as a physiotherapist session in primary care is more accessible and cheaper than multidisciplinary treatment. If it were further possible to reduce the dosages of outpatient active treatment sessions by using additional psychosocial elements and still receive positive outcomes, the treatment cost could possibly be further reduced. Thus, treatment could be offered to a larger population, with a shorter waiting time. On the effectiveness of general PCBI compared to active treatment, there are mixed results with conflicting evidence. Some studies indicate a beneficial effect in favour of PCBI. While others report no difference between PCBI and conservative activity treatments in terms of effectiveness. Positive results of low-dose PCBI have been achieved with cognitive functional therapy. The last systematic review on PCBI covered the scientific literature until 2015 and identified promising effects. A systematic review with meta-analysis, which covered group-based physiotherapy-led behavioral psychological interventions identified a small but significant long-term pain reduction compared to other active treatments. As it covered only group-based interventions and did not differ in dose, the results can only indicate that outpatient treatment is a promising alternative to other active treatments. The third systematic review in this area of research focused on cognitive-behavioural based interventions delivered by a physiotherapist. The authors found high-quality evidence for reduced disability compared to no treatment, usual care, or other active treatment. However, they also had a broad focus and included all types of controls, making a direct comparison difficult. A preliminary search of PROSPERO, MEDLINE, the Cochrane Database of Systematic Reviews and the JBI Evidence Synthesis was conducted and no current or underway systematic reviews have focused on low dosages of PCBI compared to conservative exercise therapy. Therefore, the objective of this review is to understand whether a low-dose (and therefore less costly) PCBI with a maximum of 15 treatment hours is more effective than exercise treatment alone for patients with nonspecific CLBP. Special interest is in long-term outcomes (\>1 year). # Review question Is a low-dosed primary care biopsychosocial intervention (PCBI), consisting of an active physical component and at least one psychological, social, or occupational component, with a maximum of 15 treatment hours more effective in reducing pain intensity and improving physical function than other active outpatient physical treatment approaches for adult patients with nonspecific CLBP? # Inclusion criteria ## Participants Studies with adult participants (18 years or older) who suffer from nonspecific chronic low back pain are included. Nonspecific chronic low back pain is defined as pain below the costal margin and the gluteal region, with or without radiation to one or both legs, that persists for 12 weeks or more. Trials are excluded when the sample includes participants with acute or subacute LBP (unless subacute LBP subjects comprised 15% or less of the total study population (≥85% should be CLBP), or results of patients with CLBP are presented separately), participants with specific low back pain due to a known, specific pathology (i.e., stenosis, spondyloarthritis, ankylosing, fractures, infection, or spinal cord compression) or women suffering from pregnancy-related back pain. ### Intervention(s) This review will consider studies that evaluate low-dosed outpatient biopsychosocial interventions. Following previous systematic reviews, biopsychosocial is defined as a multicomponent intervention that includes an active component (exercise, physical activity, or physiotherapy) and at least one psychological, social, or occupational component. As many different types of exercise have shown to be beneficial for patients with CLBP, there is no clear consensus on a specific type of exercise. Therefore, we will include all different types of exercises. They must involve specific activities, postures, or movements with the aim of improving physical health in chronic low back pain patients. Among others, this can include muscle or core strengthening, flexibility and mobilisation exercises, aerobic exercises, Pilates, Yoga or other specific exercises. A low dosage is defined as a maximum of 15 face-to-face treatment hours. Unsupervised home exercise does not count toward treatment hours. Included are individually (one-to-one) and group-supervised (two or more participants) therapy sessions. The active component must be delivered face-to-face. Interventions must be delivered in primary care (referral by general practitioner, physiotherapist in local facilities, primary care practice in a hospital, other outpatient healthcare professions). Mono- and multidisciplinary delivered interventions will be included. The psychological, social, or occupational component(s) can have other types of delivery (telephone, web-based). These components aim to improve (knowledge of) cognitive, emotional, social, or lifestyle factors or coping responses. Examples are cognitive behavioral therapy (CBT), acceptance and commitment therapy, pain neuroscience education (PNE), motivation or self-regulation training. According to the CLBP treatment guidelines, first-line interventions that are not recommended and feature only passive therapies (e.g., acupuncture, electrotherapy, traction, massage), spinal surgery, or sole pharmacological treatment will be excluded. Studies with completely independent, online, or telephone-based interventions (all components not equal to face-to-face), or interventions delivered inpatient (e.g., in pain clinics or rehabilitation centres) will be excluded. Studies in an occupational setting will also be excluded. ### Comparator(s) This review will consider studies that compare low-dosed (\<15h) outpatient biopsychosocial interventions with physical treatment with an active component such as exercise, physical activity or usual physiotherapy treatment. The comparison can be delivered in any type of delivery mode (face-to-face, as home training, web-based, app, etc.). Excluded are control groups that are formed from a waiting-list, undergo a complete biopsychosocial intervention (as defined above), get invasive therapy, or receive a sole pharmacological intervention. ### Outcomes This review will consider studies that include clinical outcomes of physical functioning/disability and pain intensity. According to the recommendation for core outcomes in clinical trials with participants with nonspecific low back pain, the outcomes of interest are physical functioning (measured with the Oswestry Disability Index (ODI) version 2.1a or the Roland-Morris Disability Questionnaire (RMDQ) of 24-items) and pain intensity (measured with the numerical rating scale (NRS) or visual analogue scales (VAS)). Outcomes of secondary interest are HRQoL measured with SF-12, SF-36 (dimensions: physical function, general health, mental health) or 10-item PROMIS Global Health short form) and all adverse events. ## Types of studies We will only include randomised controlled trials (RCTs) published in full text in peer reviewed journals. We include trials from the journals’ inception time to December 31, 2021 that were published in English or German language and enrolled adults with CLBP in a biopsychosocial intervention. # Methods The proposed systematic review will be conducted in accordance with the JBI methodology for systematic reviews of effectiveness evidence. According to the guidelines, our systematic review protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO 2022) on 27 February 2022 (registration number CRD42022302771). We followed the Preferred Reporting Items for Systematic review and Meta-Analyses Protocols (PRISMA-P) guideline and the available checklist. ## Search strategy The search strategy will aim to locate published studies. In this review, a three-step search strategy will be used. First, an initial limited search of MEDLINE (Ovid) and CINAHL (EBSCO) will be undertaken to identify articles on the topic. The text words contained in the titles and abstracts of relevant articles, and the index terms used to describe the articles, will be used to develop a full search strategy for: - CINAHL. - Cochrane Central Register of Controlled Trials (CENTRAL). - Ovid Medline. - Physiotherapy Evidence Database (PEDro). - PubMed. - Web of Science. The databases will be additionally searched for relevant systematic reviews and meta-analyses. Titles, abstracts, key words, and reference lists will be scanned to refine search terms. The search strategy, including all identified keywords and index terms, will be adapted for each included database and/or information source. The reference list of all included sources of evidence will be screened for additional studies. A forward reference search of all included studies will be performed using the R package *citationchaser*. Studies published in English or German will be included. Studies published since their inception date will be included. ## Study selection Following the search, all identified citations will be collated and uploaded to Mendeley Desktop 1.19.8/2020 (Mendeley Ltd., London, UK) and duplicates will be removed. Following a pilot test, titles and abstracts will then be screened by two independent reviewers (MH and PR) for assessment against the inclusion and exclusion criteria of the review in PICO Portal. Potentially relevant studies will be retrieved in full and their citation details will be imported into the PICO Portal. The full text of the selected citations will be assessed in detail against the inclusion and exclusion criteria by two independent reviewers (MH and PR). Reasons for exclusion of full text articles that do not meet the inclusion criteria will be recorded and reported in the systematic review. Any disagreements that arise between the reviewers at each stage of the selection process will be resolved through discussion, or with at least one additional reviewer. The results of the search and the study inclusion process will be fully reported in the final systematic review and presented in a Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) flow diagram. ## Assessment of methodological quality Eligible studies will be critically appraised by two independent reviewers (MH and PR) at the study level using standardised critical appraisal instruments from JBI for experimental studies. The authors of papers will be contacted to request missing or additional data for clarification, where required. Any disagreements that arise will be resolved through discussion, or with a third reviewer. The results of critical appraisal will be reported in narrative form and in a table. Following critical appraisal, studies that do not meet a certain quality threshold will be excluded. This decision will be based on the JBI 13-item checklist for randomised controlled trials. Studies will be excluded if one or more of the following checklist questions is answered with no: - Was true randomization used for assignment of participants to treatment groups? (Question 1 / selection bias) - Were treatment groups similar at baseline? (Question 3 / selection bias) - Were treatment groups treated identically other than the intervention of interest? (Question 7 /performance bias) - Were outcomes measured in the same way for treatment groups? (Question 10 / measurement bias) The methodological quality will be assessed into the three categories: Low risk of bias (\> = 9 questions answered with yes), some concerns (\> = 6 questions answered with yes) or high risk of bias (\<6 questions answered with yes). ## Data extraction Data will be extracted from studies included in the review by an independent reviewer (MH) using the data extraction module of the PICO Portal. The data extracted will be checked by a second review author (PR). Any disagreements that arise between the reviewers will be resolved through discussion or with a third reviewer. The data extracted will include specific details on (1) study characteristics (number of participants, age, sex, length of follow-up), (2) type of interventions (delivery, dosage, content), (3) baseline and follow-up patient- reported outcome measures of significance to assess pain intensity and disability, and (4) summary of findings. In case there are multiple publications regarding one RCT, all available publications will be checked and relevant data extracted. To avoid double counting that would bias the meta-analysis, the outcome measure data for eligible interventions will be selected based on the follow-up time. In case of reporting the same outcomes of the same RCT in different reports, the report with the higher methodological quality will be selected. The authors of papers will be contacted to request missing or additional data, when required. In cases where authors are uncontactable, a web- based tool will be used to extract raw data from the graphs. Any disagreements that arise between the reviewers will be resolved through discussion or with a third reviewer. ## Data synthesis A descriptive synthesis of outpatient biopsychosocial interventions with active treatment as a comparator will be performed. Studies will, where possible, be pooled in a statistical meta-analysis using RevMan Web from the Cochrane Collaboration. Effect sizes will be expressed as standardised mean differences (SMD), and their 95% confidence intervals will be calculated for analysis. Statistical analyses will be performed with a random effects model using the *DerSimonian and Laird method*. This model not only accounts for the statistical heterogeneity of the included studies, but allows for generalisation beyond these studies. Subgroup analyses will be performed when there is sufficient data to investigate different types (individual versus group setting) or length of follow-up (short, medium, long-term) of biopsychosocial interventions. After completion of the review and based on the findings of the included studies, we may perform additional subgroup analyses if they are believed to contribute valuable information. Following the meta-analysis, a sensitivity analysis will be performed to assess the impact of risk of bias on the results. In the sensitivity analysis, studies classified as having a high risk of bias will be omitted. Heterogeneity between the studies will be evaluated graphically with forest plots and funnel plots (if there are 10 or more studies included in the meta- analysis) and statistically with the chi-squared and I squared tests. As the effect sizes will be expressed in SMDs, statistical tests for funnel plot asymmetry will not be performed. The level of heterogeneity will be interpreted according to the latest version of the Cochrane Handbook for Systematic Reviews of Interventions, which provides the following guidance for the I2 statistic: ’0% to 40%: might not be important; 30% to 60%: may represent moderate heterogeneity; 50% to 90%: may represent substantial heterogeneity; and 75% to 100%: considerable heterogeneity. Should meta-analysis due to clinical and methodological heterogeneity not be possible, a narrative synthesis will be performed. The findings will then be presented in narrative form, including tables and figures to aid in data presentation, where appropriate. ## Assessing certainty in the findings The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach for grading the certainty of evidence will be followed and a Summary of Findings (SoF) will be created using GRADEpro GDT (McMaster University and Evidence Prime, ON, Canada). This will be carried out by two independent reviewers (MH and PR) at the outcome level. Any disagreements that arise between the reviewers will be resolved through discussion or with a third reviewer. The authors of papers will be contacted to request missing or additional data for clarification, where required. The SoF will present the following information, where appropriate: Patients or population, setting, intervention, comparison, number of participants, anticipated absolute effects for treatment and control, and a ranking of the quality of the evidence based on the risk of bias, directness, heterogeneity, precision, and risk of publication bias of the review results. The outcomes reported in the SoF will be pain intensity and functional limitations. # Discussion To our knowledge, this will be the first systematic review and meta-analysis of narrowly defined low-dosed PCBI in populations with nonspecific chronic low back pain. Due to the prevalence of CLBP throughout the world and its huge economic burden on individuals and health systems, we believe that this research will be interesting for decision makers worldwide. Also, an impact on treatment decisions is expected. With the focus on long-term outcomes, we take into account that low back pain symptoms show a similar pattern of improvement in the short term in a wide range of interventions. Using not only pain intensity, but also disability and HRQoL as outcomes of interest, we further take the nature of CLBP as a chronic condition into account. Our aim is to evaluate whether the outcomes can be long- lastingly improved to restore daily functioning as much as possible. -In doing this, our aim is to assess whether low-dosed PCBI can improve CLBP treatment outcomes. Depending on the results, the treatment and access options of further health programmes could be improved for a wide audience. If low-dosed biopsychosocial interventions are more effective than active treatments alone, our objective is to further identify which combination of biopsychosocial ingredients delivers the best outcomes to make effective treatment procedures accessible to a broad population at reduced costs due to minimised treatment time. Due to the worldwide burden of CLBP, we expect the results of this systematic review to be of considerable interest to health care professionals, academics, guideline developers, and decision makers. We will widely distribute the findings through academic publications, conference presentations, and communication with healthcare providers. # Supporting information This systematic review is to count towards the dissertation of MH through Hannover Medical School (MHH). ***Abbreviation*** ***Explanation*** CLBP Nonspecific chronic low back pain GRADE Grading of Recommendations, Assessment, Development and Evaluation HRQoL Health-related quality of life LBP Low back pain MBR Multidisciplinary biopsychosocial rehabilitation NRS Numerical rating scale ODI Oswestry Disability Index PCBI Primary care biopsychosocial intervention PRISMA Preferred Reporting Items for Systematic Reviews and Meta-analyses RCT Randomised controlled trials RMDQ Roland-Morris Disability Questionnaire SF-12 Short Form 12 SF-36 Short Form 36 SMD Standardised mean differences SoF Summary of Findings VAS Visual analogue scales 10.1371/journal.pone.0273983.r001 Decision Letter 0 Özden Fatih Academic Editor 2022 Fatih Özden This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 4 Aug 2022 PONE-D-22-15969The effectiveness of low-dosed outpatient biopsychosocial interventions compared to active physical interventions on pain and disability in adults with nonspecific chronic low back pain: a protocol for a systematic review with meta-analysis PLOS ONE Dear Dr. Hochheim, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Both reviewers has now completed their suggestions. Please find these comments below. Your paper should be revised and resubmitted. Please submit your revised manuscript by Sep 18 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Fatih Özden, PhD Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: [http://journals.plos.org/plosone/s/supporting- information. ](http://journals.plos.org/plosone/s/supporting-information. ) \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Does the manuscript provide a valid rationale for the proposed study, with clearly identified and justified research questions? The research question outlined is expected to address a valid academic problem or topic and contribute to the base of knowledge in the field. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Is the protocol technically sound and planned in a manner that will lead to a meaningful outcome and allow testing the stated hypotheses? The manuscript should describe the methods in sufficient detail to prevent undisclosed flexibility in the experimental procedure or analysis pipeline, including sufficient outcome-neutral conditions (e.g. necessary controls, absence of floor or ceiling effects) to test the proposed hypotheses and a statistical power analysis where applicable. As there may be aspects of the methodology and analysis which can only be refined once the work is undertaken, authors should outline potential assumptions and explicitly describe what aspects of the proposed analyses, if any, are exploratory. Reviewer \#1: Partly Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 3\. Is the methodology feasible and described in sufficient detail to allow the work to be replicable? Descriptions of methods and materials in the protocol should be reported in sufficient detail for another researcher to reproduce all experiments and analyses. The protocol should describe the appropriate controls, sample size calculations, and replication needed to ensure that the data are robust and reproducible. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors described where all data underlying the findings will be made available when the study is complete? The [PLOS Data policy](https://journals.plos.org/plosone/s/materials-and- software-sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception, at the time of publication. The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above and, if applicable, provide comments about issues authors must address before this protocol can be accepted for publication. You may also include additional comments for the author, including concerns about research or publication ethics. You may also provide optional suggestions and comments to authors that they might find helpful in planning their study. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This study protocol aimed to investigate whether a low-dose (and therefore less costly) PCBI with a maximum of 15 treatment hours is more effective than exercise treatment alone for patients with nonspecific CLBP. The strength of this study was to conduct analysis using a rigorous method of systematic reviews. However, there were some concerns in this study. First, in the Abstract and the Types of studies, the authors had better revise the search date to July 2022. I consider the latest search date is better for a comprehensive search. Second, in the Abstract and the Search strategy, they had better add Embase as one of the databases they will search. Cochrane Handbook for Systematic Reviews of Interventions Version 6.3, 2022 has recommended the search for CENTRAL, MEDLINE, and Embase in Cochrane reviews in the section 4.3.1.1 Introduction to bibliographic databases (Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA (editors). Cochrane Handbook for Systematic Reviews of Interventions version 6.3 (updated February 2022). Cochrane, 2022. Available from www.training.[cochrane.org/handbook](http://cochrane.org/handbook).). Third, in the Abstract and the Search strategy, they had better add ClinicalTrials.gov and WHO ICTRP as the databases they will search. These databases are important to search ongoing studies. Fourth, they had better add “all adverse events” to outcomes. Fifth, they had better revise the search strategies. I consider that the standard search strategy for systematic reviews is “Participants” AND “Interventions” AND the RCT filter. The standard search strategy includes two “AND”. However, the search strategy in S1 Appendix includes four “AND” in \#34 in Medline (Ovid). The search strategy may have the possibility to miss eligible studies by using too many “AND”. Therefore, I recommend they consult an information specialist, a search coordinator, or a librarian to develop search strategies for all databases to conduct a comprehensive search. Reviewer \#2: The age range considered is from 18 years where chronic LBP might not be relevant also the exercises program are not defined well. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Masahiro Banno, MD, PhD Reviewer \#2: **Yes: **Dr. PREM KUMAR BHOJARA \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0273983.r002 Author response to Decision Letter 0 14 Aug 2022 1\. First, in the Abstract and the Types of studies, the authors had better revise the search date to July 2022. I consider the latest search date is better for a comprehensive search. Dear Dr. Banno, Thank you very much for your comments. We really appreciate that you took the time and effort to critically read our protocol. We completely agree with you, that a Systematic Review (SR) should include the latest search date possible. We started this project at the beginning of November 2021. To be totally honest, we submitted this protocol in February to another journal. However, after three months of waiting time, we were informed, that they no longer publish protocols of SRs, which is why we resubmitted to PLOS One in a second step. As you might imagine, this was rather unfortunate for us, but we decided to continue to work on the SR, as we believed in the strength and accuracy of our project. By now, the final SR has progressed quite far, which is why we are rather reluctant to change the search date now. Second, in the Abstract and the Search strategy, they had better add Embase as one of the databases they will search. Cochrane Handbook for Systematic Reviews of Interventions Version 6.3, 2022 has recommended the search for CENTRAL, MEDLINE, and Embase in Cochrane reviews in the section 4.3.1.1 Introduction to bibliographic databases (Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA (editors). Cochrane Handbook for Systematic Reviews of Interventions version 6.3 (updated February 2022). Cochrane, 2022. Available from www.training.[cochrane.org/handbook](http://cochrane.org/handbook).). Thank you very much for your suggestion. We would like to answer in detail of why we did not include Embase. First, as you might have noticed, we follow the JBI methodology for systematic reviews of effectiveness evidence and not the Cochrane Handbook. In JBIs guideline Embase is listed as a potential not an obligatory source. Second, a more practical reason: in contrast to most other sources, Embase has a restricted access. Unfortunately, we did not have access to Embase through our institution. To compensate for the deficit of non-existent access, we included the Cochrane Central Register of Controlled Trials (CENTRAL). Luckily, CENTRAL records are taken from Embase, which is why we believed that this information source is included at least indirectly. Third, in the Abstract and the Search strategy, they had better add ClinicalTrials.gov and WHO ICTRP as the databases they will search. These databases are important to search ongoing studies. Thank you very much for your suggestion. The focus of this review was to find published studies on this topic, which is why we did not add ClinicalTrials and WHO ICTRP explicitly. We are aware that we may not be able to identify a risk of publication bias in the meta-analysis in focusing on published studies only. However, as we included CENTRAL, we include ClinialTrials and WHO ICTRP anyways, as those records are also present in CENTRAL. Fourth, they had better add “all adverse events” to outcomes. Thank you for this suggestion. We added the outcome. Fifth, they had better revise the search strategies. I consider that the standard search strategy for systematic reviews is “Participants” AND “Interventions” AND the RCT filter. The standard search strategy includes two “AND”. However, the search strategy in S1 Appendix includes four “AND” in \#34 in Medline (Ovid). The search strategy may have the possibility to miss eligible studies by using too many “AND”. Therefore, I recommend they consult an information specialist, a search coordinator, or a librarian to develop search strategies for all databases to conduct a comprehensive search. Thank you very much for this comment. We thought about your suggestion and your concerns for a long time. However, we decided against changing the search term at this stage. This is for multiple reasons. We tested and optimized the search strategy for several months. In this process, we spoke to experienced SR authors, followed search strategy guidelines 1,2 and analysed the search strategy of similar articles very closely. You are absolutely right that most SRs use three sets of terms: population/health condition, intervention, and study type. Amorataris and Riitano have also suggested that outcomes can be included as well. But the norm does not mean that this has to be like this all the time. As explained in the Cochrane Handbook, reviews with complex interventions can deviate. Among others, the handbook suggests to break a concept into two or more subconcepts, use iterative searches, and use citation searching on key articles. We believe that a (true) biopsychosocial intervention with a combination of physical, psychological, or social components is a rather complex intervention allowing us to deviate from the strict three-term logic. We would like to take the time to explain our search strategy in detail. \# 1 to 5 focuses on the Population \# 6 to 14 focuses on the physical part of the intervention. \# 26 to 32 focuses on the psychosocial part of the intervention. \# 16 to 22 restricts the search on primary care. \# 23 to 25 restricts on study type (RCT). Therefore, we have four 'AND' as we divide the intervention concept into two parts and restrict our search for primary care. A restriction on primary care has already been seen in similar reviews (for example: van Erp, R.M.A., Huijnen, I.P.J., Jakobs, M.L.G., Kleijnen, J. and Smeets, R.J.E.M. (2019), Effectiveness of Primary Care Interventions Using a Biopsychosocial Approach in Chronic Low Back Pain: A Systematic Review. Pain Pract, 19: 224-241. <https://doi.org/10.1111/papr.12735>). Arguably, we could have combined the physical and psychosocial part of the intervention in one term. We chose not to, to reach a reasonable precision in our results. As you might have seen, we included very broad concepts of physical and psychological types of intervention. As suggested by Cochrane, we tested our search strategy iteratively. A mere combination of the different parts resulted in a very high number of identified studies. From previous/similar reviews and hand-search, we identified key-articles a priori to check if they were identified by our search string. A reduction to two subconcepts still identified all key articles and reduced the number of identified studies significantly, as we included a stricter way of selection. As you described righteously, we could not be sure if we still had a comprehensive search strategy. This is why we included citation searching of key articles, as well as citation searching and forward searching of all identified articles. In doing so, we believe that we still have enough sensitivity in our search strategy without blowing up the results unreasonably. As mentioned above, we have rather progressed in the SR by now. Of course, a SR is only as good as the included literature. To erase remaining doubts that arose after your comment, we validated our search strategy. We validated our strategy with the strategy of van Erp et al (as cited above). As they had a very similar research question, we used their strategy and compared the identified results with our results. As they used two “ANDs” and had a broader definition of the intervention, we expected additional search results. Indeed, as an addition to our 2.132 records after removal of duplicates, we found 411 more records on PubMed. We selected all of them and found that none of those additional articles met our scope or inclusion criteria. These are the reasons, why we would like to stick to the search strategy as we are certain that we were able to identify all relevant articles. Reviewer \#2: The age range considered is from 18 years where chronic LBP might not be relevant also the exercises program are not defined well. Dear Dr. Bhojara, Thank you very much for taking the time and effort to read our protocol and the constructive feedback you gave us. It is true that the prevalence of CLBP increases with age. Also, many interventions target participants between 30 and 60 years of age. However, CLPB is also seen in younger individuals (18 to 29 years). Estimates range from a prevalence of 2 to 12%3. As CLBP is prevalent in younger individuals as well, it could be possible that studies include patients from the age of 18 onwards. As we already have quite strict inclusion criteria, we did not want to restrict further based on age. Thank you for your suggestion to explain exercises better. We updated this section in the manuscript. 10.1371/journal.pone.0273983.r003 Decision Letter 1 Özden Fatih Academic Editor 2022 Fatih Özden This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 19 Aug 2022 The effectiveness of low-dosed outpatient biopsychosocial interventions compared to active physical interventions on pain and disability in adults with nonspecific chronic low back pain: a protocol for a systematic review with meta- analysis PONE-D-22-15969R1 Dear Dr. Hochheim, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, Fatih Özden, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Does the manuscript provide a valid rationale for the proposed study, with clearly identified and justified research questions? The research question outlined is expected to address a valid academic problem or topic and contribute to the base of knowledge in the field. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 2\. Is the protocol technically sound and planned in a manner that will lead to a meaningful outcome and allow testing the stated hypotheses? The manuscript should describe the methods in sufficient detail to prevent undisclosed flexibility in the experimental procedure or analysis pipeline, including sufficient outcome-neutral conditions (e.g. necessary controls, absence of floor or ceiling effects) to test the proposed hypotheses and a statistical power analysis where applicable. As there may be aspects of the methodology and analysis which can only be refined once the work is undertaken, authors should outline potential assumptions and explicitly describe what aspects of the proposed analyses, if any, are exploratory. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Is the methodology feasible and described in sufficient detail to allow the work to be replicable? Descriptions of methods and materials in the protocol should be reported in sufficient detail for another researcher to reproduce all experiments and analyses. The protocol should describe the appropriate controls, sample size calculations, and replication needed to ensure that the data are robust and reproducible. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors described where all data underlying the findings will be made available when the study is complete? The [PLOS Data policy](https://journals.plos.org/plosone/s/materials-and- software-sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception, at the time of publication. The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above and, if applicable, provide comments about issues authors must address before this protocol can be accepted for publication. You may also include additional comments for the author, including concerns about research or publication ethics. You may also provide optional suggestions and comments to authors that they might find helpful in planning their study. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors have revised the manuscript thoroughly. I have no additional comments about the manuscript. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Masahiro Banno, MD, PhD \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0273983.r004 Acceptance letter Özden Fatih Academic Editor 2022 Fatih Özden This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 23 Aug 2022 PONE-D-22-15969R1 The effectiveness of low-dosed outpatient biopsychosocial interventions compared to active physical interventions on pain and disability in adults with nonspecific chronic low back pain: a protocol for a systematic review with meta- analysis Dear Dr. Hochheim: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Fatih Özden Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Patients with end-stage renal disease (ESRD) show much lower survival rate compared to general population of the same age. Peritoneal dialysis (PD) is an established treatment modality for patients with ESRD. In several retrospective and prospective cohort studies, predictors of outcome in patients treated with PD have been investigated. The survival of patients on PD is associated with various clinical factors such as demographic findings, comorbidities, nutritional markers, and peritoneal clearance. For each 1-year increase in age, the risk of death increases by 4%, and patients with diabetes have a 30% increased risk of death compared with non-diabetic patients. Cardiovascular disease accounts for most deaths in dialysis patients (approximately 50%). In addition, malnutrition is a negative predictor of survival in patients with ESRD. Fluid status and volume homeostasis are important in patients on PD, and a number of studies have shown that fluid overload is prevalent in patients on PD. Inadequate fluid removal in this population contributes to hypertension and is associated with an increased risk of cardiovascular disease, hypoalbuminemia, and systemic inflammation. However, this effect was found to depend mainly on the contribution of residual glomerular filtration rate (GFR). Regardless of how long the residual renal function can be maintained in patients on PD, patients will ultimately lose their residual renal function completely. Prevalent PD patients might have higher mortality rate when they become anuric during dialysis. The purpose of this study was to identify the factors that predict survival in anuric patients on PD. # Materials and methods ## Patients and data collection We conducted a prospective observational cohort study on Korean dialysis patients at the Clinical Research Center for End-Stage Renal Disease (CRC for ERSD, NCT00931970). Patients who were at least 20 years old and who began treatment with maintenance PD because of ESRD within 3 months were included for the study. Patients scheduled to receive kidney transplantation within 3 months were excluded. Between September 2008 and June 2011, 505 anuric patients on continuous ambulatory peritoneal dialysis (CAPD) were recruited into the study with the mean follow-up duration of 21.4±7.9 months. Anuria was defined as a 24-h urine output lower than 100 mL. Demographic variables at enrollment included age, gender, duration of dialysis, primary renal disease, comorbidities, laboratory data, and dialysis information. Comorbid conditions included chronic lung disease, coronary artery disease, peripheral vascular disease, cerebrovascular disease, diabetes mellitus, congestive heart failure, arrhythmia, connective tissue disease, peptic ulcer disease, mild liver disease, and moderate to severe chronic liver disease. Laboratory data were available for hemoglobin, serum blood urea nitrogen (BUN), creatinine, albumin, total cholesterol, triglyceride, ferritin, and high-sensitivity C-reactive protein (hs-CRP). PD information included membrane transport status, UF volume, weekly dialysate Kt/Vurea, body mass index (BMI), and blood pressure. The erythropoiesis-stimulating agents used in this study were epoetin-α and darbepoetin, using a conventional ratio of 200:1 to convert from epoetin-α to darbepoetin. Comorbidities, laboratory data, and dialysis information were followed at 3 and 6 months after the start of renal replacement therapy, and then at 6-month intervals thereafter. The present study used the data at the recruitment period of anuric PD patients. Patient’s death, a primary end point of the study, was reported within 1 month after the event and ascertained by data from Statistics Korea. The study had local ethical committee clearance at all sites. Written informed consent was obtained from all studied subjects before inclusion. The Institutional Review Board of Kyungpook National University Hospital approved the study protocol, and all clinical investigations were conducted in accordance with the guidelines of the 2008 Declaration of Helsinki. ## Statistical analysis Results of continuous variables were expressed as mean ± standard deviation (SD) for normal distribution. For categorical variables, the results were expressed as frequencies and percentages. Comparisons of continuous variables and categorical data between the two study groups were tested by the *t*-test and Chi-square test. To increase the power of the statistical analysis, we constructed a propensity score-matching model. The model included age, comorbid diabetes at baseline, and dialysis vintage. Propensity score matching was conducted using a 2:1 matching procedure, whereby each non-survived anuric patient on PD was matched with two survived anuric patients on PD. The predictors for survival were analyzed using a multivariate Cox’s proportional-hazard model in which all the significant variables from the univariate analysis were included. Proportional hazard assumptions were tested using Schoenfeld residuals. Survival curves were generated by the Kaplan-Meier method. For the comparison of two scales as predictors for hypoalbuminemia, a receiver operating characteristic (ROC) curves analysis was performed. We performed subgroup analyses in patients categorized by age (\<50 years, ≥50 years), gender, diabetes, dialysis vintage (\<5 years, ≥5 years), UF volume (\<1000 mL/day, ≥1000 mL/day), hemoglobin (\<10 g/dL, ≥10 g/dL), serum creatinine (\<10 mg/dL, ≥10 mg/dL), total cholesterol (\<177.5 mg/dL, ≥177.5 mg/dL), ferritin (\<100 mg/dL, ≥100 mg/dL), hs-CRP (\<0.1 mg/dL, ≥0.1 mg/dL). All statistical analyses were performed using SAS system for Windows, version 9.2 (SAS Institute Inc., Cary, NC). Differences were regarded as statistically significant when *p* \< 0.05. # Results Demographic data of the study population are shown in. Among the 505 patients on CAPD in the study, the mean age was 50.29±12.94 years, and the duration of PD was 71.50±50.34 months. Diabetes was the most common cause of ESRD (39.56%), followed by hypertension (28.40%), glomerulonephritis (25.24%), and others (6.80%). By the end of follow-up, 61 (12.07%) patients had died, and 444 (87.93%) patients were alive. Compared to non-survived patients, survived patients had a lower number of risk factors including older age (49.04±12.66 vs. 59.44±11.23, p\<0.0001), diabetes as the cause of renal failure (35.65% vs. 66.04%, p\<0.0001), and cardiovascular diseases such as atrial fibrillation and coronary artery disease (p\<0.05). In addition, survived patients had significantly higher serum creatinine (11.40±7.23 vs. 9.90±3.17, p = 0.0053) and serum albumin (3.75±1.62 vs. 3.44±0.45, p = 0.0016) levels than non-survived patients. Univariate Cox regression analysis showed an increasing hazard ratio (HR) for patient survival with increasing patient age, decreasing duration of dialysis, and the presence of accompanying coronary artery disease or diabetes. Additionally, as UF volume, serum creatinine, and serum albumin increased, the HR decreased. Multivariate analysis conducted for these data showed increased HR with significant differences according to patient age (HR = 1.049, p = 0.0034) and diabetes (HR = 2.144, p = 0.0421), and with a decreased HR according to dialysis vintage (HR = 0.976, p = 0.0393) and serum albumin (HR = 0.471, p = 0.0480). Propensity score analysis was conducted with ratio of 2:1 of survived patients to non-survived patients. Dialysis vintage, age at entry, and diabetes were used as variables for adjustment in the propensity score matching. The analysis was performed in 61 non-survived patients and 122 propensity score-matched survived patients. After propensity score matching, survived patients showed higher serum albumin (3.63±0.46 vs. 3.44±0.45, p = 0.0111) and total cholesterol (184.79±45.83 vs. 169.90±41.81, p = 0.0390) levels with significant differences. However, other variables did not show any significant differences between the two groups. In addition, the univariate Cox regression analysis for patient survival showed a decreasing risk with increasing dialysis vintage (HR = 0.982, p = 0.0477), serum albumin (HR = 0.444, p = 0.0083) and total cholesterol (HR = 0.993, p = 0.0371), while no significant differences were observed for the other variables. Multivariate analysis showed a decreasing risk with increasing dialysis vintage (HR = 0.906, p\<0.0001) and serum albumin (HR = 0.347, p = 0.0094), though total cholesterol level showed no significant difference between the two groups (HR = 0.992, p = 0.0600). Analysis using the ROC curve for predicting death according to serum albumin level showed that death could be predicted with a sensitivity of 59.4% and a specificity of 63.2% using a cutoff serum albumin value of 3.6 g/dL. The areas under the curves of serum albumin for predicting survival was 0.658, demonstrating that serum albumin level could be considered as a fair predictor of survival.. When the serum albumin level of 3.6 g/dL was set as the standard, significant differences in the survival rate were observed in unmatched and matched groups (p = 0.0005 and p = 00297). The positive association between high albumin level and survival was significantly greater for patients with older age (≥50 years), no diabetes, low UF volume (\<1000 mL/day), and low levels of serum creatinine (\<10 mg/dL), total cholesterol (\<177.5 mg/dL), ferritin (\<100 ng/mL), and hs-CRP (\<0.1 mg/dL). In the context of albumin levels greater than 3.6 g/dL, a significantly increased survival rate was observed regardless of the sex, duration of dialysis, or hemoglobin level of the patients. # Discussion In this study, based on the results of multicenter prospective observation studies on dialysis patients in Korea, we implemented analyses of the factors that might influence prognosis in anuric patients on PD. Using propensity score matching, analyses with a ratio of 2:1 of survived patients to non-survived patients were conducted according to the differences in patient age, duration of dialysis, and the presence of diabetes. Serum albumin level was found to be an independent predictor for survival in anuric patients on PD. This is the largest prospective study ever published which investigated survival predictors in anuric PD patients. The increasing dialysis vintage is associated with declining body weight and deterioration of nutritional status, which may affect survival of patients on PD. Because the present study included only the anuric PD patients at the recruitment, the exact duration of dialysis until anuria could not be evaluated. It is difficult to draw a definite conclusion from the univariate and multivariate analysis with duration of dialysis. Nevertheless, the duration of dialysis could be correlated with duration on dialysis of patients when they became anuric, especially when the duration is short. Therefore, relationship between dialysis duration and survival shown in our study might suggest that faster the patient becomes anuric, more likely the patient would not survive long. This has been reported in many studies suggesting preserving residual renal function is an important factor for longer survival in PD patients. Nutritional markers are known to be as an important factor for survival in patients on PD, and among them, serum albumin concentration is often used as a surrogate measure of malnutrition. However, the utility of albumin as a marker of malnutrition in PD has remained controversial. The association between other measures of nutrition and albumin concentration is inconsistent, which may be due to multiple factors. Albumin is a negative acute-phase reactant and its production is suppressed in inflammatory states. In addition to inflammation, peritoneal albumin loss and volume overloading can lead to hypoalbuminemia in patients on PD. Although serum albumin concentration is not an effective isolated marker of nutritional status because of its multiple confounding factors, our study showed that serum albumin level may be considered an important and meaningful factor to facilitate the prediction of prognosis in anuric patients on PD. Jansen et al. also reported that several risk factors such as low serum albumin as well as old age, comorbidity, and dialysis vintage, were significantly associated with worse patient survival. However, they studied with relatively small number of anuric patients, to which comparison with propensity matching could not be applied. Moreover, we found that serum albumin concentration may differently affect patient survival according to age, status of diabetes, UF volume, serum creatinine level, total cholesterol level, ferritin level, and hs-CRP level. To ascertain the cause of the varying effects of hypoalbuminemia, further studies that elucidate the relationship between albumin and these factors are warranted. Increased serum creatinine level has been associated with good survival, whereas lower serum creatinine level has been associated with increased mortality in hemodialysis patients. These findings suggest that serum creatinine level reflects muscle mass and that low muscle mass resulting from malnutrition is associated with poor outcomes. In a previous study, the association between serum creatinine level and survival was observed in a retrospective PD cohort. However, that previous study has a limitation in that residual renal function, which can directly affect serum creatinine level, was unable to be examined. Residual renal function may be better preserved and may contribute more significantly to total creatinine clearance, particularly in patients with a shorter PD duration. Given that residual GFR is an important predictor of survival, the association between serum creatinine and survival may be modified by residual GFR. Since our present study targeted anuric patients with PD, it was possible to exclude the influence of residual renal function on serum creatinine. In the unmatched cohort analysis, survived patients showed increased serum creatinine and a statistically significant difference in the survival rate, which increased with increasing serum creatinine levels and was confirmed in the Kaplan-Meier survival curves. These findings are in accordance with the results of previous studies. However, no significant difference in serum creatinine level was observed between the two groups in the propensity score analysis. This implies that serum creatinine may be used as a marker of nutritional status in patients with ESRD, but that it is not appropriate for predicting survival, especially anuric PD patients. Further studies are needed to verify the association between survival rate and serum creatinine level in anuric patients on PD. The optimal UF volume of anuric patients on PD remains unknown. The European Best Practice Guideline Working Group on PD recommends the minimum net UF of 1 L/day in anuric patients on PD. However, the International Society for Peritoneal Dialysis states that no definite target for UF can be suggested from the previous studies. In the same context, our study revealed no significant difference in survival rate according to the UF volume was observed in the propensity score analysis. While current guidelines do not adequately address the optimal UF target for anuric patients on PD, optimizing therapy to meet UF goals remains an important clinical target. Bhaskaran et al. performed a retrospective analysis of anuric peritoneal dialysis patients in Canada and were unable to find a significant effect of Kt/Vurea on the relative risk of death. However, multivariate analysis of a prospective cohort study of anuric patients in Hong Kong showed a significant effect of Kt/Vurea on survival advantage. Maarten et al. showed that peritoneal UF rate, Kt/Vurea \<1.5/week, and creatinine clearance \<40 L/week/1.73 m² were associated with an increase in the relative risk of death. Furthermore, they suggested that a Kt/Vurea of 1.7/week and a creatinine clearance of at least 45 L/week were reasonable targets, which is in line with the results of the ADEMEX study in which a further increase in these solute transport levels was not found to lead to better patient survival. The Kidney Disease Outcomes Quality Initiative (KDOQI) provides specific guidelines for PD adequacy since higher small solute clearance is associated with improved survival on PD. According to our results, variations in peritoneal small solute clearances did not correspond to improved survival. Further studies that assess factors other than small-solute clearances are necessary to determine their effects on survival. Numerous investigations have recently monitored certain biomarkers such as cardiac troponins and hs-CRP in addition to clinical data such as diabetes and hypertension in order to more clearly identify patients with coronary artery disease and enable the risk stratification of patients with ESRD. Several studies have shown that the elevated hs-CRP is related with cardiovascular disease and patient’s death, whereas other studies have reported no association between hs-CRP and survival. In the present study, significant differences in the prognosis of patients between the two groups according to blood pressure, presence of cardiomegaly in LV hypertrophy on EKG, and CXR were not observed, and similar results were shown in the propensity score analysis. In addition, in terms of hs-CRP level, there was no significant difference between the two groups, and a similar pattern was shown in the propensity score analysis. Previous study demonstrated the relationship between diminished residual renal function and increased level of CRP. However, there are several studies which failed to prove the correlation. Therefore, the question of whether hs-CRP is predictive of clinical outcomes in patients on PD irrespective of peritoneal clearance and residual renal function remains controversial. Further research is necessary to confirm the relevance of hs-CRP level in the prognosis of anuric patients on PD. However, this study showed that low hs-CRP levels decreased the hazard ratio for survival in anuric patients on PD, with higher serum albumin levels. In conclusion, our study showed that in the absence of residual renal function, serum albumin level is an independent factor that influences the survival of patients on PD. Additionally, patients with relatively older age, no diabetes, peritoneal UF of less than 1000 mL/day, and low serum creatinine, total cholesterol, serum ferritin, and hs-CRP levels, survival rate was found to increase along with increasing serum albumin level. The statistical significance of the results was fortified through propensity score matching that was conducted using prognosis factors gleaned from previous studies of patients on PD. However, further large-scale research is necessary to confirm the relevance of cardiovascular factors, nutritional prognostic factors, and peritoneal clearance in predicting the prognosis of anuric patients on PD. [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work.

# Introduction Head and neck cancer (HNC) refers to cancer of the nasal cavity and paranasal sinuses, the oral cavity, the salivary glands, pharynx and larynx. The worldwide incidence of HNC is some 650,000 cases per year; in Europe, this incidence reaches 140,000 cases, whereas in Spain HNC incidence is around 10,000 cases annually. It is more common in men than in women; the mean age at diagnosis is 50 years. The most common risk factors for HNC are tobacco use, alcohol consumption and infection with human papillomavirus. Treatment for HNC commonly involves surgery, radiotherapy (RT), chemotherapy, or a combination of these. Intraoperative procedures causing stretching, compression or burning (from electrocauterization) of soft and neural tissues can lead to neuropraxia or axonal injury, as can post-operative scarring, haemorrhages and infections potentially impairing motor function in the affected region and surrounding areas with an impact on oral and oropharyngeal functioning. When lymph node metastases are suspected, a neck dissection (ND) may be performed, either selective ND (SND), modified radical (MRND), or radical ND (RND) depending on requirements, in addition to the resection of the primary tumor surgery. SND, which involves cervical lymphadenectomy, preserves one or more of the lymph node groups that are usually removed in RND. In MRND, all the local lymph nodes are removed, but one or more non-lymphatic structures are preserved (e.g., the spinal accessory nerve, the internal jugular vein or the sternocleidomastoid muscle). RND involves the extirpation of all ipsilateral lymph node groups from the lower border of the mandible to the clavicle, as well as the removal of the spinal accessory nerve, the internal jugular vein and the sternocleidomastoid muscle. Radiotherapy may induce the formation of collagen, leading to a thickening of the dermis and, therefore, fibrosis, which may cause a loss of function of the masticatory system. Muscle contractures may also appear as a consequence of radiation-induced fibrosis in the neck and shoulder regions, which reduces their motor functioning. Adequate skeletal support and muscle function are essential to swallowing. If the oral cavity and pharyngeal muscles are damaged by surgery or radiotherapy, dysphagia, may result. The toxic effects of chemotherapy—which is commonly prescribed in the treatment of locally advanced disease may also hinder patient recovery and swallowing ability. Swallowing impairments related to the presence of trismus (maximum mouth opening \[MMO\] ≤35 mm) have been studied in survivors of head and neck cancer (sHNC), but it is unknown whether any associations exist between MMO, temporomandibular dysfunction (TMD), cervical and shoulder motor function, and swallowing impairments in sHNC. Therefore, the primary aim of this case-control study is to test if there are differences between sHNC and healthy age- and sex-matched controls for MMO, TMD, cervical and shoulder motor function, and swallowing impairments. The secondary aim is to analyse in sHNC the association between MMO, TMD, cervical and shoulder motor function, and swallowing impairments. # Methods ## Study subjects The sHNC were recruited at the Virgen de las Nieves University Hospital, Granada (Spain). All met the following inclusion criteria: 1) age ≥18 years, 2) curative treatment completed in the previous 6–36 months, 3) shoulder and/or cervical dysfunction present, and 4) a tumor located (before treatment) in the nasal cavity, either paranasal sinus, nasopharynx, oral cavity, oropharynx, hypopharynx, or larynx. The exclusion criteria were: 1) having a metastasis or active neoplasm, 2) previous neck and/or shoulder impairments, and 3) cognitive impairment. The control group was formed by healthy age-and sex matched volunteers who responded to announcements for the study. They were excluded if they reported a history of cervical, shoulder and/or TMJ pain, a history of trauma, or if they had any systemic disease. The present study was approved by the Biomedical Investigation Ethics Committee, Granada, Spain (CEi-GRANADA Ref: 0045-N-16) and conducted in accordance with the Declaration of Helsinki. All patients gave written informed consent before being formally enrolled. ## Subject demographic and clinical data Demographic (age, sex, tobacco and alcohol consumption) and clinical data (cancer location and tumor stage at diagnosis, time since diagnosis, affected side, kind of curative cancer treatment) were recorded at the appointment with the patient. Affected side was described as ipsilateral side and unaffected side as contralateral side. Data on smoking habit (non-smoker, smoker, or ex-smoker) and alcohol consumption (none, monthly, weekly and daily) were also recorded. ## Assessment During the assessment MMO, cervical active range of motion (AROM) and muscle strength, and shoulder AROM and muscle strength were measured by objective tests. TMD, shoulder pain and disability, and swallowing impairments were measured by questionnaires. Also a visual analogue scale (VAS) was used to measure the experience of swallowing difficulty. ## Maximum mouth opening MMO was measured (mm) once as the inter-incisor distance (with the patient sitting) using a sliding caliper. ## Temporomandibular dysfunction TMD was assessed using the Fonseca Anamnestic Index (FAI), a commonly employed screening tool for TMD. It consists of 10 self-administered questions with three possible responses: yes, sometimes or no. A total score of 0 reflects no TMD, 1 indicates mild dysfunction, 2 moderate dysfunction, and 3 severe dysfunction. ## Cervical function The cervical active range of motion (AROM) was measured by examining cervical flexion, extension, inclination and rotation (towards both sides) using a cervical range of motion device (Performance Attainment Associates©, Spine Products, Roseville, MN, USA). During testing patients sat in an upright position. Cervical muscle strength was determined using the deep cervical flexor endurance test (DCFET). Briefly, subjects started in a supine position with the examiner’s hands under their head. They then performed an upper-cervical extension (i.e., making a double chin), raising the head as little as possible from the examiner’s hands. The time elapsed from when the patient raised the head until 1) the adopted posture could no longer be maintained, 2) the patient’s head rested on the examiner’s hands for more than 1 s, or 3) the patient started to feel pain, was recorded. This test has an intraclass coefficient (ICC) of 0.82–0.91. ## Shoulder function Shoulder AROM was measured (both sides) using a two-arm goniometer with a 360° protractor examining shoulder flexion, abduction, external and internal rotation with the patient lying in the supine position. In each test, patients were instructed to move the joint from a neutral position to the end of their range (i.e., until pain or stiffness appeared), avoiding compensation movements. Each movement was examined once and the angle reached recorded. The ICC of the goniometer used is excellent (0.94). Shoulder muscle (upper trapezius) strength was measured using the Daniels and Worthingham’s muscle test scale, scoring from absence of contraction (0) to normal muscle response (5). Tests were again performed bilaterally. Briefly, patients sat in an upright position with both arms resting by the trunk, and pushed as strongly as possible with their shoulder towards the ceiling, against the examiner’s hand (with the score determined by the examiner) This test has an ICC of 0.63–0.98. The shoulder pain and disability index (SPADI) was determined for all subjects. The SPADI is a 13-item, self-administered questionnaire with two subscales: pain and disability. Subjects report the pain and disability experienced in the previous week. Each item is rated on a 0–10 scale, from no pain/no dysfunction (0) to maximum pain/impossible (10). The scores for pain and disability are calculated from the sum of the corresponding items divided by the maximum score possible and multiplied by 100. To obtain the total score, the mean of the pain and disability scores is calculated. The SPADI has been validated for use in populations with shoulder pain and has an ICC of ≥0.89. The SPADI Spanish version has been validated for general use, and has been used for populations with HNC. ## Swallowing impairments The Eating Assessment Tool (EAT-10) was used to evaluate self-reported swallowing impairments. This tool rates 10 items on a 5-point Likert scale from 0 (no impairment) to 4 (severe problem). The sum of the scores for these 10 items provides the overall score; impaired swallowing is reflected by a score of ≥3. The ICC for this test is 0.72–0.91 for a wide range of populations with swallowing disorders, including patients with HNC. Swallowing difficulty was also assessed using a 10 cm-long visual analogue scale (VAS) (0 = no difficulty, 10 = impossible to swallow) (10). Patients were also asked where they felt the problem existed when swallowing (pre-oral \[i.e., mouth opening\], oral, pharyngeal, or all three places). ## Statistical analysis Continuous data were expressed as means and standard deviations, ordinal and categorical data as numbers and percentages. The distribution of all variables was determined using the Kolmogorov-Smirnov test. Differences between sHNC and healthy age- and sex-matched controls in continuous and normal distributed data were analyzed with the independent T-test. The Mann-Whitney U test was used for ordinal and non-normal distributed continuous data. Differences on categorical variables between sHNC and healthy controls were analyzed by the Chi² test. Pearson’s correlation coefficient was determined to identify associations between normally distributed variables; Spearman’s correlation coefficient was determined when any variable in a pair was not normally distributed. Correlation analysis for cervical and shoulder AROM variables were performed using the mean values for the left and right sides. Significance was set at p\<0.05. All calculations were performed using SPSS 25.0 software (IBM Corp., Armonk, NY, USA). All p-values lower than 0.05 were considered statistically significant. Tests were performed with software SPSS 25.0 (Chicago, Illinois). # Results ## Demographic and clinical data Thirty-two sHNC, 12 women and 20 men, and 32 healthy age- and sex-matched controls were recruited; their mean age was 58.8±11.9 for the sHNC group and 58.4±12 for the control group. Eleven HNC and 19 controls did not smoke, 4 sHNC and 4 controls were smoker at the assessment time, and 17 sHNC were ex-smoker compared to 9 ex-smoker controls. No statistically significant differences were found for the smoking habits, but we did find statistically significant difference over alcohol consumption between groups (*p* = 0.004), with the sHNC group drinking less often than the control group. Twenty eight sHNC received surgery, 11 of whom also received radiotherapy, and 16 of whom also received chemoradiotherapy (CRT). Four received CRT only. Twenty two sHNC underwent ND. The mean time between diagnosis and assessment was 21.1±10.7 months. Tables and summarize the subjects' demographic and clinical characteristics. ## Maximum mouth opening MMO was statistically significant different between sHNC and healthy-matched controls (*p* = 0.002). On average sHNC reached 34.5mm (±13.3) which is scored as the presence of trismus, whereas the control group reached 44.1mm (±7.4). AMMO for 17 sHNC and 5 controls was scored equal or below 35mm, therefore this was scored as trismus presence. Fifteen sHNC and 27 controls scored the absence of trismus with a MMO at least 35mm. ## Temporomandibular dysfunction FAI was scored as no dysfunction for 12 sHNC and 24 controls, as light dysfunction for 7 sHNC and 8 controls; whereas 6 sHNC scored moderate dysfunction and 7 sHNC scored severe dysfunction. ## Cervical function sHNC revealed a significantly smaller cervical extension and inclination to both the affected and unaffected sides compared with controls (*p* \< 0.05), whereas there were no statistically significant differences for the cervical flexion (*p* = 0.561) neither for cervical rotation, affected (*p* = 0.077) and unaffected sides (*p* = 0.194). sHNC reached statistically significant lower times on the DCFET while performing than the control group (*p* \< 0.001). ## Shoulder function Flexion and abduction on the affected side of the shoulder was significantly lower in sHNC than controls (*p* \< 0.05), but there were no statistically significant differences for shoulder flexion and abduction on the unaffected side, neither for the external and internal rotation to both sides (*p* \> 0.05). Daniels and Worthingham’s muscle testing scale was scored below 5 for 12 sHNC whereas 1 control subject scored 4. Maximum score was reached for 20 sHNC and 31 controls. There was statistically significant difference between groups on the affected side (*p* = 0.008). SPADI scores for pain, disability as well as the total score differed significantly between both groups (*p* \< 0.001). Mean scores for sHNC were 30.1, 17.1 and 0.3 for the pain, disability and total subscales respectively, whereas 6.1, 2.7 and 0.04 were the scores reached on these subscales for the healthy controls. ## Swallowing function Patients reached a mean of 16.7 on the EAT-10 questionnaire, which showed a statistically significant difference (*p* \< 0.001) with the control group. The mean values from the VAS for the difficulty when swallowing were also statistically significant different between both groups (*p* \< 0.001). Swallowing difficulty differences were statistically significant (*p* \< 0.001) between sHNC and healthy controls. These swallowing difficulties were most reported by sHNC at the pharynx (40.6%), followed by the oral cavity (25%) and pre-oral (12.5%). Three patients (9.4%) reported disturbance at the 3 regions (pre-oral, oral and pharynx). ## Correlation between motor function and swallowing impairments shows the results for the measured variables. highlights the significant correlations (direct and inverse) detected. MMO correlated directly with the Daniels and Worthingham’s muscle test score on the affected side (p = 0.030), and inversely with the total SPADI and SPADI disability scores (p\<0.05). The FAI correlated inversely with cervical rotation (p = 0.016) and directly with both SPADI subscales, the total SPADI score, and the EAT-10 score (p\<0.05). Cervical flexion correlated directly with shoulder flexion and shoulder abduction (p\<0.05). Cervical inclination correlated directly with shoulder abduction and shoulder external rotation (p\<0.05), whereas cervical rotation showed a direct correlation with the Daniels and Worthingham’s muscle test score on the unaffected side (p = 0.003). The DCFET correlated directly with shoulder abduction (p = 0.003). The SPADI disability and pain indices correlated inversely with cervical extension and cervical inclination respectively (p\<0.05). Finally, the EAT-10 score correlated inversely with MMO (p\<0.001), cervical inclination (p = 0.002), cervical extension (p = 0.002) and the Daniels and Worthingham’s muscle test score on the affected side (p = 0.007), and directly with TMD (measured as FAI) (p = 0.045). No other correlations were found. # Discussion The aims of the current study were to study the correlation between MMO, TMD, motor function related to the cervical and shoulder regions and swallowing function in sHNC and to describe the differences on AMMO, TMD, cervical and shoulder functioning and swallowing function between sHNC and healthy matched controls. Our results showed that MMO, TMD, cervical and shoulder motor function, and swallowing are considerably affected in sHNC compared to healthy controls. Moreover, correlations between MMO, TMD, cervical and shoulder function and swallowing function were found in sHNC. ## Survivors of head and neck cancer compared to healthy controls MMO was found to be lower in patients compared to healthy controls. Trismus (restricted mouth opening) is a common complaint after treatment for HNC, appearing in about a quarter of all. Indeed, mouth opening decreases by some 20% after treatment, especially with RT —the consequence of damage and tissue fibrosis, possibly induced by apoptosis in response to radical‐mediated DNA damage. The lower mean values of the sHNC compared to the control group for the cervical and shoulder AROM variables obtained in the present work agree with reductions consistently reported for populations with HNC (including sHNC). Extension and inclination to both sides on the cervical region were lower in sHNC. Flexion and abduction of the shoulder on the affected side were significantly lower in sHNC. Indeed, prospective cohort studies have indicated reduced AROM for both the cervical and shoulder areas at one year and five years after treatment. This may be the result of surgical stretching, compression or tissue burning. Extensive surgery, such as RND, may reduce shoulder function via the sacrifice of the spinal accessory nerve. In addition, the fibrosis that appears in radiated regions can reduce the AROM. The loss of strength evidenced on this study in sHNC compared to healthy controls is also related to curative treatment: radiated-induced fibrosis reduces normal mobility of the regions affected; this may explain the decrease in the deep cervical flexor musculature. After resection or skeletonization of the accessory nerve, resection of the fascia surrounding the muscle or its devascularization, upper trapezius fibers may be no longer functional. The decreased AROM added to the loss of strength may lead to a subjective perception of pain and disability during daily life activities, as shown in this study with the SPADI questionnaire. These results are in concordance with the most common shoulder impairments previously reported after ND: pain, shoulder drop and loss of AROM. Swallowing function was decreased in sHNC compared to healthy controls. This finding is in accordance with previous studies that evidenced swallowing impairments in patients treated for head and neck cancer. RT causes harm to muscles involved in swallowing, leads inflammatory responses that induces fibrosis in time, atrophy, sensory loss, and thus, may result in dysphagia. In this study, the most common reported location for swallowing disturbance was the pharynx, but this may be due to the clinical characteristics of our sample, as 35% of the sHNC participating on this study presented their tumor at hypopharynx or larynx levels. ## Correlation between motor and swallowing functions Greater mobility and strength in the cervical region were found to be associated with the same in the shoulder. The cervical and shoulder regions are strongly connected via the origins and insertions of different muscles and by nerve branches involving the brachial plexus. Moreover, previous research had stated that a surgical procedure in the cervical region involving cervical nerve roots might result in upper-extremity motor dysfunction. The TMD (FAI) and SPADI results also correlated directly with each other; the greater the perception of TMD, the greater the shoulder pain and disability perceived. Although there may be less biomechanical interaction between the temporomandibular and shoulder regions than between the cervical and shoulder regions, this result suggests that, in sHNC, a relationship exists between loss of function in the former pair and the perception of pain and disability. The correlation between MMO and cervical function shows the importance of an optimal cervical AROM in adequate mouth opening (and perhaps vice versa). A reduced MMO in sHNC is common due to radiation-induced fibrosis of the masticatory and cervical muscles. It is known that the severity of TMD is related to the severity of cervical region disorders, although this has not been studied specifically in sHNC. An earlier prospective study reported the presence of TMD in patients diagnosed with HNC as possibly due to bruxism brought on by the anxiety and fear associated with a cancer diagnosis. This is the first study to show an association between swallowing function and cervical AROM and shoulder strength in sHNC; a poorer cervical AROM and reduced shoulder strength are related to an increased perception of swallowing impairment. A previous cross-sectional study reported a reduction in shoulder AROM in dependent older adults to be associated with dysphagia. No clear information exists regarding the relationship between swallowing and shoulder muscle function, but it may be that a loss of strength in the shoulder and cervical regions affects the position of the larynx and consequently its movement during swallowing. One cross-sectional study suggests that patients with head and neck cancer are at risk of reduced physical functioning due to their undertaking lower levels of physical activity; when physical effort is also reduced due to treatment, swallowing difficulties might be intensified. ## Strengths and limitations A strength of this study is its use of objective and subjective methods of assessment; this allowed objective data (e.g., MMO, AROM, DCFET results) to be correlated with subjects’ perception of their difficulties. The small sample size, however, did not allow for regression analyses to check for differences due to tumor location, tumor size or the curative treatment received. Finally, the present sHNC had completed their treatment in the previous 6 to 36 months. In future work with larger samples it might be advisable to stratify patients in terms of the time elapsed since treatment ended and include the association of oncology treatment parameters with different levels of disability. ## Clinical implications The present results suggest that physiotherapy to improve the MMO and cervical and shoulder AROM and strength in sHNC may help reduce swallowing difficulties. The associations detected between different body regions indicate treatment strategies may need to involve the face, cervical and shoulder regions. Mouth opening is mainly treated via the use of jaw-mobilizing devices and exercises once any radiotherapy is completed. The cervical region may be treated with stretching exercises and cervical traction (horizontal plane), while the shoulder should be subjected to passive and active range of motion exercises to prevent adhesive capsulitis. Both regions should be treated as soon as possible after any surgery. ## Future research Future studies should explore the correlations between the measured variables in larger samples of sHNC with different clinical characteristics. The results obtained may allow for clinical trials of specific treatments aimed at reducing swallowing difficulty. # Conclusion sHNC present lower MMO, higher perception of TMD, lower cervical and shoulder function which are inter-related to each other’s, besides greater swallowing impairments compared to healthy controls. The degree of swallowing impairment perceived by sHNC is associated with a lower MMO, higher perception of TMD, poorer cervical AROM (specifically cervical extension and inclination) and reduced shoulder strength. These impairments may be induced by the surgical procedure and the side effects of RT and chemotherapy. Physiotherapy might help improve these variables, reducing the perception of swallowing difficulty. The authors thank all individuals who participated in this study. We are also grateful to Mr Adrian Burton for assistance with the English language. This work was part of a PhD thesis conducted in the Clinical Medicine and Public Health Doctoral Studies of the University of Granada, Spain. [^1]: The authors have declared that no competing interests exist.

# Introduction Surveys of butterfly species have often been considered good surrogates for surveys of total biodiversity (e.g. in Malaysia). This is because of their role in food webs - caterpillars consume large quantities of plants and are themselves consumed by other animals in large numbers - and because, relative to most other animal groups, collecting and identifying adult butterflies is considered easy. This is particularly so in Peninsula Malaysia where butterflies have received intensive taxonomic attention. The “Butterflies of the Malay Peninsula” have been the subject of a series of comprehensive field guides, beginning with Distant in 1882–1886, and followed by four editions of Corbet and Pendlebury’s classic checklist, first published in 1934 and most recently revised by Eliot in 1992. Butterflies have benefitted and suffered from intensive taxonomic attention. In many cases a preponderance of names exists for the same species and names are often used incorrectly (see list of synonyms). During a recent survey of butterflies in Southern Thailand, 150 km north of the Malaysian border, fewer than 50% of the observed butterflies were identified to species. Adding to these difficulties is widespread but inconsistent use of butterfly subspecies names and concepts. Butterfly surveys in Peninsula Malaysia have not been consistent in using or ignoring subspecies names. This can make a big difference to biodiversity surveys - if we consider species as the biodiversity unit there are 793 units in Peninsula Malaysia, but if subspecies is considered the biodiversity unit, the number rises to 930. Butterfly trinomials have traditionally been used to recognize ‘moderate’ morphological differentiation correlated with disjunct geographical distributions. However, non-discrete morphological variation and the application to contiguously distributed populations, often make subspecies boundaries ambiguous. Following Tobias et al. ’s recommendations for avian subspecies delimitation, Braby et al. recently suggested standardized phenotypic criteria for subspecies delimitation in butterflies. Although considered desirable, Braby et al. refrained from setting criteria based on DNA characters, citing a lack of data. However, they did acknowledge that under their concept, subspecies are genetically distinct, but not reciprocally monophyletic according to mitochondrial DNA, noting that lineages possessing a diagnostic morphological character and also showing reciprocal monophyly are probably better regarded as distinct species. This criterion of concordance for species delimitation is in line with “state-of-the-art” practice in taxonomy i.e. the MTMC (Mitochondrial Tree Morphological Character congruence) of Miralles and Vences. Mitochondrial DNA barcodes – are increasingly being used as a supplementary taxonomic identification tool in surveys of Lepidoptera. However, DNA barcoders have often ignored subspecies names, –, and have used personalized alphanumeric codes for biodiversity units discovered below the traditionally recognized species boundary (e.g. *Hamadryas* feroniaECO01). These units used to account for previously overlooked (and possibly cryptic) diversity have come to be known as “dark taxa” and the correspondence between subspecies, recognized by morphological differentiation, and dark taxa is often difficult to resolve (e.g. does *H.* feroniaECO01 = *H. feronia farinulenta*?). Most GenBank and BOLD records do not include subspecies names, meaning it is impossible to tell if the authors of the DNA sequence could determine which subspecies the butterfly belonged to or not. It may be possible to narrow down subspecies identity based on locality, but locality is often missing, or imprecise, for GenBank records too. The aim of this study was to build a DNA barcode reference library for the true butterflies (species from the families – Papilionidae, Pieridae, Nymphalidae, Lycaenidae, Riodinidae) of Peninsula Malaysia from specimens in the Museum of Zoology, University of Malaya (UMKL) collection. We tested the capacity of the library to function as an accurate identification tool for species, screening for signatures of misidentifications, of multiple species sharing identical or very similar DNA barcodes, and of currently unrecognised diversity within the collection. Given the inconsistency in using or ignoring subspecies names in surveys of butterflies, we also explored the value of attaching subspecies names to records in DNA barcode reference libraries. The new DNA barcode library for Peninsula Malaysia was analysed in conjunction with publicly available DNA barcodes from other Asia-Pacific localities to test the ability of the DNA barcodes to discriminate subspecies. Are butterfly subspecies distinctive biodiversity units that can be distinguished by their DNA barcodes and if so, what differentiates them from species? This is an important question. Twenty- eight native butterfly species are currently protected under Malaysian law but in other jurisdictions subspecies can also have legal status. # Materials and Methods ## Building a DNA Barcode Reference Library for the True Butterflies of Peninsula Malaysia The UMKL butterfly collection comprises three thousand specimens with representatives of around 30% of the known fauna of Peninsula Malaysia. DNA barcodes were obtained by sampling dry legs from specimens in UMKL. Sampling was restricted to a few specimens per species, including morphologically and geographically diverse specimens where possible. Taxonomy and nomenclature follows our scratchpad, and reflects taxonomic decisions since Eliot. The legs were sent to the Canadian Centre for DNA Barcoding for DNA barcode assembly following standard high-throughput protocols for insects. Details of the specimens and DNA barcodes (including GenBank accessions) are available on BOLD in the public dataset: DS-BUTMAY and in. We performed an initial screen of the dataset by blasting each new DNA barcode against the full database of BOLD. In cases where new DNA barcodes matched DNA barcodes assigned to a different species name (with \>98% similarity) we reexamined the specimens’ morphology to determine the accuracy of the original identifications (provided in the “Taxonomy Note” field of the specimen records on BOLD). Following this initial screen we subsequently noted cases where specimens currently with different species names have identical or similar DNA barcodes (with \>98% similarity) and cases where specimens currently with the same species name have dissimilar DNA barcodes (≤98% similarity). The genetic distances referred to are all K2P corrected (Kimura 2-parameter; as provided by BOLD). We used 2% as the basis for our screening following the example of previous DNA barcoding studies which have demonstrated that although there is no expectation for a universal threshold of genetic distances between or within species, 2% provides a useful heuristic upon which to base deeper investigation. ## Testing if Subspecies can be Distinguished by their DNA Barcodes By blasting the UMKL DNA barcodes against the full BOLD database we determined which species in the dataset have DNA barcodes on BOLD from other researchers. When a subspecies name was not provided we derived a subspecies name for these DNA barcodes by searching published accounts of the DNA sequences (i.e. journal articles or authors’ websites, e.g. –) and by making inferences based on the reported geographical distribution of the subspecies. Note that many DNA barcodes come from GenBank with poorly reliable data, especially imprecise geographical origin, or are “Private” or “Early Release” on BOLD and not publicly viewable, but which nevertheless contribute to a BOLD identification. Where a species from UMKL was determined to be present on BOLD with DNA barcodes from multiple subspecies we then examined if the subspecies were distinguishable based on a “Tree Based Identification” (Neighbor-Joining) in BOLD. Specifically, we observed if each subspecies: i) shared identical DNA barcodes with another subspecies; ii) had unique DNA barcodes but which did not form an exclusive cluster on the tree provided by BOLD; iii) had unique DNA barcodes which formed an exclusive cluster. # Results and Discussion ## A DNA Barcode Reference Library for Identification of True Butterflies in Peninsula Malaysia A DNA barcode was obtained from 458 of 561 specimens (82%) submitted for analysis, accounting for 233 species. While similar to that reported for other Lepidoptera DNA barcoding studies, considering that the oldest specimen submitted for analysis was 20 years old the success rate seemed low for a relatively recent collection. This could serve as a warning for those attempting to build a DNA barcode library from tropical museum collections (but see) and has prompted a review of specimen storage conditions at UMKL. An approach that has been suggested is to freeze newly collected butterflies and store them as frozen tissue vouchers rather than the traditional pinning and drying of specimens. DNA extraction, amplification and sequencing using ‘Lep’ primers was highly efficient with fresh (\<3 yrs) material. Further mining of public and private collections coupled with targeted field sampling should gradually move the library to completion and increase the number of representatives per species. However, in view of the hyper-diversity of Peninsula Malaysia this is a challenge compared to temperate regions (e.g. the 180 butterfly species of Romania). Screening the new DNA barcode dataset against the full BOLD database followed by reexamination of morphology revealed that about 15% of specimens in the UMKL collection were originally misidentified. Many of these were nymphalids from the subfamily Satyrinae and the tribe Heliconiini within the Heliconiinae. One noteworthy case was a pierid originally identified as *Delias barcasa dives* and collected at Genting Highlands, Pahang, in 2011. DNA barcoding conclusively assigned the specimen to *Delias agostina* (99.3% similarity with DQ082779 from Chiang Mai in northern Thailand) confirming the presence of the species in Peninsula Malaysia. *Delias agostina* is not included in the plates of D’Abrera but is featured in the Corbet and Pendlebury *Delias* key with “Burma” printed in bold and in the species checklist with an asterisk, indicating resident status as unconfirmed. Successive screening also revealed several cases of multiple species within the same genus showing identical or similar DNA barcodes. UMKL DNA barcodes for *Danaus melanippus hegesippus* shared 99.1% similarity with a “Private” *D. genutia* DNA barcode from Australia (subspecies not given but probably *D. g. alexis*) which in turn was \>2% distant from UMKL *D. g. intermedius* DNA barcodes. The Australian subspecies has previously been treated as a distinct species. Interestingly, the phylogenetic sister of *D. melanippus, D. affinis* (according to), was not the closest matching species, being \>2.9% from *D. melanippus* and \>2.6% from *D. genutia*. UMKL DNA barcodes recorded under *Euploea modesta modesta* matched closely (\<99.8%) with GenBank DNA barcodes from India recorded under *E. core* and “Early-Release” DNA barcodes (98.8%) recorded under *E. alcathoe* and *E. core* from Australia and Papua New Guinea. *E. m. modesta* is found in India, *E. m. lugens* in Australia and Papua New Guinea. Similarly, the single short UMKL DNA barcode (307 bp) for *E. camaralzeman malayica* matched closely (99.6%) with a “Published” DNA barcode for *E. core* from Thailand and matched 100% to other “Early-Release” *E. core* DNA barcodes on BOLD. Furthermore, the UMKL DNA barcodes for *E. doubledayi evalina* matched 100% to *E. algea* (KC306717) from India and yet another “Early-Release” *E. core* from Australia. There was a further distinct cluster of *E. core* on BOLD containing DNA barcodes from Australia and Thailand which was distant from all the UMKL *Euploea*. One UMKL DNA barcode recorded under *Euploea eunice leucogonis* and collected from Genting Highlands, Pahang, in 2012 was distant (3.3%) from the two other UMKL *E. eunice leucogonis* DNA barcodes, which themselves were similar (99.2%) to *E. kluji* from India (KC306728) but relatively distant (98.0%) from *E. kluji* from Southern Thailand (HQ962260). The morphologically similar, and one time synonym, *E. leucostictos* formed a distinct sister to this cluster. As wittily noted by Corbet and Pendlebury (2nd edition) in the legend to Plate 23, “it is easier to ascertain the country of origin of a (*Euploea*) specimen than to determine its specific identity”, the genus is notorious for being taxonomically difficult. Any taxonomic interpretation is further complicated by reports of hybrids and the fact that species are commonly reared for butterfly parks (and released). There may be a tendency for collectors to assign difficult specimens to the most common species - *E. core* - accounting for its appearance in many places in this screening. Identification of *Eurema* species, a genus found abundantly in disturbed and undisturbed habitats alike, is also notoriously difficult. UMKL DNA barcodes recorded under three species of *Eurema* (*E. ada iona*, *E. hecabe contubernalis*, *E. lacteola lacteola)* showed low divergence amongst themselves and also with various *Eurema* species from various Asia-Pacific localities. The DNA barcodes all sat within the same BIN (Barcode Index Number); the system on BOLD which clusters DNA barcodes into operational taxonomic units closely corresponding to traditionally recognized species. A review of *Eurema* in Peninsula Malaysia is currently underway by our research group. Whether *Eurema* as an example of ‘barcode sharing’ is actually a reflection of the difficulty assigning these small yellow butterflies to species on the basis of wing patterns remains to be seen. *Loxura atymnus fuconius* and *L. cassiopea cassiopea* are morphologically similar and the UMKL specimen of *L. atymnus fuconius* was originally recorded under *L. cassiopea cassiopea*. However, these species cannot be confused as the wing patterns, when studied carefully, and the DNA barcodes, although close (1.7% distant and in the same BIN), are characteristic for each species. In the UMKL dataset the single representative of *Polyura athamas athamas* was distant from the *P. a. uraeus* DNA barcodes (2.1%) and closer to *P. hebe* (1.7%). Like Eliot we are hesitant to draw conclusions about the species status of these two taxa, in our case because of the small number of specimens available in UMKL and because only a short DNA barcode (307 bp) was generated for the *P. a. athamas* specimen. However, these taxa are easily distinguished as the wing patterns and the DNA barcodes, although close, are characteristic for each taxon. UMKL DNA barcodes recorded under four species of *Tanaecia* (*T. aruna aruna*, *T. iapis puseda*, *T. munda waterstradti*, *T. pelea pelea*) sat in the same BIN along with three DNA barcodes from Thailand, also representing multiple species. The taxonomy of this genus is difficult, with species specific diagnostic characters mostly from the male genitalia, (not studied here), and needs further investigation. Non-monophyly of *Charaxes bernardus* has been reported before, with *C. marmax* nested within *C. bernardus* on the molecular phylogenetic tree of Aduse-Poku et al.. We found that *C. durnfordi durnfordi* and *C. bernardus crepax* shared identical DNA barcodes, despite very distinctive wing patterns. This interesting and rare pattern deserves further study and may reflect the complex biogeographical history of this genus or mitochondrial introgression. UMKL DNA barcodes recorded as *Mycalesis mineus macromalayana* sat in a BIN with GenBank DNA barcodes for *M. mineus* from India, but the BIN also contained DNA barcodes from GenBank recorded under *M. visala*, *M. intermedia* and *M. perseoides*. Also present were unpublished *M. mineus* and *M. panthaka* DNA barcodes from China. Like the other genera above the Malaysian *Mycalesis* have a long history of taxonomic difficulty. UMKL DNA barcodes recorded as *Tirumala septentrionis septentrionis*, the only common *Tirumala* in Peninsula Malaysia, sat in a BIN with “Early Release” *T. hamata* DNA barcodes from Australia and Papua New Guinea. *T. septentrionis septentrionis* overwintering in Taiwan has previously been treated as *T. hamata septentrionis*. *T. limniace*, a similar looking species, DNA barcodes from India were also in the BIN and may be misidentifications. UMKL *Troides helena cerberus* DNA barcodes matched closely (\>98.8%) with GenBank and BOLD *T. oblongomaculatus* from Indonesia. These closely related species have been treated historically as a single species. *T. oblongomaculatus,* a “relic race of uncertain status”, has been reported to hybridize, including with taxonomically distant species. UMKL DNA barcodes recorded under *Ypthima horsfieldi humei* shared close similarity (\>99.5%) with *Ypthima nebulosa* DNA barcodes from Thailand. *Y. nebulosa* has not been reported for Peninsula Malaysia but according to Corbet and Pendlebury is likely to be found in the region suggesting the specimens in UMKL require further evaluation. Screening against BOLD highlighted 27 other species with unique DNA barcodes but which were \<2% distant from other species. These represented borderline cases for the screening threshold which were nevertheless allocated to different BINs by BOLD and cases associated with short sequence lengths or suspected misidentified DNA barcodes on GenBank/BOLD. Within the new Peninsula Malaysia dataset, only three species showed wide (\>2%) conspecific distances: *Euploea eunice*, *Polyura athamas* (see above) and *Hebomoia glaucippe*. DNA barcodes for *H. g. anomala* found on Pulau Aur, Johor, a small island off the east coast of mainland Peninsula Malaysia, were 4.2% distant from the DNA barcode for *H. g. aturia* from the mainland which clustered closely with BOLD DNA barcodes from Thailand, most likely *H. g. aturia*, and different subspecies from Taiwan and China. The differences in wing pattern between these two groups are readily apparent with the Pulau Aur butterflies exhibiting a deeper yellow upperside. *H. g. anomala* was described as a distinct species by Pendlebury in 1939. Compared with the levels of cryptic diversity discovered in other DNA barcoding surveys three species showing wide conspecific distances is relatively few, suggesting that the long history of taxonomic study of the butterflies of Peninsula Malaysia has led to a relatively accurate account of species diversity. Furthermore, two of the three cases, *Polyura* and *Hebomoia*, were associated with subspecies, one of which had previously been treated as distinct species. There were no other cases of species being represented by more than one subspecies in the UMKL collection. Perhaps the one case of truly unrecognized diversity within the Peninsula Malaysia dataset was the distinct DNA barcodes within *Euploea eunice leucogonis* and this deserves further study to determine if this is truly the exception. Following correction of morphological misidentifications in UMKL, the DNA barcodes for 78% of the 233 species were unique (with non-overlapping conspecific and interspecific distances for multiple representatives) when compared with conspecifics and closest matches on BOLD. Excluding outliers - confirmed or probable misidentified DNA barcodes on BOLD and conspecific distances associated with divergent subspecies or cryptic species diversity - the number of distinct species rises to 92%, validating the capacity of the DNA barcode reference library for rapid and effective assignment of true butterflies to species. The few cases of ‘barcode sharing’ that remain provide stimulus for subsequent studies. Considering the importance of butterflies as bioindicators and conservation flagships we are particularly encouraged by the potential of DNA barcoding to enable local species inventories, without the need for lethal sampling, but with much higher accuracy and precision than can be achieved by observing butterflies on the wing, or even by traditional morphological identification (considering the misidentifications in UMKL). ## Can Subspecies be Distinguished by their DNA Barcodes and What Differentiates them from Species? There were 1189 DNA barcodes on BOLD for the 233 UMKL species and we determined that 80 species were represented by multiple subspecies. Of the 192 subspecies, 86 were represented by singletons and 87% of these singletons had unique DNA barcodes. Of the 106 subspecies represented by multiple DNA barcodes 81% had unique DNA barcodes not shared with other subspecies and 66% formed exclusive clusters on identification trees. Because many of the subspecies were represented by singletons and “Early-Release” or “Private” DNA barcodes on BOLD, it is outside the scope of this study to examine how many of the subspecies would be reciprocally monophyletic for mtDNA in phylogenetic (maximum or statistical parsimony) analyses. However, under current levels of representation, most subspecies are genetically distinct for mtDNA which is in accordance with the expectations of the butterfly subspecies concept of Brady et al.. How this pattern changes or stabilizes as BOLD continues to grow will clarify the nature of the relationship between DNA barcodes and subspecies more accurately. The results suggest that as subspecies move from singletons to multiple representatives the number of subspecies with unique DNA barcodes could decrease (87% versus 81%;). Because the butterfly DNA barcodes available on BOLD came from a range of local surveys or phylogenetic studies, the geographic coverage was patchy and no biogeographic patterns were apparent from the analysis. However, it was not uncommon for UMKL DNA barcodes to be similar to conspecific DNA barcodes from India or China, at the extremities of the Asia- Pacific region while distinct subspecies were from one of the region’s many islands. Of the subspecies with unique DNA barcodes many were highlighted when we screened the UMKL DNA barcode dataset against the full BOLD database using the \>2% conspecific distance threshold. Conspecific genetic distances of this magnitude, i.e. distances typically seen between species, would normally warrant “dark taxon” status in DNA barcoding studies and some of these cases have in fact been highlighted by previous studies (e.g. several species from Western Ghats, India). Historical studies have likewise highlighted the morphological distinctiveness of these taxa as implied through their current disparate subspecies designations. Many of these subspecies had previously been treated as distinct species, and the DNA barcode data supports a re-evaluation of their status. Similarly, ‘unrecognized’ lepidopteran diversity revealed through DNA barcoding in the other surveys had been recognized previously, although as subspecies taxa, or as sunken or forgotten names. This may reflect the challenge of meshing Linnaean taxonomy with DNA taxonomy systems. In these cases above, consistent application of subspecies names in DNA barcode reference libraries would negate the need for dark taxon designation. Following a reverse MTMC, DNA barcoding could provide a means of testing, through concordance, if subspecies, established on the basis of moderate morphological differentiation between localities, are of sufficient evolutionary independence to merit species status. Note that 13 other species had specimens with unique DNA barcodes but which were \>2% distant from conspecific DNA barcodes. However, these further cases were due to confirmed or suspected misidentified DNA barcodes on GenBank/BOLD (See). ## DNA Barcode Reference Libraries and Subspecies In this study we present a preliminary DNA barcode library for a major component of the true butterfly species of Peninsula Malaysia. The majority of species and subspecies sampled possessed unique DNA barcodes. Although there is no fixed threshold of genetic distances clearly differentiating conspecific from interspecific distances, BOLD identification trees generally show a discernible pattern of low conspecific distances compared to interspecific distances, so can enable effective assignment of unknown DNA barcodes to species, especially when examined in conjunction with the BIN system. Unlike assignment to a species, assignment of an unknown DNA barcode to a subspecies using a BOLD identification tree would not be easily accomplished. The genetic distances between most conspecific subspecies are small and indistinguishable from distances between members of the same subspecies. Although the majority of subspecies with multiple representatives formed exclusive clusters on Neighbor-joining trees in our analyses, forming an exclusive cluster cannot logically guide taxonomic assignments in the absence of other discernible patterns - exclusive clusters are present at, and between, all taxonomic levels on a tree. Those subspecies that show ‘large’ inter-taxa distances probably warrant full species status. Conversely, there are undoubtedly cases where subspecies names are applied to groups that probably do not warrant taxonomic recognition. For example, considering the similarity of *Loxura atymnus* and *L. cassiopea* the necessity for finer taxonomic divisions is dubious. Subspecies designations as shorthand for geographically and morphologically differentiated groups provide a useful heuristic for assessing how such groups correlate with clustering patterns of DNA barcodes, especially as the number of DNA barcodes per species in reference libraries increases. Considering this, we feel there is significant value in attaching subspecies names to records in DNA barcode databases. A beneficial addition to BOLD would be the facility to allow data contributors to specify subspecies names while still recognising that members of different subspecies are conspecific for the purpose of progress statistics and other analytical tools. # Supporting Information A. Sasekumar, Rosli Ramli, Thary Gazi, and Evelyn Leong provided invaluable assistance during this study. Anonymous reviewer 3 provided helpful comments which greatly improved this manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JJW MSA. Performed the experiments: JJW KWS. Analyzed the data: JJW KWS. Contributed reagents/materials/analysis tools: JJW MSA. Wrote the paper: JJW.

# Introduction The ability to quickly, accurately, and precisely classify the stages of development for the nematode *Caenorhabditis elegans* is crucial for many aspects of worm research, including monitoring development, stock maintenance, crossing, and stage synchronization. In addition, the measurement of developmental stages within a population is useful in many studies. For example, mutations in genes, such as the *daf-2* insulin-like receptor pathway, alter normal development and lead to inappropriate arrest in the dauer stage Further, a class of genes known as heterochronic genes result in abnormal development by disrupting the normal patterns of cell division and differentiation steps in specific stages. Consequently research focused on these classes of genes have used visual assays to classify the stages of worm development and then to draw conclusions about both the individual genes involved in these pathways and gene- gene interactions between them. One common method for identifying and measuring the numbers of dauer stage animals within a population has been the use of sodium dodecyl sulfate (SDS) selection to kill non-dauer worms, and followed by recording the number of dauer and non-dauer larvae identified. However, this method does not distinguish between other stages of development (L1, L2, L3, L4 and adult), and it is not able to identify partially-formed dauers, that show many but not necessarily all of the features for the dauer larva, in the count. A second approach is the use of a stage-specific reporter gene, such as the *col-19p*:*gfp* reporter which allows adult animals to be readily visualized by fluorescence. But, this approach requires the prior identification of a gene showing stage-specific expression, and the ability to generate a reporter that exhibits both highly specific and readily visible fluorescent protein expression. A third approach is the use of imaging techniques and computer software, such as the WormSizer ImageJ plug-in, to measure worm size in an objective manner. While the use of measurements can demonstrate differences in the developmental rate or body size of mutants, determining if the differences reflect alterations in specific growth and/or developmental processes could be more challenging using this approach alone. Consequently, visual assessment is an attractive way to obtain detailed information about larval development within a population both quickly and without the need for specialized strains or equipment. However, a drawback to this method is the potential for variability arising from differences in how individual raters perceive each stage of development. The issue of potential disagreement between raters on the developmental stage of the same worm, also known as inter-rater agreement, has been addressed and analyzed in other fields such as psychology and medicine. However, there is little mention of formal statistical modeling of inter-rater agreement in *C*. *elegans* publications using measurements of developmental stage. Given both concerns about data reproducibility and the interest in inter-rater agreement in other fields, we sought to apply statistical analysis to this basic aspect of worm research to explore the operating parameters of visual scoring. To address the variability between different researchers evaluating stage of development by eye, it is necessary to generate a dataset consisting of the developmental stage assigned to each animal within a sample of worms by multiple raters. The developmental stage assessment is an outcome consisting of an ordered categorical variable, given that the different stages are recorded as discrete qualities, with a natural order based on the progression between L1, L2, dauer, L3, L4, and adult. A computationally simple and commonly used statistical method for describing inter-rater agreement utilizing this type of ordered categorical data is through the calculation of the kappa statistic \[, \]. This statistic is a single value that is interpreted as the amount of agreement exceeding that which is expected by chance alone. While the kappa statistic is a popular type of analysis due to the ease of computation, and the apparent simplicity of interpretation, there are also a number of criticisms that make use of the kappa statistic problematic. Important problems with the kappa statistic are two-fold. First, the kappa statistic sometimes may give rise to paradoxical or misleading results with certain arrangements of data, and second, it combines the two independent components of agreement (accuracy and precision) into one parameter which precludes a clear description of how the raters disagree. Latent variable modeling provides a more insightful method for analyzing inter-rater agreement for ordered categorical measurements. This approach has the advantage of providing information concerning the accuracy and precision of each rater. However, it should be noted that in most applications, the true developmental stage of each worm is unknown, so that only relative accuracy can be assessed. The basic concept of agreement among measurements is relevant whether the measurements are of a discrete, ordinal nature or whether they are continuous and quantitative. A simple but limited approach to assessing agreement between two continuous and quantitative measurements consists of making a plot of the pairwise differences versus the corresponding pairwise averages. In the medical literature, this approach was championed by Bland and Altman and the plots are often referred to as Bland-Altman plots. If the pairwise differences are normally distributed, then it is easy to compute the “limits of agreement” that demarcate where 95% of the differences will tend to fall. If the limits of agreement are very narrow with regards to the outcome, then for all intents and purposes, the measurements made by the two methods are interchangeable. When this is not the case, then it is necessary to characterize what part of the difference is systematic and what part is random. This is important because the systematic error can be removed by determining the appropriate calibration equation. Unfortunately, when the limits of agreement are too large to be considered negligible, the limitations of this approach are readily apparent because with only two measurements, it is impossible to apportion the error into systematic and random parts. As Bland and Altman (and many others) have pointed out, the usefulness of conducting a measurement error study of two methods without repeats of at least one of the measurements is questionable. To have a useful study, at least one of the methods must be repeated (Bland and Altman suggested that both methods be repeated) or if neither method is repeated, at least three methods must be simultaneously compared. Because of the discrete nature of ordinal measurements, Bland-Altman plots are of limited usefulness for studying agreement among the raters of worm development stages. However, the relevance and importance of apportioning the error into its systematic and random components persists. This knowledge will guide efforts to remove the tendency for raters to disagree. For example, the prescription for removing disagreement among raters will differ if there is little systematic error but high imprecision (tendency for large random errors) versus the reverse situation. To overcome the limitations of the kappa statistic and Bland-Altman plots and to provide an analysis of the precision and accuracy of the individual raters, we describe how the common factor model with thresholds can be applied to the ratings of worm developmental stage performed by members of a *C*. *elegans* research group. We then further use the model to perform a detailed evaluation the precision and accuracy of the individual raters as a means of studying the reliability and limitations of visual scoring of worm development in a research setting. Our methods and findings are relevant for understanding the performance of visual scoring of development as an experimental outcome, and provide a means to evaluate the assessment of other phenotypes that depend on rater performance in *C*. *elegans* research. # Results ## Development of the common factor model The common factor model approach assumes that all of the raters are jointly rating the same worm, and hence this model is applicable to the experimental design used in this work. The common factor model permits (assuming there are three or more raters and multiple worms) the decomposition of the rater differences into the part due to systematic error and the part due to random error. This decomposition is indirect in the sense that the scores are discrete ordinal categories, i.e., the L1, L2, dauer, L3, or L4 larval stages, and an underlying continuous scale must be assumed for the purposes of the model. illustrates the common factor model for an example involving three raters. The common factor is envisioned as the true worm stage but on a continuous scale, which is denoted by μ. Each rater judges the stage on his or her own continuous scale, shown as χ₁, χ₂, and χ₃ in the model. Both variables μ and χ are latent, meaning that they are not directly observable, but are included in the model since they underlie the actual observable values. The true stage of worm development, μ, is assumed to be normally distributed with a mean of zero and a standard deviation of one. This standard deviation is denoted by the curved arrow adjacent to μ in. Each rater then assesses worm development using their individual rating scale, denoted by χ₁, χ₂, or χ₃. The variable χ₁, χ₂, or χ₃ is related to μ via a linear function with an intercept of zero and a slope of ρ₁, ρ₂, and ρ₃, respectively. This results in χ having a standard deviation of 1-ρ², which is denoted by the curved arrows. The discrete ordinal ratings for an individual worm that are assigned by a rater are derived from each rater’s underlying latent continuous scale by applying the rater’s thresholds, which lie along the normal distribution. Because there are five categories, there will be four thresholds between them. The stages of development fall in regions along a standard normal distribution, and values for these threshold cutoffs are estimated from the observed ratings. ## Summary of rater data To use the common model to compare inter-rater agreement, we asked seven different individuals, who had a range of experience in *C*. *elegans* research, to score 60 worms with regards to developmental stage. To ensure that all raters scored the same worms, as required by the assumptions of the model, we used files consisting of both still and video images of each worm, and the individuals were asked to score each animal with regards to the stage of development (L1, L2, dauer, L3, or L4). Review of the ratings showed that 42% of the time there was complete agreement among the raters, 35% of the time only a single rater differed from the group by a single stage, 12% of the time multiple raters differed with a split between them of a single stage, and 12% of the time multiple raters differed with a split of two stages. We then determined the frequency with which each rater scored the animals into each developmental stage. For the L1 stage, the highest and lowest frequencies were 0.36 and 0.20, for L2 these were 0.13 and 0.20, 0.10 and 0.23 for dauer, 0.08 and 0.20 for L3, and 0.18 and 0.30 for L4 (Observed column). Hence, the largest discrepancy in scoring was present for the L1 stage, and among raters, individuals 1 and 4 showed the largest discrepancy. Additionally, raters 3, 4, 5, and 7 tended to place a larger fraction of worms in the extreme larval stages (L1 or L4) compared to the other individuals. The discrepancy between individual raters can also be seen using pairwise error plots showing the results of one rater on the x-axis versus another rater on the y-axis. We used jittering to add a small amount of noise to the horizontal and vertical coordinates for the points to facilitate the visualization of data points that would otherwise be overlapping. In these graphs, any points falling away from the diagonal line correspond to discrepancies between the two raters, whereas points falling along the diagonal line demonstrate agreement. The pairwise plots show many points along the diagonal indicating agreement. No discrepancies were greater than two stages with most differences being just one stage (e.g., L1 larva versus L2) as opposed to two stages. The points tend to align best for the extreme stage (L1 and L4), whereas most of the discrepancies appear in the middle stages (L2, L3, and dauer). These plots also demonstrate that for each pair of raters, there is variability regarding whether the disagreements fall above or below the diagonal line of agreement. For the pairs of raters where most of the discordant points lie either above or below the diagonal, this is suggestive of a directional bias of one rater compared to the other. For instance, in the plot comparing rater 2 versus rater 5 , the points fall roughly equally above or below the diagonal line, while in the plot comparing rater 3 versus rater 7, most of the discordant points lie below the diagonal line, suggesting a tendency of rater 3 to score the animals in later developmental stages compared to rater 7. Visualizing the data in these pairwise plots can suggest a bias or directionality to the disagreements as discussed above, and also can suggest random differences. The problem is that these plots cannot indicate how much of the observed difference is due to systematic error and how much is due to random error. In order to fully describe these two types of error among the seven raters, it is necessary to model the data, as described below. ## Fitting the data to the common factor model with thresholds To further explore inter-rater agreement in terms of the bias and random error of the raters, we fit a common factor model similar to that shown in to the ordinal measurements, but with the seven raters instead of the three raters depicted. The parameters in this model are the standardized factor loadings ρ₁ through ρ₇, as well as four threshold values for each rater, c_i1 through c_i4. In this model, the factor loadings can be interpreted as the correlation between each rater’s stage assignments and the true developmental stage of the animals since all of the latent variable μ, which represents the true developmental stage of the animal has been standardized to have a mean of zero and a standard deviation of one. Further, the use of a standardized latent variable constrains the residual error standard deviations σ_*i* to be equal to $\sqrt{1 - \rho^{2}}$. The thresholds can be interpreted as dividing the continuous latent scales into the ordered categories representing each developmental stage. In this model, there are seven factor loadings, one for each rater, and 28 (4 × 7) thresholds for a total of 35 parameters to be estimated. ## Factor loading estimates We estimated each of the parameters including the factor loadings along with their corresponding 95% confidence intervals and the results are shown in. The residual error standard deviations that describe the rater imprecision were computed using the factor loadings using the formula in the previous section. Because of the inverse relationship between the factor loadings and the residual error standard deviations, a higher factor loading value results in a lower residual error standard deviation. The residual error standard deviation describes the raters imprecision while, conversely the factor loading describes the rater’s precision. At the same time, the factor loading describes the latent measurement scale for each rater, i.e., the size of the measurement unit. That is, differences in *ρ*_*i* among the raters imply different scales for the raters. On the latent scale, the marginal distribution for each rater has a standard deviation of one and a mean of zero. As a result, the residual error standard deviation $\sqrt{1 - \rho^{2}}$ represents the uncertainty in the rater's latent continuous measurement conditional on the true latent value ρ and thus describes the imprecision, or lack of repeatability, of the rater's measurement process. The larger the residual error standard deviation or imprecision, the more likely that repeated measurements of an animal would fall into different ordinal categories. In order to compare raters in terms of imprecision, the differences in the raters’ scales (*ρ*_*i*) must be taken into account. This is easily accomplished simply by dividing the residual error standard deviation by the corresponding factor loading: $$\frac{\sqrt{1 - \rho_{i}^{2}}}{\rho_{i}}\ = \ \sqrt{\frac{1}{\rho_{i}^{2}} - 1}$$ and is referred to as the “scale-adjusted imprecision.” As the factor loading approaches 0, the scale-adjusted imprecision increases without bound. When the factor loading equals 1, the scale-adjusted imprecision equals 0 indicating perfect precision. As shown in, the raters in this study had scale-adjusted imprecision standard deviations that ranged from 0.045 to 0.192, with rater 5 being the most precise and rater 1 being the least precise. Overall, the order of rater precision (best to worst), based on lowest to highest residual error values is rater 5, 4, 3, 7, 2, 6, and then 1, which correspond to residual error values of 0.045, 0.077, 0.101, 0.135, 0.15, 0.156, 0.192, respectively. To facilitate comparison of the residual errors between raters, and show the ratio of the residue error estimates between pairs of raters (columns divided by rows). The rater who stands out is rater 5, who has a considerably lower residual error estimate (as much as three times lower than other raters). Raters 1 and 2 have the highest residual error rates, and these differ the most with raters 3, 4, and 5. The magnitude of these error rates are discussed in more detail below. ## Threshold estimates The factor loading estimates indicate how closely each rater’s assessment of developmental stage for a worm correlates with its true developmental stage and represents only random error, providing a description as to how consistently or reliably an individual rater classifies each worm. By contrast, the thresholds represent systematic differences among the raters in the location and width defining the ordinal stages on the latent continuous scale. In this model, threshold 1 divides the L1 stage from stage L2 stage on the latent continuous scale. Threshold 2 divides L2 from dauer, and so forth. The estimated thresholds based on the fitted model are shown in and graphed in. Based on these estimated thresholds, it is possible to determine the proportion of worms a particular rater would be expected to classify into a particular stage of development providing an alternative way of comparing the performance of the raters. Each proportion is simply the area under the unit-standard normal curve corresponding to the particular stage. The expected proportions of each stage of development for each rater are shown in (Expected column). Comparing the expected proportions from rater to rater gives an indication of where particular raters systematically differ in how they classify worms. shows a visual heatmap display of the differences in expected proportion between raters (columns minus rows), with red indicating a positive difference and green indicating negative, and the brighter the color, the greater the difference. This display provides a means of quickly assessing the rater or raters that stand out as being considerably different from the others in the scoring of a particular stage of development. In our group data, for the L1 stage, rater 4 and rater 1 stand out as showing the highest and lowest proportion of animals being assigned to this stage, respectively. For the L2 stage, all raters are relatively similar, for the dauer stage, rater 2 and 4 have a relatively large difference in animals assigned to this stage, and for the L3 and L4 stages, rater 6 stands out as assigning considerably more animals to the L3 stage and correspondingly fewer to the L4 stage. ## Assessing the relevance of rater differences While the estimated factors discussed above provide a means of comparing different aspects of how each rater in a group scores worm developmental stage, it is important to consider whether these differences make a notable change in the experimental data collected by each rater. Specifically, one would want to know how much imprecision or bias may be shown by a rater without compromising the ability to produce reliable measurements of stages of worm development. To make this determination, one should consider both the residual error of each rater as well as the threshold estimates for each stage of development. If a rater were to make two independent, repeated measurements of the same worm, the standard deviation of the expected difference in the continuous latent measurements would be about 1.41 times the residual error standard deviation. These values would be for raters 1 through 7, 0.27, 0.21, 0.14, 0.11, 0.06, 0.22 and 0.19, respectively. These values may then be compared to the differences between the thresholds for a given rater. The smaller the threshold differences are compared to $\sqrt{2 - {2\rho}^{2}}$, then the more likely imprecision is to result in repeated judgments that produce differences in the worm's developmental stage. This propensity varied depending on rater and stages. However, we found that for all of the raters, the threshold widths tended to be large compared to $\sqrt{2 - {2\rho}^{2}}$, which suggests that rater bias plays a larger role in discrepancies in scoring than does imprecision. But, there are a few instances when threshold width was similar to $\sqrt{2 - {2\rho}^{2}}$, such as the interval between threshold 2 and threshold 3 for rater 4. In these circumstances imprecision can play a larger role in the observed differences than seen elsewhere. To investigate the impact of rater bias, it is important to consider the differences between the raters’ estimated proportion of developmental stage. For the L1 stage rater 4 is approximately 100% higher than rater 1, meaning that rater 4 classifies worms in the L1 stage twice as often as rater 1. For the dauer stage, the proportion of rater 2 is almost 300% that of rater 4. For the L3 stage, rater 6 is 184% of the proportion of rater 1. And, for the L4 stage the proportion of rater 1 is 163% that of rater 6. These differences between raters could translate to unwanted differences in data generated by these raters. However, even these differences result in modest differences between the raters. For instance, despite a three-fold difference in animals assigned to the dauer stage between raters 2 and 4, these raters agree 75% of the time with agreement dropping to 43% for dauers and being 85% for the non-dauer stages. Further, it is important to note that these examples represent the extremes within the group so there is in general more agreement than disagreement among the ratings. Additionally, even these rater pairs might show better agreement in a different experimental design where the majority of animals would be expected to fall in a specific developmental stage, but these differences are relevant in experiments using a mixed stage population containing fairly small numbers of dauers. ## Evaluating model fit To examine how well the model fits the collected data, we used the threshold estimates to calculate the proportion of worms in each larval stage that is predicted by the model for each rater. These proportions were calculated by taking the area under the standard normal distribution between each of the thresholds (for L1, this was the area under the curve from negative infinity to threshold 1, for L2 between threshold 1 and 2, for dauer between threshold 2 and 3, for L3 between 3 and 4, and for L4 from threshold 4 to infinity). We then compared the observed values to those predicted by the model. The observed and expected patterns from rater to rater appear roughly similar in shape, with most raters having a larger proportion of animals assigned to the extreme categories of L1 or L4 larval stage, with only slight variations being seen from observed ratios to the predicted ratio. In addition, model fit was assessed by comparing threshold estimates predicted by the model to the observed thresholds, and similarly we observed good concordance between the calculated and observed values. # Discussion The aims of this study were to design and carry out an experiment for measuring inter-rater agreement among *C*. *elegans* researchers, to evaluate these data using a common factor latent variable model with thresholds, to draw conclusions about relative accuracy and precision among a group of raters, and to generate a protocol for measuring inter-rater agreement that other labs can follow. The major findings were that: 1) it is possible to quickly obtain data from multiple raters assessing the same sample of worms for five stages of development; 2) that visual assessment of pairwise error plots reveals some degree of differences among the seven raters, especially for the L2, dauer and L3 stages; and 3) that in the one factor model, raters tend to have high reliability (precision) but there are larger differences between raters in the threshold values for the stage boundaries (accuracy). ## Measuring inter-rater agreement in *C*. *elegans* researchers using visual assessment The goal of our experimental design was to minimize any sources of variability in the animals used for scoring that would then be confounded by rater error. To ensure that each rater was making an assessment under the same visual conditions, a video recording of a magnified population of worms was used instead of having the raters look at live worms under a microscope. This both presented an identical animal for scoring to each rater, and enabled each worm to be given a unique identifier to eliminate the possibility that raters were assessing a different worm, or inadvertently scored the same worm twice. One criticism of this design is that by using a pre-recorded image, this could take away certain aspects of a visual assessment a rater may use when looking under a microscope, such as adjusting the amount zoom and focus. These adjustments may contribute to rater’s ability to make an accurate judgment, and hence this study design does not evaluate these aspects of rater behavior that might facilitate reaching a judgment. However, our approach helps to promote consistency in the worm sample being assessed by the observers, and was likely essential to make gathering data from multiple raters feasible. By providing both a still image and movie, it is hoped that the range of views and behaviors available would be at least somewhat similar to what is available to a researcher making an assessment in the lab. ## Sources of inter-rater differences Our findings suggest that differences in accuracy between the raters play a larger role than imprecision. In considering the threshold values, which characterize the differences in accuracy, a large number of comparisons may be made between the seven raters. However, in our study, we found that the distinction between the L1 and L2 and the L2 and dauer stages seemed to be most problematic as shown by the larger spread in the threshold values for the overall group. Also contributing to these differences were specific large differences between pairs of raters. Our study was not focused on the origins of the differences in rater behavior, but given the finding that accuracy was more important than precision. Subsequent work that seeks to understand the visual cues involved in scoring would be of value. It is possible that the raters use the parameters of size, morphology of the gonad or other internal structures, and animal behavior to varying degrees in determining the developmental stage of the worms, and that the individual’s approach and weighting produces the observed differences. There may also be knowledge and experience differences that contribute as some of the raters in our group routinely staged animals in their research while others had other areas of expertise. Regardless of the cause, the model provides specific information for how particular raters could adjust how they categorize worms, in order to agree more closely with the other raters in their lab. For example, rater 4 could review how to distinguish the younger larval stages from the older stages, due to his/her tendency to assign more worms to the L1 stage. Furthermore, rater 6 has a much higher threshold 4 value compared to the other raters, and rater 1 has a considerably lower threshold 1 value. This information would suggest that rater 6 tends to rate worms that have reached the L4 stage in earlier stages, and that rater 1 tends to rank L1 worms into higher categories. These raters could be mindful of these tendencies during subsequent scoring sessions. Despite both computational and model complexity, this assessment is a worthwhile means of obtaining detailed and informative descriptions of inter-rater agreement, information that is not available by oversimplified and often misleading methods such as the computation of kappa coefficients. The results obtained can both provide confidence with regards to the success of the research team in accurately scoring the parameter of interest, and additionally the modeling approach provides insights into specific ways to improve the reliability of the raters. # Materials and Methods ## Collecting worm rater data To obtain a mixed stage population of worms containing dauer larvae as well as L1, L2, L3 and L4 stages, wild type N2 worms were grown at 20°C on nematode growth agar (NGA). This population was then supplemented with dauer larvae that were transferred from an *eak-4; tatn-1* double mutant stock, grown at 25°C. The *eak-4; tatn-1* double mutant will produce approximately 90% dauer larva when grown at 25°C. Movies focused on small groups of worms within the mixed-stage NGA plate were obtained at 40X magnification using a Zeiss Stemi 2000 stereomicroscope with a color video camera attached. The I.C. capture 2.0 software was used to record the movies and still images. There were 9 separate 5 to 30 second movie files made, and each movie containing 3 to 10 worms that a rater would score. To ensure that all raters were scoring the same individual worms, the worms in these movies were each given a unique identifying number, and these were labeled in a separate image of a still from the first frame of each movie. The still images and movies were scored by seven individuals who had some amount of *C*. *elegans* research experience. Before rating the worms, each rater was given a five-minute refresher tutorial by the author, explaining the qualifying features of each stage of development, and showing an example video containing each stage. Raters then were given a score-card, and for each movie file, were asked to identify the particular worm to be rated, and then play the video as needed in order to make a decision on the stage of development, and recording the result. This decision was typically reached after only a second or two of viewing the worm, and often a rating was reached after only viewing the still frame from the movie. These data were tabulated with a column for each rater. A total of seven raters were recruited for this study, and each rater was asked to judge the larval stage of 60 different worms. Hence each rater had 60 observations, with the larval stage of development coded by ordinal values one through five. ## Describing Agreement Using a Calibration Model with Thresholds Although rating a worm for its stage of development depends on complex judgments, the resulting measurements are on a cruder scale compared to measurements of the worms’ length or weight. Length and weight are measured on a continuous, quantitative scale while determinations of the developmental stage are discrete measurements on an ordinal scale. Determining the agreement among measurements of length or weight made by different methods on the same set of items requires a measurement error model. The measurement error model describes how the common factor (the true values of the common items each method measures) influences each observed measurement. The measurement error model describes both the systematic differences (accuracy or relative bias) among the methods and the differences due to random error (imprecision). The relative bias between any two methods can then be removed by calibration. The same basic approach will work with ordinal ratings but the relative bias and random error are assumed to occur on an underlying continuous scale represented by continuous latent variables. The continuous latent variables then are translated to the observed discrete ordinal ratings via rater-specific threshold models. The calibration model for three raters *i* = 1, 2, and 3 making measurements on a continuous scale for the *j*^th worm, can be described by a set of three simultaneous equations: $$X_{1j}\ = \ \alpha_{1} + \beta_{1}\mu_{j} + \epsilon_{1j}$$ $$X_{2j}\ = \ \alpha_{2} + \beta_{2}\mu_{j} + \epsilon_{2j}$$ $$X_{3j}\ = \ \alpha_{3} + \beta_{3}\mu_{j} + \epsilon_{3j}$$ Here, the true quantity is denoted by μ. The measurements are denoted by *X*. The systematic bias is described by α and β, and the random error by *∈*. These equations relate the measurements to the underlying, unknown true values. Because μ appears in all three equations, it is possible to relate *X*₁ with *X*₂, *X*₁ with *X*₃, and *X*₂ with *X*₃ –the μ drops out. That is, we can compare the raters even though the true values are never known. We can also describe the random error even though no item is measured more than once. (In effect, conditional on μ, we effectively have repeats for each method.) A path diagram can be used to illustrate the calibration model, as shown in. In this diagram, the quantity to be measured (the “true” value—a latent variable) is represented by the circle and the measurements by squares. The true value “causes” the observed measurements so there are single-sided arrows pointing from μ to each measurement. The *i*^*th* rater distorts the true values and this distortion is quantified by α_*i* and *β*_*i*. The double-sided arrow pointing to the circle for μ denotes the standard deviation σ for the true values. In a similar fashion, the standard deviations for each *∈*_*i* are denoted as σ_i next to the rectangle for each measurement *X*_i. In this model, the scale bias parameter β_i and the random error parameter *∈*_*i*. When the measurements (*X*_i) are ordinal, the calibration model can still be used but the ordinal categories (*k* = 1, …, 5) need to be modeled as outcomes based on an underlying (unobserved) normally distributed variable χ_i (mean of zero, standard deviation of one). We assume that the continuous true values μ are normally distributed with mean 0 and variance 1. It is convenient to constrain the unconditional normal distributions for each χ to a mean of zero and standard deviation of one. These constraints result in the following model: $$\chi_{1j}\ = \ \rho_{1}\mu_{j} + \epsilon_{1j}$$ $$\chi_{2j}\ = \ \rho_{2}\mu_{j} + \epsilon_{2j}$$ $$\chi_{3j}\ = \ \rho_{3}\mu_{j} + \epsilon_{3j}$$ The intercepts vanish and the factor loadings (β) are equivalent to correlation coefficients and have been denoted by ρs. These parameters represent the correlation between each unobserved continuous rating χ with the unobserved true value μ. The constraints effectively remove any systematic error between the true values and the unobserved χ. Any systematic error is transferred to the *k*^th threshold for subject *i t*_ik. The area under the marginal distribution of χ_i below *t*_i1 is the proportion of observations *X*_i for the first category labeled zero (0). Similarly, the area under the marginal distribution between *t*_i1 and *t*_i2 is the proportion of observations *X*_i for the second category labeled one (1), and so on. If there are *m* ordered categories, then there are *m-1* thresholds for each rater. Because the χ_i are latent variables, additional constraints are required to allow them to be identified. This requires that scale parameter ρ_i and the residual error *∈*_*i* standard deviation σ_i being related so that: $$\sigma_{i}\ = \ \sqrt{1 - \rho_{i}^{2}}$$ This constraint effectively means that ρ_i represents the inverse of the random error or precision (so that σ_i is the imprecision) while the thresholds *t*_ik represent the systematic error (bias). This threshold model can be fitted using structural equation modeling software and the OpenMx R package was used. Full information maximum likelihood was used to estimate the model parameters. The one factor ordinal model with seven raters and four ordinal categories was adapted from code available from (<http://openmx .psyc.virginia.edu/docs/OpenMx/latest/FactorAnalysisOrdinal_Matrix.html>). A detailed annotated version of the code used to produce the final results is shown in the supplemental material. ## Computational challenges in modeling worm rater data There are definite benefits to fitting this type of model to the data, however, one challenge of this approach is the potential technical difficulty in achieving model convergence via the necessity of using an iterative solution. When iterative procedures are used, the software used to implement the procedure typically tries to assess whether or not convergence has been achieved. It should be remembered that this assessment is fallible. Just because no warning message was generated does not necessarily imply convergence and conversely, a warning message does not necessarily imply that convergence was not reached. In all cases, the output from iterative estimation procedure must be verified for convergence regardless of any warning messages or not. Iterative solutions typically require starting values for each parameter. The better the starting values, the more likely convergence will be achieved (and achieved in fewer iterations). Providing good starting values is akin to knowing the answer before having the answer. Many times determining good starting values is difficult so that random starting values are used. When using random starting values, it is customary to use a number of sets. When the common factor model (or similar) is used with continuous measurements, convergence is typically easier and faster to achieve. Threshold models needed for ordinal measurements are more challenging so that it is a good idea to run a larger number of sets of random starting values. The output of these random starts are collected and ordered by likelihood function value. Ideally, the outcome with the highest likelihood value should be the maximum likelihood solution. Often, with a large number of random sets the highest likelihood value will occur multiple times. Also, often very small differences in the likelihood value are connected with parameter estimates that are virtually identical, all leading to the conclusion that the maximum likelihood solutions has been achieved. Checking the parameter estimates to the observed data summaries (as we did) also helps confirm a reasonably correct solution. Additional techniques are to use software that implements the iterative procedure in an alternative fashion, and seeing that the solutions are either identical or very similar. We used all of these techniques to ensure the maximum likelihood solution was achieved. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AAF RAB JMB GMM ALF. Performed the experiments: AAF ALF. Analyzed the data: AAF RAB JMB GMM. Contributed reagents/materials/analysis tools: AAF RAB. Wrote the paper: AAF RAB GMM ALF.

# Introduction Among all types of cancer, breast cancer has both the highest incidence (24%) and highest mortality (15%) in women around the world. Mammography uses low- energy X-rays to identify abnormalities in the breast. For women who are at average risk for breast cancer, most of the benefit of mammography results from biennial screening during ages 50 to 74 years. Of all age groups, women aged 60 to 69 years are most likely to avoid death from breast cancer through mammography screening. The sensitivity and specificity of mammography screening for breast cancer are reported to be 77–78% and 89–97%, respectively. Although breast cancer screening with mammography is considered effective in reducing breast cancer-related mortality, interpreting mammograms is a delicate task and prone to errors, with at least 25% of detectable cancers being missed. Detecting subtle regions such as microcalcifications and focal asymmetric density (FAD) in particular pose difficult hurdles for physicians. Several computer-aided detection (CAD) systems have been developed to overcome this problem and provide physician support. Initially, studies showed that a single-reading with CAD systems could be an alternative to double-reading. However, studies have since concluded that the cost-effectiveness of screenings had not improved, mainly because of the low specificity of traditional CAD systems. Recently, the application of convolutional neural networks, one field of deep learning (DL), has led to dramatic improvements in visual object recognition, detection, and segmentation. In this study, we adopted to create a detection- based DL model that could detect all the findings that breast cancer can present, including not only masses, but also architectural distortion and microcalcifications. While masses can be segmented, other findings are difficult to segment because it is difficult to accurately delineate the boundary between normal and abnormal areas. Therefore, we thought that a bounding box detection AI model was the most suitable for our study. Models using DL have routinely surpassed the performance of traditional methods due to their automated feature extraction. These dramatic improvements have caught the eye of researchers in several fields, including mammography. In addition to those that detect breast cancer, there are studies to predict the risk of breast cancer from mammography. For patients with breast cancer, there are models which estimate the expression of receptors involved in chemotherapy selection, and those that predict pathological types. Sensitivity for studies detecting breast cancer was found to be in the range of 0.76–0.97, with a mean number of false positive indications per image (mFPI) of 0.48–3.56. Sensitivity and mFPI are often used to evaluate the detection model, where the mFPI is the average number of false positive lesions displayed by the model for a single image. There is a trade-off between sensitivity and mFPI, since the greater the number of false positive lesions presented by the model, the higher the sensitivity. For this reason, a higher sensitivity with a lower mFPI is desirable in a model intended to help physicians interpret mammograms for the benefit of their patients. The purpose of the present study was to train and validate a state-of-the-art DL-based model to detect breast cancer with higher performance than existing models. # Methods ## Study design First, a DL-based model for detecting breast cancer on mammograms was trained and validated using retrospectively collected mammograms annotated by the radiologists with the locations of malignant lesions. Second, the model was tested with independent datasets for the detection of breast cancers. The Ethical Committee of Osaka City University Graduate School of Medicine comprehensively reviewed and approved the protocol of this study. Since the mammograms had been acquired during daily clinical practice, the need for informed consent was waived by the ethics board. We have created this article in compliance with the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement. There are two possible ways to label mammograms when developing an AI model for breast cancer screening. The mammograms can be labelled using BI-RADS grading or pathology. The advantage of the former is that a large dataset of mammograms can be prepared since pathology results are not required, but on the other hand, BI- RADS grading is known to be more subjective than the pathology result. In other words, if we created an AI model with BI-RADS as a label, the AI model may output false positives for mammograms that have a high grading in BI-RADS but are not pathologically breast cancer. ## Datasets To train, validate, and test the DL-based model, four datasets were used: a hospital development dataset, a hospital test dataset, a clinic test dataset, and the Digital Database for Screening Mammography (DDSM) dataset. Mammograms for the hospital development dataset and the hospital test dataset were retrospectively collected from patients who were surgically diagnosed with breast cancer at Osaka City University Hospital, which provides secondary care. Mammograms in the clinic test dataset were collected from patients who underwent mammography screening at Medcity21 Clinic, a provider of preventive medicine. The hospital development dataset and hospital test dataset were collected consecutively from January 2006 through December 2016 and from January 2017 through December 2017, respectively. The clinic test dataset was collected consecutively from April 2014 through March 2017. Malignant mammograms were collected from both sides of patients with bilateral breast cancer and the affected side of patients with unilateral breast cancer for the hospital test, hospital development, and clinic test datasets. Nonmalignant mammograms for the hospital development and hospital test datasets were collected from the healthy side of patients with unilateral breast cancer. The mammograms were diagnosed as nonmalignant in preoperative screening by five surgeons who specialized in breast surgery. Nonmalignant mammograms in the clinic test dataset were collected from both sides of healthy patients, and the healthy side of patients who had pathologically diagnosed unilateral breast cancer. Nonmalignancy was then confirmed with 2 years of follow-up mammograms by two radiologists who had 18 years and 10 years of experience interpreting mammography. Since the study included breast cancer patients who visited each institution for the first time, none of the datasets had overlaps. Both left and right mediolateral oblique (MLO) and craniocaudal (CC) images were collected, if available. ## Ground truth labelling Malignant lesions on the affected side of mammograms in the hospital development dataset were annotated by two radiologists who had 6 years and 5 years of experience interpreting mammography. Mammograms were annotated with bounding boxes and labelled as mass, calcification, distortion, and FAD with reference to ultrasound, radiological, biopsy, and surgical reports. When there was disagreement between the radiologists, consensus was achieved by discussion. In addition, they could consult with a third expert if needed. Mammograms with no findings in the affected side were excluded. The density of the mammary glands on all mammograms was assessed by the same radiologists according to the BI-RADS in consensus. This assessment was performed on a mammogram basis, rather than a patient basis. All malignant findings (mass, calcifications, FAD, and architectural distortion) of each cancer were merged into one bounding box. Mammograms with multiple breast cancers would have multiple bounding boxes. Malignant lesions on the affected side of mammograms in the hospital test dataset and the clinic test dataset were annotated in the same manner as the hospital development dataset by two radiologists who had 6 years and 12 years of experience interpreting mammography. Ground truth labelling for the publicly available DDSM development dataset was as follows. The Curated Breast Imaging Subset of the DDSM (CBIS-DDSM) is an updated and standardized version of the DDSM. In this dataset, all mammograms include pathologically verified breast cancer; a segmentation of malignant findings is included. Malignant mammograms were collected from both sides of patients with bilateral breast cancer and the affected side of patients with unilateral breast cancer from the CBIS-DDSM. Bounding boxes were created from the longest diameter in the vertical and horizontal directions of the malignant segmentation. All malignant findings (mass, calcifications, FAD, and architectural distortion) of each cancer on the same mammogram were merged into one bounding box. Mammograms with multiple breast cancers would have multiple bounding boxes. ## Training and validation of the model A DL-based model was developed using RetinaNet to detect lesions and evaluate the probability of breast cancer in mammograms. RetinaNet is a regression-based, unified framework with a backbone and two subnetworks which detect and classify objects. The backbone network used in our study was ResNet152 with a feature pyramid network. The ResNet has four downsampling levels and the FPN has five upsampling levels, each with 256 channels. The backbone network computes convolutional feature maps of an entire input mammogram. The first subnetwork, called “class subnet,” classifies the output of the backbone network as either malignant or not malignant. The second subnetwork, called “box subnet,” performs convolutional bounding box regression. This network adopted focal loss for class subnet and L1 loss for box subnet. Focal loss focuses training on a sparse set of hard examples and prevents the vast number of easy negatives from overwhelming the detector during training. RetinaNet is tuned to classify sites outside the adenoma bounding box as background. For example, mammary glands in a different location from the breast cancer on mammograms will be treated as a true negative. Through these processes, the model extracts features that are unique to breast cancer. For structural details, see ; the source code is available online. This model was built in the TensorFlow framework. The RetinaNet-based model was trained and validated with both malignant and nonmalignant mammograms from the hospital and DDSM development datasets. The images and bounding boxes for the training and validation of the RetinaNet were prepared as follows: (i) Mammograms were downscaled to 800 pixels on the longest side while maintaining the aspect ratio. This pixel size was the minimum value of the longest side of the mammograms in the development datasets, so we downsized larger images in order to be able to include as many images as possible. (ii) The shorter side of the mammograms was padded black to 800 pixels. (iii) Bounding boxes were also resized to match each downscaled malignant mammogram. The mammograms and bounding boxes in the two development datasets were divided into training and validation with five-fold cross-validation. The RetinaNet was trained for 100 epochs, and the learning parameters when the value of the validation-loss function was the lowest was adopted. The learning progress of the DL-based model was monitored by both the value of the validation-loss function and the sensitivity of detection for breast cancers when the intersection over union (IoU) was set to 0.5. As optimizers, SGD and Adam were evaluated with their default parameters. All images were augmented using random rotation from –0.1 radians to 0.1 radians, with a random shift of 10% (80 pixels), a random shear of 10% (80 pixels), and random scaling from –10% (–80 pixels) to 10% (80 pixels), then flipped vertically and horizontally. The model was programmed to display bounding boxes on the area of suspected cancer in a mammogram, along with a malignancy likelihood ratio from 0 to 1. The model can adjust the number of boxes that are presented as well as the cut-off of the malignancy likelihood ratio of the proposed boxes. (S1 Fig) We have trained other AI models as well. Descriptions of these models are available in the supplementary materials in. ## Model performance test A lesion-based performance test was performed on the hospital and clinic test datasets. The test was performed as follows: (1) All mammograms were prepared as described for the training and validation of the model, steps (i) to (iii). (2) The trained DL-based model with the lowest validation-loss value was applied to these processed mammograms. (3) The overlap of the bounding box presented by the model and the radiologist annotated ground truth was calculated; this is known as the IoU. When the IoU was 0.3 or higher, the model had correctly identified the known malignancy. This IoU was chosen based on the results of a previous study. Until every ground truth was detected, the model continued to present the boxes from highest model-estimated malignancy to lowest, lowering the threshold of malignancy for presented boxes. These boxes and the malignancy likelihood ratios presented by the model were used to evaluate the detection performance. Additionally, an image-based performance test was performed on the hospital and clinic test datasets to assess the model’s ability to discriminate between malignancy and nonmalignancy. The DL-based model’s threshold of malignancy was determined by the Youden Index for this evaluation. The test was performed as follows: (1) All mammograms were prepared as described for the training and validation of the model, steps (i) to (iii). (2) The model was applied to these processed mammograms in the test datasets. (3) A malignant mammogram with annotations with an IoU greater than or equal to 0.3 for a ground-truth lesion was defined as a true positive image, a malignant mammogram with annotations with an IoU less than 0.3 for a ground-truth lesion was defined as a false negative image, a nonmalignant mammogram with no annotations on a mammogram was defined as a true negative image, and a nonmalignant mammogram with one or more annotations was defined as a false positive image. ## Statistical analysis In the lesion-based performance test, we evaluated whether the bounding boxes proposed by the model accurately identified malignant lesions in mammograms using the free-response receiver operating characteristic (FROC) curves. In the FROC, the vertical axis shows sensitivity; the horizontal axis shows mFPI. Thus, the FROC curve shows sensitivity as a function of the number of false positive lesions. Sensitivity was defined as the number of true positive lesions that the model presented divided by the number of all true positive lesions. The mFPI was defined as the number of false positive lesions that the model presented divided by the number of all mammograms in the dataset. Additionally, in the image-based performance test, we evaluated the model using the partial area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Two of the authors (D.U. and D.K.) performed all analyses using R, version 3.6.0. The FROC curves were plotted by R. All statistical inferences were performed with a two-sided 5% significance level. ### Patient and public involvement There was no direct patient or public involvement in this study. # Results ## Datasets The hospital development dataset included 3179 images (897 patients; age range, 25–97 years; mean age ± standard deviation, 58 ± 12 years) after excluding 367 images (170 MLO and 197 CC images) with no malignant findings. There were 1448 malignant and 1731 nonmalignant images. There were 1412 digital and 1767 scanned film images. Regarding breast density, 472 images were almost entirely in fat, 993 in scattered fibroglandular tissue, 999 in heterogeneously dense tissue, and 715 in extremely dense tissue. The malignant findings were as follows: 812 masses, 703 calcifications, 389 FAD, and 520 architectural distortions. The publicly available DDSM development dataset included a total of 1457 malignant images each with one bounding box. All images were collected from the CBIS-DDSM. In total, 4636 mammograms (2905 malignant and 1731 nonmalignant images) from the hospital and DDSM development datasets were used to develop the model. The hospital test dataset included a total of 491 images (139 patients; age range, 33–92 years; mean age ± standard deviation, 59 ± 13 years) after excluding 49 images (22 MLO and 27 CC images) without malignant findings on the affected mammograms. In total, there were 225 malignant and 266 nonmalignant images. Among these 491 images, there were 327 digital and 164 scanned film images. Regarding breast density, 74 images were almost entirely in fat, 180 in scattered fibroglandular tissue, 161 in heterogeneously dense tissue, and 76 in extremely dense tissue. In total, 230 breast cancers were detected in 225 malignant images (two malignant cancers were detected in five patients). The malignant findings were as follows: 103 masses, 83 calcifications, 74 FAD, and 93 architectural distortions. The clinic test dataset included a total of 2821 images (865 patients; age range, 32–84 years; mean age, 52 ± 8 years) after excluding 1358 images with no follow-up and one CC image with no malignant findings. There were 37 malignant and 2784 nonmalignant images. All images were digital. Regarding breast density, 435 images were almost entirely in fat, 962 in scattered fibroglandular tissue, 983 in heterogeneously dense tissue, and 441 in extremely dense tissue. No mammograms showed multiple cancers. The malignant findings were as follows: six masses, 19 calcifications, 11 FAD, and six architectural distortions. A flowchart of the eligibility criteria of the hospital and clinic datasets is shown in. Detailed demographic information of the development and test datasets is provided in Tables and, respectively. ## Model development The DL-based model was trained and validated on the two development datasets with five-fold cross-validation. The highest performance was observed when the optimizer used was Adam. The validation-loss function minima was obtained at 52 epochs. ## Model performance test The lesion-based performance of the DL-based model had a sensitivity of 1.00 with 0.47 mFPI in the hospital test dataset, and 1.00 with 0.45 mFPI in the clinic test dataset. The partial AUC with an mFPI of 1.0 was 0.93 (0.90–0.95) in the hospital dataset and 0.93 (0.90–0.96) in the clinic test dataset. Every malignancy detected was the lesion with the highest likelihood ratio in the mammogram. In cases in which there were two malignant findings in one mammogram, both lesions detected were the ones with the highest and second highest probability of malignancy. The most difficult cancers for the model to detect in the hospital and clinic test datasets are shown in. Although these lesions had the highest probability of malignancy in the mammograms, the malignancy likelihood ratios were the lowest of all true positive lesions (0.24 in the hospital test dataset and 0.33 in the clinic test dataset). Results applying other AI models are available in the supplementary materials in. The image-based performance showed that the accuracy, sensitivity, specificity, PPV, and NPV were 0.86 (0.83–0.89), 0.84 (0.79–0.89), 0.88 (0.83–0.91), 0.85 (0.80–0.90), and 0.87 (0.82–0.90), respectively, in the hospital test dataset, and 0.85 (0.84–0.87), 0.84 (0.68–0.94), 0.85 (0.84–0.87), 0.07 (0.05–0.10), and 1.00 (0.99–1.00), respectively, in the clinic test dataset. # Discussion The results of the present study indicated that the proposed DL-based model could accurately detect all breast cancers on mammograms with 0.47 mFPI in the hospital test dataset and 0.45 mFPI in the clinic test dataset. To our knowledge, the model developed in this research represents state-of-the-art performance for detecting breast cancer. In examining relevant prior research, we found fourteen studies proposing DL- based models designed for detecting breast cancers on mammograms (not only for classifying lesions as malignant or nonmalignant). Specifically, McKinney *et al*. achieved a multi-localization receiver operating characteristic of the partial AUC of 0.048 with a false positive rate of 10%. Even though they also used both normal and malignant images to train their model, our model has a lower mFPI and detects and classifies lesions at the same time rather than separately. Two studies had performance comparable to our model. The reported lesion-based sensitivity in these studies was 0.76–0.97, with an mFPI of 0.48–3.56. Ribli *et al*. achieved a sensitivity of 0.9 with a 0.3 mFPI for detecting breast cancer, while Jung *et al*. achieved a sensitivity of 0.86–1.00 with a 0.5–3.0 mFPI for detecting only mass lesions of breast cancer. Our model achieved a higher sensitivity and a lower mFPI than have been reported previously. Although it is difficult to compare the model performance because of the differences in the test datasets, possible explanations for the performance of our model are the size and composition of the development dataset and the DL architecture. Our model was developed with 4636 mammograms (2905 malignant and 1731 nonmalignant images), while Ribli *et al*. (2843 mammograms) and Jung *et al*. (116–632 mammograms) developed their models using only malignant mammograms. It is possible that development with a larger number, as well as both malignant and nonmalignant images, resulted in a lower mFPI due to our model learning more about normal features. With respect to the DL architecture, our model was developed using RetinaNet based on ResNet-152. RetinaNet is particularly useful when images for each of the classes (here malignant and nonmalignant) are likely to present in uneven numbers. Additionally, the variety of mammograms used to develop the model likely prevented overfitting. Overfitting is a result of learning that corresponds too closely to a particular development dataset and may therefore fail to fit additional data. In the present study, two datasets from different institutions were used, as were both converted-film and digital images. With regard to the image-based performance of our DL-based model, it was relatively difficult for our DL-based model to detect malignant findings in denser breast tissues and calcifications. Similar results have been reported in other studies. This is reasonable because the development datasets were annotated by radiologists, then the DL-based model extracted and learned features from these datasets. In other words, the performance of the model depends on the quality and quantity of the developing datasets. Another hypothesis for these difficulties is that malignant findings in denser mammary glands and calcifications are so subtle that they might have been lost when the mammograms were resized during the development process. Decreasing the compression ratio when developing model is worth investigating in the future. Since our trained model is open source, it is possible to efficiently re-train a part of the trained model with new mammograms which are closer to the cohort of intended use. Different countries and institutions have different cohorts of mammograms which may differ from those used to train the model for this study. Others may achieve better use of our trained model by fine-tuning it to fit their own purposes. The study described here is not without limitations. We found that the clinic test dataset was largely dominated by normal cases, but still not as many as the real screening cohort. The number of false positives may be higher in the real screening cohort and its impact should be considered. We developed and tested a model for the automated detection of breast cancer from mammograms using DL with RetinaNet. Our model was able to detect all breast cancers in the test datasets, regardless of type or tissue density, with a comparatively small mFPI. The trained model is open source and can be used worldwide. Our model is available free of charge with Apache License 2.0. # Supporting information We are grateful to Yoshikazu Hashimoto, Mitsuhiro Inomata, and Hiroko Osaki for technical assistance in regard to deep learning. We would also like to thank MedCity21 of the Osaka City University Hospital Advanced Medical Center for Preventive Medicine for joining our study. We thank Shannon Walston for proofreading our manuscript. [^1]: D.U. reports grants from Wellness Open Living Labs, LLC, during the conduct of the study; N.O. reports grants and personal fees from Bayer, grants and personal fees from Eisai, personal fees from Sanofi, personal fees from Aska, outside the submitted work; T.T. reports personal fees from Taiho Pharmaceutical Co., Ltd., personal fees from Chugai Pharmaceutical Co., Ltd., personal fees from Kyowa Hakko Kirin Co., Ltd., personal fees from Eisai Co., Ltd., personal fees from Pfizer Japan Inc., personal fees from Novartis Pharma K.K., personal fees from AstraZeneca K.K., personal fees from Takeda Pharmaceutical Co., Ltd., outside the submitted work; S.K. reports personal fees from Chugai Pharmaceutical Co., Ltd., personal fees from Eisai Co., Ltd., personal fees from Pfizer Japan Inc., personal fees from Novartis Pharma K.K., personal fees from Asahi Kasei Pharma Co., Ltd., personal fees from Medicon Inc., outside the submitted work; M.M. reports personal fees from Taisho Pharma Co., Ltd., personal fees from MOCHIDA PHARMACEUTICAL CO., LTD, personal fees from TSUMURA & CO., personal fees from Otsuka Pharmaceutical Co., Ltd., personal fees from Mylan EPD G.K., outside the submitted work; A.Y., S.N., T.M., S.H., M.S., T.S., K.K., H.T., K.M., T.H., A.S., and Y.M. have nothing to disclose.

# 1 Introduction Diseases that spread by transmission between individuals can give rise to epidemic waves that pass through a population. One infected person can infect several others who are susceptible to the infection, characterized by the basic reproduction number *R*₀, initially typically generating an exponential growth of the number of infections. The number of infections reaches a peak and later dies down when there is a sufficient number of individuals that have gained immunity after they recovered from the infection so that further growth is hampered. The fraction of immune individuals reached at the point when the epidemic starts to recede is called herd immunity. There are big uncertainties as to when and why an epidemic reaches its peak and the levels of herd immunity required. Simple models of infections dynamics predict that for an initially fast growing epidemic most of the population will become infected before the epidemic dies down. It was noted early by William Farr when investigating smallpox and other epidemics that epidemics appear to follow a general time course in the form of a skewed bell shaped curve. They first grow fast, reach a peak and then die down quickly, typically much before the majority of a population has been affected. The fact that an epidemic dies down is usually attributed to the fact that there exists some degree of immunity in the population. The uncertainty about when the peak of an epidemic is reached and why an epidemic dies out even if there remains a large number of still susceptible individuals reveals that the factors that limit an epidemic are not well understood. Furthermore, the effectiveness and impact of mitigation measures such as social distancing to counter a fast growing epidemic are not known. Simplified models of infection dynamics, such as the classic Susceptible- Infected-Removed (SIR) model have been used for a long time to describe the dynamics of epidemics spreading through a population. Such models capture key features of the epidemic as a nonlinear wave with qualitative properties that match observed bell-shaped dynamics of epidemic waves. However, more quantitatively, such models exhibit the robust feature that a quickly growing epidemic does not stop unless the majority of a susceptible population has reached immunity after going through the infection. This raises the question whether important factors are missing in these simple and elegant models. To understand at what conditions and at what levels epidemic waves become self- limiting and die down remains an important challenge. This aspect is also key to understand the role and effectiveness of social distancing measures to influence dynamics of an epidemic wave. Simple epidemic models treat the population as effectively consisting of identical individuals. However, individuals in a population can differ widely. The importance of population heterogeneity was put forward to understand smallpox epidemic which could not be captured by simple models. Such heterogeneity has been taken into account by adding details such as introducing several compartments to a model or by introducing distributions of susceptibility or infectiousness. It was suggested that population heterogeneity reduces effective herd immunity levels. In this paper, we present a generalization of the SIR model that takes into account effects of population heterogeneity. We show here that effects of heterogeneity can be added without losing the simplicity of the SIR model and keeping its mathematical structure. We introduce a single new parameter, the susceptibility exponent *α*, which characterizes a generic power-law heterogeneity in the distribution of infection susceptibilities of the population. Power laws are often found in nonlinear and complex systems. In the present context, power laws could be expected for example based on a variability of immune responses of different individuals which could imply a wide variability in the efficiency of the transmission of an infection. Furthermore, population heterogeneity could be relevant at very different scales, from the immune response of cells to the behaviors of individuals that affect infection rates. Such as broad range of relevant scales could give rise to approximately scale free properties or power laws. In the heterogeneous SIR model proposed here, the qualitative behaviors of the epidemic wave are unchanged. However, as a function of the parameter *α*, the wave can become self-limited at much lower levels of infected individuals as compared to the classic SIR model. In the limit of large *α* we recover the classic SIR model of homogeneous populations. For smaller *α* we find that the number of infections at the peak and the cumulative number of infections after the epidemic has passed can be strongly reduced. Our work has implications for the concept of herd immunity and clarifies that herd immunity cannot be discussed independently of population heterogeneity. We discuss the dynamics of the SARS-CoV-2 pandemics using the heterogeneous SIR model applied to data on reported infection numbers and COVID-19 associated deaths in Germany. We estimate parameter values including the susceptibility exponent *α* and show that the time course observed in Germany is compatible with different scenarios ranging from a homogeneous population strongly affected by mitigation to a self-limited epidemic wave in a heterogeneous population where social distancing measures play a minor role. # II. The Susceptible-Infected-Removed model The Susceptible-Infected-Removed (SIR) model captures key features of a spreading epidemic as a mean field theory based on pair-wise interactions between infected and susceptible individuals. This model captures generic and robust features without aiming to describe specific details. In the presence of *I* infected individuals in a population of *N* individuals, the infection can be transmitted to susceptible individuals. They stay infectious during an average time *γ*⁻¹ after which they no longer contribute to infections. The number of susceptible individuals *S* and the number of infected individuals *I* obey $$\begin{array}{rcl} \overset{˙}{S} & = & {- \beta\overline{x}\frac{IS}{N}} \\ \end{array}$$ $$\begin{array}{rcl} \overset{˙}{I} & = & {\beta\overline{x}\frac{IS}{N} - \gamma I\quad,} \\ \end{array}$$ where the dots denote time derivatives. The infection rate is denoted *β* and can in general depend on time *t*. This time dependence could correspond to seasonal changes or mitigation measures. The infection rate can be modulated by the average infection susceptibility $\overline{x}$ which we introduce to capture effects of population heterogeneity discussed below. In the classical SIR model $\overline{x} = 1$. The number of removed (or recovered) individuals is given by *N* − *S* − *I* and the cumulative number of infections is *C* = *N* − *S*. A key parameter is the basic reproduction number $$\begin{array}{r} {R_{0} = \frac{\beta}{\gamma}\quad,} \\ \end{array}$$ which denotes the average number of new infections generated by an infected individual. The growth rate of infections is $\overset{˙}{I}/I = \lambda(t) = \gamma\left( R(t) - 1 \right)$, where $R(t) = \beta\overline{x}S/(N\gamma)$ is a time dependent reproduction number. The time course of an epidemic is often provided as the number of new cases per day. This corresponds to the rate of new infections per unit time $$\begin{array}{r} {J = \beta\overline{x}\frac{IS}{N}} \\ \end{array}$$ with $J = \overset{˙}{C} = - \overset{˙}{S}$ and *R* = *J*/(*γI*). ## A. Infection dynamics in homogeneous populations In the simple case of a homogeneous population, all individuals have the same degree of susceptibility, *x* = 1 and the population average of *x* is $\overline{x} = 1$ independent of time. This is the classic SIR model. An example for a solution to these equations for homogeneous population $\overline{x} = 1$ and constant *β* is given in. The corresponding time dependent reproduction number is presented in. The number of infections first grows exponentially with growth rate $$\begin{array}{r} {\lambda_{0} = \gamma\left( R_{0} - 1 \right)\quad.} \\ \end{array}$$ As the number of susceptible decreases, the epidemic reaches a peak number of infected *I*_max = *I*(*t_I*) at time *t* = *t*_*I* with $\overset{˙}{I}\left( t_{I} \right) = 0$ and *R*(*t*_*I*) = 1. At this peak, a fraction *S*_*I*/*N* = 1/*R*₀ of individuals remain susceptible. The cumulative number of infections *C*_*I* at the maximum of *I* thus obeys $$\begin{array}{r} {\frac{C_{I}}{N} = 1 - \frac{1}{R_{0}}\quad.} \\ \end{array}$$ is the classic herd immunity level which is the fraction of immune individuals in the population beyond which the epidemic can no longer grow. Finally the epidemic dies down exponentially with rate $$\begin{array}{r} {\lambda_{\infty} = \gamma\left( \frac{R_{0}S_{\infty}}{N} - 1 \right)\quad,} \\ \end{array}$$ where $$\begin{array}{r} {\frac{S_{\infty}}{N} = - \frac{1}{R_{0}}W\left( - R_{0}e^{- R_{0}} \right)} \\ \end{array}$$ is the fraction of susceptible individuals that remain after long times, see Appendix A. Here *W*(*z*) denotes Lambert W-function. The total fraction of infections over the course of the epidemic is *C*_∞/*N* = 1 − *S*_∞/*N*. For a classic SIR model with homogeneous population we have for *R*₀ = 2.5, a herd immunity level *C*_*I*/*N* of 60% of the population, see. After the infection has passed *C*_∞/*N* ≃ 89% of the population have been infected, see (green lines). The fraction of the population that become infected increase for larger *R*₀. The SIR model thus suggests that for *R*₀ \> 2 the epidemic wave exceeds a majority of the population before the epidemic begins to die out. ## B. Infection dynamics with population heterogeneity Not all individuals are the same and for some susceptible individuals the probability of infection per time is lower than for others. This can be captured by a distribution of susceptibilities *x*. We denote *s*(*x*)*dx* the number of individuals with susceptibility between *x* and *x*+ *dx*. The total number of susceptible individuals is then $S(t) = \int_{0}^{\infty}dxs(x,t)$. For each sub-population *s*(*x*) with susceptibility *x*, the number of susceptible individuals decreases as $$\begin{array}{r} {\partial_{t}s = - \beta xs\frac{I}{N}\quad,} \\ \end{array}$$ which for the whole population implies with average susceptibility $$\begin{array}{r} {\overline{x}(t) = \frac{1}{S(t)}\int_{0}^{\infty}dx\; xs(x,t)\quad,} \\ \end{array}$$ which is in general time dependent. This time dependence can be discussed by introducing the variable *τ* that increases monotonically during the epidemic wave. It starts with *τ* = 0 and is defined via the equation $\overset{˙}{\tau} = \beta I/N$, which implies that it reaches a final value when the epidemic has decayed. Therefore *τ* can be interpreted as a measure of how far the epidemic has progressed. can then be written as ∂_*τ* *s* = −*xs*, and the number of susceptible individuals is $$\begin{array}{r} {S(\tau) = \int_{0}^{\infty}dxs_{0}(x)e^{- \tau x}\quad,} \\ \end{array}$$ where *s*₀(*x*) is the initial susceptibility distribution at time *t* = *t*₀ with average $\overline{x} = 1$, see Appendix B. ## C. Infection dynamics with generic power law heterogeneity The dynamics of epidemic waves depends on the shape of the initial distribution *s*₀(*x*). Here, we consider distributions that have the special property of shape invariance under the dynamics of epidemics. This property is satisfied by a gamma distribution $$\begin{array}{r} {s_{0}(x) \sim x^{- 1 + \alpha}e^{- \alpha x}\quad,} \\ \end{array}$$ which is governed by a power-law at small *x*, characterized by the exponent *α* \> 0, and a cut off at large *x*, see in Appendix C. The distribution *s*₀(*x*) has average $\overline{x} = 1$ and variance 1/*α*. Indeed, we have $s(x,t) = {\overline{x}}^{- 1 + \alpha}s_{0}\left( x/\overline{x} \right)$, where the time dependence enters via $\overline{x}(t)$, see Appendix C. This shape invariance implies that the gamma distribution is maintained at all times and is not merely an initial condition. Furthermore, starting with any initial distribution that exhibits a power law *s*₀(*x*)∼*x*^−1+*α* at small *x*, the distribution will converge for large *τ* to the shape invariant gamma distribution, which therefore is an attractor of the dynamics, see Appendix B. Note that in the limit of large *α*, we recover the classic SIR model for a homogeneous population. For small *α*, the population is strongly heterogeneous. By inserting the distribution given in into we obtain $$\begin{array}{r} {S(\tau) = \frac{N - I_{0}}{\left( 1 + \frac{\tau}{\alpha} \right)^{\alpha}}\quad,} \\ \end{array}$$ as detailed in Appendix C. The average susceptibility is $$\begin{array}{r} {\overline{x} = \frac{1}{1 + \frac{\tau}{\alpha}}\quad,} \\ \end{array}$$ which starts from $\overline{x} = 1$ for *τ* = 0 and decreases for increasing *τ*, thus dampening the epidemic. We can now express the dynamics given in Eqs and as two dynamic equations for *I*(*t*) and *τ*(*t*) which read $$\begin{array}{rcl} \overset{˙}{I} & = & {I\beta\left( 1 - \frac{I_{0}}{N} \right)\left( 1 + \frac{\tau}{\alpha} \right)^{- (\alpha + 1)} - \gamma I} \\ \end{array}$$ $$\begin{array}{rcl} \overset{˙}{\tau} & = & {\beta I/N\quad.} \\ \end{array}$$ with initial values *I*(0) = *I*₀, *τ*(0) = 0 and *S*(0) = *N* − *I*₀. The number of susceptible individuals at time *t* is then given by *S*(*t*) = *S*(*τ*(*t*)). An example of a time course of this model for *α* = 0.1 is shown in. We can discuss how the shape of the epidemic wave depends on the parameter *α*. The epidemic starts out with exponential growth of infected individuals at rate λ₀ = *γ*(*R*₀ − 1), with *R*₀ = *β*/*γ*. The time dependent reproduction number is $$\begin{array}{r} {R = {\overline{x}}^{1 + \alpha}R_{0}\quad.} \\ \end{array}$$ When the reproduction number drops to *R* = 1, the number of infected reaches a maximum $$\begin{array}{r} {\frac{I_{\text{max}}}{N} = 1 - \frac{1}{R_{0}} - \frac{1}{R_{0}}(1 + \alpha)\left( R_{0}^{\frac{1}{1 + \alpha}} - 1 \right)\quad.} \\ \end{array}$$ Beyond the herd immunity level given by the cumulative number of infections at the maximum of *I* $$\begin{array}{r} {\frac{C_{I}}{N} = 1 - R_{0}^{- \frac{\alpha}{\alpha + 1}}\quad,} \\ \end{array}$$ the reproduction number *R* drops below 1 and the epidemic dies down. In Eqs– we have considered the limit of small *I*₀/*N* for simplicity. The general derivation is given in Appendix C. In the limit of large *α*, these expressions converge to those obtained for the homogeneous SIR model, see Appendix C. The remaining fraction of susceptible individuals at the peak and after the epidemic has passed is shown as a function of *α* in. This reveals that as *α* is reduced, the fraction of the population reached by the epidemic decreases and can become very small for small *α*. At the same time the infections are more spread out over time and a larger fraction occurs after the peak when *α* is reduced, see. An important case is a strongly heterogeneous population. For small *α* ≪ 1, we obtain simple analytical expressions for the behavior of the system, see Appendix E. In this limit we have $I_{\text{max}}/N \simeq \alpha\left( \ln R_{0} + 1/R_{0} - 1 \right)$ and *C*_*I*/*N* ≃ *α* ln *R*₀. An important quantity is the rate *J* of new cases per time. For small *α* it takes the maximal value $$\begin{array}{r} {\frac{J_{\text{max}}}{N} \simeq \gamma\alpha\left( \left( R_{0} - 2 \right)e^{\frac{1}{R_{0}} - 1} + 1 \right)} \\ \end{array}$$ The final number of susceptible individuals is given by $$\begin{array}{r} {\frac{S_{\infty}}{N} = {\overline{x}}_{\infty}^{\alpha}\quad,} \\ \end{array}$$ where for small *α* the average susceptibility after the infection has passed is ${\overline{x}}_{\infty} \simeq - 1/\left( R_{0}W_{- 1}\left( - e^{- 1/R_{0}}/R_{0} \right) \right)$. Here *W*₋₁(*z*) denotes the −1 branch of Lambert *W* function. We finally have for small *α* $$\begin{array}{r} {\lambda_{\infty} = \gamma\left( R_{0}{\overline{x}}_{\infty} - 1 \right)\quad.} \\ \end{array}$$ A key result is that for small *α* the herd immunity level can be much below the classical value suggested by the SIR model. For example for *R*₀ = 2.5 and *α* = 0.1, we have *I*_max/*N* ≃ 2.8%, and a fraction *C*_*I*/*N* ≃ 8% of infected individuals required for herd immunity, much lower than is usually suggested. The total number of infected at long times is *C*_∞/*N* ≃ 14%, see Appendix C. The reason for the small amplitude of the epidemic wave in a heterogeneous population as compared to the amplitude for a homogeneous population is the drop of the average infection susceptibility $\overline{x}$. This drop occurs because in a heterogeneous population the most susceptible individuals are first removed, thereby lowering the average susceptibility. # III. Application to the SARS-CoV-2 epidemic in Germany We analyze the dynamics of the SARS-CoV-2 epidemic in Germany using public data provided by the Robert Koch institute. These daily reports provide the numbers of reported positive tests for each day, but also the dates of reporting of those infections which later turn out as fatal. The total number of new reported infections per day $J_{\text{rep}}^{t}$ (red symbols) are shown in together with the number of reported infections per day that were later fatal (blue symbols), which we denote $J_{\text{rep}}^{f}(t)$. Both sets of data can be interpreted as proxies for the rate *J* of new cases per day up to an unknown factor. They show qualitatively similar behavior, a rapid growth and a decline after passing a maximum. However there are quantitative differences, in particular the growth rates at early and late times, given by the slopes of the data in a single logarithmic plot are different, see. The number of new cases per day that are later fatal $J_{\text{rep}}^{f}(t)$ is related to the number of new infections per day as $J_{\text{rep}}^{f}(t) = Jf$, where *f* denotes the infection fatality rate, the fraction of infections that are fatal, which we consider to be constant for simplicity. ## A. Comparison to heterogeneous SIR model The calculated number of new cases per day *J* obtained as solution to Eqs and for a heterogeneous population and scaled by the factor *f* to match the data of fatal cases are shown in as solid blue lines. These lines are shown together with the number $J_{\text{rep}}^{f}$ of new fatal cases per day as blue symbols. The factor *f* was determined such that the cumulated cases per day $J_{\text{rep}}^{f}/f$ matches the cumulative number of cases *C* on June 15. The time axis is chosen such that the model matches the data. From a fit of the model to the data we obtain the parameter estimates *R*₀ ≃ 2.67 and *γ* ≃ 0.146 day⁻¹. Good fits to the data are found for a range of *α* sufficiently small, about *α* \< 0.2. The resulting infection fatality rates *f* vary as *α* is changed. Using *α* = 0.05 corresponds to an infection fatality rate *f* ≃ 0.13%. It could be larger or smaller if a different value of *α* was used. This would not significantly affect the quality of the fit as long as *α* \< 0.2. The calculated time courses *I*(*t*), *J*(*t*) and *C*(*t*) corresponding to these fits to $J_{\text{rep}}^{f}$ are shown in (e). The dependence of the average susceptibility $\overline{x}$ on time and the function *τ*(*t*) are shown in (f). The increase of *τ* with time represents the advance of the epidemic. It reaches a final value at long times as discussed in Appendix C. The inset in (f) shows the shape of the distribution of susceptibility in the population at different stages characterized by different values of *τ*. It is surprising that the model fits the data of fatal cases with just two fit parameters while yielding a reasonable infection fatality rate. This is further clarified when using the fit values of *R*₀ and *γ* to calculate λ₀ ≃ 0.24 day⁻¹, slightly smaller than the estimate given in. Using, we also find λ_∞ ≃ −0.069 day⁻¹, very close to the estimate from the data. The data of all reported cases can also be captured by the model for small *α*, see red symbols and red lines. This fit is not as close and the parameter values are different, see. Our comparison of the model to the data shows that the model captures the time course of fatal cases surprisingly well for the case of strong heterogeneity for infection fatality rates that fall in the range of estimates from immunological studies. ## B. Quantification of the shape of the epidemic wave In order to understand how the shape of the wave of infections constrains the possible parameter values of *R*₀, *γ* and *α*, we consider in addition to the initial growth rate λ₀ and the final decay rate λ_∞ two coefficients describing the epidemic dynamics near its peak, using the expansion $$\begin{array}{r} {\ln J(t) \simeq \ln J_{\text{max}} + \frac{A_{2}}{2}\left( t - t_{J} \right)^{2} + \frac{A_{3}}{6}\left( t - t_{J} \right)^{3}\quad,} \\ \end{array}$$ where the linear term disappears by definition at the maximum *J*_max = *J*(*t_J*) at time *t*_*J*. The coefficients $A_{2} = \overset{¨}{J}\left. /J \right|_{t = t_{J}}$ and $A_{3} = \dddot{J}\left. /J \right|_{t = t_{J}}$ can be obtained for the homogeneous and heterogeneous SIR model, see Appendix C, D and E. shows dimensionless combinations of these values as a function of *R*₀ for different *α* ranging from the homogeneous case *α* → ∞ to strongly heterogeneous with *α* → 0 as solid lines of different color. The values obtained from the fits shown in are indicated as dashed lines together with shaded regions corresponding to estimated uncertainty ranges of these values. We find that the ratio $A_{2}/\lambda_{\infty}^{2}$, which is independent of *γ* depends only weakly on *α*. We can therefore use it to estimate *R*₀, see. Using *A*₂ ≃ −0.01 day⁻² and λ_∞ ≃ −0.07 day⁻¹ determined from the data of fatal cases, we have $A_{2}/\lambda_{\infty}^{2} \simeq 2.0$ leading to the estimate *R*₀ ≃ 2.5, see. This estimate can now be used to infer bounds on *α*. The ratio λ_∞/λ₀ ≃ −0.3 is only consistent with *R*₀ ≥ 2.6 and the lower value corresponds to the limit of small *α*, see. This reveals that *α* ≪ 1 must be small and that the classic SIR model with homogeneous population is not consistent with this data. We can now estimate *γ* using the small *α* limit. For *R*₀ ≃ 2.6, we have *γ* ≃ 2λ_∞ ≃ 0.14 day⁻¹, see. The data does not provide information about the true total number of infections. Therefore the precise value of *α* remains unknown. We can use estimates from immunological studies estimating the number of infections to determine *α*. This suggests a range of about 0.01 \< *α* \< 0.15, corresponding to 0.65%\>*f* \> 0.04%. also shows the estimated ranges for data on all reported cases in red. For this case the inferred values of *R*₀ is larger and the consistency with the data is less strong. ## C. Effects of mitigation and social distancing measures During an epidemic conditions can change over time. For example, mitigation by social distancing measures, quarantining or seasonal changes could affect how quickly an infection spreads on average from person to person. Given that such changes are global, they may be captured by a time-dependence of the rate *β*(*t*). In the following, we discuss scenarios of mitigated epidemics, starting from a reference point with an initial infection rate *β*₀ prior to mitigation. We use this reference to define the herd immunity *C*_*I* of the population via. The herd immunity level depends on the basic reproduction number *R*₀ = *β*₀/*γ* and on the population heterogeneity *α*. For immunity levels above herd immunity, *C* ≥ *C*_*I*, the population is stable after mitigation measures are completely relaxed and *β* is restored to its original value *β*₀. We examine three different scenarios with a comparable total number of infections. These scenarios are shown in. They are characterized by different levels of immunity relative to herd immunity at July 1 and thus differ in the future behaviors beyond this time. Starting in all scenarios with *R*₀ = 2 and using *γ* = 0.24, the model follows the initial growth at rate λ₀ of the reported cases. If *β* is kept constant, *β* = *β*₀, the model deviates from the data at later times, see dotted lines in. If *β* is permitted to change in time, almost any reported time course could be described by the model. We use the data to infer a time course of *β*(*t*) such that the model follows the data, see Appendix G. The inferred values of *β* are shown as circles in. In order to fit the model to the data in different mitigation scenarios, we use a piecewise linear modulation of *β*. The time dependence of *β*(*t*) that resulted from these fits are shown in as solid lines. The value of *β* decreases sharply at the onset of mitigation. After this decrease it stays roughly constant or increases at constant rate, thus relaxing mitigation. The magnitude of maximal mitigation and the two slopes of *β*(*t*) were used as fit parameters. In the case of early mitigation, fast reduction of *β* suppresses the epidemic before any appreciable progress towards herd immunity was made. Mitigation needs to be strong and sustained to be compatible with the data. By July, the population reaches only about 6−7% of herd immunity in this case. Note that this is the only scenario of the classic SIR model with a homogeneous population (*α* → ∞) that could be compatible with the data. For heterogeneous populations with *α* ≲ 0.2, scenarios with milder mitigation and with infection levels closer to herd immunity are compatible with the data, see. A case of moderate mitigation with *α* = 0.1 is shown in. The population in this case reaches by July 1st about ∼45% of herd immunity. A sustained mitigation is needed to account for the data, albeit with smaller magnitude compared to the first case. If the epidemic starts slightly earlier (3 days for the case shown in as compared to (d-f)), the population reaches ∼95% of herd immunity by July 1st. Here, mitigation has the effect to reduce the cumulative number of infections as compared to a non-mitigated case (*C*/*N* = 5.8% compared to 11% by July 1st). This reduction of cumulative infections *C* results from a reduction of the number of infectious individuals *I* at the point when herd immunity is reached. In the absence of mitigation, *I* reaches its maximum when *C* = *C*_*I*, whereas mitigation can reduce *I* to small numbers as herd immunity is reached, preventing further infections. The minimal number of infections that can be achieved by temporary mitigation is *C*_*I*, which is up to 50% smaller than the long lime limit *C*_∞ in an unmitigated epidemic. The scenarios of temporary reduction of *β* could capture the mitigation effects of social distancing measures. To relate the inferred time dependence of *β* to measures of social activity, we show in, *β*(*t*) together with mobility data from Ref. for comparison, see Appendix H. This mobility data shows a sharp decline and a slow but steady return to the initial state roughly in line with inferred changes of *β*(*t*). The three scenarios differ in the fraction of herd immunity they reach by July 1 and therefore in their future trajectories. However, *β*(*t*) was adjusted by a fitting procedure such that all scenarios are consistent with the data on reported infections. This reveals that it can be difficult to distinguish effects of heterogeneity leading to a time dependent average susceptibility $\overline{x}$ from mitigation effects corresponding to time-dependent *β*. Indeed our analysis shows that changes in mitigation strength can be compensated to some degree by changes of heterogeneity described by *α*. # IV. Conclusions and perspectives We have presented a generalization of the classic Susceptible-Infected-Removed model for epidemic waves, which adds one new parameter to the model that captures population heterogeneity by a power-law exponent *α*. This exponent describes the power law that characterizes the distribution of susceptibility in the population *s*(*x*) ∼ *x*^−1+*α* for small *x*. A special case for such distributions is the gamma distribution. Gamma distributions have been used before to describe heterogeneous populations and an approach similar to the one presented here has been proposed in where also the shape invariance of gamma distributions is mentioned. Here, we have shown that gamma distributions have the special properties that they are both shape invariant under the dynamics and attractors of the dynamics for power-law distributions. This implies that for each *α* there exists a class of distributions with the same power law at small *x* which share the same limiting dynamics and distribution. The generalization of the SIR model introduced here captures the effects of these power laws by the parameter *α* in a generic way. Note that this generalization does not change the simplistic nature of the SIR model and does not change its numerical or analytical complexity. For *α* \> 1, population heterogeneity is weak and in the limit of large *α*, one recovers the classic SIR model of homogeneous population, see. For *α* \< 1 population heterogeneity plays a key role in limiting the peak of the epidemic wave. We show that as a result of strong population heterogeneity (small *α*), the wave peaks when only a small minority of individuals have been infected, see. The herd immunity level, the point where the epidemic dies down spontaneously becomes very small for small *α*, see. Thus our model shows that for small *α*, an epidemic wave can die out after reaching only a small fraction of the population even though a majority of the population is still susceptible. In this case the population is stable with respect to introducing new infected individuals because the average susceptibility $\overline{x}$ has dropped significantly, see. Many properties of the nonlinear wave in this generalized model can be obtained exactly as a function of *α* and in the limit of small *α*. Numerical solutions can be generated quickly and efficiently. In a heterogeneous population the average susceptibility $\overline{x}$ stays almost constant at early stages of the epidemic where the number of new cases grows exponentially with rate λ₀ = *γ*(*R*₀ − 1). At this stage the dynamics is the same as in the classic model and independent of *α*. However, $\overline{x}$ then drops rather quickly and the epidemic waves thus reaches its peak and dies down, see. This sudden drop in average susceptibility results from a shift of the distribution of susceptibility. The most susceptible individuals are removed from the dynamics at higher rates than those with low susceptibility. This leads to a rapid reduction of the average susceptibility until it has dropped to a low value where the time dependent reproduction number *R* falls below 1, see. The wave then dies down at rate λ_∞ and the average susceptibility approaches a final value ${\overline{x}}_{\infty}$. Thus the qualitative behavior of the classic SIR model is unchanged and the key parameters, the recovery rate *γ* and the basic reproduction number *R*₀ have the same values and properties. However, the power-law distribution of susceptibility can dramatically change the peak of the epidemic and alters the precise shape of the wave. The dynamics effectively shifts the edge of the susceptibility distribution, see inset in, which changes the stability of the population from prone to an exponentially growing wave to a stably decaying wave without requiring a large number of infections. The simple SIR model does not aim to capture details such as the population structure, the geography or human travel. In the spirit of statistical physics it is based on the idea that the collective behaviors of many individuals give rise to an emergent epidemic wave with robust and generic features that can be captured by a simplified model that focuses on key aspects. Here we show that power-law heterogeneity is a key factor that should not be left out. Note that our approach to take into account population heterogeneity can also be applied to extensions and generalizations of the SIR model such as the SEIR and the SIRS models. We apply our model to data on the SARS-CoV-2 epidemic in Germany in 2020. The data from Germany provides time stamps on reporting dates of infections and reporting dates of infections that later are fatal. Surprisingly, the data for fatal cases is well described by the heterogeneous SIR model with constant parameters and small *α* but not by the classic SIR model with constant parameters. In the case of SARS-CoV-2, immunological data suggests that only a minority of the population exhibits antibodies. This is consistent with a fit of our model to the data using a small value of *α*. The data on all reported cases can also be captured by the model, but the fit is less convincing. Comparing the data on all reported cases to the data on the time course of cases that are fatal reveals some differences. Clearly the fatal cases represent a different sampling as these cases correspond to predominantly old individuals and therefore measure a different quantity. However, starting from all reported cases and then using the fatal outcome as a second criterion could reduce biases due to changes in testing rates, testing strategies as well as testing errors. An epidemic wave does not progress under constant conditions but is subject to changes such as mitigation measures and seasonal effects. We use our model in comparison with the data from Germany to investigate different scenarios of mitigation that correspond to different level of immunity in the population. In the case of a homogeneous population the data can only be accounted for if mitigation is strong and suppresses the epidemic far below herd immunity. This scenario further requires mitigation to be sustained and it leads to a fragile and unstable state when mitigation measures are relaxed. In the case of heterogeneous populations, intermediate scenarios are possible which stay below herd immunity or just reach herd immunity, see. In the latter example shown in, mitigation effectively reduces the total number of infections by keeping immunity just at herd immunity level, leading to a stable state where mitigation can be safely relaxed. This is a desirable outcome because the number of infections could be reduced by mitigation by up to 40% without the need of sustaining mitigation, see. When discussing rapidly evolving epidemics such as SARS-CoV-2, herd immunity is often not considered to be reachable as it is predicted to require an unacceptably high fraction of cumulative infections. Interestingly, the picture changes dramatically if a strongly heterogeneous population is considered. In this case herd immunity can be reached rather quickly while a large majority of the population is still susceptible. This raises the question of what are the features that are variable and that give rise to heterogeneity and how widely they are expected to vary in the population. One possibility is that differences of susceptibility stem from differences in the abilities of immune systems of susceptible individuals to react to a new pathogen. In addition to adaptive immunity related to the presence of specific antibodies, many individuals show a T-cell response to SARS-CoV-2. This response could for example be due to less specific cross reactions related to earlier encounters with related viruses. Here we have focused on data from Germany until July 2020 because it provides detailed information that is not available in most countries. Furthermore, Germany has relatively few reported infections and deaths per capita. Our work shows that even this rather mild manifestation of the epidemic can be captured by a heterogeneous SIR model with mild mitigation. The data of newly infected cases with fatal outcome can even be explained in a strongly heterogeneous population without considering any mitigation effects at all. This implies that in order to quantify the effects of mitigation, population heterogeneity has to be taken into account. In order to disentangle effects from heterogeneity and from mitigation the combination of different types of information is important. For example, analyzing in different countries the circumstances under which different sero-prevalence levels or multiple epidemic waves are observed could be key to understand the roles of mitigation and heterogeneity. # Appendix A: Properties of the homogeneous SIR model For a homogeneous population with $\overline{x} = 1$, the SIR model given in Eqs can be written as $$\begin{array}{rcl} \overset{˙}{I} & = & {\left( 1 - \frac{I_{0}}{N} \right)\beta Ie^{- \tau} - \gamma I} \\ \end{array}$$ $$\begin{array}{rcl} \overset{˙}{\tau} & = & {\beta\frac{I}{N}} \\ \end{array}$$ with *S* = (*N* − *I*₀)exp(−*τ*), *I*(0) = *I*₀ and *τ*(0) = 0. We therefore have $dI/d\tau = \overset{˙}{I}/\overset{˙}{\tau} = \left( N - I_{0} \right)e^{- \tau} - N/R_{0}$, where *R*₀ = *β*/*γ*. For constant *β* this implies $$\begin{array}{r} {I(\tau) = I_{0}e^{- \tau} + N\left( 1 - e^{- \tau} \right) - \frac{N\tau}{R_{0}}\quad.} \\ \end{array}$$ We can eliminate *τ* and find $$\begin{array}{r} {\frac{I}{N} = 1 - \frac{S}{N} + \frac{1}{R_{0}}\ln\frac{S}{N - I_{0}}\quad.} \\ \end{array}$$ The maximum *I*_max = *I*(*τ_I*) with *I*′(*τ*_*I*) = 0, where the prime denotes a *τ*-derivative, occurs for $$\begin{array}{r} {\tau_{I} = \ln\left( R_{0}\left( 1 - \frac{I_{0}}{N} \right) \right)\quad.} \\ \end{array}$$ Therefore we the maximal number of infected individuals reads $$\begin{array}{r} {I_{\text{max}} = N\left\lbrack 1 - \frac{1}{R_{0}} - \frac{1}{R_{0}}\ln\left( R_{0}\left( 1 - \frac{I_{0}}{N} \right) \right) \right\rbrack\quad.} \\ \end{array}$$ At the maximum *I*_max, we have *dI*/*dS* = 0, which implies $$\begin{array}{r} {\frac{S\left( \tau_{I} \right)}{N} = \frac{1}{R_{0}}\quad.} \\ \end{array}$$ At long times, the infection dies out when *I*(*τ*_∞) = 0 with (1 − *I*₀/*N*)exp(−*τ*_∞) = 1 − *τ*_∞/*R*₀ and *S*_∞ = (*N* − *I*₀)exp(−*τ*_∞). We therefore have *S*_∞/*N* = 1 + ln(*S*_∞/(*N* − *I*₀))/*R*₀ and $$\begin{array}{r} {\frac{S_{\infty}}{N} = - \frac{1}{R_{0}}W\left( - R_{0}e^{- R_{0}}\left( 1 - \frac{I_{0}}{N} \right) \right)\quad,} \\ \end{array}$$ where *W*(*z*) denotes the 0-branch of Lambert W-function. The time dependent solution *τ*(*t*) can be obtained from *Ndτ*/*I*(*τ*) = *βdt* via $$\begin{array}{r} {\int_{0}^{\tau}\frac{d\tau^{\prime}}{1 - \left( 1 - I_{0}/N \right)e^{- \tau^{\prime}} - \tau^{\prime}/R_{0}} = \beta t\quad.} \\ \end{array}$$ To discuss empirical data, we consider the time-course of the rate of new cases *J* = *βIS*/*N*. We have $\overset{˙}{J}/J = \beta K$ with $$\begin{array}{r} {K = \left( 1 - \frac{I_{0}}{N} \right)e^{- \tau} - \frac{1}{R_{0}} - \frac{I}{N}\quad.} \\ \end{array}$$ The maximum *J*_max = *J*(*τ_J*) is reached for *K*(*τ*_*J*) = 0, which implies $$\begin{array}{r} {\tau_{J} = W_{- 1}\left( - \left( 1 - \frac{I_{0}}{N} \right)2R_{0}e^{- (R_{0} + 1)} \right) + R_{0} + 1\quad,} \\ \end{array}$$ where *W*₋₁(*z*) denotes the −1 branch of the Lambert *W*-function with *W*(*z*)*e*^*W*(*z*) = *z*. At the maximum *J*_max of *J* we have $\overset{˙}{S}I + \overset{˙}{I}S = 0$ and therefore $$\begin{array}{rcl} {S\left( \tau_{J} \right)} & = & {- \frac{1}{2R_{0}}W_{- 1}\left( - 2R_{0}e^{- 1 - R_{0}} \right)} \\ \end{array}$$ $$\begin{array}{rcl} {I\left( \tau_{J} \right)} & = & {S\left( \tau_{J} \right) - \frac{1}{R_{0}}} \\ \end{array}$$ and finally $$\begin{array}{r} {J_{\text{max}} = \beta S\left( \tau_{J} \right)\left( \frac{S\left( \tau_{J} \right)}{N} - \frac{1}{R_{0}} \right)\quad.} \\ \end{array}$$ Near the maximum of the rate *j*_max with $\overset{˙}{J} = 0$ we have $A_{2} = \overset{¨}{J}\left. /J \right|_{t = t_{J}}$ and $A_{3} = \dddot{J}\left. /J \right|_{t = t_{J}}$. For the homogeneous SIR model, we have $$\begin{array}{rcl} A_{2} & = & {- \frac{\gamma^{2}}{2}\left\lbrack 1 + W_{- 1}\left( - 2R_{0}e^{- 1 - R0} \right) \right\rbrack\left\lbrack 2 + W_{- 1}\left( - 2R_{0}e^{- 1 - R0} \right) \right\rbrack} \\ \end{array}$$ $$\begin{array}{rcl} A_{3} & = & {\frac{\gamma^{3}}{4}\left\lbrack 2 + W_{- 1}\left( - 2R_{0}e^{- 1 - R0} \right) \right\rbrack^{2}\quad.} \\ \end{array}$$ # Appendix B: Distributions of infection susceptibility in the population For an initial distribution *s*₀(*x*) of susceptible individuals with susceptibility *x*, we define $S(\tau) = \int_{0}^{\infty}dxs_{0}(x)e^{- \tau x}$. We can then write the dynamics of the epidemic spreading given in Eqs and as two equations for *I*(*t*) and *τ*(*t*) $$\begin{array}{rcl} \overset{˙}{I} & = & {- \beta\frac{I}{N}\frac{dS}{d\tau} - \gamma I} \\ \end{array}$$ $$\begin{array}{rcl} \overset{˙}{\tau} & = & {\beta\frac{I}{N}} \\ \end{array}$$ with initial values *I*(0) = *I*₀, *τ*(0) = 0 and *S*(0) = *N* − *I*₀. The number of susceptible at time *t* is then given by *S*(*t*) = *S*(*τ*(*t*)). Defining the cumulant-generating function *Γ*(*τ*) = −ln *S*(*τ*), we have $$\begin{array}{r} {\overline{x} = \frac{d}{d\tau}\Gamma} \\ \end{array}$$ and the nth cumulant of *x* is for *n* \> 1 given by $$\begin{array}{r} {\left\langle x^{n} \right\rangle_{c} = ( - 1)^{n + 1}\frac{d^{n}}{d\tau^{n}}\Gamma\quad.} \\ \end{array}$$ The classic case with homogeneous population then corresponds to *s*₀(*x*) = (*N* − *I*₀)*δ*(*x*), see Appendix A. Here we consider distributions which exhibit a power-law behavior for small *x* with *s*₀ ∼ *x*^−1+*α*. For *α* \> 0 the power law must be cut off at large *x* \> *x*₀ for the distribution to be normalizable. From ∂_*τ* *s* = −*xs*, we have *s*(*x*, *τ*) = *s*₀(*x*)*e*^−*τx* which for *τ* ≫ 1/*x*₀ approaches *s*(*x*, *τ*)∼*x*^−1+*α* *e*^−*τx*. The moments of this distribution can be obtained from the cumulant generating function $\Gamma(\tau) = - \ln\int_{0}^{\infty}dxx^{- 1 + \alpha}e^{- \tau x} = - \alpha\ln\tau + \text{const.}$ The cumulants of this distribution are 〈*x*^*n*〉_*c* = *ατ*^−*n*(*n* − 1)!. Using $\overline{x} = \alpha/\tau$, the susceptibility thus approaches for large *τ* the limiting distribution $$\begin{array}{r} {s(x,\tau) \sim x^{- 1 + \alpha}e^{- \alpha x/\overline{x}}\quad,} \\ \end{array}$$ which is the gamma distribution. # Appendix C: The generalized SIR model with population heterogeneity We have shown in Appendix B that for susceptibility distribution with a power law at small *x* the gamma distribution is an attractor of the dynamics. We therefore choose at time *t* = 0 a gamma distribution with average $\overline{x} = 1$ as initial condition. It is given by $$\begin{array}{r} {s_{0}(x) = \left( N - I_{0} \right)\frac{\alpha^{\alpha}}{(\alpha - 1)!}x^{- 1 + \alpha}e^{- \alpha x}\quad.} \\ \end{array}$$ Here (*α* − 1)! denotes Euler’s gamma function. Note that in the limit of large *α*, this approaches the homogeneous case with *s*₀(*x*)≃*δ*(*x* − 1). We then have *S*(*τ*) = (*N* − *I*₀)(1 + *τ*/*α*)^−*α* and *S*′ = −(*N* − *I*₀)(1 + *τ*/*α*)^−(1+*α*), where the prime denotes a derivative with respect to *τ*. As time evolves, the shape of the distribution *s*(*x*, *t*) is time independent. Indeed, with ∂_*τ* *s* = −*xs* we have *s*(*x*, *τ*) = *s*₀(*x*)*e*^−*τx* and thus $$\begin{array}{r} {s(x,\tau) = \left( N - I_{0} \right){\overline{x}}^{- 1 + \alpha}f\left( x/\overline{x} \right)} \\ \end{array}$$ with $\overline{x}(\tau) = (1 + \tau/\alpha)^{- 1}$. The time-invariant distribution is then given by $$\begin{array}{r} {f(z) = \frac{\alpha^{\alpha}}{(\alpha - 1)!}z^{- 1 + \alpha}e^{- \alpha z}\quad.} \\ \end{array}$$ The dynamic equation of the heterogeneous SIR model read $$\begin{array}{rcl} \overset{˙}{I} & = & {I\beta\left( 1 - \frac{I_{0}}{N} \right)\left( 1 + \frac{\tau}{\alpha} \right)^{- (\alpha + 1)} - \gamma I} \\ \end{array}$$ $$\begin{array}{rcl} \overset{˙}{\tau} & = & {\beta I/N\quad.} \\ \end{array}$$ Defining $I^{\prime} = \overset{˙}{I}/\overset{˙}{\tau}$, we have $$\begin{array}{r} {I^{\prime} = \left( N - I_{0} \right)\left( 1 + \frac{\tau}{\alpha} \right)^{- (\alpha + 1)} - \frac{N}{R_{0}}\quad.} \\ \end{array}$$ For constant *β*, we then have by integrating over *τ* $$\begin{array}{r} {I = I_{0}\left( 1 + \frac{\tau}{\alpha} \right)^{- \alpha} + N\left( 1 - \left( 1 + \frac{\tau}{\alpha} \right)^{- \alpha} \right) - \frac{N\tau}{R_{0}}\quad.} \\ \end{array}$$ The maximum of *I* is reached for *τ* = *τ*_*I* with *I*′ = 0 and thus $$\begin{array}{r} {\left( 1 + \frac{\tau_{I}}{\alpha} \right)^{\alpha + 1} = \left( 1 - \frac{I_{0}}{N} \right)R_{0}\quad.} \\ \end{array}$$ We thus obtain $$\begin{array}{r} {\frac{I_{\text{max}}}{N} = 1 - \frac{1}{R_{0}} - \frac{1}{R_{0}}(1 + \alpha)\left\lbrack \left( \left( 1 - \frac{I_{0}}{N} \right)R_{0} \right)^{\frac{1}{1 + \alpha}} - 1 \right\rbrack\quad.} \\ \end{array}$$ The herd immunity level *C*_*I*/*N* is obtained by using *C*_*I* = *N* − *S*_*I*, where *S*_*I* is derived using together with. It follows that $$\begin{array}{r} {\frac{C_{I}}{N} = 1 - \frac{N - I_{0}}{N}\left( \left( 1 - \frac{I_{0}}{N} \right)R_{0} \right)^{- \frac{\alpha}{\alpha + 1}}\quad.} \\ \end{array}$$ The epidemic ends at long times for *I*(*τ*_∞) = 0, for which $$\begin{array}{r} {\left( 1 - \frac{I_{0}}{N} \right)\left( 1 + \frac{\tau_{\infty}}{\alpha} \right)^{- \alpha} = 1 - \frac{\tau_{\infty}}{R_{0}}} \\ \end{array}$$ with *S*_∞ individuals that remain susceptible. This quantity obeys $$\begin{array}{r} {R_{0}\frac{S_{\infty}}{N} + \alpha\left( 1 - \frac{I_{0}}{N} \right)^{\frac{1}{\alpha}}\left( \frac{S_{\infty}}{N} \right)^{- \frac{1}{\alpha}} = R_{0} + \alpha\quad.} \\ \end{array}$$ We then find $$\begin{array}{r} {\frac{S_{\infty}}{N} = \frac{R_{0} + \alpha}{R_{0}}F\left( \frac{\alpha}{R_{0} + \alpha}\left\lbrack \frac{R_{0}\left( 1 - \frac{I_{0}}{N} \right)}{R_{0} + \alpha} \right\rbrack^{\frac{1}{\alpha}},\frac{1}{\alpha} \right)\quad,} \\ \end{array}$$ Where the function *F*(*z*, *ν*) is defined as the inverse of the function *x*^*ν*(1 − *x*) via the condition *F*^*ν*(1 − *F*) = *z*. Finally, using *Ndτ*/*I*(*τ*) = *βdt*, the time dependent solution *τ*(*t*) can be written as $$\begin{array}{r} {\int_{0}^{\tau}\frac{d\tau^{\prime}}{1 - \left( 1 - I_{0}/N \right)\left( 1 + \tau^{\prime}/\alpha \right)^{- \alpha} - \tau^{\prime}/R_{0}} = \beta t\quad.} \\ \end{array}$$ # Appendix D: Dynamics of the rate of daily new cases Data on the dynamics of the epidemic typically provides information about new cases reported per day. We therefore consider the rate of nex cases *J* = −*βIS*′/*N*, where the prime denotes a *τ* derivative. Using *I*′ = −*S*′ − *N*/*R*₀, we have $$\begin{array}{r} {I(\tau) = I_{0} + S(0) - S(\tau) - \frac{N\tau}{R_{0}}\quad.} \\ \end{array}$$ We then write $$\begin{array}{r} {\frac{\overset{˙}{J}}{J} = \beta K} \\ \end{array}$$ with $$\begin{array}{r} {K = - \frac{S^{\prime}}{N} - \frac{1}{R_{0}} + \frac{I}{N}\frac{S^{\prime\prime}}{S^{\prime}}\quad.} \\ \end{array}$$ We then have $$\begin{array}{r} {K^{\prime} = - \frac{S^{\prime\prime}}{N} + \frac{I^{\prime}}{N}\frac{S^{\prime\prime}}{S^{\prime}} + \frac{I}{N}\frac{S^{\prime\prime\prime}S^{\prime} - S^{\prime\prime 2}}{S^{\prime 2}}} \\ \end{array}$$ The maximum of *J* occurs at *τ* = *τ*_*J* with *K*(*τ*_*J*) = 0. We thus have $A_{2} = \left( \overset{¨}{J}/J \right)|_{t = t_{j}} = \beta\overset{˙}{K}$ and $A_{3} = J^{- 1}\left( d^{3}J/dt \right)|_{t = t_{j}} = \beta\overset{¨}{K}$. Plugging in *S*(*τ*) = *N*(1 + *τ*/*α*)^*α* for the heterogeneous SIR model with *I*₀ ≪*N*, yields $$\begin{array}{rc} K & {= \left( 1 + \frac{\tau}{\alpha} \right)^{- 1}\left( \left( 1 + \frac{\tau}{\alpha} \right)^{- \alpha} - \frac{\alpha + 1}{\alpha}\frac{I}{N} \right) - \frac{1}{R_{0}}} \\ \end{array}$$ $$\begin{matrix} & {= \frac{1}{\alpha + \tau}\left( \left( 1 + \frac{\tau}{\alpha} \right)^{- \alpha}(2\alpha + 1) - (\alpha + 1)\left( 1 - \frac{\tau}{R_{0}} \right) \right) - \frac{1}{R_{0}}} \\ \end{matrix}$$ At the maximum in *J*, we have *K* = 0, yielding $$\begin{array}{r} {\frac{2\alpha + 1}{(\alpha + 1)\left( \frac{\alpha}{R_{0}} + 1 \right)} = \left( 1 + \frac{\tau_{J}}{\alpha} \right)^{\alpha}\left\lbrack 1 - \frac{\alpha^{2}}{(\alpha + 1)\left( \alpha + R_{0} \right)}\left( 1 + \frac{\tau_{J}}{\alpha} \right) \right\rbrack} \\ \end{array}$$ Using the function *F*(*z*, *ν*) defined by *F*^*ν*(1 − *F*) = *z*, we can solve this for *τ*_*J*: $$\begin{array}{r} {\tau_{J} = \frac{(\alpha + 1)\left( \alpha + R_{0} \right)}{\alpha}F\left( \frac{2\alpha + 1}{(\alpha + 1)\left( \frac{\alpha}{R_{0}} + 1 \right)}\left\lbrack \frac{\alpha^{2}}{(\alpha + 1)\left( \alpha + R_{0} \right)} \right\rbrack^{\alpha},\alpha \right) - \alpha} \\ \end{array}$$ This allows us to compute *A*₂ and *A*₃ $$\begin{array}{rc} \left. A_{2} = \frac{\overset{¨}{J}}{J} \middle| {}_{\tau = \tau_{J}} = \beta\overset{˙}{K} = \right. & {{\frac{\gamma^{2}(1 + \alpha)}{\left( \alpha + \tau_{J} \right)^{2}}\left( 1 + \frac{\tau_{J}}{\alpha} \right)^{- 2\alpha}\left\lbrack \left( R_{0} - \tau \right)\left( 1 + \frac{\tau_{J}}{\alpha} \right)^{\alpha} - R_{0} \right\rbrack}\left\lbrack - (1 + 2\alpha)R_{0} + \left( \alpha + R_{0} \right)\left( 1 + \frac{\tau_{J}}{\alpha} \right)^{\alpha} \right\rbrack} \\ \end{array}$$ $$\begin{array}{cl} \left. A_{3} = \frac{\dddot{J}}{J} \middle| {}_{\tau = \tau_{J}} = \beta\overset{¨}{K} = \right. & {\frac{\gamma^{3}(1 + \alpha)}{\left( \alpha + \tau_{J} \right)^{3}}\left( 1 + \frac{\tau_{J}}{\alpha} \right)^{- 3\alpha}\left\lbrack \left( R_{0} - \tau_{J} \right)\left( 1 + \frac{\tau_{J}}{\alpha} \right)^{\alpha} - R_{0} \right\rbrack} \\ & \left\lbrack - \left( \alpha + R_{0} \right)\left( \alpha + 2R_{0} - \tau_{J} \right)\left( 1 + \frac{\tau_{J}}{\alpha} \right)^{2\alpha} \right. \\ & {\quad + (1 + \alpha)R_{0}\left( 1 + \frac{\tau_{J}}{\alpha} \right)^{\alpha}\left( 3\alpha + 2(2 + \alpha)R_{0} - (1 + 2\alpha)\tau_{J} \right)} \\ & \left. \quad - 2(1 + \alpha)(1 + 2\alpha)R_{0}^{2} \right\rbrack \\ \end{array}$$ The parameters λ₀, λ_∞, *A*₂ and *A*₃ can be obtained from linear and cubic fits to the logarithm the number of daily reported cases *J*_rep. For these fits, time intervals corresponding to initial exponential growth (*T*_*i*), peak *T*_*p* and final decay *T*_*f* need to be defined. We use the time point $t_{m}^{\text{rep}}$, where *J*_rep reaches its maximum as a reference point relative to which the intervals are given by: $$\begin{array}{rc} {T_{i} =} & \left\lbrack t_{m}^{\text{rep}} - 3\Delta t,t_{m}^{\text{rep}} - \Delta t \right\rbrack \\ \end{array}$$ $$\begin{array}{rc} {T_{p} =} & \left\lbrack t_{m}^{\text{rep}} - \Delta t,t_{m}^{\text{rep}} + \Delta t \right\rbrack \\ \end{array}$$ $$\begin{array}{rc} {T_{f} =} & \left\lbrack t_{m}^{\text{rep}} + \Delta t,t_{m}^{\text{rep}} + 3\Delta t \right\rbrack \\ \end{array}$$ These time intervals are further reduced depending on the used data and Δ*t* such that all time points before the last day with *J*_rep = 0 prior to $t_{m}^{\text{rep}}$ and after the first day *J*_rep = 0 after $t_{m}^{\text{rep}}$ are excluded. The fits in and the dashed horizontal lines in Figs and correspond to fits with Δ*t* = 19days. The shaded areas in Figs and depict the range of parameter values one obtains for fits with 10 days ≤ Δt ≤ 20days. # Appendix E: Small *α* limit for heterogeneous populations For small *α* the system reaches a well defined limiting dynamics that can be expressed analytically. We start from *I*(*τ*) for small *I*₀/*N* $$\begin{array}{r} {\frac{I}{N} = 1 - \left( 1 + \frac{\tau}{\alpha} \right)^{- \alpha} - \frac{\tau}{R_{0}}} \\ \end{array}$$ which for small *α* becomes $$\begin{array}{r} {\frac{I}{N} \simeq \alpha\ln\left( 1 + \frac{\tau}{\alpha} \right) - \frac{\tau}{R_{0}}\quad.} \\ \end{array}$$ The maximum of *I* occurs at *τ* = *τ*_*I* when *I*′ = 0 or *τ*_*I*/*α* ≃ *R*₀ − 1. We thus have $$\begin{array}{r} {\frac{I_{\text{max}}}{N} \simeq \alpha\left( \ln R_{0} + \frac{1}{R_{0}} - 1 \right)\quad.} \\ \end{array}$$ Similarly, using *C*_*I*/*N* = 1 − (1 + *τ*_*I*/*α*)^−*α*, we find for small *α* $$\begin{array}{r} {\frac{C_{I}}{N} \simeq \alpha\ln R_{0}} \\ \end{array}$$ At long times, we have *I*(*τ*_∞) = 0, where (1 + *τ*_∞/*α*)^−*α* = 1 − *τ*_∞/*R*₀. For small *α* this implies *α* ln(1 + *τ*_∞/*α*)≃*τ*_∞/*R*₀ and thus $S_{\infty} = \left( N - I_{0} \right){\overline{x}}_{\infty}^{\alpha}$ and $\lambda_{\infty} = \beta{\overline{x}}_{\infty} - \gamma$, where $$\begin{array}{r} {{\overline{x}}_{\infty}^{- 1} = - R_{0}W_{- 1}\left( - \frac{e^{- 1/R_{0}}}{R_{0}} \right)\quad.} \\ \end{array}$$ In the limit of small *α*, *u* = *τ*/*α* is finite. The limiting function *u*(*t*) for small *α* can be expressed as $$\begin{array}{r} {\int_{0}^{u}\frac{du^{\prime}}{\ln\left( u^{\prime} + 1 \right) - u^{\prime}/R_{0} + i_{0}} = \beta t\quad,} \\ \end{array}$$ where *i*₀ = *I*/(*αN*) in the limit *α* = 0. The number of susceptible then becomes $$\begin{array}{r} {\frac{S}{N} \simeq 1 - \alpha\ln(1 + u)\quad.} \\ \end{array}$$ Finally we discuss the maximum of the rate of new cases *J* = *J*_max. We have $\overset{˙}{J}/J = \beta K$, where $$\begin{array}{r} {K = \left( 2 + \frac{1}{\alpha} \right)\left( 1 + \frac{\tau}{\alpha} \right)^{- (\alpha + 1)} - \frac{1}{R_{0}} + \frac{\alpha + 1}{\alpha}\;\frac{1 - \tau/R_{0}}{1 + \tau/\alpha}} \\ \end{array}$$ At the maximum of *J*, *τ* = *τ*_*J* with $$\begin{array}{r} {(2\alpha + 1)\left( 1 + \frac{\tau_{J}}{\alpha} \right)^{- \alpha} = \frac{\alpha}{R_{0}}\left( 1 + \frac{\tau_{J}}{\alpha} \right) + (\alpha + 1)\left( 1 - \frac{\tau_{J}}{R_{0}} \right)} \\ \end{array}$$ Defining ${\overline{x}}_{J} = \left( 1 + \tau_{J}/\alpha \right)^{- 1}$ we have in the limit of small *α* $$\begin{array}{r} {{\overline{x}}_{J} = \exp\left( \frac{1}{R_{0}} - 1 \right)\quad.} \\ \end{array}$$ The value of *J* at the maximum is $$\begin{array}{r} {\frac{J_{\text{max}}}{N} = \alpha\gamma R_{0}{\overline{x}}_{J}\left( - \ln{\overline{x}}_{J} - \frac{1 - {\overline{x}}_{J}}{R_{0}{\overline{x}}_{J}} \right)\quad.} \\ \end{array}$$ We determine $A_{2} = \overset{¨}{J}/J = \beta^{2}\overset{˙}{K}$ and $A_{3} = \dddot{J}/J = \beta^{3}\overset{¨}{K}$, with $\overset{˙}{K}/\beta = K^{\prime}I/N$ and $\overset{¨}{K}/\beta^{2} = K^{\prime\prime}I^{2}/N^{2}$. We then find $$\begin{array}{rcl} A_{2} & = & {- \gamma^{2}R_{0}^{2}{\overline{x}}_{J}^{2}\left( - \ln{\overline{x}}_{J} - \frac{1 - {\overline{x}}_{J}}{R_{0}{\overline{x}}_{J}} \right)} \\ \end{array}$$ $$\begin{array}{rcl} A_{3} & = & {2\gamma^{3}R_{0}^{3}{\overline{x}}_{J}^{3}\left( - \ln{\overline{x}}_{J} - \frac{1 - {\overline{x}}_{J}}{R_{0}{\overline{x}}_{J}} \right)^{2}\quad.} \\ \end{array}$$ We also have $$\begin{array}{r} {\frac{A_{2}^{2}}{A_{3}} = \frac{\gamma R_{0}{\overline{x}}_{J}}{2}} \\ \end{array}$$ and $$\begin{array}{r} {\frac{A_{3}^{2}}{A_{2}^{3}} = 4\left( - \ln{\overline{x}}_{J} - \frac{1 - {\overline{x}}_{J}}{R_{0}{\overline{x}}_{J}} \right)\quad.} \\ \end{array}$$ # Appendix F: Mitigation in the heterogeneous SIR model We now consider the case where the rate of infections *β*(*t*) becomes time dependent because of overall changes of conditions such as seasonal effects or measures of social distancing. Using *I*′ = −*S*′+ *N*/*R*₀, we have $\lambda = \overset{˙}{I}/I = - \beta S^{\prime}/N - \gamma$ and the reproduction number $$\begin{array}{r} {R = - \frac{\beta}{\beta_{0}}R_{0}\frac{S^{\prime}}{N}.} \\ \end{array}$$ where *β*₀ = *β*(*t* = 0) and *R*₀ = *β*₀/*γ*. The epidemic can be mitigated by a reduction of *β* over time. However if the mitigation is relaxed the epidemic can grow again. As the epidemic advances, *τ* increases as $\overset{˙}{\tau} = \beta I/N$. Growth of infection number is no longer possible for *τ* \> *τ*_*I* with $$\begin{array}{r} {- S^{\prime}\left( \tau_{I} \right) = \frac{1}{R_{0}}} \\ \end{array}$$ Thus the condition *τ* \> *τ*_*I* defines herd immunity conditions where the epidemic can no longer grow. If mitigation sets in early, before *τ* = *τ*_*I*, the epidemic is slowed and it takes more time to reach herd immunity. in this case a new wave starts after mitigation is relaxed. If mitigation occurs for *τ* \> *τ*_*I*, mitigation facilitates the decay of infections by reducing λ_∞ = −(*β*_∞/*β*₀)*R*₀ *S*′(*τ*_∞) − *γ* as compared to the value λ_∞ = −*R*₀ *S*′(*τ*_∞) − *γ* without mitigation. # Appendix G: Inferring *β*(*t*) from reported cases For a given time course of infections, there always exists a function *β*(*t*) such that the SIR model follows this time course. We first consider the classic SIR model. A change in the rate of new infections *J* = *βIS*/*N* can be decomposed in three different contributions, $$\begin{array}{r} {\frac{\text{d}}{\text{d}t}\ln J = \frac{\overset{˙}{\beta}}{\beta} + \frac{\overset{˙}{I}}{I} + \frac{\overset{˙}{S}}{S}.} \\ \end{array}$$ In the case of an early mitigation, *S* ≈ *N* and thus $\overset{˙}{S}/S \approx 0$. Together with, we find $$\begin{array}{r} {\frac{\text{d}}{\text{d}t}\ln J = \frac{\text{d}}{\text{d}t}\ln\beta + \beta - \gamma.} \\ \end{array}$$ This provides a differential equation for ln *β* if ln *J*(*t*) is given, which does not require knowledge of the amplitude of *J*. We infer *β*(*t*) for each day, using the initial value *β*(0) = 0.48days⁻¹ at March 15. We use an iterative scheme to calculate the rate for the next day as $$\begin{array}{r} {\ln\beta(i + 1) = \ln\beta(i) + \ln J_{\text{obs}}(i + 1) - \ln J_{\text{obs}}(i) - \beta(i) + \gamma,} \\ \end{array}$$ where $\ln J_{\text{obs}}(i) = (1/7)\sum_{\Delta t = - 3}^{3}J_{\text{rep}}\left( i + \Delta t \right)$ is a running average over seven days of the number of reported cases. For the two scenarios of a later mitigation, the heterogeneous SIR model was considered with *J* = *βI*(1 − *I*₀/*N*)(1 + *τ*/*α*)^{−(*α*+ 1)}. We then have $$\begin{array}{r} {\frac{\text{d}}{\text{d}t}\ln J = \frac{\text{d}}{\text{d}t}\ln\beta + \frac{\text{d}}{\text{d}t}\ln I - \frac{\alpha + 1}{\alpha}\left( 1 + \frac{\tau}{\alpha} \right)^{- 1}\frac{\beta I}{N}\quad.} \\ \end{array}$$ Again, this equation can be used to construct an iterative scheme to infer *β*(*t*). For given initial number of infected individuals on March 15 *I*(0), we can iteratively obtain the subsequent values as $$\begin{array}{rcl} {\ln\tau(i + 1)} & {= \ln\tau(i) +} & {\frac{\beta(i)I(i)}{N\tau(i)},} \\ \end{array}$$ $$\begin{array}{rcl} {\ln I(i + 1)} & {= \ln I(i) +} & {\beta(i)\left( 1 - \frac{I_{0}}{N} \right)\left( 1 + \frac{\tau(i)}{\alpha} \right)^{- (\alpha + 1)} - \gamma,} \\ \end{array}$$ $$\begin{array}{ccl} {\ln\beta(i + 1)} & {= \ln\beta(i) +} & {\ln\frac{J_{\text{obs}}(i + 1)}{J_{\text{obs}}(i)} - \ln\frac{I(i + 1)}{I(i)}} \\ & & {+ \frac{\alpha + 1}{\alpha}\left( 1 + \frac{\tau(i)}{\alpha} \right)^{- 1}\frac{\beta(i)I(i)}{N}\quad.} \\ \end{array}$$ The starting value of *τ*(0), can be derived by inverting for *I*(*τ*(0)) = *I*(0). # Appendix H: Mobility data Data concerning the changes in mobility of the population has been provided by Google. The data reports the changes compared to a baseline of visits and length of stay at different places. The baseline depends on the specific day of the week and refers to the median value, for the corresponding day of the week, during the 5-week period Jan 3–Feb 6, 2020. shows these changes for Germany for a representative number of categories. These categories are defined in as follows: “Grocery and pharmacy: Mobility trends for places like grocery markets, food warehouses, farmers markets, specialty food shops, drug stores, and pharmacies. Transit stations: Mobility trends for places like public transport hubs such as subway, bus, and train stations. Retail and recreation: Mobility trends for places like restaurants, cafes, shopping centers, theme parks, museums, libraries, and movie theaters. Residential: Mobility trends for places of residence. Workplaces: Mobility trends for places of work. The residential category shows a change in duration while the other categories measure a change in total visitors.” 10.1371/journal.pone.0239678.r001 Decision Letter 0 Gontis Vygintas Academic Editor 2020 Vygintas Gontis This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 3 Sep 2020 PONE-D-20-24977 Power-Law Population Heterogeneity Governs Epidemic Waves PLOS ONE Dear Dr. Jülicher, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Comments of reviewers are valuable and give you an opportunity to improve the paper. Please submit your revised manuscript by Oct 18 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Vygintas Gontis, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Thank you for inlcuding your funding statement;     "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript." At this time, please address the following queries: Please clarify the sources of funding (financial or material support) for your study. List the grants or organizations that supported your study, including funding received from your institution.State what role the funders took in the study. If the funders had no role in your study, please state: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.”If any authors received a salary from any of your funders, please state which authors and which funders.If you did not receive any funding for this study, please state: “The authors received no specific funding for this work.” Please include your amended statements within your cover letter; we will change the online submission form on your behalf. 3\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: In the paper under review, the authors give a detailed study of the classical SIR model for epidemics in the case when the heterogeneity of the susceptible population is allowed. The main focus is made on the case where the initial susceptibility distribution has gamma density with parameter \$\\\alpha\$. This, in particular, leads to the time-dependent reproduction number \$R(t)\$, decreasing as hyperbolic function, i.e.\\\\ \$R(t)=(\\\bar x(t))^{1+\\\alpha} R_0\$. The authors provide quantitative properties of the model, such as the herd immunity level, the final size of epidemics, etc. An important conclusion is that, in the heterogeneous population, the herd immunity level can be much lower than in homogeneous case (typically 60\\\\%). In my opinion, the subject and results of the paper are very interesting, the paper is well written and I recommend it for publishing in the journal after minor revision. Attached, please find a list of comments. Reviewer \#2: In this manuscript, the authors propose a generalized SIR model taking into account the heterogeneity of the population, i.e., a distribution of the susceptibility of being infected, in terms of a parameter alpha, which is the power-law exponent of the susceptibility distribution when small values of alpha are considered. In other words, when alpha -\> infinity, the classical SIR model is recovered and for the case of alpha -\> 0, the heterogeneity of the population is incorporated into the SIR model. The key-result of their work is that the population herd immunity is earlier achieved in the case of a heterogeneous population, implying in a lower number of infected people and fatalities. The authors employ their model to analyze the case of the Covid-19 spread in Germany and discuss the importance of taking the population's heterogeneity in the effectiveness of mitigation actions, such as social distancing. Below, I raise some points to be addressed: \- In panel (a) and (d) of Fig. 1, it is not quite clear to me why the number of infected people is lower for the heterogeneous SIR model. Also, what is the justification for using the specifically values of R0 = 2.5, gamma = 0.13 day^-1 and alpha = 0.1? I suggest that the authors make these points clear; \- For future works, I believe it would be interesting to explore the very same analysis here employed in terms of the heterogeneity of the population in the light of the SIRS model, since it would be interesting to analyse how the population's susceptibility distribution is affected in the case where Recovered people can become susceptible of being infected again. Perhaps it would interesting to mention this in the main text; \- On page 6, the authors mention about the Lambert W function and points to a more detailed discussion about it in Appendix A. At this point, I believe it is worth adding a couple of references for clarity both in the main text and in Appendix A about the Lambert W function, just for the sake of completeness; \- The authors should correct on page 3, third paragraph, first sentence, the typo "the" twice in the sentence "In the heterogeneous SIR model proposed here, the qualitative behaviors of the the epidemic wave are unchanged."; \- I suggest that the first sentence of page 8 "Eq. (9) can be then be written as" to be corrected to "Eq. (9) can then be written as..."; \- On page 8, the authors write "The dynamics of epidemic waves depends on the shape of the initial distribution s0(x). Here, we consider distributions that have the special property of shape invariance under the dynamics of epidemics. This property is satisfied by a gamma distribution". I suggest that the authors state if only the gamma distribution satisfies such a condition and, if it is not the case, then it would be interesting to add a sentence about the consideration of other distributions as well; \- In Fig 4 panel (a), I suggest that the authors state why their solutions of the heterogeneous SIR model do not incorporate the initial increase in the number of infected people and the small increase between Jun and Jul for both the number of infections and fatalities. The same holds for panel (c) in the case of the initial growth of both the cumulative number of infected people and fatalities. In panel (f), it would be interesting to discuss what is the meaning of tau saturating at the specific value of \~0.2, as well as about the meaning of the average susceptibility x saturating close to the \\\tau curve? What does this mean? I suggest that the authors briefly discuss about these points; \- Based on their discussions, the achievement of the herd immunity is key regarding the fade of the disease spread. Based on their discussion about and heterogeneous population, do the authors have any suggestions for public policies in order to minimize the number of infections and, consequently, the number of fatalities?; \- On page 18, the authors write "We show that as a result of strong population heterogeneity (small alpha), the wave peaks when only a small minority of individuals have been infected, see Fig. 1 (d)-(f)." This is indeed true. However, upon analysing panels (a), (d), and (g) of Fig. 6, one notices that the model solution fit of the data (red solid line) present lower reported cases decrease rate than in the scenario without mitigation (red dotted line). This is particularly true after June. How can this be explained? It seems that, although the maximum is earlier achieved, the decrease rate is lower than in the case without mitigation. I suggest that the authors include a few sentences to discuss about this; \- Section "Discussion" seems more like "Conclusions and Perspectives"; \- I believe that the number of references could be improved in this work since there are a lot of discussions throughout the manuscript that deserves more important references. In summary, the work is relevant since in reality not everyone is equally susceptible to being infected and thus the authors' consideration of a susceptibility distribution among the population is solid and can indeed improve the understanding of the epidemics dissemination. I do recommend publication after minor revisions. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0239678.r002 Author response to Decision Letter 0 9 Sep 2020 Response to Referee 1 \>In the paper under review, the authors give a detailed study of the classical SIR \>model for epidemics in the case when the heterogeneity of the susceptible \>population is allowed. The main focus is made on the case where the initial \>susceptibility distribution has gamma density with parameter \$\\\alpha\$. \>This, in particular, leads to the time- dependent reproduction number \$R(t)\$, \>decreasing as hyperbolic function, i.e.\\\\ \$R(t)=(\\\bar x(t))^{1+\\\alpha} R_0\$. \>The authors provide quantitative properties of the model, such as the herd \>immunity level, the final size of epidemics, etc. An important conclusion is that, \>in the heterogeneous population, the herd immunity level can be much lower \>than in homogeneous case (typically 60\\\\%). \> \>In my opinion, the subject and results of the paper are very interesting, the \>paper is well written and I recommend it for publishing in the journal after \>minor revision. Attached, please find a list of comments. We thank the referee for a careful reading of our manuscript \>1. In the models for SARS-CoV-2 epidemics, the SEIR model is in widespread \>use, where E stands for the ”exposed” compartment. I wonder, if it is possible \>to relate it with the generalized SIR model in (1)–(2) with time- varying \>parameters? In our work we focus on the simpler SIR model where the distinction between exposed and infected is not made. The additional “exposed” state E in the SEIR model introduces a time delay between exposure to a pathogen and the onset of infectiousness. This extra delay due to the exposed state could be captured by an exponential memory kernel in Eq (2) with an extra relaxation time describing the delay. However, the main properties of the epidemic wave such as herd immunity levels are only weakly affected by the extra short delay in the SEIR model. In our work we focus on the SIR model because it captures essential features of an epidemics in a minimal model. In our revised manuscript we now relate to the SEIR model in the discussion. \>2. The model in eqs. (1)–(2) is not very clearly explained. The classical SIR \>model assumes three components – susceptible (S), infected (I) and recov- \>ered (R). Whereas eqs. (1)–(2) of the paper does not include part R. Also, \>it is slightly unclear the meaning of x ̄. It would be useful to explain what does \>it mean ”dimensionless average susceptibility”. Usually the coefficient at IS/N \>is called ”infection rate” and can admit values larger (or smaller) than 1. We use the symbol R for the time dependent reproduction number rather than the number of recovered individuals. Note that the number of recovered individuals is given by \#recovered = N-S-I. We now add this information. In the revised manuscript we have rewritten the explanation of the model in order to be more clear about beta and \\\bar x and we now call beta the infection rate. \>3. p. 5. It is mentioned that time-varying β could correspond to seasonal \>changes or mitigation measure. For illustrative purposes, it would be nice to \>see concrete forms of β(t) in such cases. While in the first parts of the manuscript we consider constant beta, we discuss in section III.C different mitigation scenarios where we choose specific forms of beta(t) in order to discuss effects of mitigation during this year’s SARS-Cov2 epidemic, see Fig. 6. \>4. It is well-known that SIR-type compartment models are rather sensitive to \>initial conditions. It would would be good to see some discussion on this \>sensitivity. At least, how the graphs will change if you take other values than \>I0 = 10. The initial condition I0 is not important for our work and a change in I0 would essentially only shift the absolute time of the initial point if the time of the wave peak is known. Once the wave is progressing and reaches its peak, its behaviour depends only very weakly on the initial conditions if the initial number of infected is small compared to the population size. Note that in the appendices we do provide exact expressions for many quantities as a function of I0/N, revealing the role of initial conditions. However for all plots and results shown in the manuscript the initial number I0 is completely unimportant. Because the plots for other choices of I0 would not be distinguishable from the ones shown we prefer not to show plots with other values of I0. Another reason not to discuss different I0 is that at the early time of the epidemics when I(t) went from say 10 to 20 we have absolutely no information about the epidemic wave and therefore it does not to help discussing wether it started with I0=2 very early or with I0=15 several days later. \>5. p. 7, l. -9. Since s depends on both x and t, I would recommend to write \>S(t) = ∫ \\\int_0^∞ dxs(x, t). Similarly, in eq. (10), as the right-hand side \>depends on t, better to write x ̄(t) = (1/S(t))\\\int_0^∞ dx xs(x, t). We have changed these expression as the referee suggests. \>6. p. 7, l. -2. The meaning of the variable τ (”a measure for how far the \>epidemic has advanced”) should be explain better. The variable tau emerges from a trick to solve the equations. It does not have a physical meaning but it has some clear and important properties. In the revised manuscript we now explain better the variable tau. \>7. p, 8, eq. (12). Strictly speaking, eq. (12) gives a density function, not \>distibution. Also, you may wish to add that α \> 0 and to write a precise form \>s_0(x)=…(The symbol ’∼’ usually means asymptotic equivalence.) Also, I \>wonder if it would be possible to introduce additional flexibility in the initial \>susceptibility by introducing two-parameter gamma distribution with \>density s_0(x)=….. We agree with the referee and have added that alpha \>0. Note that we thought about all these points carefully and we have good reasons to present the work in the way we did. We have used the term distribution in the context of our aim to make the paper more easily accessible for a broader readership including non-experts for whom the term probability distribution is usually better known. For this reason we have also kept the equations in the main part of the manuscript rather light, but we provide all the precise and full expressions in the Appendices for expert readers. The appendices carry a lot of substance and should not be seen as secondary. Note that the precise form of the initial density function s_0 is provided in Appendix C in Eq. (C1). In our revised manuscript we now refer to the appendix to clarify what the normalization factor is. We have checked that the full 2-parameter gamma distribution does not provide any further flexibility. The reason is that we choose without of loss of generality \\\bar x=1 at the initial time point. With the second parameter of the gamma distribution we can change this initial value but this change can be absorbed by changing the infection rate beta. So in fact one can see beta as the second parameter. But since beta already exists in the classical SIR model we keep it and only add one new parameter alpha describing the ratio of variance and mean. \>Minor comments: \>p. 3, l. -11. Delete ’the’. Done \>p. 8, eq. (13). Please add more details how this equality is obtained. We added some information and now refer to Appendix C for details. p\. 9. Please add more details how eqs. (18) and (19) are obtained. We added a reference to the Appendix C where the details are provided. \>p. 22. It seems that the Lambert W -function should be defined by \>W (z)eW (z) = z. Done Response to Referee 2 \>In this manuscript, the authors propose a generalized SIR model taking \>into account the heterogeneity of the population, i.e., a distribution of the \>susceptibility of being infected, in terms of a parameter alpha, which is \>the power-law exponent of the susceptibility distribution when small values \>of alpha are considered. In other words, when alpha -\> infinity, the \>classical SIR model is recovered and for the case of alpha -\> 0, the \>heterogeneity of the population is incorporated into the SIR model. The \>key-result of their work is that the population herd immunity is earlier \>achieved in the case of a heterogeneous population, implying in a lower \>number of infected people and fatalities. The authors employ their model \>to analyze the case of the Covid-19 spread in Germany and discuss the \>importance of taking the population's heterogeneity in the effectiveness \>of mitigation actions, such as social distancing. \>Below, I raise some points to be addressed: \>- In panel (a) and (d) of Fig. 1, it is not quite clear to me why the number \>of infected people is lower for the heterogeneous SIR model. Also, what \>is the justification for using the specifically values of R0 = 2.5, \>gamma = 0.13 day^-1 and alpha = 0.1? I suggest that the authors \>make these points clear; The figure just shows the fact that the number of infected people is lower in the heterogeneous SIR model to clearly show this point. The reasons are the lowered herd immunity levels given by Eq (19) resulting from the drop in \\\bar x as shown in Fig. 1 f. In appendix C we provide an exact analysis of the nonlinear dynamics that reveals these surprising properties. In order to explain this better, we now clarify after Eq (22) that the drop of \\\bar x is the reason for a reduced herd immunity level and resulting lower infection numbers. The parameter values are here just for illustrative purposes. The qualitative behaviors do not depend on the parameter choice within broad ranges. However the values chosen are rounded versions of values we found are relevant to the current SARS-Cos2 epidemics are typical values used in the current epidemics. gamma = 0.13 corresponds to individuals being infectious during one week, and this parameter does not affect the shape but only the duration of the wave. alpha=0.1 was chosen so that the difference between panels (a) and (d) is clearly visible but such that the maximum of I(t) can still be seen in (d). \>- For future works, I believe it would be interesting to explore the very \>same analysis here employed in terms of the heterogeneity of the population \>in the light of the SIRS model, since it would be interesting to analyse how \>the population's susceptibility distribution is affected in the case where \>Recovered people can become susceptible of being infected again. \>Perhaps it would interesting to mention this in the main text; There are many open and interesting questions that one can address with our approach. We agree that it will be very interesting to look into effects where recovered individuals becomes infected again. In the revised manuscript we now mention the SIRS model in the discussion. \>- On page 6, the authors mention about the Lambert W function and \>points to a more detailed discussion about it in Appendix A. At this point, \>I believe it is worth adding a couple of references for clarity both in the \>main text and in Appendix A about the Lambert W function, just for the \>sake of completeness; We have added a reference on the Lambert W function. \>- The authors should correct on page 3, third paragraph, first sentence, \>the typo "the" twice in the sentence "In the heterogeneous SIR model \>proposed here, the qualitative behaviours of the the epidemic wave are \>unchanged."; Thanks - corrected \>- I suggest that the first sentence of page 8 "Eq. (9) can be then be \>written as" to be corrected to "Eq. (9) can then be written as…"; Thanks - done \>- On page 8, the authors write "The dynamics of epidemic waves depends \>on the shape of the initial distribution s0(x). Here, we consider distributions \>that have the special property of shape invariance under the dynamics of \>epidemics. This property is satisfied by a gamma distribution". I suggest that \>the authors state if only the gamma distribution satisfies such a condition and, \>if it is not the case, then it would be interesting to add a sentence about the \>consideration of other distributions as well; To our knowledge the gamma distribution is the only distribution that is shape invariant under the dynamics. However to be cautious we avoid claiming this fact as we do not have a proof. Our work and our results do not depend on the gamma distribution being the only shape invariant distribution. \>- In Fig 4 panel (a), I suggest that the authors state why their solutions of \>the heterogeneous SIR model do not incorporate the initial increase in the \>number of infected people and the small increase between Jun and Jul for \>both the number of infections and fatalities. The same holds for panel (c) in \>the case of the initial growth of both the cumulative number of infected \>people and fatalities. In Fig. 4 we simply compare fits of our model to data. The fits show that the data based on reported cases that lead to deaths (blue), which probably measures more reliably the progression of serious illnesses, is better captured by the model than simply using the reported number of reported cases (red). A possibility for the difference to the data early and in June is that the blue data gives a better representation of the epidemics than the red data. However we have refrained form making such statements because all available data have problems and there are many reasons why there could be serious biases and errors in the data. In our manuscript we note on p. 13 that it is surprising that the model fits both types of data rather well even though we only have three time independent parameters (note the classical SIR model cannot account for this data). \>In panel (f), it would be interesting to discuss what is the meaning of tau \>saturating at the specific value of \~0.2, as well as about the meaning of \>the average susceptibility x saturating close to the \\\tau curve? What does \>this mean? I suggest that the authors briefly discuss about these points; The final value of tau is explained in Appendix C where we show that tau reaches a fixed final value when the epidemics dies out that is described exactly by the implicit equation (C11) in the revised manuscript. In our revised manuscript we now mention in the main text on p. 7 that tau reaches a final value and we add a reference to appendix C when discussing panel (f) of Fig. 4. Note that the relationship between tau and \\\bar x is given in Eq. (14). The similarity of \\\bar x and \\\tau at the end of the wave is a coincidence and consistent with Eq. (14). \>- Based on their discussions, the achievement of the herd immunity is key \>regarding the fade of the disease spread. Based on their discussion about \>and heterogeneous population, do the authors have any suggestions for \>public policies in order to minimize the number of infections and, consequently, \>the number of fatalities?; We have on purpose refrained from a discussion of public policy. Here we want to focus on the concepts and the science. The science presented here is clear and rigorous. Implications for policy are much less rigorous and depend on many other factors and can be coloured by opinions. We think that our work has many implications for public policy and that it will stimulate and be useful for future discussions but we do not want to weaken our work by adding elements that are uncertain. \>- On page 18, the authors write "We show that as a result of strong \>population heterogeneity (small alpha), the wave peaks when only a small \>minority of individuals have been infected, see Fig. 1 (d)-(f)." This is indeed \>true. However, upon analysing panels (a), (d), and (g) of Fig. 6, one notices \>that the model solution fit of the data (red solid line) present lower reported \>cases decrease rate than in the scenario without mitigation (red dotted line). \>This is particularly true after June. How can this be explained? It seems that, \>although the maximum is earlier achieved, the decrease rate is lower than \>in the case without mitigation. I suggest that the authors include a few \>sentences to discuss about this; It is correct that in the case of the epidemics that is dying out because of mitigation (Fig. 6), we find that the rate of decay of cases is slower than without mitigation. This is similar to the idea to “flatten the curve”, i.e. mitigation reduces the maximal number of infections but broadens the wave and thus makes it slower. However this feature is not completely general and we prefer not to enter this discussion. \>- Section "Discussion" seems more like "Conclusions and Perspectives"; We have changed the discussion to “Conclusions and Perspectives” \>- I believe that the number of references could be improved in this work \>since there are a lot of discussions throughout the manuscript that deserves \>more important references. We have now added reference \[28\] on Lamberts W function and Ref. \[35-37\] for the SEIR and the SIRS model. We think that all statements that need backing by references have been referenced. \>In summary, the work is relevant since in reality not everyone is equally \>susceptible to being infected and thus the authors' consideration of a susceptibility \>distribution among the population is solid and can indeed improve the \>understanding of the epidemics dissemination. I do recommend publication after \>minor revisions. We thank the referee for a careful reading of our manuscript 10.1371/journal.pone.0239678.r003 Decision Letter 1 Gontis Vygintas Academic Editor 2020 Vygintas Gontis This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 11 Sep 2020 Power-Law Population Heterogeneity Governs Epidemic Waves PONE-D-20-24977R1 Dear Dr. Jülicher, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, Vygintas Gontis, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Congratulations, After careful analysis of the article revised version and Response to Reviewers, we have decided to accept the paper for the publication in Plos One. Reviewers' comments: 10.1371/journal.pone.0239678.r004 Acceptance letter Gontis Vygintas Academic Editor 2020 Vygintas Gontis This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 17 Sep 2020 PONE-D-20-24977R1 Power-Law Population Heterogeneity Governs Epidemic Waves Dear Dr. Jülicher: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Vygintas Gontis Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# 1. Introduction Since the initiation of China’s reform and opening-up policy, the country has experienced substantial economic growth and improvement in the living standards of its people. However, this progress has come at a cost, characterized by high pollution and energy consumption, leading to the emergence of serious environmental issues such as global warming. With the economy’s continuous development, Chinese people have raised their expectations for a higher quality of life, necessitating the exploration of approaches that promote high-quality economic development while prioritizing environmental protection. The 19th National Congress of the Communist Party of China in 2017 acknowledged that the primary contradiction within Chinese society was unbalanced and inadequate development, along with the ever-growing needs of the people for an enhanced standard of living. The Congress proposed a shift in the economic development mode, emphasizing the need for swift high-quality economic growth. The concept of high-quality economic development encompasses diverse fields, necessitating coordinated efforts across all domains. Scholars have extensively researched the primary challenges hindering high- quality development in China, which include uneven and insufficient economic growth. This challenge is manifested in the central and western regions of the country, which exhibit slower development than the eastern region, which has better resources and a thriving economy. To address this issue, inclusive finance has emerged as a viable solution aimed at providing effective financial services to marginalized groups such as small and micro-enterprises, low-income earners, and the poor. In recent years, digital inclusive finance has gained prominence, driven by advancements in internet technology, promoting information dissemination, improving the coverage of inclusive finance, and facilitating poverty eradication efforts. Digital inclusive finance has the potential to identify relevant groups accurately, provide them with the necessary financial support, and ultimately promote harmonious social development and sustainable economic growth. Additionally, digital inclusive finance can contribute to the growth in household consumption, which is one of the three key drivers of economic development. Previous economic models in China were characterized by low- and medium-end consumption. However, with continued economic development, people have become increasingly interested in high-quality consumption. Consequently, household consumption has become a crucial engine of economic growth, with its continuous expansion facilitating development across all economic sectors, boosting related industries, and generating employment opportunities. Furthermore, as household consumption continues to expand, it drives the optimization and upgrading of consumption structures. This upgrading process is marked by increasing demand for quality consumer goods and heightened attention to factors such as brand, culture, and service. These developments are important for the promotion of high-quality economic development in China. This study aims to investigate the relationship between digital inclusive finance, consumer consumption, and high-quality economic development in China using a spatial econometric model with panel data from 30 provinces and cities. In addition to examining the linear relationships between these variables, a semi-parametric spatial lag model is employed to investigate potential non- linear relationships. This study also analyzes an analysis of the mediating and moderating effects of consumer consumption on the relationship between digital inclusive finance and high-quality economic development. Finally, to test the robustness of the results, the study also replaces the 0–1 weight matrix with an economical weight matrix. The findings of this study are expected to provide valuable insights into the impact of digital inclusive finance on consumer consumption and high-quality economic development contributing to the ongoing efforts to develop a high-quality economy in China. This study contributes significantly to the existing literature in several ways. First, it examines the association between digital inclusive finance, household consumption, and high-quality economic development from both linear and nonlinear perspectives. The semi-parametric spatial lag model used in this study enables a more comprehensive analysis of the dynamic change process, adding to the credibility of this study. Second, this study focuses primarily on investigating the relationship between digital inclusive finance, household consumption, and high-quality economic development and endeavors to explore the paths of influence among these factors, offering guidance for future scholars interested in exploring these pathways. Finally, this study expands upon prior research by not only treating inclusive finance and consumption as explanatory variables, but also examining consumption as both an intermediary and moderating variable, providing a more nuanced understanding of the relationship among these three factors. # 2. Theoretical analysis and research hypotheses ## 2.1 Digital inclusive finance and high-quality economic development In 2005, inclusive finance was introduced to promote equal access to financial services for all customers. With the advancement of Internet technology, digital inclusive finance has emerged. While there is limited research on the impact of digital inclusive finance on high-quality economic development, some scholars have examined the relationship between digital finance, digital economy, and high-quality development. For instance, Xiong Wang (2022) found that digital inclusive finance can help small and medium-sized technology-based enterprises alleviate financing constraints, while Jianwei Li (2022) noted that it can effectively alleviate SME financing constraints, with private and family businesses benefiting in particular. Other studies have investigated the role of inclusive finance in promoting economic growth through an analysis of the endogenous growth theory (Jun He, 2019) and the impact of the digital economy on high-quality economic development from various perspectives (Ren Baoping, 2022; Zhan Wenqing, 2022; YAN Tao, 2022; Liu Jiaqi, 2022; Ge Heping, 2021). Wu (2021) investigated the impact of the Internet on energy-saving efficiency and found that it can promote energy conservation and emission reduction through technological progress, human capital, openness, and energy structure. Zhao (2022) showed that the digital economy can enhance the technological development and human capital of cities, with significant spillover effects across regions. Li (2022) and Du (2022) demonstrated that digital finance can reduce environmental pollution by promoting industrial structure upgrading and rationalization, with marketization and government support playing important roles in this process. Han (2022) examined the link between finance and innovative development and found that financial development can attract human capital and promote innovation within and across regions. These studies provide valuable insights into the complex relationship between digital technology and economic and social development, highlighting the multifaceted nature of this phenomenon and its potential to foster high-quality growth and sustainability. Based on these findings, it is evident that digital inclusive finance can promote enterprise innovation, meet the capital needs of enterprise survival and development, and optimize and adjust industrial structures, thereby improving economic development. This study recommends the following actions to enhance the impact of digital inclusive finance on high-quality economic development: **Hypothesis 1:** the development of high-quality economies can be effectively promoted by digital inclusive finance. ## 2.2 Digital inclusive finance and consumer consumption Goods and services are essential components of consumer consumption, and involve monetary or virtual expenditures to meet people’s basic needs for survival, development, and enjoyment. Researchers have studied the increasing consumption of consumersfrom the perspective of digital inclusive finance. Ang (2011) studied the impact of financial policy on private consumption in India and found that financial repression policies are linked to a reduction in private consumption fluctuations, while a more open financial system can help suppress it. Grossman (2014) discovered that digital finance could reduce potential power consumption by providing convenience to consumers. Cioacă (2020) demonstrated that internet technology can reduce income gaps and promote resident consumption when deeply integrated into society. Based on spatial correlation and data from 265 Chinese prefecture-level cities from 2011 to 2018, Wang (2022) found that residents could effectively maximize their consumption potential with digital inclusive finance. Compared with the western region, the eastern and central regions showed a more significant impact on driving consumption levels and the spatial spillover effect. Additionally, digital inclusive finance can drive an increase in consumer consumption in first and second-tier cities compared to third and fourth-tier cities. Based on China’s provincial data from 2011 to 2017, Jiang (2020) empirically found that digital inclusive finance improves consumer consumption levels and optimizes consumption structures by reducing the income gap between urban and rural areas and optimizing the industrial structure. Yi (2018) observed that the inclusion of digital finance in promoting economic growth is stronger for rural, middle-income, and underdeveloped households, and is influenced by the education level and cognitive ability of the household head. Li (2022) contended that digital inclusive finance, as a distinctive form of finance, is a vital driver in augmenting household consumption. Empirical evidence confirms that it has a substantial effect on boosting rural consumption. From a climate change perspective and with reference to the historical temperature of cities, He (2022) found that digital financial inclusion can not only raise farmers’ consumption but also promote consumption upgrading to a certain extent. Moreover, digital financial inclusion can enhance the ability of individuals to withstand climate change. Zhao (2022) asserts that with the advancement of mobile payments in China, digital inclusive finance can invigorate residents’ consumption and enhance the happiness of vulnerable groups. Luo (2022) scrutinizes the association between digital financial inclusion and consumption inequality in China and demonstrates that digital financial inclusion can promote consumption among low-income individuals and diminish the income gap. According to Cheng (2022), consumption has played an increasingly substantial role in propelling the economy since China entered a new era. Rural income and consumption are crucial components of high-quality economic development. Therefore, the inclusion of digital finance and stable financial policies contributes to consumer consumption levels and consumption structures by increasing residents’ incomes and the spatial spillover effect. The following conclusions were drawn from the above analysis: **Hypothesis 2:** Digital inclusive finance can effectively promote consumer consumption. ## 2.3 Digital inclusive finance, consumer consumption and high-quality economic development China’s current economic development strategy is characterized by high-quality economic development, which not only focuses on the total amount of economic development but also develops an economic system that emphasizes quality. Scholars are studying innovative, coordinated, green, open, and shared developments to achieve this goal. Consumer consumption is considered an efficient means to achieve high-quality economic development through the use of digital inclusive financing. The Keynesian demand theory stresses that insufficient consumption demand will causes overcapacity and restrains economic growth. Rostow believed that only through consumption can industrial development and upgrading be stimulated to promote economic development at a more advanced level. Through theoretical analysis, Wang (2017) suggested through a theoretical analysis that China’s sustainable economic development needs to ensure economic growth through consumption, which is also an inevitable choice for future development in China. Zhao (2020) found via model testing that China’s total consumption maintained a positive correlation with its economic growth and that the economic scale would continue to grow with the growth of total consumer consumption. Through empirical analysis, Ma (2007) found that consumer consumption has a significant guiding and driving effect on China’s economic growth, and that the change in the consumer consumption structure will also force the upgrading and adjustment of the industrial structure. Popescu (2010) and Lan(2022) argued that while the economy is important, maintaining a high quality of life requires healthy consumption patterns. Shu (2021) emphasized that the COVID-19 pandemic has significantly impacted China’s economy and highlighted the need to promote economic transformation, expand domestic demand, and achieve sustainable development driven by consumption. Xing (2023) found that consumption upgrading can improve agricultural green total factor productivity and promote the long-term, high-quality development of alcoholometers through technical efficiency and progress. Finally, Zhou (2022) contended that to cope with the pressure of sustainable economic development, it is crucial to improve consumption levels, and that digital payment methods can more effectively increase rural consumption and promote the upgrading of the consumption structure. The existing research has confirmed that consumer consumption is crucial to economic growth. Additionally, the fusion of financial technology and financial development has improved range, precision, and efficiency leading to an increase in consumer income levels. In the context of "double-loop" development, China is increasingly demanding expanding domestic demand, breaking away from dependence on foreign trade, and promoting economic development. This study proposes a conclusion based on the above analyses. **Hypothesis 3:** Consumer consumption mediates the relationship between digital inclusive finance and high-quality economic development. A schematic illustrating the mechanism of digital inclusive finance, household consumption, and high-quality economic development is presented in. In summary, the analysis of existing literature reveals two main gaps. Firstly, previous research has primarily focused on the relationship between digital inclusive finance, household consumption and high-quality economic development, with limited studies examining the linkages between the three variables, resulting in gaps in related research. Secondly, scholars in prior studies have not adequately considered the impact of digital inclusive finance on surrounding regions or the nonlinear relationship between the variables. Thus, this paper addresses these gaps by building upon previous studies and providing a more detailed examination of the relationship between digital inclusive finance, household consumption, and high-quality economic development. Consequently, this research makes a substantial contribution to the theoretical understanding and policy recommendations of the topic at hand. # 3. Model setting and data description ## 3.1 Model construction ### 3.1.1 Model setting based on spatial panel regression China’s rapid economic growth, driven by reforms and opening-up, has been accompanied by unbalanced and uneven development among provinces, leading to a spatial clustering of high-quality economic development at the national level. In this study, we employ a spatial econometric model to investigate the effects of digital inclusive finance on consumer consumption and high-quality economic development, we employ a spatial econometric model in this study. Specifically, we use a spatial Durbin model based on the test results of the model, with the following specifications: (1)-(3): $${CEE}_{it} = \beta_{1}W_{i}{DIF}_{it} + \beta_{2}{URE}_{it} + \beta_{3}{GBE}_{it} + \beta_{4}{CPI}_{it} + \beta_{5}{EOI}_{it} + \beta_{6}{CPE}_{it} + u_{i} + \gamma_{i} + \varepsilon_{it}$$ $$\varepsilon_{it} = \lambda W_{i}\varepsilon_{t} + v_{it}$$ $${HQE}_{it} = \beta_{1}W_{i}{DIF}_{it} + \beta_{2}{URE}_{it} + \beta_{3}{GBE}_{it} + \beta_{4}{CPI}_{it} + \beta_{5}{EOI}_{it} + \beta_{6}{CPE}_{it} + u_{i} + \gamma_{i} + \varepsilon_{it}$$ $$\varepsilon_{it} = \lambda W_{i}\varepsilon_{t} + v_{it}$$ $${HQE}_{it} = \beta_{1}W_{i}{CEE}_{it} + \beta_{2}W_{i}{DIF}_{it} + \beta_{3}{URE}_{it} + + \beta_{4}{GBE}_{it} + \beta_{5}{CPI}_{it} + \beta_{6}{EOI}_{it} + \beta_{7}{CPE}_{it} + u_{i} + \gamma_{i} + \varepsilon_{it}$$ $$\varepsilon_{it} = \lambda W_{i}\varepsilon_{t} + v_{it}$$ Where *u*_*i* represents the individual effect, *γ*_*i* is the time effect, *λ* is the spatial autocorrelation coefficient, *β*₁⋯*β*₇ are the regression coefficients, *W*_*i* is the spatial weight matrix, *ε*_*it* and *v*_*it* is the error terms. To better analyze the nonlinear relationship between digital inclusive finance, consumer consumption, and high-quality economic development, this paper proposes a semi-parametric spatial lag model with the following settings: (4)—(5): $${CEE}_{it} = \delta{\sum\limits_{j \neq i}{W_{ij}{CEE}_{it}}} + g({DIF}_{it}) + \beta_{1}{URE}_{it} + \beta_{2}{GBE}_{it} + \beta_{3}{CPI}_{it} + \beta_{4}{EOI}_{it} + \beta_{5}{CPE}_{it} + u_{i} + \gamma_{i} + \varepsilon_{it}$$ $${HQE}_{it} = \delta{\sum\limits_{j \neq i}{W_{ij}{HQE}_{it}}} + g({DIF}_{it}) + \beta_{1}{URE}_{it} + \beta_{2}{GBE}_{it} + \beta_{3}{CPI}_{it} + \beta_{4}{EOI}_{it} + \beta_{5}{CPE}_{it} + u_{i} + \gamma_{i} + \varepsilon_{it}$$ Where *δ* is the spatial lag utility coefficient, and *g*(DIF_it) is the non-parametric part. By taking the partial derivative of this formula, the nonlinear relationship between the development of digital inclusive finance on consumer consumption and high-quality economic development is obtained. ### 3.1.2 Model setting based on mediating effect and moderating effect It is essential to consider the mediating role of consumer consumption in the relationship between digital inclusive financing and high-quality economic development This study uses a mediating effect model to determine whether consumer consumption influences economic quality development. The specific models are shown in Eqs–. Additionally, to evaluate whether the role of consumer consumption interferes with the relationship between digital inclusive finance and economic quality development, a moderating effect model was constructed as follows: $${CEE}_{it} = \beta_{1}{DIF}_{it} + \beta_{2}{URE}_{it} + \beta_{3}{GBE}_{it} + \beta_{4}{CPI}_{it} + \beta_{5}{EOI}_{it} + \beta_{6}{CPE}_{it} + u_{i} + \gamma_{i} + \varepsilon_{it}$$ $${HQE}_{it} = \beta_{1}{DIF}_{it} + \beta_{2}{URE}_{it} + \beta_{3}{GBE}_{it} + \beta_{4}{CPI}_{it} + \beta_{5}{EOI}_{it} + \beta_{6}{CPE}_{it} + u_{i} + \gamma_{i} + \varepsilon_{it}$$ $${HQE}_{it} = \beta_{1}{CEE}_{it} + \beta_{2}{DIF}_{it} + \beta_{3}{URE}_{it} + \beta_{4}{GBE}_{it} + \beta_{5}{CPI}_{it} + \beta_{6}{EOI}_{it} + \beta_{7}{CPE}_{it} + u_{i} + \gamma_{i} + \varepsilon_{it}$$ $${HQE}_{it} = \beta_{1}{CEE}_{it} + \beta_{2}{DIF}_{it} + \beta_{3}{CEE}_{it}*{DIF}_{it} + \beta_{4}{GBE}_{it} + \beta_{5}{CPI}_{it} + \beta_{6}{EOI}_{it} + \beta_{7}{CPE}_{it} + u_{i} + \gamma_{i} + \varepsilon_{it}$$ ## 3.2 Variable selection ### 3.2.1 Measurement of the level of high-quality economic development Indicators of high-quality economic development should reflect many aspects, such as the continuous improvement of people’s lives and the steady development of the country’s economy. This study adopts the methods proposed by Zhang (2022), Shen (2022), and Wei (2022) to select 35 basic indicators to measure the level of high-quality economic development. The specific indicators are listed in, which presents an indicator system for high-quality economic development. ### 3.2.2 Variable description We selected the digital inclusive finance development index and consumer consumption as explanatory variables. The degree of development and the balance of digital inclusive finance are used to reflect the level of development. Additionally, a high-quality economic development index was chosen as the explanatory variable. To ensure the accuracy of our empirical model, we included several control variables: the urbanization rate, the general budget expenditure, consumer price index, investment efficiency, and capital production efficiency. summarizes a summary of the selected variables. The 14th Five-Year Plan and the Outline of the Long-term Goals for 2035 both emphasized the need to develop China’s urbanization to a higher level. This urbanization drive is expected to boost the building material market and promote economic growth. As a result, the level of urbanization can influence consumer consumption and high-quality economic development. Therefore, we included the urbanization rate as one of the control variables in this study. General budget expenditure refers to the planned allocation and use of centralized budget revenue by the state. This covers various fields such as capital construction, science and technology development, support for rural areas, and cultural and educational undertakings. This expenditure plays a significant role in promoting infrastructure construction, industrial upgrading, the integrated development of urban and rural areas, and other dimensions of high-quality economic development. Therefore, it was considered a significant control variable in this study. The Consumer Price Index reflects the changes in the prices of goods and services that are related to residents’ daily lives. An increase in the CPI indicates rising prices, which can have a significant impact on residents’ daily lives and economic development if sustained over time. Investment efficiency measures the benefits of capital construction investments from perspective of the national economy perspective. This can effectively avoid inefficient and ineffective allocation of production factors, resulting in a waste of resources. Therefore, improving investment efficiency is crucial for sustainable economic development. Capital production efficiency can be effectively measured as the output generated per unit of capital over time. This indicates that as output steadily increases, the production efficiency of capital improves, resulting in enhanced economic and social development. This reflects an improvement in the quality of economic development over time. # 4. Empirical analysis ## 4.1 Data selection and processing The main focus of this study was on panel data from 30 provinces, autonomous regions, and municipalities in China, covering the period from 2011 to 2020. Data from Tibet, Hong Kong, Macao, and Taiwan were excluded. Any missing data were filled in using the trendupdating method. summarizes the descriptive data. ## 4.2 Measurement of high-quality economic development index The improved entropy weight method was used to derive an index of high-quality economic development after calculating 35 specific indicators in five dimensions. The results of high-quality economic development indices for China’s provinces and cities are presented in and, respectively, which provide a comprehensive summary of the specific results. and illustrate the varying levels of high-quality economic development across China’s provinces and cities. As of 2020, Guangdong Province (0.579), Beijing (0.571), and Zhejiang Province (0.541) had the highest index values for high- quality economic development. In contrast, the three provinces and cities with the lowest index values were the Ningxia Hui Autonomous Region (0.413), Shaanxi Province (0.413), and Gansu Province (0.404). The level of high-quality economic development of different regions in China varies depending on their geographical location. The eastern region, located in a relatively superior geographic area, ranks first in terms of high-quality economic development. The central region has achieved steady economic growth rates and high-quality development levels in the eastern region. However, the western regions have relatively slow economic development owing to their geographical location and environment, resulting in a lower level of high-quality economic development than the eastern and central regions. Generally, the eastern regions exhibit a higher level of high-quality economic development than the western and central regions at the national level. ## 4.3 Spatial econometric regression results ### 4.3.1 Spatial autocorrelation test Prior to conducting the spatial econometric regression, it was necessary to examine the presence of spatial autocorrelation in the variables. Spatial autocorrelation is commonly assessed using the Moran’s index, which measures the degree of similarity between the values of a variable in adjacent regions. A Moran’s index greater than zero indicates a positive spatial correlation, whereas a negative value indicates a negative spatial correlation. As shown in, the Moran indices for high-quality economic development, digital inclusive finance, and consumer consumption were positive and statistically significant from 2011 to 2020, suggesting a strong spatial association between these variables in China’s provinces and cities. Therefore, this study employed spatial econometric techniques to explore these relationships. Based on the scatter plot of the Moran index in 2020 (Figs –), it is evident that there was a spatial correlation between high-quality economic development, consumer consumption, and digital inclusive finance among provinces and cities in China during the epidemic period. This positive correlation can be attributed to the government’s scientific and reasonable epidemic prevention measures, which minimized the impact of the epidemic on the economy. ## 4.4 Empirical results of spatial panel regression In this study, the spatial weights were calculated using a 0–1 matrix, where a value of 1 indicates adjacency between two provinces and 0 indicates otherwise. To determine the appropriate spatial econometric model to use, we conducted a joint significance test as well as the LR and LM tests. The p-value of R-LM (lag) was 0.343 when studying the impact of digital inclusive finance on quality economic development quality, and the other variables are significant. This suggests that the spatial error model can be inferred preliminarily. Subsequently, based on the results of both the LR and WALD tests, the spatial Durbin model was recommended as it is more general than spatial error or spatial lag models. Although the selected models of the LM, LR, and LM tests were not consistent, the spatial Durbin model was recommended. Therefore, we recommend using the spatial Durbin model with time and double-fixed variables after conducting Hausman and joint significance tests. When examining the impact of digital inclusive finance on consumer consumption, the LM (lag) and R-LM (lag) p-values were 0.382 and 0.522, respectively. However, both the LR and WALD tests were significant, suggesting the use of the spatial Durbin model. As the selected models of the LM, LR, and LR tests were not consistent, the spatial Durbin model was recommended. This is because it is more general than the spatial error and spatial lag models. After conducting the Hausman and joint significance tests, we found that the spatial Durbin model with time and individual double fixation should be used. When examining the impact of digital inclusive finance and consumer consumption on high-quality economic development, the p-value of R-LM (lag) was 0.378, and the remaining variables were significant. Subsequently, the LR and WALD tests found that both are significant, so spatial Durbin model can be used. Based on the Hausman test results, there was no significance found in the p-value of 0.150, and therefore, the random effects model was used to carry out the research. When examining the impact of digital inclusive finance on high-quality economic development (as shown), a positive correlation was found between digital inclusive finance and high-quality economic development. Specifically, an increase of 1% in digital inclusive finance resulted in a 4.3% increase in high- quality economic development, verifying Hypothesis H1. Additionally, an increase of 1% in the general budget for the control variables resulted in a 5.6% increase in high-quality economic development, suggesting a positive correlation between economic development and the general budget, with a higher budget level leading to better economic development. The reason for this may be that the shift in government spending from the general public budget to public investment, scientific and technological innovation, and environmental protection in China has prompted the improvement of infrastructure, increase of income, and enhancement of social welfare. As an important tool for adjusting income distribution, general public expenditure also contributes to the adjustment of the tax system and the reduction of the individual burden. This may explain why China has been able to achieve the goal of high-quality economic development, as general public expenditure plays a significant role in this process. When studying the impact of digital inclusive finance on consumer consumption, it was found that consumer consumption increased by 6.9% when digital inclusive finance increased by 1%, indicating a positive correlation between digital inclusive finance and consumer consumption. A higher level of digital inclusive finance leads to higher consumer consumption and vice versa, thus verifying H2. Among the control variables, investment efficiency was positively correlated with consumer consumption. Specifically, a 1% increase in investment efficiency will resulted in a 2.5% increase in consumer consumption. This may be due to the fact that an increase in investment efficiency raises the overall wealth level of society, increasing income expectations and boosting consumption. Meanwhile, the level of general budget expenditure was negatively correlated with consumer consumption, with a 1% increase in general budget expenditure resulting in a 2.3% decrease in consumer consumption. This is due to effective investment efficiency as firstly, improvement in investment efficiency can directly contribute to economic growth and enhance people’s income and wealth levels but also inevitably influence the consumption level and consumption structure. Secondly, a high level of investment efficiency can help stabilize the economy, enhance market confidence, and encourage investors to spend freely. When examining how digital inclusive finance and consumer consumption contribute to high-quality economic development, it was found that a 1% increase in digital inclusive finance will results in a 3.066% increase in high-quality economic development and a 1% increase in consumer consumption leads to an 8.161% increase in high-quality economic development, which is significant at the 1% level. Therefore, digital inclusive finance and consumer consumption are positively related to high-quality economic development, with a stronger driving effect on high-quality economic development. Furthermore, the lower the level of digital inclusive finance and consumer consumption, the weaker the impact on high-quality economic development. Among the control variables, investment efficiency is negatively correlated with the consumer price index, possibly due to the increased efficiency of investment in certain manufacturing industries, the replacement of manual labor with more intelligent manufacturing equipment, and a decrease in expected income from wages. Additionally, the rise in the consumer price index affects consumer spending to some extent. shows the model’s three spatial spillover effects, revealing a direct effect of 0.069 for digital inclusive finance on high-quality economic development, with a total effect of 0.109, but no significant indirect utility. This indicates that digital inclusive finance promotes economic development directly by impacting high quality, but not through regional spillover effects. According to Model (2), the direct effect of digital inclusive finance on consumer consumption was 0.044 and the total effect was 0.071, with no significant indirect utility. This suggests that digital inclusive finance enhances consumer consumption through a direct effect, but not through regional spillover effects. Based on the results of Model (3), it can be seen that the direct effect of digital inclusive finance on high-quality economic development was 0.030 and the total effect was 0.036, with no significant indirect effect. Additionally, the direct effect of consumer consumption on high-quality economic development was 0.083 and the total effect was 0.095, both of which were statistically significant at the 5% level. However, the indirect effect was not significant, indicating that the direct effect of consumer consumption on high-quality economic development was insignificant at the 5% level. ## 4.5 Semi-parametric spatial lag model The nonlinear relationship between digital inclusive finance and consumer consumption is shown in as an inverted U-shaped curve. The curve shows that When the level of digital inclusive finance development is below 2.4, it promotes positive consumer consumption. However, when high-quality economic development falls between the range of 2.4–2.6, the influence of digital inclusive finance on consumer consumption becomes weakens. On the one hand, digital inclusive finance provides consumers with more efficient and convenient financial services, enabling them to access more financial support, thereby promoting economic and consumption growth. On the other hand, the expansion of digital financial inclusion may lead to inappropriate marketing strategies and risky loans by financial institutions, resulting in a higher debt ratio for consumers, which can negatively affect their consumption levels. Nevertheless, the overall impact of digital inclusive financing on consumer consumption remains positive and mostly. shows a nonlinear relationship between digital inclusive finance and high- quality economic development, characterized by an inverted U pattern. The figure illustrates that digital financial inclusion has a positive effect on economic quality development until it reaches a level of 2.3, beyond which digital financial inclusion begins to have a negative impact on economic quality development. In the initial stage of economic development, income augmentation through consumption can boost economic growth and drive the upgrade of the consumption structure. However, as consumer demand saturates, certain consumption patterns may have negative societal impacts, such as excessive borrowing and overconsumption. These behaviors can lead to financial distress, financial crises, and other problems that can be detrimental to sustainable economic growth. Overall, the influence of digital inclusion on high-quality economic development appears to gradually diminish and ultimately becomes negative beyond a certain threshold. ## 4.6 Mediating effect This study examined the mediating effect of consumer consumption on the relationship between digital inclusive finance and the quality of economic development. We conducted a secondary test using bootstrap statistics to ensure robustness of the results. As shown in, all the regression results were significant at the 5% level. However, relying solely on ordinary regression alone could lead to biased conclusions. Therefore, we also considered the spatial spillover effects between regions to make the regression results more descriptive. We used bootstrapping to test for the presence of a mediating effect, which was indicated if the 95% confidence interval did not include 0. Our results showed that there is a mediating effect, as neither the direct nor indirect effects included zero in the bootstrap tests. Thus, our results support H3, indicating that digital inclusive finance influences the quality of economic development through the mediating effect of consumer consumption. This study introduces digital inclusive finance as an interaction term with consumer consumption to test the moderating effect, while considering the mediating effect. As shown in, the negative coefficient of the interaction term and the significantly positive coefficient of consumer consumption suggest that consumer consumption weakens the positive contribution of digital inclusive finance to high-quality economic development. The positive contribution of digital inclusive finance is more significant when consumer consumption is low. As the level of consumer consumption increases, the positive contribution of digital inclusive finance decreases. These findings indicate that consumer consumption has a significant substitution effect on the positive contribution of digital inclusive finance to high-quality economic development ## 4.7 Robustness test To ensure the robustness of our findings, we conducted a robustness test by using a geographical distance matrix instead of a 0–1 matrix for the weight matrix. summarizes the results, which demonstrated that digital inclusive finance has a significant positive effect on high-quality economic development at the 1% significance level and a positive effect on consumer consumption at the 10% significance level. Significantly, the numerical values were consistent with those obtained from the to 0–1 matrix, suggesting that our spatial econometric analysis is robust. As shown in Figs and, when comparing the bias diagram of digital inclusive finance and the bias diagram of digital inclusive finance on high-quality economic development by changing the spatial weight matrix and using the 0–1 matrix, it can be found that the two bias diagrams are essentially the same. This indicates that the empirical results are not affected by changing the spatial weight matrix in the semi-parametric lag model, and thus the spatial lag model used in this paper is robust. # 5. Research conclusions and suggestions The study aimed to investigate the correlation between digital inclusive finance, consumer consumption, and quality economic development in 30 provinces and cities in China from 2011 to 2020. The results suggest that: (1) There is a nonlinear relationship between digital inclusive finance and high-quality economic development, forming an inverted "U" shape. (2) Digital inclusive finance has a significant positive effect on public consumption, and the relationship between digital inclusive finance and public consumption also follows a nonlinear inverted "U" shape. (3) Consumer consumption mediate the relationship exists between digital inclusive finance and high-quality economic development. However, there is a significant substitution relationship between consumer consumption and digital inclusive finance in the development of a high- quality economy. Based on these findings, the following recommendations are proposed. First, the regional network infrastructure should be improved to ensure the accessibility of digital financial inclusion products for a broader population. Second, financial literacy and knowledge should be popularized to increase overall financial participation by strengthening financial education and encouraging diversified product usage to gain value from capital break-even. Moreover, optimizing the environment for the development of digital inclusive finance is critical, including formulating relevant policies, standardizing market orders, and establishing a credit system to ensure sustainable and healthy development. Only through the implementation of these measures can we realize the positive spillover effects of digital financial inclusion on the economy and society. Consumption is a crucial driver of economic growth, and plays an essential role in national development. Residents’ consumption demand is a significant source of economic growth. Improving the distribution system, strengthening the social security system, increasing residents’ income levels, and enhancing their purchasing power are necessary to promote consumption upgrades and growth. Moreover, the government can support consumer consumption by implementing preferential policies and other means of stimulating economic growth. Furthermore, optimizing the consumption structure can promote the upgrading and optimization of industrial structures. The structure and patterns of household consumption can directly affect an enterprise’s production and supply. Therefore, guiding consumers to adopt sustainable and green products and services can promote industrial upgradation, reduce environmental pollution and resource waste, and increase society’s overall well-being. Additionally, consumption increases residents’ sense of happiness and fulfillment, thereby promoting social harmony and stability. Thus, the government should strengthen consumer protection, maintain market order, crack down on illegal acts that infringe on consumers’ rights and interests, enhance their confidence, and protect their rights and interests to play a positive role in social harmony and stability. The government should also formulate corresponding policies to strengthen consumption-oriented industrial policies and support and encourage the development of green industries, such as environmental protection, energy conservation, low carbon, and new energy. This will guide consumers towards a sustainable and low-carbon lifestyle and enhance social awareness and a sense of responsibility for environmental protection. This study examined the relationships between digital inclusive finance, household consumption, and high-quality economic development. However, it has several limitations: (1) Although the paper highlights the connection between digital inclusive finance, household consumption, and high-quality economic development, there may be additional transmission paths among the three. Moreover, there may not be a simple one-way causal relationship between variables but a more intricate two-way causal relationship. Therefore, a more precise definition of the relationship between these two is required. (2) Indicators of high-quality economic development must be explicit. As China has recently proposed high-quality economic development, it has not yet formed a completely unified index system; therefore, the indicators should be more accurate. Given these concerns, future research will include more theoretical logic on the relationship between digital inclusive finance, resident consumption, and high-quality economic development. The Granger causality test was used to determine the causal relationships between them. Finally, as China develops, a high-quality economic development indicator system can be defined more accurately, allowing the establishment of a more precise indicator system. # Supporting information 10.1371/journal.pone.0285695.r001 Decision Letter 0 Shahzad Umer Academic Editor 2023 Umer Shahzad This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 3 Apr 2023 PONE-D-23-05574Digital inclusive finance, Consumer consumption and high-quality economic developmentPLOS ONE Dear Dr. Zhang, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by May 18 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <[email protected]>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Umer Shahzad, PhD Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <h ttps://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_b ody.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: <https://www.youtube.com/watch?v=_xcclfuvtxQ> 3\. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match.  When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section 4\. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: <https://www.youtube.com/watch?v=_xcclfuvtxQ> \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: No \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: In this study, this study utilizes panel data from 30 Chinese provinces and cities spanning 2011 to 2020 to explore the impact of digitized inclusive finance on consumer consumption and high-quality economic development through a spatial econometric model. But this issue has been extensively studied. Overall, the innovation and research value of this research is insufficient. Language style is colloquial. Moreover, there are many irregular errors in this work. Some comments are listing below: 1\. The major defect of this study is the debate or argument is not clear stated in the introduction session. Hence, I would suggest the author to enhance your theoretical discussion and arrives your debate or argument. 2\. Introduction. The logic of the introduction writing still needs to be strengthened, how to introduce from Digital inclusive finance, Consumer consumption to high-quality economic development. The authors need to elaborate on the core concepts of the article, explain the definition of the core concepts, and explain the practical necessity of studying high-quality economic development. 3\. The author needs to supplement the literature on the Digital inclusive finance, Consumer consumption and high-quality economic development. A summary of the research gaps in the existing literature allows the reader to understand the differences in the manuscripts. 4\. The mechanism analysis section seems so brief that the logical relationships of some variables are not accurately expressed. 5\. The author should provide more discussion of economic reasons for each regression result, not just describe the result. Moreover, there is not much discussion of the findings and how they link to the rest of the paper. 6\. The literature review does not cover some recent studies. Recently, some scholars have published quality papers on similar topic. Please see the following studies in this regard to strengthen your introduction and literature review. Does the digital economy promote industrial green transformation? Evidence from spatial Durbin model. How does digital finance affect industrial transformation. Going green in China: How does digital finance affect environmental pollution? Mechanism discussion and empirical test. How does financial development environment affect regional innovation capabilities? New perspectives from digital finance and institutional quality. Does the internet development put pressure on energy-saving potential for environmental sustainability? Evidence from China 7\. The summary section needs to reduce the description of the research background and add detailed research conclusions. 8\. A stronger motivation should be given or the contribution of this work should be clearly stated. 9\. The author needs to replace all Chinese references with English references. 10\. It is necessary to provide a mechanism analysis figure in the manuscript. 11\. In the manuscript, I did not see the reference section. I hope the author will supplement them in the revised version. 12\. It would be appropriate to indicate future research directions and limitations of this at the end of the conclusion section just before references. 13\. The language style is so colloquial. Please improve the use of English as well as the writing style throughout the paper, including the abstract and the main text. Please seek help of a native speaker or professional editorial services. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <[email protected]>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0285695.r002 Author response to Decision Letter 0 20 Apr 2023 Q1. The major defect of this study is the debate or argument is not clear stated in the introduction session. Hence, I would suggest the author to enhance your theoretical discussion and arrives your debate or argument. Modification instructions: Thank you very much for your suggestions. The author has rewritten the introduction as follows: Since the initiation of China's reform and opening-up policy, the country has experienced substantial economic growth and improvement in the living standards of its people. However, this progress has come at a cost, characterized by high pollution and energy consumption, leading to the emergence of serious environmental issues such as global warming. With the economy’s continuous development, Chinese people have raised their expectations for a higher quality of life, necessitating the exploration of approaches that promote high-quality economic development while prioritizing environmental protection. The 19th National Congress of the Communist Party of China in 2017 acknowledged that the primary contradiction within Chinese society was unbalanced and inadequate development, along with the ever-growing needs of the people for an enhanced standard of living. The Congress proposed a shift in the economic development mode, emphasizing the need for swift high-quality economic growth. The concept of high-quality economic development encompasses diverse fields, necessitating coordinated efforts across all domains. Scholars have extensively researched the primary challenges hindering high- quality development in China, which include uneven and insufficient economic growth. This challenge is manifested in the central and western regions of the country, which exhibit slower development than the eastern region, which has better resources and a thriving economy. To address this issue, inclusive finance has emerged as a viable solution aimed at providing effective financial services to marginalized groups such as small and micro-enterprises, low-income earners, and the poor. In recent years, digital inclusive finance has gained prominence, driven by advancements in internet technology, promoting information dissemination, improving the coverage of inclusive finance, and facilitating poverty eradication efforts. Digital inclusive finance has the potential to identify relevant groups accurately, provide them with the necessary financial support, and ultimately promote harmonious social development and sustainable economic growth. Additionally, digital inclusive finance can contribute to the growth in household consumption, which is one of the three key drivers of economic development. Previous economic models in China were characterized by low- and medium-end consumption. However, with continued economic development, people have become increasingly interested in high-quality consumption. Consequently, household consumption has become a crucial engine of economic growth, with its continuous expansion facilitating development across all economic sectors, boosting related industries, and generating employment opportunities. Furthermore, as household consumption continues to expand, it drives the optimization and upgrading of consumption structures. This upgrading process is marked by increasing demand for quality consumer goods and heightened attention to factors such as brand, culture, and service. These developments are important for the promotion of high-quality economic development in China. This study aims to investigate the relationship between digital inclusive finance, consumer consumption, and high-quality economic development in China using a spatial econometric model with panel data from 30 provinces and cities. In addition to examining the linear relationships between these variables, a semi-parametric spatial lag model is employed to investigate potential non- linear relationships. This study also analyzes an analysis of the mediating and moderating effects of consumer consumption on the relationship between digital inclusive finance and high-quality economic development. Finally, to test the robustness of the results, the study also replaces the 0-1 weight matrix with an economical weight matrix. The findings of this study are expected to provide valuable insights into the impact of digital inclusive finance on consumer consumption and high-quality economic development contributing to the ongoing efforts to develop a high-quality economy in China. This study contributes significantly to the existing literature in several ways. First, it examines the association between digital inclusive finance, household consumption, and high-quality economic development from both linear and nonlinear perspectives. The semi-parametric spatial lag model used in this study enables a more comprehensive analysis of the dynamic change process, adding to the credibility of this study. Second, this study focuses primarily on investigating the relationship between digital inclusive finance, household consumption, and high-quality economic development and endeavors to explore the paths of influence among these factors, offering guidance for future scholars interested in exploring these pathways. Finally, this study expands upon prior research by not only treating inclusive finance and consumption as explanatory variables, but also examining consumption as both an intermediary and moderating variable, providing a more nuanced understanding of the relationship among these three factors. 2\. Introduction. The logic of the introduction writing still needs to be strengthened, how to introduce from Digital inclusive finance, Consumer consumption to high-quality economic development. The authors need to elaborate on the core concepts of the article, explain the definition of the core concepts, and explain the practical necessity of studying high-quality economic development. Modification instructions: Thank you very much for your suggestions. The author has rewritten the introduction as follows: Since the initiation of China's reform and opening-up policy, the country has experienced substantial economic growth and improvement in the living standards of its people. However, this progress has come at a cost, characterized by high pollution and energy consumption, leading to the emergence of serious environmental issues such as global warming. With the economy’s continuous development, Chinese people have raised their expectations for a higher quality of life, necessitating the exploration of approaches that promote high-quality economic development while prioritizing environmental protection. The 19th National Congress of the Communist Party of China in 2017 acknowledged that the primary contradiction within Chinese society was unbalanced and inadequate development, along with the ever-growing needs of the people for an enhanced standard of living. The Congress proposed a shift in the economic development mode, emphasizing the need for swift high-quality economic growth. The concept of high-quality economic development encompasses diverse fields, necessitating coordinated efforts across all domains. Scholars have extensively researched the primary challenges hindering high- quality development in China, which include uneven and insufficient economic growth. This challenge is manifested in the central and western regions of the country, which exhibit slower development than the eastern region, which has better re-sources and a thriving economy. To address this issue, inclusive finance has emerged as a viable solution aimed at providing effective financial services to marginalized groups such as small and micro-enterprises, low-income earners, and the poor. In recent years, digital inclusive finance has gained prominence, driven by advancements in internet technology, promoting information dissemination, improving the coverage of inclusive finance, and facilitating poverty eradication efforts. Digital inclusive finance has the potential to identify relevant groups accurately, provide them with the necessary financial support, and ultimately promote harmonious social development and sustainable economic growth. Additionally, digital inclusive finance can contribute to the growth in household consumption, which is one of the three key drivers of economic development. Previous economic models in China were characterized by low- and medium-end consumption. However, with continued economic development, people have become increasingly interested in high-quality consumption. Consequently, household consumption has become a crucial engine of economic growth, with its continuous expansion facilitating development across all economic sectors, boosting related industries, and generating employment opportunities. Furthermore, as household consumption continues to expand, it drives the optimization and upgrading of consumption structures. This upgrading process is marked by increasing demand for quality consumer goods and heightened attention to factors such as brand, culture, and service. These developments are important for the pro-motion of high-quality economic development in China. This study aims to investigate the relationship between digital inclusive finance, consumer consumption, and high-quality economic development in China using a spatial econometric model with panel data from 30 provinces and cities. In addition to examining the linear relationships between these variables, a semi-parametric spatial lag model is employed to investigate potential non- linear relationships. This study also analyzes an analysis of the mediating and moderating effects of consumer consumption on the relationship between digital inclusive finance and high-quality economic development. Finally, to test the robustness of the results, the study also replaces the 0-1 weight matrix with an economical weight matrix. The findings of this study are expected to provide valuable insights into the impact of digital inclu-sive finance on consumer consumption and high-quality economic development contributing to the ongoing efforts to develop a high-quality economy in China. This study contributes significantly to the existing literature in several ways. First, it examines the association between digital inclusive finance, household con-sumption, and high-quality economic development from both linear and nonlinear perspectives. The semi-parametric spatial lag model used in this study enables a more comprehensive analysis of the dynamic change process, adding to the credibil-ity of this study. Second, this study focuses primarily on investigating the relation-ship between digital inclusive finance, household consumption, and high-quality economic development and endeavors to explore the paths of influence among these factors, offering guidance for future scholars interested in exploring these pathways. Finally, this study expands upon prior research by not only treating inclusive finance and consumption as explanatory variables, but also examining consumption as both an intermediary and moderating variable, providing a more nuanced understanding of the relationship among these three factors. 3\. The author needs to supplement the literature on the Digital inclusive finance, Consumer consumption and high-quality economic development. A summary of the research gaps in the existing literature allows the reader to understand the differences in the manuscripts. Modification instructions: Thank you very much for your suggestions. The author has rewritten the introduction as follows: 2.1 Digital Inclusive Finance and High-quality Economic Development In 2005, inclusive finance was introduced to promote equal access to financial services for all customers. With the advancement of Internet technology, digital inclusive finance has emerged. While there is limited research on the impact of digital inclusive finance on high-quality economic development, some scholars have examined the relationship between digital finance, digital economy, and high-quality development. For instance, Xiongying Wang (2022) found that digital inclusive finance can help small and medium-sized technology-based enterprises alleviate financing constraints, while Jianwei Li (2022) noted that it can effectively alleviate SME financing constraints, with private and family businesses benefiting in particular. Other studies have investigated the role of inclusive finance in promoting economic growth through an analysis of the endogenous growth theory (Jun He, 2019) and the impact of the digital economy on high-quality economic development from various perspectives (Ren Baoping, 2022; Zhan Wenqing, 2022; YAN Tao, 2022; Liu Jiaqi, 2022; Ge Heping, 2021). Wu (2021) investigated the impact of the Internet on energy-saving efficiency and found that it can promote energy conservation and emission reduction through technological progress, human capital, openness, and energy structure. Zhao (2022) showed that the digital economy can enhance the technological development and human capital of cities, with significant spillover effects across regions. Li (2022) and Du (2022) demonstrated that digital finance can reduce environmental pollution by promoting industrial structure upgrading and rationalization, with marketization and government support playing important roles in this process. Hang (2022) examined the link between finance and innovative development and found that financial development can attract human capital and promote innovation within and across regions. These studies provide valuable insights into the complex relationship between digital technology and economic and social development, highlighting the multifaceted nature of this phenomenon and its potential to foster high-quality growth and sustainability. Based on these findings, it is evident that digital inclusive finance can promote enterprise innovation, meet the capital needs of enterprise survival and development, and optimize and adjust industrial structures, thereby improving economic development. This study recommends the following actions to enhance the impact of digital inclusive finance on high-quality economic development: Hypothesis 1: the development of high-quality economies can be effectively promoted by digital inclusive finance. 2.2 Digital inclusive finance and consumer consumption Goods and services are essential components of consumer consumption, and involve monetary or virtual expenditures to meet people's basic needs for survival, development, and enjoyment. Researchers have studied the increasing consumption of consumers from the perspective of digital inclusive finance. Ang (2011) studied the impact of financial policy on private consumption in India and found that financial repression policies are linked to a reduction in private consumption fluctuations, while a more open financial system can help suppress it. Grossman (2014) discovered that digital finance could reduce potential power consumption by providing convenience to consumers. Cioacă (2014) demonstrated that internet technology can reduce income gaps and promote resident consumption when deeply integrated into society. Based on spatial correlation and data from 265 Chinese prefecture-level cities from 2011 to 2018, Wang (2022) found that residents could effectively maximize their consumption potential with digital inclusive finance. Compared with the western region, the eastern and central regions showed a more significant impact on driving consumption levels and the spatial spillover effect. Additionally, digital inclusive finance can drive an increase in consumer consumption in first and second-tier cities compared to third and fourth-tier cities. Based on China's provincial data from 2011 to 2017, Jiang (2020) empirically found that digital inclusive finance improves consumer consumption levels and optimizes consumption structures by reducing the income gap between urban and rural areas and optimizing the industrial structure. Yi (2018) observed that the inclusion of digital finance in promoting economic growth is stronger for rural, middle-income, and underdeveloped households, and is influenced by the education level and cognitive ability of the household head. Li (2022) contended that digital inclusive finance, as a distinctive form of finance, is a vital driver in augmenting household consumption. Empirical evidence confirms that it has a substantial effect on boosting rural consumption. From a climate change perspective and with reference to the historical temperature of cities, He (2022) found that digital financial inclusion can not only raise farmers' consumption but also promote consumption upgrading to a certain extent. Moreover, digital financial inclusion can enhance the ability of individuals to withstand climate change. Zhao (2022) asserts that with the advancement of mobile payments in China, digital inclusive finance can invigorate residents' consumption and enhance the happiness of vulnerable groups. Luo (2022) scrutinizes the association between digital financial inclusion and consumption inequality in China and demonstrates that digital financial inclusion can promote consumption among low-income individuals and diminish the income gap. According to Cheng (2022), consumption has played an increasingly substantial role in propelling the economy since China entered a new era. Rural income and consumption are crucial components of high-quality economic development. Therefore, the inclusion of digital finance and stable financial policies contributes to consumer consumption levels and consumption structures by increasing residents' incomes and the spatial spillover effect. The following conclusions were drawn from the above analysis: Hypothesis 2: Digital inclusive finance can effectively promote consumer consumption. 2.3 Digital Inclusive Finance, Consumer Consumption and High-quality Economic Development China's current economic development strategy is characterized by high-quality economic development, which not only focuses on the total amount of economic development but also develops an economic system that emphasizes quality. Scholars are studying innovative, coordinated, green, open, and shared developments to achieve this goal. Consumer consumption is considered an efficient means to achieve high-quality economic development through the use of digital inclusive financing. The Keynesian demand theory stresses that insufficient consumption demand will causes overcapacity and restrains economic growth. Rostow believed that only through consumption can industrial development and upgrading be stimulated to promote economic development at a more advanced level. Through theoretical analysis, Wang (2017) suggested through a theoretical analysis that China's sustainable economic development needs to ensure economic growth through consumption, which is also an inevitable choice for future development in China. Zhao (2020) found via model testing that China's total consumption maintained a positive correlation with its economic growth and that the economic scale would continue to grow with the growth of total consumer consumption. Through empirical analysis, Ma (2007) found that consumer consumption has a significant guiding and driving effect on China's economic growth, and that the change in the consumer consumption structure will also force the upgrading and adjustment of the industrial structure. Popescu (2010) and Li (2021) argued that while the economy is important, maintaining a high quality of life requires healthy consumption patterns. Su (2021) emphasized that the COVID-19 pandemic has significantly impacted China's economy and highlighted the need to promote economic transformation, expand domestic demand, and achieve sustainable development driven by consumption. Xing (2022) found that consumption upgrading can improve agricultural green total factor productivity and promote the long-term, high-quality development of alcoholometers through technical efficiency and progress. Finally, Zhou (2022) contended that to cope with the pressure of sustainable economic development, it is crucial to improve consumption levels, and that digital payment methods can more effectively increase rural consumption and promote the upgrading of the consumption structure. The existing research has confirmed that consumer consumption is crucial to economic growth. Additionally, the fusion of financial technology and financial development has improved range, precision, and efficiency leading to an increase in consumer income levels. In the context of "double-loop" development, China is increasingly demanding expanding domestic demand, breaking away from dependence on foreign trade, and promoting economic development. This study proposes a conclusion based on the above analyses. Hypothesis 3: Consumer consumption mediates the relationship between digital inclusive finance and high-quality economic development. A schematic illustrating the mechanism of digital inclusive finance, household consumption, and high-quality economic development is presented in Figure 1. Figure1 Mechanism analysis diagram In summary, the analysis of existing literature reveals two main gaps. Firstly, previous research has primarily focused on the relationship between digital inclusive finance, household consumption and high-quality economic development, with limited studies examining the linkages between the three variables, resulting in gaps in related research. Secondly, scholars in prior studies have not adequately considered the impact of digital inclusive finance on surrounding regions or the nonlinear relationship between the variables. Thus, this paper addresses these gaps by building upon previous studies and providing a more detailed examination of the relationship between digital inclusive finance, household consumption, and high-quality economic development. Consequently, this research makes a substantial contribution to the theoretical understanding and policy recommendations of the topic at hand. 4\. The mechanism analysis section seems so brief that the logical relationships of some variables are not accurately expressed. Modification instructions: Thank you very much for your suggestions. The author rewrote the mechanism of the article, and the result is as the answer to question 3. 5\. The author should provide more discussion of economic reasons for each regression result, not just describe the result. Moreover, there is not much discussion of the findings and how they link to the rest of the paper. Modification instructions: Thank you very much for your suggestions. The author added the discussion part of the article according to the reviewer's request. 6\. The literature review does not cover some recent studies. Recently, some scholars have published quality papers on similar topic. Please see the following studies in this regard to strengthen your introduction and literature review. Does the digital economy promote industrial green transformation? Evidence from spatial Durbin model. How does digital finance affect industrial transformation. Going green in China: How does digital finance affect environmental pollution? Mechanism discussion and empirical test. How does financial development environment affect regional innovation capabilities? New perspectives from digital finance and institutional quality. Does the internet development put pressure on energy-saving potential for environmental sustainability? Evidence from China. Modification instructions: Thank you very much for your suggestions. The author has added the above documents to the paper. 7\. The summary section needs to reduce the description of the research background and add detailed research conclusions. Modification instructions: Thank you very much for your suggestions. The author reduced the description of the background in the conclusion. 8\. A stronger motivation should be given or the contribution of this work should be clearly stated. Modification instructions: Thank you very much for your suggestions. According to the reviewer's requirements, the author added the description of the contribution part in the paper, which is as follows: This study contributes significantly to the existing literature in several ways. First, it examines the association between digital inclusive finance, household consumption, and high-quality economic development from both linear and nonlinear perspectives. The semi-parametric spatial lag model used in this study enables a more comprehensive analysis of the dynamic change process, adding to the credibility of this study. Second, this study focuses primarily on investigating the relationship between digital inclusive finance, household consumption, and high-quality economic development and endeavors to explore the paths of influence among these factors, offering guidance for future scholars interested in exploring these pathways. Finally, this study expands upon prior research by not only treating inclusive finance and consumption as explanatory variables, but also examining consumption as both an intermediary and moderating variable, providing a more nuanced understanding of the relationship among these three factors. 9\. The author needs to replace all Chinese references with English references. Modification instructions: Thank you very much for your suggestions. According to the requirements of reviewers, all Chinese literature is replaced with English literature in this paper 10\. It is necessary to provide a mechanism analysis figure in the manuscript. Modification instructions: Thank you very much for your suggestions. According to reviewer's requirement, mechanism analysis diagram is added in this paper. 11\. In the manuscript, I did not see the reference section. I hope the author will supplement them in the revised version. Modification instructions: Thank you very much for your suggestions. According to the requirements of reviewers, the supplement of literature review has been completed. 12\. It would be appropriate to indicate future research directions and limitations of this at the end of the conclusion section just before references. Modification instructions: Thank you very much for your suggestions. According to the reviewer's request, the author pointed out the future research direction and limitations at the end of the article. This study examined the relationships between digital inclusive finance, household consumption, and high-quality economic development. However, it has several limitations: (1) Although the paper highlights the connection between digital inclusive finance, household consumption, and high-quality economic development, there may be additional transmission paths among the three. Moreover, there may not be a simple one-way causal relationship between variables but a more intricate two-way causal relationship. Therefore, a more precise definition of the relationship between these two is required. (2) Indicators of high-quality economic development must be explicit. As China has recently proposed high-quality economic development, it has not yet formed a completely unified index system; therefore, the indicators should be more accurate. Given these concerns, future research will include more theoretical logic on the relationship between digital inclusive finance, resident consumption, and high-quality economic development. The Granger causality test was used to determine the causal relationships between them. Finally, as China develops, a high-quality economic development indicator system can be defined more accurately, allowing the establishment of a more precise indicator system. 13\. The language style is so colloquial. Please improve the use of English as well as the writing style throughout the paper, including the abstract and the main text. Please seek help of a native speaker or professional editorial services. Modification instructions: Thank you very much for your suggestions. The author asked native speakers or professional editors to polish the language, and the proof of polish is as follows: 10.1371/journal.pone.0285695.r003 Decision Letter 1 Shahzad Umer Academic Editor 2023 Umer Shahzad This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 28 Apr 2023 Digital inclusive finance, Consumer consumption and high-quality economic development PONE-D-23-05574R1 Dear Dr. Zhang, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <[email protected]>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <[email protected]>. Kind regards, Umer Shahzad, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0285695.r004 Acceptance letter Shahzad Umer Academic Editor 2023 Umer Shahzad This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 10 May 2023 PONE-D-23-05574R1 Digital inclusive finance, consumer consumption and high-quality economic development Dear Dr. Zhang: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <[email protected]>. If we can help with anything else, please email us at <[email protected]>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Umer Shahzad Academic Editor PLOS ONE [^1]: NO authors have competing interests.

# Introduction Lactoferrin (Lf) is a protein involved in a large array of immune system activities in mammals that all lead to host protective effects. In neutrophils, Lf is synthesized and stored in the secretory granules waiting for an external signal to be released; which is provided within the inflamed tissues. There, Lf is massively released, so that its iron-scavenging properties can be directed against microbes together with its direct microbicidal activity. The presence of high levels of Lf in inflammatory diseases indicates a possible use of Lf as a clinical marker. Therefore, Lf clearly belongs to the innate nonspecific immune system, but also acts as a modulator of the inflammatory process. Lf binds to neutrophil membranes and promotes the activation and phagocytosis of neutrophils. Lf was also reported as a promoter of motility, superoxide production, and release of proinflammatory molecules such as nitric oxide, Tumour Necrosis Factor-α (TNF-α), and Interleukin-8 (IL-8) from human neutrophils, monocytes and macrophages. It has also been reported to act *in vitro* as a chemo-attractant for human neutrophils, and other cells. Since inflammation may cause harmful systemic effects, there is a crucial need to regulate the immune process so that the response is commensurate. It is assumed that Lf, among others, exerts such a regulation of the immune response. The LPS- binding ability of Lf contributes to downregulating the activity and recruitment of innate immune cells. Lf was also shown to have anti-inflammatory properties, mainly by preventing the production and release of cytokines that induce recruitment and activation of immune cells at inflammatory sites. The ability of Lf to bind iron makes the protein also a powerful anti-oxidant. Thus, Lf may chelate ferric ion and prevent the formation of hydroxyl radicals and subsequent lipid peroxidation. Our laboratory has shown that neutrophils from allergic patients release ROS in response to allergens in an IgE-mediated mechanism. This mechanism was also involved in the induction of the expression of the key inflammatory enzyme cyclooxygenase-2, a process which requires formation of hydroxyl radicals through the Fenton reaction. In regard to allergy, Lf also seems to play important anti-inflammatory roles. Allergy is a process that involves the activation of lymphocytes, macrophages, mast cells, basophils, eosinophils, neutrophils and others. Interestingly, Lf is overexpressed in patients with allergies, and *in vivo* studies showed Lf protection against skin and lung allergies. Furthermore, the ability of Lf to destabilize tryptase, chymase, and cathepsin G, potent proinflammatory proteases released from mast cells, has been demonstrated. These authors also showed *in vitro* an inhibition of anti-IgE induced histamine and tryptase release from human mast cells by Lf. Finally, Lf decreases the recruitment of eosinophils, and reduces pollen antigen-induced allergic airway inflammation in a murine model of asthma. There are 3 defined types of IgE receptors, all previously described in neutrophils (FcεRI, FcεRII/CD23, and galectin-3). We have previously shown that neutrophils isolated from allergic patients produce a functional response to those Ags that produce clinical symptoms. There is increasing evidence of the participation of neutrophils in allergic processes in general, and in asthma in particular. Despite the evidence that Lf is involved in asthma allergic processes, it is unknown whether neutrophils can be one of the main cellular sources of this key inflammatory mediator directly in response of an IgE-mediated stimulus. Here we show for the first time the ability of human neutrophils to release Lf in response to allergens which induce positive skin prick tests and specific IgE in asthmatic patients. The amount of released Lf was correlated with serum specific IgE levels and the severity of allergic asthma symptoms. # Methods ## Ethics Statement The Hospital Universitario Virgen Macarena ethics committee approved the study and each subject gave written informed consent **(Ref: C.I. 1772)**. ## Materials The allergens (Ags) were commercially available Ag extracts, including D₁ (*Dermatophagoides pteronyssinus*), G₃ (*Dactylis glomerata*), T₉ (*Olea europaea)*, and W₆ (*Artemisia vulgaris*). They were purchased from Diater (Madrid, Spain). Platelet Activating Factor (PAF-C16, 1-o-hexadecyl-2-acetyl-sn-glycero-3-phosphorylcholine) was from Sigma (Madrid, Spain). Ficoll-Hypaque, phosphate-buffered saline (PBS), RPMI 1640, heat-inactivated foetal bovine serum, L-glutamine, penicillin and streptomycin were purchased from Lonza (Verviers, Belgium). Mouse monoclonal antibody (mAb) anti-FcεRI (α chain) clone AER-37 (CRA 1) was purchased from eBioscience (San Diego, CA, USA). Mouse mAb anti-FcεRII (CD23) clone 9P.25 was purchased from IZASA-Immunotech (Barcelona, Spain). Mouse mAb anti-galectin-3 clone A3A12 was from Abcam (Cambridge, UK). All culture reagents (including Ags) used in this work had endotoxin levels of ≤ 0.01 ng/ml, as verified by the Coatest *Limulus* lysate assay (Chromogenix, Mölndal, Sweden). ## Patients and controls The studied groups included adult atopic patients with bronchial asthma, and healthy non-atopic volunteer controls. Asthma severity was classified based on a current guideline, into four groups: intermittent, mild persistent, moderate persistent, and severe persistent. The patients with allergic rhinitis were diagnosed on the basis of criteria previously described. All subjects were lifelong non-smokers. Asthma was diagnosed on the basis of criteria previously described in detail. The patients had positive skin prick test Diater and specific IgE (HYTEC 288, Hycor Biomedical Inc.-IZASA, Barcelona, Spain) to one common allergen (house-dust mites and pollens). The subjects received neither treatment nor specific hyposensitization. The asthmatic patients were not allowed to take short-acting β₂-agonists within the 4 hours before challenge of neutrophils *in vitro*. Oral bronchodilators were withheld for 48 hours, and none of the subjects had taken corticosteroids, cromolyn sodium, or nedocromil sodium in the previous week. The healthy controls had no history of allergy or bronchial symptoms, and had negative skin prick test Diater and specific IgE (HYTEC 288) to a battery of inhalant allergens (house-dust mites, pollens, molds, and animal danders). No subjects had a history of infection within the previous 6 weeks. ## Cell isolation and culture Highly purified human peripheral blood neutrophils were isolated as previously described. Briefly, after isolation neutrophil preparations were further purified, depleting CD9⁺ cells (eosinophils), CD203c⁺ cells (basophils), and CD14⁺ cells (monocytes) using a magnetic cell- sorting system (MACS) by four rounds of incubation with mouse anti-human CD9, anti-human CD203c, and anti-human CD14 Abs, and then with anti-mouse IgG micromagnetic beads. The purity of the neutrophil preparations was determined as previously described. This purification method reduced contaminating eosinophils to 0.001–0.004% of the final cell population. The purity of neutrophils was on average \>99%. Monocytes, basophils and lymphocytes were not detected in the neutrophil preparations as previously showed. Neutrophils were cultured in RPMI 1640 medium supplemented with 10% heat-inactivated foetal bovine serum, 2 mM L-glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin, and maintained at 37°C in an atmosphere of 95% O₂ and 5% CO₂. None of the reagents affected the viability of the cells at the concentrations used in this work, as confirmed by the Trypan Blue dye-exclusion test. ## Lung function FEV₁ was measured using a dry spirometer (Vitalograph, Buckingham, UK). The best value of three manoeuvres was expressed as a percentage of the predicted value. ## Dissociation of neutrophil-bound immunoglobulins Ig molecules were dissociated from the cell-surface of neutrophils as described previously. Briefly, after isolation, neutrophils were resuspended in 5 ml of a solution of 0.13 M NaCl, 0.005 M KCl, and 0.01 M lactic acid which was adjusted with 1N NaOH to pH 3.9 and incubated on ice for 5 min. An equal volume of PBS was then added to the treated cells, and the mixture was centrifuged, washed with PBS, and resuspended in RPMI 1640 medium. After treatment, neutrophils were cultured with the different agents. ## Analysis of released Lactoferrin Lf was determined in the culture supernatant of 5 x 10⁶ cells cultured in 1 ml, using an ELISA kit from Oxis International Inc. (Portland, OR, USA), following manufacture's instructions. The sensitivity of the assay is 1 ng/ml. 5 μg/ml PAF was used as positive control. ## Statistical analysis All statistical analyses were performed using GraphPad Prism version 4.00 for Windows (GraphPad Software, San Diego, CA, USA). Normality distribution was first examined using Shapiro-Wilk normality test before analysis of statistical significance. Data are expressed as means ± S.E.M. A one-way ANOVA was used to make comparisons between groups. Regression analysis was performed using Pearson rank correlation coefficients. A level of p\< 0.05 was considered significant. # Results ## Lf is released in response to allergens by human neutrophils from allergic asthmatic patients The effect of specific Ags to which the patients were sensitized upon Lf release into the culture supernatant was examined by ELISA. As shown in , the specific Ags to which the patients were sensitized (n = 15) stimulated Lf release by neutrophils as confirmed by ELISA. As also shown in, neutrophils released Lf in a dose- and time-dependent manner after exposure to the same Ags that induce clinical allergic symptoms in the patients. Lf release was detectable from 10 min and continued to increase up to 30 min. Lf release was detected at an Ag dose of 5 μg/ml, reaching the maximum at a dose of 10 μg/ml. No effect was observed at any dose of irrelevant Ags, Ags to which the patients were not sensitized. shows that neutrophils from allergic asthmatic patients (n = 17) cultured with an Ag to which the patients were sensitized released a significantly higher Lf amount than neutrophils from healthy donors (n = 17) or from asthmatic patients (n = 17) cultured with an Ag to which the patients were not sensitized (p\<0.001). No statistical differences were observed between unstimulated cells, cells from healthy donors and cells from allergic subjects cultured in the presence of Ags to which the patients were not sensitized (p = 0.818). Nevertheless, Lf release was clearly detected when neutrophils from donors from all groups were treated with PAF. ## Lf is released in response to Abs against FcεRI and galectin-3 by neutrophils from allergic asthmatic patients Experiments were next performed to examine whether the engagement of IgE receptor/s could mimic the effect observed over the Lf release in response to Ags. The effect of agonist Abs against FcεRI (CRA1) or galectin-3 (A3A12) promoted the release of Lf into the culture medium (FcεRI\>galectin-3 (p\<0.001)), while anti-FcεRII (9P.25) had no effect (p = 0.788). A non-specific mouse IgG Ab was used as a negative control. We further investigated the participation of IgE in Ag-induced release. Lf release was not detected when IgE molecules were stripped from the neutrophil surface prior to Ag challenge, but was detected when neutrophils were incubated with PAF. The amount of LF released by neutrophils from patients suffering of intermittent asthma without symptoms in the previous 4 weeks (n = 17) , mild persistent asthma (n = 17), moderate asthma (n = 17), and severe asthma (n = 17), after incubation with allergens, was evaluated in relation to the patients’ levels of specific serum IgE. A significant correlation was observed between LF secretion and specific serum IgE. Next experiments were performed to determine whether the release of Lf from neutrophils is related with asthma or rhinitis. No statistical differences were observed between the levels of released Lf by neutrophils from intermittent allergic asthma patients without symptoms in the 4 previous weeks (n = 17) and by neutrophils from allergic rhinitis patients without symptoms in the 4 previous weeks (n = 17) (p = 0.635). We also analyzed whether Lf release was related with the level of symptoms. As shown in, all groups of asthmatics had a higher Lf level than those of control subjects. Significant between-group differences were detected for Lf, with a progressive increase in Lf concentration that was related to asthma severity. # Discussion In this study, we have shown that in response to Ag challenge, neutrophils from allergic asthmatic patients release Lf in an Ig-dependent manner. This is supported by the following evidence: a) *in vitro* challenge of highly purified neutrophils from allergic asthmatic patients with Ags induced the release of Lf; b) anti-IgE receptors Abs mimic the effect of Ags upon Lf release, and c) Ag- mediated Lf release is cancelled by stripping the IgE molecules from the neutrophil surface, suggesting an involvement of IgE/IgE receptors. Lf is secreted in an iron-free form from epithelial cells into most exocrine fluids, and particularly into milk. To the best of our knowledge, none of the identified cellular sources of Lf were found to produce and release this important inflammatory mediator upon Ag challenge. Thus, the present data provide the first evidence that a human cell produces and releases Lf as a consequence of direct Ag challenge. Lf is of particular relevance to allergic asthma and rhinitis. Compared with normal subjects, increased concentrations of Lf have been detected in basal nasal washes of allergic rhinitis patients during pollen season or Ag nasal challenge and in induced sputum and bronchoalvelolar lavage fluid from asthmatic patients. Despite the evidence that Lactoferrin (Lf) is involved in asthma allergic processes, it is unknown whether neutrophils can be one of the main cellular sources of this key inflammatory mediator directly in response of an IgE-mediated stimulus. In the context of the allergic asthma, the involvement of neutrophils remains controversial. Asthma is an inflammatory disease with a complex immunopathology involving several different cell types and mediators. Eosinophils have been considered the most important cells in the pathophysiology of asthma and other allergic diseases. However, the interest in neutrophils as important mediators of the asthmatic airway inflammation has been renewed because they are the first cells to enter the airway in response to an allergen challenge and because the presence of airway eosinophilia does not fully explain this pathologic process (28). In this sense, the presence of the three IgE- receptors has been described in neutrophils. Here we show that FcεRI and galectin-3 are involved in the process. We previously showed that these receptors are also involved in the release of IL-8, eosinophil cationic protein (ECP) and NFAT2 nuclear translocation. The Ab against FcεRII/CD23 had no effect on Lf release, however we previously found that the same Ab has an effect on L-selectin and CD66b expression, and in the production of histamine and metalloproteinase-9 by human neutrophils. The role of galectin-3 is controversial. Recent studies in murine models using galectin-3 gene transfer indicate that galectin-3 is anti-inflammatory, however, a large number of *in vivo* and *in vitro* studies suggest that galectin-3 is pro-inflammatory. Galectin-3-deficient mice develop significantly less airway hyperresponsiveness and dermatitis after allergen challenge and a lower T_H2 response. Galectin-3-deficient mast cells exhibit impaired mediator release and defective JNK expression. However, lack of galectin-3 drives response to Paracoccidioides brasiliensis toward a T_H2-biased immunity. In this sense, we found that Lf is produced upon activation of galectin-3, and because Lf is involved in ROS production, and the release of other toxic mediators, our data suggest that it has a pro-inflammatory role. In summary, we present evidence of a new mechanism of Lf release by human neutrophils from allergic asthmatic patients. In addition, the levels of serum specific IgE, and the severity of asthma symptoms correlated well with the Ag- mediated Lf release, therefore pointing out a possible use of Lf as a follow up tool to measure the progression of the allergic condition and/or the effectiveness of a treatment. In conclusion our data support a role of neutrophils as a source of Lf in Ag-induced rhinitis and asthmatic reactions. # Supporting Information [^1]: David Rodriguez and Ricardo Palacios are employed by Laboratorios Diater, Madrid, Spain, a commercial company. However, they did not provide any kind of support in the form of salaries for authors or any form of commercial contribution and only a scientific contribution. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. There are no patents, commercial products in development or marketed products to declare. [^2]: Conceived and designed the experiments: JM. Performed the experiments: LF-D AV-R IV CC RA. Analyzed the data: JM MP PB DR RP. Wrote the paper: JM AV-R MP PB DR RP.

# Introduction ## Brief background on NSCLC Globally, lung cancer is the second most common newly diagnosed cancer and the leading cause of cancer death. In 2012, there were an estimated 1.8 million new lung cancer cases and almost 1.6 million deaths. Non-small cell lung cancer (NSCLC) accounts for approximately 83% of all newly diagnosed lung cancers, and most patients (70%) are diagnosed with advanced disease. Despite treatment advancements, overall survival remains poor with 5-year survival estimates globally ranging from 10% to 20%. Lung cancer imposes a substantial economic and societal burden. In Europe, lung cancer–related premature mortality cost an estimated €17 billion in 2008. In the United States, annual lung cancer treatment expenditures were estimated at \$13.1 billion USD in 2014 and lost productivity due to premature lung cancer deaths was an additional estimated \$36.1 billion USD. ## Rationale Systemic therapy can provide a meaningful clinical benefit for patients with advanced NSCLC, and several new therapies have been approved since 2010. However, there is a paucity of published information describing how these therapies are used in real-world clinical practice, especially for second and later lines of therapy. Increased understanding of how these treatments are used in routine clinical practice and the associated clinical outcomes may provide insights into how these therapies benefit patients, inform strategies that support development of new therapies, and ultimately decrease the global economic and societal burden of NSCLC. ## Objectives The primary objective of this study was to describe treatment patterns and survival outcomes among patients who received second-line treatment for advanced NSCLC in routine clinical practice. The study also describes treatment patterns and available information on survival for the subset of patients who received third-line treatment. ## Approach To achieve these objectives, we conducted a systematic literature review and qualitative evidence synthesis of observational studies. # Methods ## Literature search strategy This systematic review was conducted in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. A review protocol does not exist. A tiered search string that included a combination of keywords and medical subject headings (MeSH;) was used to search the following databases: BIOSIS Previews, Current Contents Search, Embase, Gale Group PROMT, International Pharmaceutical Abstracts, Medline, and SciSearch. ## Study selection criteria Observational cohort studies that included detailed information on second- and third-line treatment patterns in patients with advanced NSCLC were eligible for inclusion. Only English-language studies published in peer-review journals between January 1, 2010, and March 1, 2017, were considered. This date range was intended to capture current real-world data on advanced NSCLC treatment patterns in order to account for changes in treatment guidelines related to newly approved therapies and biomarker-guided treatment decisions. Studies that described only one type of therapy, case reports, clinical trials, conference abstracts, and Delphi panels were excluded. Additionally, studies that introduced selection bias with patient selection criteria like requiring platinum-based chemotherapy or a specific mutation were excluded since the intent of the qualitative review was to describe the treatment patterns in a broad, unselected patient population. ## Study selection process Article titles and abstracts were initially screened by one reviewer (JD) to identify articles that potentially fulfilled the study selection criteria. Full- text articles were independently evaluated for inclusion by 2 reviewers (JD and MM). Additional publications were identified through examination of references cited by the included publications and were included if they also fulfilled the selection criteria. ## Data extraction Two independent reviewers extracted key information from each publication, including study location, time period, methods, patient and tumor characteristics, treatment regimens by line of therapy, survival outcomes, response rates, and biomarker testing information. If multiple publications described a single study, the extracted data were combined, and each publication was referenced to reflect this circumstance. Discrepancies during study data extraction were resolved by consensus between the reviewers with reference to the articles. If data differed between 2 publications describing the same study, data from the more recent publication were used. ## Main study measures The main study measures included the proportion of treatment regimens used in second- and third-line advanced NSCLC treatment and overall treatment rates in first-, second-, and third-line therapy. Treatment lines were defined based on the reporting authors’ definitions. Single-agent chemotherapies included systemic anti-cancer therapies other than targeted therapies that were prescribed as monotherapy. Combination chemotherapies comprised 2 or more systemic anti-cancer therapies, including both chemotherapies and targeted therapies. Single-agent targeted therapies included any targeted therapy prescribed as monotherapy. Investigational drugs included any drugs administered as part of a clinical trial. In studies that did not explicitly report treatment proportions among patients receiving second- or third-line treatment, the proportion was estimated based on the number of patients receiving a specific treatment regimen and the total number of patients receiving second- or third- line treatment. Survival outcomes were reported based on the original definitions described in each study. ## Assessment of bias Risk of bias and methodological quality of the studies were assessed using the 2013 RTI framework created for the Agency for Healthcare Research and Quality for observational studies. Two authors independently performed this assessment. # Results The literature search yielded 1,329 citations, and 10 additional citations were identified through examination of references cited by the included publications. After applying the selection criteria, 15 articles describing 12 different study cohorts were included in the qualitative data synthesis. Of the 12 study cohorts, 7 were retrospective medical record reviews or database analyses and 5 were prospective single-center or multicenter cohorts. The study cohorts included patients from Europe (7), North America (3), Asia (1) and South America (1). In general, there was minimal risk of bias within the studies, although several studies had issues that could lead to confounding of the reported outcomes, either due to the length of time during which the study was conducted or missing information on key study variables such as histology or use of oral therapies. Overall, the study results were judged as believable after accounting for limitations. ## Patient characteristics Median age at advanced NSCLC diagnosis ranged from 59 to 68 years in 10 studies that reported this information. Males comprised a larger proportion of the study population in all cohorts. There were 2 studies that included only patients with a single NSCLC histology—one with only squamous patients, and one with only non- squamous patients. Among the other studies, adenocarcinoma was the most common NSCLC histology, with the proportion of adenocarcinoma patients ranging from 33% in Brazil to 77% in Japan. Two studies enrolled only stage IV patients. The proportion of patients who never smoked was highest in the Japanese cohort. Five studies reported performance status after first-line therapy or at initiation of second-line therapy, with most patients having a Karnofsky Performance Status of 70 or better or an Eastern Cooperative Oncology Group score of 0 or 1. ## Treatment patterns ### Second-line treatment and describe the distribution of second-line treatment regimens in each of the studies. The proportion of patients receiving second-line therapy among the studies varied depending on how the study cohort was selected and what treatments were included. In studies that followed patients from initial NSCLC diagnosis, the proportion of patients who received second-line treatment ranged from 8% in a population-based Canadian study that did not include oral therapies (epidermal growth factor receptor \[EGFR\] tyrosine kinase inhibitors \[TKIs\]) to 53% in a German study at a single institution. In 10 studies that provided detailed information on prescribed treatment regimens, single-agent chemotherapy was the most commonly used second-line treatment regimen except in one study from Japan. In the Japanese study, single- agent targeted therapy was the most commonly prescribed second-line treatment. Docetaxel, pemetrexed, and gemcitabine were among the top 3 most frequently used single-agent chemotherapeutics across the studies. In Japan gefitinib was the most frequently used second-line single-agent targeted-therapy. Outside of Asia, the most frequently used single-agent targeted therapy was erlotinib. Pemetrexed was one of the most commonly prescribed second-line treatments in the studies that included patients treated after its initial approval in 2004. The study conducted in Canada reported that pemetrexed use increased to 70% among treated second-line patients in 2008, when government funding was approved for pemetrexed. Second-line treatment with EGFR-TKIs increased in the European studies after initial approval of erlotinib in 2005. This uptake occurred prior to the widespread introduction of EGFR mutation testing, which was not required at the time of initial erlotinib approval. Platinum-based doublet chemotherapy was administered in all 7 studies that described the use of second-line combination chemotherapy regimens. The most frequent use of combination chemotherapy was reported in a study that included only patients with squamous histology. ### Third-line treatment Nine of the 12 studies reported detailed information on patients who received third-line therapy. In these studies, approximately 30% of patients received third-line treatment. Single-agent chemotherapy accounted for about 50% of third-line treatments in general, and targeted therapy accounted for almost 40%. Docetaxel, gemcitabine, vinorelbine, and pemetrexed were the most frequently administered third-line chemotherapies. Erlotinib was the most common single- agent targeted therapy in the third-line setting in all countries. ### Other lines of therapy All 12 studies reported details about the distribution of first-line treatment regimens. The most frequently administered first-line treatment across all countries was platinum-doublet chemotherapy. Three European studies reported use of targeted therapy in the first-line setting. In an Italian cohort, gefitinib, bevacizumab, and erlotinib were administered to 6%, 4%, and 1% of first-line treated patients, respectively. Erlotinib was administered to 3% of first-line patients in a German cohort. Only 4 patients (0.4%) in a French cohort received an EGFR-TKI in first-line therapy. In Japan, fourth-line chemotherapy was reported in 17.7% of the patients who received first-line chemotherapy. The top 3 anti-cancer treatment regimens in the fourth-line setting were single-agent S-1 (22%), docetaxel (21%), and gefitinib (21%). In Brazil, 2.5% of all patients with stage IV NSCLC (4.3% of the patients who received first-line chemotherapy) received fourth-line chemotherapy. ## Survival outcomes Eleven of 12 studies included in this review presented overall survival (OS) data. Five studies reported OS from time of metastatic NSCLC (mNSCLC) diagnosis. Three of these studies reported the median OS among patients with advanced NSCLC who received second-line therapy. Median OS from mNSCLC diagnosis date ranged from 8.7 months among patients who received only first- and second-line (ie, no third-line) treatments in a single-center German cohort to 17 months among patients who received at least second-line treatment in a single-center Brazilian cohort. In 5 studies that described median OS from time of second-line treatment initiation in routine clinical practice, median OS ranged from 4.6 months (95% CI, 3.8–5.7) to 12.8 months (95% CI, 10.7–14.5). Median OS estimates were close to 8 months in 2 studies that reported this information specifically for patients with adenocarcinoma in the second-line setting. Median OS estimates were reported from the time of third-line treatment initiation in 4 studies and median OS ranged from 3.8 (95% CI, 2.6–5.4) to 12.0 months (95% CI, 9.3–14.2) ## Biomarkers Two studies reported biomarker testing and treatments prescribed based on biomarker status, one in Italy (LIFE) and one in the US that included only patients with adenocarcinoma. In the US study, EGFR testing frequency increased significantly from 2.3% between 2007 and 2009 to 32% in the first 6 months of 2011. The LIFE study, which was conducted in 2011–2012, reported that 60% of Italian patients received biomarker testing. In this study, biomarker testing was more frequent among patients who were younger, female, never-smokers, and those with adenocarcinomas. Both studies provided evidence of biomarker-driven therapy choices within their cohorts. The LIFE study reported that 26 of 37 (70%) patients who received EGFR- TKIs as first-line therapy were known to have EGFR-activating mutations prior to first-line treatment. In the US cohort, 50% of 24 patients with known *EGFR* mutations received erlotinib as second-line treatment compared with 17% of the 89 *EGFR* wild-type patients. The use of erlotinib was significantly more common in patients with *EGFR* mutations compared with *EGFR*-wild type patients (*P* \<.001). # Discussion The objective of this study was to describe real-world treatment patterns and survival outcomes for patients with advanced NSCLC who received second-line or later treatments, through a review of recently published observational studies. To our knowledge, there are no systematic reviews that summarize this information. By qualitatively reviewing this information from multiple countries, we provide a broad overview of how patients are managed in this setting around the world. Overall, the retrieved studies showed how newly approved therapies are rapidly integrated into clinical practice, and how clinical practice evolves in response to therapeutic advances. In addition, most studies demonstrated a high level of clinician adherence to international treatment guidelines, such as those from European Society of Medical Oncology and the US National Comprehensive Cancer Network. Among the studies that focused solely on patients with non-squamous NSCLC, single-agent pemetrexed, docetaxel, or EGFR-TKIs were the most common treatment choices, which aligns with current guidelines. However, there were examples of non-adherence to guidelines, namely use of platinum-doublet chemotherapies as second-line treatment. This was highest in the all-squamous cohort, which is consistent with the limited number of approved second-line treatments for this particular subgroup of patients with advanced NSCLC that were available at the time these studies were conducted. Moreover, while specific patient-level information was not available from these studies, it is possible that use of platinum-doublet chemotherapy in the second-line setting could be motivated by oncologists’ perceptions of the superiority of these particular treatment regimens. Survival after initiation of second-line treatment was reported in most studies, and was generally consistent with outcomes reported from pemetrexed, docetaxel, and erlotinib clinical trials. The consistency of survival outcomes suggests that clinical trial results may be generalizable to real-world patient populations. Additionally, the data confirm the limited survival benefit of treatments that were available at the time these studies were conducted. An exception to this was the lower survival among second-line patients in Germany reported by Zietemann and Duell compared with outcomes reported from other European studies with similar outcome definitions. No explicit reason for the lower survival reported in this study was identified based on patient and tumor characteristics, or the study setting. The range of survival outcomes reported by the studies included in this review exemplifies the variation that can occur across multiple studies and may be due to individual-level patient differences. In addition, these survival differences highlight the importance of examining population-level data and outcomes from multiple observational studies to understand the range of outcomes experienced in routine clinical practice and the limited ability to draw conclusions from studies that do not report detailed patient-level data. In most studies that reported the proportion of patients with advanced NSCLC treated across lines of therapy, between one-third to one-half of those who received first-line treatment also received second-line therapy. None of these studies provided information to explain why patients did not receive later lines of therapy. However, Gridelli et al reported that poor performance status and older age were the most common reasons patients did not receive first-line treatment. These reasons, in addition to a poor response to first-line treatment, may be why patients with advanced NSCLC do not receive later lines of therapy. Newer treatments such as immunotherapies may offer these patients an opportunity to receive treatment because of their more favorable toxicity profiles compared with existing treatments. Additional studies that examine why patients are not treated could help identify opportunities to increase the proportion of patients with advanced NSCLC who derive benefit from treatment. There was limited information reported on biomarker-guided therapy decisions in the studies included in this review, likely due to the time periods covered by the studies. In the studies covering more recent time periods, the rapid uptake seen in the use of targeted therapies and biomarker testing is encouraging. It reveals that biomarker-driven treatment decisions are being integrated into routine clinical practice. This information provides a positive outlook for the uptake of newer targeted therapies in the second- and third-line settings. The rapid uptake of anaplastic lymphoma kinase (ALK) inhibitors such as crizotinib and ceritinib in patients with advanced NSCLC harboring *ALK* mutations provides further evidence that biomarker-directed therapy is an important treatment option for these patients. Biomarker testing may impact uptake of newly approved targeted therapies. More information on how and when biomarker testing is performed in routine clinical practice is needed, since it is unclear in the included studies whether biomarker testing was more commonly performed prior to first-line versus second- line treatment. Across the European studies, erlotinib use increased despite approval of *EGFR* mutation testing in Europe after erlotinib approval. This suggests a trend among clinicians of using new therapies when clinically appropriate, despite limitations in access to biomarker testing. These trends imply positive uptake of new immunotherapies like programmed death-ligand 1/programmed death 1 (PD-L1/PD-1) checkpoint inhibitors despite evolving biomarker testing guidelines. A few limitations should be considered when interpreting the information provided in this review. A meta-analysis could not be conducted due to the heterogeneity of the study designs and data presentation in the included studies. There were substantial differences in patient selection procedures, data collection methods, and outcome measure definitions. Thus, a qualitative data synthesis was the most appropriate way to summarize these studies. Moreover, the review was robust and systematic, which renders the search and presentation of results transparent and reproducible. Additionally, publication bias could influence the evidence presented in this review; however, studies were found from multiple countries. Finally, conclusions from the results presented herein are limited by the observational design of the included studies. # Conclusions Real-world studies of second- and third-line treatment patterns in advanced NSCLC provide insights into how evidence from clinical trials impacts clinical practice. The studies included in this qualitative systematic review demonstrate the limited clinical benefit of the second- and third-line treatments for advanced NSCLC that were available at the time these studies were conducted. Within recent years, several novel therapies have been approved for use in these settings, and it will be important to determine the benefit that these and other new treatments may provide to patients with advanced NSCLC. # Supporting information Thanks to scientific communications leaders Cindy Yun and Alicia Chung of Genentech, Inc., for medical writing support. This study was sponsored by F. Hoffmann-La Roche, Ltd. Medical writing assistance for this manuscript was provided by Health Interactions, Inc, paid for by F. Hoffmann-La Roche, Ltd. [^1]: We have the following interests to report: This study was sponsored by F. Hoffmann-La Roche, Ltd. Ms. Davies is employed by Hoffmann-La Roche, Ltd. and Dr. Waterkamp and Dr. McCusker are employed by F. Hoffmann-La Roche AG. Ms. Davies reports support for manuscript preparation from Roche-Genentech, during the conduct of the study; Employee; outside the submitted work. Dr. Gridelli reports personal fees from Roche, personal fees from Eli Lilly, outside the submitted work. Dr. McCusker reports support for manuscript preparation from Roche-Genentech, during the conduct of the study; Employee; stock ownership from Roche-Genentech, outside the submitted work. Dr. Waterkamp reports Employee of Roche/Genentech, outside the submitted work. Decision to publish and medical writing support was provided by scientific communication leaders Cindy Yun and Alicia Chung who are employees of Genentech Inc. Support in preparation of the manuscript was contracted to Health Interactions, Inc, and paid for by F. Hoffman-La Roche, Ltd. There are no patents, products in development or marketed products to declare. This does not alter our adherence to all the PLOS ONE policies on sharing data and materials. [^2]: **Conceptualization:** JD MEM. **Formal analysis:** JD MEM. **Funding acquisition:** JD. **Investigation:** JD MEM. **Methodology:** JD MEM. **Project administration:** MEM. **Resources:** JD MEM. **Supervision:** MEM. **Validation:** MEM. **Visualization:** JD MEM. **Writing – original draft:** JD MEM. **Writing – review & editing:** JD MP CG FD DW MEM.

# Introduction In eukaryotic organisms as diverse as yeasts and mice, lifespan is extended by caloric restriction or mutations that inactivate conserved growth signaling pathways (reviewed in). Conserved elements of these pathways include RAS proteins and members of the AKT/PKB family of kinases that function in glucose signaling in yeast and insulin/insulin-like growth factor (I-(IFG-1)) signaling in higher eukaryotes, as well as elements of TOR-dependent nitrogen signaling pathways. Mutations that promote, rather than inhibit, signaling in these pathways shorten lifespan. The mechanisms by which caloric restriction or mutational inactivation of growth signaling pathways extend lifespan remain unclear. Both induce stress responses that depend on heat shock and other proteins, including proteins that mitigate oxidative stress. The induction of oxidative stress responses, in particular, might promote longevity by protecting against reactive oxygen damage to lipids, protein and DNA, all of which accumulate oxidative damage during aging. Oxidative damage to DNA could be responsible for some of the increased genome instability associated with aging in all eukaryotes. However, a definitive role for oxidative damage in aging has not yet been established. For example, in mice, a mutation in the superoxide-scavenging enzyme *SOD2* elevates oxidative damage to DNA, but does not shorten lifespan. The absence of a correlation between levels of reactive oxygen and longevity has also been noted in *Drosophila*. It also remains unclear how cellular responses to other stresses might inhibit age-dependent increases in DNA damage and promote longevity. The budding yeast *Saccharomyces cerevisiae* has proved to be a valuable model organism for investigating conserved pathways regulating lifespan in all eukaryotes. Replicative lifespan in budding yeast is assessed by determining the number of times cells divide in the presence of nutrients before they senesce and die via an apoptotic-like mechanism. An important factor determining replicative lifespan is DNA replication stress–i.e., inefficient DNA replication that leads to replication fork stalling–which stimulates recombination at a replication fork barrier in the rDNA locus ( and references therein). Chronological lifespan in budding yeast is distinct from replicative lifespan and is measured by assessing the viability of cells driven into a growth- arrested state by nutrient depletion. The quiescent state of nutrient-depleted budding yeast cells shares features with postmitotic cells in higher eukaryotes, including growth arrest of a large fraction of cells with a G1 content of DNA. Nutrient-depleted budding yeast cells can survive in this quiescent state for an extended time, during which replenishing the medium restores growth. Similar to replicatively aged budding yeast cells, chronologically aged cells eventually die via an apoptotic-like mechanism that includes degradation of DNA. Most efforts to understand how alterations in nutrient-signaling pathways impact chronological lifespan in budding yeast have focused on changes in nutrient depletion-induced stress resistance, especially resistance to oxidative stress. Stress resistance in this organism is mediated in large part by the “Rim15 regulon”, which is upregulated by the Rim15 kinase in response to depletion of nutrients. The Rim15 regulon includes a number of genes that, like *RIM15*, promote longevity in the chronological aging model. This includes genes encoding the stress-responsive transcription factors Msn2 and Msn4 and genes induced by Msn2 and Msn4, such as *SOD1* and *SOD2* encoding superoxide dismutases that mitigate oxidative stress. In the presence of glucose and other nutrients, Rim15 kinase activity and the induction of stress resistance are inhibited by the Ras- cAMP glucose signaling pathway, the constitutive activation of which shortens lifespan. Furthermore, nuclear localization of Rim15, which is required for induction of the Rim15 regulon, is negatively regulated by Sch9 in the presence of glucose. Abrogation of this negative regulatory mechanism in *sch9Δ* cells extends chronological lifespan. Rim15 also is negatively regulated by the nitrogen-sensitive TOR-dependent nutrient signaling pathway, inhibition of which also extends chronological lifespan, and by Pho80-Pho85 cyclin-CDK-dependent pathways that respond to phosphate levels. Chronological lifespan extension associated with the induction of oxidative stress responses by Rim15 and other proteins is consistent with the longstanding “free radical” theory of aging, which posits oxidative damage as a major determinant of lifespan in all eukaryotes. However, Rim15 also mediates the G1 arrest induced by nutrient deprivation in budding yeast (reviewed). In addition to the G1 arrest induced when medium is depleted of nutrients, Rim15 is also required for G1 arrest when TOR-dependent nitrogen-signaling pathways are inhibited. Phosphate starvation also leads to the activation of Rim15 and entry into a G0-like state. Thus, Rim15 integrates signals from several nutrient- signaling pathways to arrest cells in the G1 phase of the cell cycle during nutrient deprivation, in addition to its induction of oxidative and other stress responses. The possibility that alterations in the Rim15-dependent G1 arrest induced by nutrient deprivation impact chronological lifespan when nutrient signaling pathways are altered has not been investigated previously. The shorter lifespan of nutrient-depleted *rim15Δ* cells is accompanied by growth arrest throughout the cell cycle, including S phase. A similar growth arrest throughout the cell cycle is observed during nutrient depletion of cells harboring the constitutively activating *RAS2*^val19 mutation, which also shortens chronological lifespan. Growth arrest of *RAS2*^val19, *rim15Δ* and other cells in S phase could be caused by a reduction in the levels of nucleotides and other factors required for efficient DNA synthesis, which would lead to replication stress. As in other eukaryotes, replication stress promotes genome instability, and apoptosis in budding yeast. Both genome instability and apoptosis are characteristic features of chronological aging in this organism, as well as in many other eukaryotes. Based on these considerations, we hypothesized that inhibition of nutrient- signaling pathways can extend the chronological lifespan of budding yeast by inducing a more efficient nutrient depletion-induced G1 arrest. This more efficient G1 arrest would protect against replication stress-induced genome instability and apoptosis by reducing the growth arrest of cells in S phase with incompletely replicated chromosomes. To test this hypothesis, we measured the frequency with which cells arrest in G1 and undergo apoptosis during nutrient depletion under conditions previously shown to alter the chronological lifespan of this organism, including caloric restriction and osmotic stress. We also examined the frequency with which nutrient depletion induces G1 arrest and apoptosis in strains defective in responses to DNA replication stress. The results of these experiments demonstrated that nutrient depletion causes replication stress in cells that fail to enter into or maintain a G1 arrest when deprived of nutrients and instead arrest growth in S phase. They suggest that, in addition to its role in replicative aging in budding yeast, replication stress is an important factor determining chronological lifespan in this organism. # Results ## Correlation between lifespan and efficiency of G1 arrest in nutrient-depleted cells *SCH9* encodes a kinase that promotes glucose and nitrogen signaling in budding yeast in parallel with the Ras-cAMP glucose-sensing pathway ( and references therein). Sch9 is an orthologue of AKT/PKB kinases that function in growth signaling downstream of IGF-1 in higher eukaryotes (reviewed). Deletion of *SCH9* enhances Rim15-dependent stress responses and extends chronological lifespan, similar to the lifespan-extending effects of some mutations in AKT homologues in higher eukaryotes. To test whether deletion of *SCH9* would increase the efficiency of G1 arrest during nutrient depletion (as predicted by our hypothesis), we compared the extent of this arrest in wild type and *sch9Δ* cells by determining budding indices (fraction of budded cells), since cells with buds have passed Start and have entered or proceeded through S phase. We also employed flow cytometry to determine the proportion of nutrient-depleted cells that arrested with a G1 content of DNA. Measurements of cellular DNA content also revealed cells undergoing apoptotic DNA degradation indicated by less than a G1 DNA content. Consistent with our hypothesis, we found that during nutrient depletion, *sch9Δ* cells arrested with a G1 content of DNA more efficiently than did wild-type cells. A more efficient G1 arrest in *sch9Δ* compared to wild type cells was indicated by the increased height and narrowness of the G1 DNA-content peak in flow cytometry profiles in parallel with a reduced budding index. As reported earlier by Fabrizio *et al.*, we found that *sch9Δ* cells had a longer chronological lifespan than wild-type cells. Also consistent with lifespan extension in *sch9Δ* cells, we found that at later times, *sch9Δ* cells suffered less apoptotic DNA degradation. Rim15 is a member of the PAS kinase family containing PAS domains (reviewed). PAS domains are highly conserved regulatory modules that respond to a variety of stimuli, including oxygen, redox status and energy levels. As described above, Rim15 integrates signals from several nutrient-signaling pathways to regulate stress responses and withdrawal from the cell cycle. In the presence of nutrients, activation of Rim15 and the induction of stress responses are inhibited by Sch9 and other elements of nutrient signaling pathways. During nutrient depletion, cells deleted of *RIM15* have a shorter lifespan and arrest throughout the cell cycle, which was indicated by a higher budding index compared to wild type cells. We also found that nutrient-depleted *rim15Δ* cells exhibit a higher budding index. In addition, we found that compared to wild type cells, *rim15Δ* cells arrest less efficiently with a G1 content of DNA and undergo more rapid apoptotic DNA degradation. Deletion of *SCH9* from *rim15Δ* cells suppresses their shortened lifespan (. Compared to *rim15Δ* cells, *sch9Δ rim15Δ* cells exhibited a more efficient G1 arrest and a reduced frequency of apoptotic DNA degradation during nutrient depletion. Ras2 plays a role in a glucose-signaling pathway that functions upstream of Rim15 in parallel with the Sch9 pathway (reviewed). Similar to deletion of *SCH9*, deletion of *RAS2* extends the chronological lifespan of nutrient- depleted budding yeast cells. We found that it also decreases budding index, indicating a tighter G1 arrest. A tighter G1 arrest in *ras2Δ* cells was detected by flow cytometry as well. In contrast, constitutive activation of Ras2 by the *RAS2*^val19 mutation shortens chronological lifespan. In agreement with an earlier observation, we found that the *RAS2*^val19 mutation decreased the efficiency of G1 arrest during nutrient depletion. A relationship between chronological lifespan and efficiency of G1 arrest during nutrient depletion was also suggested by the effects of another experimental manipulation previously shown to decrease the chronological lifespan of nutrient-depleted cells—culturing cells in defined (SC) medium rather than rich (YPD) medium. Although the basis for the decreased chronological lifespan in SC medium is not clear, nutrient-depleted cells cultured in this medium exhibit a higher metabolic rate, which correlates with shorter chronological lifespan. The shorter lifespan of DBY746 cells in SC compared to YPD medium was accompanied by a less efficient G1 arrest. It was also accompanied by extensive apoptotic DNA degradation, which was absent from cells cultured in YPD medium at the same time points. These effects were also detected in a second genetic background (W303; not shown). In combination, these observations establish a strong correlation between chronological lifespan and efficiency of G1 arrest under nutrient-limiting conditions. ## Caloric restriction and osmotic stress extend chronological lifespan in parallel with a more efficient G1 arrest during nutrient depletion Caloric restriction extends the replicative and chronological lifespans of budding yeast, as well as the chronological lifespans of most other eukaryotic organisms. To determine whether caloric restriction also induces a more efficient G1 arrest during nutrient depletion, we assessed the effects of long- term culture of budding yeast cells in medium containing a reduced concentration of glucose (0.5% instead of 2%). Reducing the concentration of glucose to 0.5% extended the chronological lifespan of cells in the DBY746 background at the same time that it reduced the budding index of these cells. Similar results were obtained in the BY4741 genetic background. FACS analysis of DNA content indicated that in both genetic backgrounds, cells cultured in 0.5% glucose more slowly underwent apoptotic DNA degradation indicated by fewer cells with less than a G1 content of DNA. Therefore, similar to mutational inactivation of Sch9-or Ras2-dependent nutrient signaling pathways, caloric restriction extends chronological lifespan of budding yeast in concert with a more efficient G1 arrest and less frequent apoptosis. Unlike cells in which lifespan was extended by deletion of *SCH9* or *RAS2*, however, calorie-restricted DBY746 cells exhibited a progressive age-dependent increase in DNA content. A progressive increase in DNA content was observed in calorie-restricted BY4741 cells as well, although this increase occurred more slowly. An increase in nuclear DNA content is not consistent with the reduced budding index of calorie-restricted cells, which indicates less frequent cell division. In mammals (including humans), caloric restriction stimulates mitochondrial biogenesis. To determine whether increased mitochondrial DNA associated with mitochondrial biogenesis might be responsible for the increased DNA content of calorie-restricted budding yeast cells, we stained cells with the DNA-specific dye DAPI. A substantial increase in the number of DAPI signals from non-nuclear, cytoplasmic DNA was apparent in nutrient-depleted DBY746 and BY4741 cells cultured for three days in medium containing 0.5% glucose compared to 2% glucose. Southern blot analysis of DNA isolated from DBY746 cells at this time point confirmed that cells cultured in medium containing 0.5% glucose contained more mitochondrial DNA relative to nuclear DNA compared to cells cultured in medium containing 2% glucose. We conclude that the increased DNA content detected by FACS under caloric restriction conditions corresponds to increased mitochondrial DNA. Microscopic inspection of DAPI-stained cells suggested that caloric restriction also increased the size of cells in both genetic backgrounds. This was confirmed by measurements of cell volume. The G1 cyclin *CLN3* is downregulated during nutrient depletion, and this downregulation occurs more rapidly in calorie-restricted cells. Downregulation of *CLN3* in cycling cells prolongs G1 and increases cell size. Therefore, the larger size of calorie-restricted cells is consistent with a tighter G1 arrest associated with more rapid downregulation of *CLN3*. Similar to caloric restriction, increasing the osmolarity of culture medium by adding sorbitol (a nonmetabolizable sugar alcohol) extends both replicative and chronological lifespans of budding yeast. The effects of osmotic stress on replicative lifespan may be mechanistically related to those of caloric restriction. In the absence of nutrient depletion, osmotic stress inhibits growth by inducing a G1 arrest. This suggested the possibility that the effects of sorbitol on chronological lifespan of nutrient-depleted cells might be related to a tighter G1 arrest. Consequently, we examined the effects of sorbitol on nutrient depletion-induced G1 arrest in parallel with effects on viability. As reported previously, adding sorbitol to medium extended the chronological lifespan of BY4741 cells to a similar extent compared to caloric restriction. Also similar to caloric restriction, lifespan extension by sorbitol was accompanied by a reduction in budding index and in the number of cells undergoing apoptotic DNA degradation. These findings extend the correlation between chronological lifespan extension, less frequent apoptosis and more efficient G1 arrest. ## Induction of DNA replication stress by nutrient depletion One of the potential benefits of the G1 arrest induced by nutrient depletion is that it would protect cells from replication stress they would suffer if they were in S phase under suboptimal conditions for replicating DNA. These conditions might include the depletion from medium of substrates required for synthesis of dNTPs. Nutrient depletion also downregulates the transcription of many genes encoding proteins required for DNA replication. These genes include *RNR1* encoding the large subunit of ribonucleotide reductase, which is required for the synthesis of dNTPs. The combined effects of reduced concentrations of substrates for DNA synthesis and downregulation of genes encoding proteins required for DNA replication is expected to produce replication stress in nutrient-depleted cells that failed to enter into or maintain a G1 arrest, and arrest growth in S phase instead. The *MEC1* gene encodes a protein required for cellular responses to both replication stress and DNA damage. *RAD9* encodes a protein required for responses to DNA damage, but not replication stress (reviewed). To determine whether nutrient depletion induces replication stress, we compared the phenotypes of nutrient-depleted wild type, *mec1-21* and *rad9Δ* cells. During nutrient depletion in YPD medium, *mec1-21* cells died more rapidly than wild- type or *rad9Δ* cells. More rapid cell death of *mec1-21* cells occurred in parallel with a less efficient arrest with a G1 content of DNA and an elevated budding index. At later time points, *mec1-21* cells also underwent apoptotic DNA degradation, which was largely absent from wild-type or *rad9Δ* cells at the same time points. Nutrient-depleted *rad9Δ* cells exhibited budding indices similar to those of wild type cells. This is consistent with the absence of a role for Rad9 in cellular responses to replication stress, in contrast to the roles of Mec1 and Rad53. However, *rad9Δ* cells died more rapidly than wild type cells, although not as rapidly as *mec1-21* cells. This suggests that nutrient depletion induces DNA damage, in addition to replication stress. *RAD53* encodes a protein that functions downstream of Mec1 in cellular responses to DNA damage and replication stress.. Similar defects in G1 arrest were observed in *rad53-21* and *mec1-21* compared to wild-type and *rad9Δ* cells during nutrient depletion in SC medium. Furthermore, apoptotic DNA degradation was accelerated in *rad53-21* and *mec1-21* cells. Apoptotic DNA degradation in wild type, *mec1-21* and *rad9Δ* strains was also accelerated in cells cultured in SC compared to YPD medium. Together with the data presented in, these findings indicate that nutrient depletion induces replication stress, which is enhanced in SC compared to YPD medium. An essential function shared by Mec1 and Rad53 is to maintain sufficient dNTP pools to complete DNA replication during S phase. Maintenance of dNTP pools requires the induction of Rnr1 activity by Mec1-and Rad53-dependent pathways that inactivate Sml1, an inhibitor of Rnr1. Although inactivation of Sml1 is required to maintain viability in *mec1Δ* and *rad53Δ* strains, Sml1 is intact in the partial loss-of-function *mec1-21* and *rad53-21* strains. To determine whether failure to maintain dNTP pools might explain the effects of the *mec1-21* and *rad53-21* mutations, we analyzed the effects of ectopic expression of *RNR1* from a high copy plasmid. In some, but not all, experiments, ectopic expression of *RNR1* in wild-type or *rad9Δ* cells enhanced arrest with a G1 content of DNA induced by nutrient depletion (“wild type” and “*rad9Δ*”;). This enhanced G1 arrest was accompanied by a reduction in the number of cells undergoing apoptotic DNA degradation at later time points ( “wild type”). This is consistent with the possibility that replication stress caused by insufficient dNTPs during nutrient depletion can trigger apoptosis. In most experiments, however, effects of ectopic expression of *RNR1* were not detected in wild-type or *rad9Δ* cells. We conclude that cells harboring wild- type Mec1 and Rad53 usually maintain sufficient levels of dNTPs at early stages of nutrient depletion to complete the replication of chromosomes before arresting in G1. The occasional effects of ectopic *RNR1* expression in wild type and *rad9Δ* cells may reflect a critical balance between the competing requirement during nutrient depletion to maintain sufficient dNTPs to complete S phase at early stages and to downregulate *RNR1* at later stages. In contrast to the occasional effects of ectopic expression of *RNR1* in wild type and *rad9Δ* cells, ectopic expression of *RNR1* in *mec1-21* cells consistently suppressed the defective G1 arrest in these cells as measured by DNA content (“*mec1-21*”), as well as bud indices. It also suppressed increased apoptotic DNA degradation in these cells during nutrient depletion ( “*mec1-21*”). In fact, ectopic expression of *RNR1* reproducibly led to a more efficient G1 arrest at earlier time points in *mec1-21* compared to wild-type cells indicated by more cells with a G1 DNA content (compare “wild type pRNR1” with “*mec1-21* pRNR1”) and fewer buds (compare “pRNR1” with “pRNR1”). This may be due to the combined effects of elevated dNTP pools resulting from ectopic expression of *RNR1* and the premature activation of late S phase-firing replication origins that occurs in cells harboring mutations in *MEC1*,. In the presence of excess dNTPs, activation of late replication origins in early-S- phase cells would accelerate exit from S phase during nutrient depletion. This may allow for exit from S phase before other factors required for DNA replication have been downregulated in response to nutrient depletion. In contrast to *mec1-21* cells, cells from which both *MEC1* and *SML1* had been deleted did not exhibit a defective G1 arrest or increased apoptosis. This is consistent with a role for Mec1 in promoting G1 arrest during nutrient depletion related to its downregulation of Sml1 and increased ribonucleotide reductase activity. Downregulation of Sml1 would lead to the induction of ribonucleotide reductase activity and increased dNTP pools required for exit from S phase before other factors required for DNA replication have been downregulated. In contrast to its effects in *mec1-21* cells, ectopic expression of *RNR1* had no detectable effect on the defective G1 arrest and increased apoptotic DNA degradation phenotypes of *rad53-21* cells (“*rad53-21*” and). This may reflect the existence of lethal defects in Rad53-dependent, but Mec1-independent regulation of histone levels in cells subjected to DNA damage or replication stress. An additional function of Rad53 that is not shared by Mec1 has been suggested by other studies as well. This includes the demonstration that *rad53Δ*, but not *mec1Δ* cells exhibit a slow-growth phenotype when Sml1 is inactivated in these cells, which suggests a Mec1-independent function of Rad53 that is not related to the regulation of nucleotide levels. We conclude that nutrient depletion-induced replication stress is mitigated in wild-type cells by Mec1-dependent induction of Rnr1 activity. Regulation of Rnr1 activity by Mec1 during nutrient depletion likely occurs in collaboration with Rad53, but in conjunction with other Mec1-and Sml1-independent effects of Rad53 that also impact nutrient depletion-induced G1 arrest. ## Ectopic expression of Cln3 abrogates G1 arrest, shortens lifespan and accelerates genome instability and apoptosis during nutrient depletion The findings described above are consistent with the hypothesis that inhibition of nutrient signaling pathways extends chronological lifespan by promoting a more efficient arrest in G1 that protects against DNA replication stress. Progression past Start in G1 and entry into S phase require the activity of the Cdc28 cyclin-dependent kinase, which is regulated by the G1 cyclin Cln3 upstream of the cyclins Cln1 and Cln2 (reviewed). The expression of all three Clns is down-regulated in nutrient-depleted cells. We next asked whether ectopic expression of *CLN3* during nutrient depletion would attenuate the G1 arrest induced by nutrient depletion and shorten chronological lifespan in the absence of increased nutrient signaling. In initial experiments, ectopic expression of *CLN3* paradoxically increased the fraction of viable cells at later time points (compared to empty vector- transformed control cells) in concert with increased apoptotic DNA degradation. At later time points in chronological aging experiments, strains with shorter chronological lifespans often exhibit a phenomenon called adaptive regrowth, which occurs stochastically in association with the release of substances from apoptosing cells into the medium. These substances promote the growth of cells that at earlier time points were growth-arrested due to nutrient depletion. Adaptive regrowth could explain the relative increase in viable cells detected at later time points in populations of cells ectopically expressing *CLN3*, despite an increased frequency of apoptosis in these cell populations. Consequently, in subsequent experiments, we measured the effects of ectopic *CLN3* expression on chronological lifespan under conditions that avoided adaptive regrowth by periodically reseeding cells in medium that had been pre- depleted of nutrients. This blocked the accumulation of substances released from apoptosing cells without altering the nutrient-depleted status of surviving cells. Pre-depleted medium was prepared by long-term (seven days) culture of *sch9Δ* cells, since these cells are less susceptible (compared to wild type cells) to apoptosis or adaptive regrowth. *CLN3*-overexpressing and vector-control cells were reseeded into this pre-depleted medium (from which *sch9Δ* cells had been removed) every other day beginning one day after cultures of these cells were first established. Analysis of wild type and *sch9Δ* cells cultured for an extended time period using this regimen indicated that periodic reseeding in pre-depleted medium did not alter the chronological lifespan of wild type cells or the tighter G1 arrest and reduced apoptotic DNA degradation that occurs in *sch9Δ* cells. Therefore, the use of pre-depleted medium does not alter the mechanisms by which nutrient signaling pathways impact chronological lifespan. In pre-depleted medium, cells ectopically expressing *CLN3* reproducibly exhibited a less efficient G1 arrest compared to cells transformed with an empty vector, and frequently arrested growth in S phase instead. The less efficient growth arrest in G1 of cells ectopically expressing *CLN3* was accompanied by accelerated apoptotic DNA degradation and a shorter chronological lifespan (note logarithmic scale). Shorter chronological lifespan associated with ectopic expression of *CLN3* during nutrient depletion was recently reported by others as well. The shorter lifespan of *CLN3*-expressing cells during nutrient depletion is similar to reports of shortened lifespan associated with ectopic expression of *CLN3* in cells undergoing growth arrest induced by inhibiting TOR signaling pathways with rapamycin. Genome instability is an important component of chronological aging in budding yeast and other organisms. To determine whether the less efficient growth arrest in G1 and shortened lifespan of nutrient-depleted cells ectopically expressing *CLN3* are accompanied by accelerated age-dependent genome instability, we asked whether these cells suffer an increase in the frequency of mutations in the *CAN1* gene, as measured by increased resistance to the toxic amino-acid analogue canavanine. We found that, similar to a previous report, nutrient depletion induced a chronological-age-dependent increase in mutation frequency in wild-type cells (“vector”). This mutation frequency was dramatically elevated in chronologically aged cells ectopically expressing *CLN3* (“pCLN3”). Therefore, in addition to shortening chronological lifespan and stimulating apoptosis, the failure to efficiently arrest growth in G1 during nutrient depletion contributes to chronological age-dependent genome instability. # Discussion ## Impact of altered nutrient signaling on chronological lifespan Most efforts to understand how alterations in nutrient signaling pathways impact the chronological lifespan of budding yeast have focused on changes in oxidative stress responses regulated by these pathways downstream of Rim15. Our findings point to an efficient Rim15-dependent growth arrest in G1 that also requires downregulation of Cln3 as an additional factor determining chronological lifespan in this organism. Caloric restriction, mutational inactivation of Sch9 or Ras2 and growth in YPD rather than SC medium enhance this G1 arrest and extend lifespan. In contrast, constitutive activation of nutrient signaling by *RAS2*^val19 or deletion of *RIM15* increases the frequency with which nutrient-depleted cells growth-arrest in S phase instead of G1 and shortens chronological lifespan. During nutrient depletion, sustained expression of *CLN3*–which is normally downregulated under these conditions–also increases the frequency with which cells growth arrest in S phase in concert with a shorter lifespan and age-dependent increases in genome instability and apoptosis. Therefore, downregulation of Cln3 is required for G1 arrest and long-term survival in nutrient-depleted cells. Importantly, the effects of ectopic *CLN3* expression occur in the absence of altered nutrient signaling that would impact oxidative stress. Therefore, changes in the efficiency of G1 arrest associated with altered nutrient signaling can impact chronological lifespan and age- dependent genome stability independently of changes in stress responses that would affect oxidative damage to DNA and other cellular constituents. How downregulation of Cln3 occurs during nutrient depletion remains unclear. Xbp1, a transcriptional repressor of *CLN* genes, may contribute to this downregulation. *XBP1* is induced by nutrient depletion, and ectopic expression of Xbp1 in cycling cells slows growth by lengthening G1 phase. Induction of *XBP1* by nutrient depletion may occur downstream of Rim15 activation and its induction of Msn2 and Msn4. The promoter for *XBP1* harbors binding sites for these and other stress factors induced by Rim15 activation during nutrient depletion. Msn2 and Msn4 antagonize PKA-dependent growth. This establishes a role for these stress response factors in growth regulation–perhaps upstream of the induction of *XBP1*-in addition to the role they play in responses to oxidative and other stresses. Nutrient depletion-induced G1 arrest is enhanced by osmotic stress, similar to the effects on G1 arrest of mutational inactivation of nutrient signaling pathways and caloric restriction. In cycling cells, osmotic stress inhibits growth by inducing a G1 arrest accompanied by the downregulation of Clns and Cdc28 activity. This G1 arrest depends on stabilization of the Cdc28 inhibitor Sic1 by the Hog1 MAP kinase. Hog1 may also downregulate the expression of Clns in osmotically stressed cells, which would contribute to Sic1 stability. The *CLN* transcriptional repressor *XBP1* is also induced by osmotic stress, and therefore may contribute to the downregulation of Clns and Cdc28 activity in osmotically stressed cells. Similar to ectopic expression of *CLN3* during medium depletion, abrogation of the Hog1-dependent G1 arrest in osmotically stressed cells is accompanied by genome instability. We think it is likely that the chronological lifespan-extending effects of osmotic stress during nutrient depletion are related to enhanced downregulation of Clns and Cdc28 activity by Hog1-and Xbp1-dependent mechanisms that leads to a tighter nutrient depletion- induced G1 arrest. ## Replication stress and chronological lifespan What is the mechanism by which a more efficient G1 arrest during nutrient depletion enhances chronological lifespan? Our results also indicate that nutrient depletion causes DNA replication stress. The induction of replication stress during nutrient depletion is indicated by the finding that cellular responses mediated by Mec1 and Rad53 (which respond to replication stress and DNA damage) are more important to the survival of nutrient-depleted cells than responses mediated by Rad9 (which responds to DNA damage, but not replication stress). At least at initial stages of nutrient depletion, Mec1 mitigates replication stress, most likely by increasing levels of dNTPs. This is indicated by the ability of ectopically expressed *RNR1* to suppress the shorter lifespan and increased apoptosis in nutrient-depleted *mec1-21* cells. Ectopic expression of *RNR1* would have the effect of increasing dNTP pools that in wild type (but not *mec1-21*) cells are upregulated by Mec1 and Rad53 in response to replication stress. Mec1 and Rad53 upregulate ribonucleotide reductase by a mechanism that leads to destabilization of the ribonucleotide reductase inhibitor Sml1. Therefore, replication stress at early stages of nutrient depletion is mostly or entirely a consequence of starvation for dNTPs. Consistent with this possibility, *mec1Δ* cells from which *SML1* has also been deleted (which leads to increased levels of dNTPs) are phenotypically similar to wild type cells. The absence of effects in *mec1Δ sml1Δ* compared to wild type cells also suggests that the checkpoint function of Mec1, which is separate from its role in regulating dNTP metabolism, does not play a role in regulating chronological lifespan. Deletion of *RAD9* also shortens lifespan and accelerates apoptosis, although not to the same extent as the *mec1-21* and *rad53-21* mutations. Therefore, nutrient depletion likely induces DNA damage in addition to replication stress. Nutrient depletion-induced DNA damage is consistent with the age-dependent accumulation of mutations during chronological aging. Importantly, the induction of DNA damage or replication stress during nutrient depletion is not inconsistent with our failure to detect a role for Mec1-dependent checkpoint responses in the survival of nutrient-depleted cells. Mec1-dependent pathways are inactivated by superoxide anions, which accumulate during nutrient depletion. Since Mec1-dependent pathways also contribute to the stability of stalled replication forks, inactivation of these pathways by superoxide anions at later stages of nutrient depletion might contribute to DNA damage in nutrient-depleted cells undergoing replication stress. Although replication stress-induced recombination in the rDNA locus is a major determinant of budding yeast replicative lifespan ( and references therein), replication stress has not been considered a factor in chronological aging in this organism. Most likely this is because during nutrient depletion, most cells appear to arrest without buds and with a G1 content of DNA, which suggests they are not in S phase. It is difficult to detect minor increases in DNA content by flow cytometry, however, and small buds are difficult to detect microscopically. Furthermore, the activation of some, but not all, nutrient signaling pathways may lead to uncoupling of events required for budding from progression past Start. Consequently, although DNA content and bud counts provide useful relative measures of the number of cells that arrest in G1 during nutrient depletion, they likely overestimate the number of G1 cells and underestimate the number of cells that arrest growth while in S phase. This view is consistent with a recent report that nutrient depletion leads to the accumulation of a substantial fraction of less dense cells, which includes all the cells that remain budded during nutrient depletion plus many cells that do not appear to have buds. These less-dense cells differentially express a number of genes encoding proteins required for the resolution of stalled DNA replication forks, suggesting they are under replication stress. These cells also more frequently undergo apoptosis compared to denser cells that do not express these genes. These phenotypes are consistent with our proposed role for nutrient depletion-induced replication stress in chronological aging. ## How does nutrient depletion cause replication stress in S phase cells? Several factors likely contribute to replication stress in cells that arrest growth in S phase during nutrient depletion. In addition to the potential depletion from medium of substrates required for synthesis of dNTPs, these factors include downregulation of genes encoding proteins required for DNA replication. Interestingly, ectopic expression of the constitutively activating *RAS2*^val19 mutation induces transcription of *CLN3*, but not transcription of *RNR1* and other DNA replication-related genes (including supplementary information). Consequently, during nutrient depletion, cells harboring the *RAS2*^val19 mutation may enter S phase with a reduced capacity to replicate DNA. Similar uncoordinated entry into S phase when only a subset of growth-regulatory pathways are active could explain an earlier report of growth-promoting, but lethal apoptosis-inducing effects of glucose added to stationary phase cultures in the absence of other nutrients. Aberrant initiation of DNA replication is another factor that likely contributes to replication stress in nutrient-depleted cells. Deregulated Cln expression inhibits the assembly of pre-replicative complexes (pre-RCs) required for initiation of DNA replication at replication origins. Inhibition of pre-RC assembly induces genome instability, as well as DNA damage and apoptosis. ## Caloric restriction and chronological lifespan In cycling budding yeast cells, the effects of caloric restriction on replicative lifespan are related to reduced signaling through PKA, TOR and Sch9-dependent pathways, because caloric restriction does not extend replicative lifespan further when these pathways have been inactivated,. Our findings are consistent with the possibility that caloric restriction during nutrient depletion extends chronological lifespan via a similar reduction in signaling through these pathways, which leads to a tighter nutrient depletion-induced G1 arrest and reduced replication stress. This tighter G1 arrest may be related to the accelerated downregulation of *CLN3* mRNA when glucose concentration is reduced. Both caloric restriction and deletion of *CLN3* suppress the shorter chronological lifespan of nutrient-depleted polyploid compared to haploid cells in parallel with a more efficient G1 arrest. This is also consistent with the possibility that caloric restriction can extend chronological lifespan by downregulating Cln3. Although deletion of *CLN3* extended the chronological lifespan of polyploid cells, it did not extend the chronological lifespan of haploid cells in this prior study or in our experiments (not shown). This might reflect the existence of redundant, Cln3-independent pathways that contribute to G1 arrest during nutrient depletion, some of which may be inoperative in polyploid cells. Whatever the explanation, our experiments clearly suggest that the increased chronological lifespan of *cln3Δ* or calorie-restricted polyploid cells reported in this earlier study was related to attenuation of replication stress. ## Replication stress and hormesis effects on aging Many of the genes induced by nutrient depletion are also induced by other stresses, including oxidative and osmotic stress, heat and DNA damage. Specific stresses often confer cross-resistance to other stresses, for reasons that are not clear. Cross-resistance to multiple stresses and the lifespan-extending effects associated with mild exposures to these stresses is the basis for the “hormesis hypothesis” of aging. This hypothesis posits a general stress response induced by caloric restriction and other stresses that protects against a variety of different stresses as an important component of longevity in all eukaryotes. The nature of this putative general stress response is not clear. In addition to nutrient depletion and osmotic stress, other stresses including oxidative stress, heat, and DNA damage —all of which also have been implicated in hormesis effects that extend lifespan—inhibit growth by inducing a G1 arrest in budding yeast. The *CLN* transcriptional repressor Xbp1 is induced by all these stresses, in addition to its induction by nutrient depletion and osmotic stress. The growth-inhibitory effects of stresses and the hormesis-like lifespan-extending effects of osmotic stress in concert with a tighter G1 arrest reported here clearly point to the existence of a general response to environmental stresses that enhances longevity by inhibiting growth, and thus replication stress and age-dependent genome instability. Inhibition of replication stress under these conditions likely contributes to cross-resistance to various stresses that underlies the hormesis hypothesis. Similar to genes induced by nutrient depletion, genes induced by other stresses are nonrandomly distributed in the budding yeast genome and are more likely to be repressed by chromatin structure in the absence of stress. This is consistent with the possibility that hormesis effects of stresses on lifespan in budding yeast are coordinately regulated by chromatin structure within clusters of stress-related genes. ## Replication stress and aging in higher eukaryotes In addition to their effects on replicative and chronological lifespan in budding yeast, caloric restriction, low levels of stress and mutations that inactivate growth signaling pathways extend the lifespans of many higher eukaryotes as well. We propose that, similar to budding yeast, these factors promote longevity in all eukaryotes by inhibiting replication stress associated with uncoordinated entry into or exit from S phase. This may be particularly important during differentiation leading to a quiescent, non-dividing state, which requires downregulation of the Cln3 homologue cyclin D1. Consistent with this model, in mice, caloric restriction extends lifespan in concert with a reduction in the number of cells in S phase in a variety of tissues containing differentiating cells, including intestinal epithelium, spleen, thymus and mesenteric lymph nodes. Other studies have also detected a relationship between caloric restriction and reduced cellular proliferation in higher eukaryotes (reviewed). In fact, accumulating evidence points to replication stress downstream of aberrant growth signaling as a determinant of lifespan in higher eukaryotes, including humans. This includes the recent discovery that replication stress- induced DNA damage and apoptosis are present in preneoplastic human cells that eventually give rise to various cancers, for which age is a dominant risk factor. Replication stress-induced DNA damage and apoptosis at early stages of neoplasia likely arise downstream of the mutational activation of genes that promote growth, similar to the constitutive activation of *RAS2* by the *RAS2*^val19 mutation in budding yeast. For example, it was recently shown that sustained mitogenic signaling induced by ectopic expression of activated *RAS* in quiescent rat fibroblasts stimulates G1 cyclin-dependent kinase activity, entry into S phase, genome instability and apoptosis, all of which correspond to phenotypes detected in the yeast experiments reported here. Similar to the *RAS2*^val19 mutation, which induces *CLN3* but not genes encoding proteins required for efficient DNA replication, the oncogenic activation of Ras and other proteins in mammalian cells may induce a subset of the complex, interacting growth-regulatory pathways normally required for progression into and through S phase. Induction of a subset of these pathways in the absence of the parallel induction of proteins required for efficient DNA replication would create replication stress. Depending on the type of cell and the status of cell cycle checkpoint pathways, DNA damage produced by replication stress could induce a number of aging phenotypes. In addition to neoplastic transformation, these include senescence of stem cells or their proliferating progeny and cell death. Age is also a dominant risk factor for many neurodegenerative diseases. Apoptosis associated with entry into S phase and inefficient DNA replication in normally quiescent postmitotic neurons has been implicated in the etiology of a number of these diseases as well ( and references therein). In fact, a causal role for the activation of TOR signaling pathways in neurodegeneration was recently demonstrated in a *Drosophila* model of neurodegenerative disease. Although this role can be explained by other models, our findings suggest replication stress as a potential explanation. In summary, the findings reported here point to DNA replication stress downstream of deregulated nutrient signaling as an important determinant of chronological aging in budding yeast. They suggest that caloric restriction and other experimental manipulations that inhibit growth—including mild stresses—can extend chronological lifespan by decreasing the frequency with which cells arrest growth in S phase during nutrient depletion. This reduces replication stress, as well as age-dependent DNA damage and genome instability caused by replication stress. Accumulating evidence suggests that replication stress induced by aberrant growth regulation is an important factor in age-related pathologies in higher eukaryotes, including humans. Although the destabilizing effect of replication stress in the rDNA locus associated with replicative aging in budding yeast has not been detected in other eukaryotes, a universal role for replication stress in aging is suggested by the fact that that in many eukaryotic organisms, mutations in RecQ helicases required for accurate and efficient DNA replication promote genome instability and premature aging (reviewed in. Our findings suggest that some of the genome instability associated with aging is related to DNA damage caused by replication stress instead of oxidative stress. Deregulated growth signaling and inappropriate entry into, or exit from, the cell cycle in nutrient-starved budding yeast cells provide a new paradigm for investigating how oncogenic activation of growth- regulatory pathways and the induction of replication stress in mammalian cells contribute to cancer and other age-related diseases. # Materials and Methods ## Yeast strains and plasmids Strain backgrounds employed in this study were DBY746 (MATa leu2-3, 112, his3Δ1 trp1-289, ura3-52, GAL⁺); W303-1A (MATa, ade2-1, ura3-1, his3-11, trp1-1, leu2-3, can1-100), JC482 (MATα, leu2, ura3, his4) and BY4741 (MATa, his3Δ1, leu2Δ0, met15Δ0, ura3Δ0). Mutant strains employed were PF102 (DBY746 sch9::URA3); MWY420 (DBY746 rim15::TRP1); rim15Δ sch9Δ (DBY746 sch9::URA3 rim15::LEU2); RAS2^val19 (JC482 RAS2^val19); Y604 (W303 mec1-21); Y301 (W303 rad53-21); HKY845 (W303 rad9::HIS3). All strains in the DBY746 background were from V. Longo (USC) except MWY420, which was constructed for this study. Strains in the JC482 background were from M. Breitenbach (University of Salzburg). Y604 and Y301 were from S. Elledge (Harvard). HKY845 was from H. Klein (NYU). The plasmid Yep-RNR1 was from E. Vallen, Swarthmore College. Yep 24 (vector control for YEp24-RNR1) was purchased from ATCC. pCCul expressing CLN3 from the CUP1 promoter and its control vector pJ16 were from W. Heideman (U. of Wisconsin). pSR17 containing rim15::TRP employed for construction of MWY420 and PMF100 containing ras2^val19 (employed in the construction of MWY421) were from V. Longo (USC). To construct MWY420, rim15Δ::TRP1 was excised as a Xho1-Sal1 fragment from pSR117 and inserted in DBY746 by transformation and selection for TRP⁺ growth. To construct MWY421, RAS2^val19 :: URA3 was excised as an EcoR1-HindIII fragment from pMF100 and inserted into W303 by transformation and selection for URA⁺ growth. ## Cell culture conditions, flow cytometry measurements of DNA content, bud measurements To assess chronological lifespan, cells from exponentially proliferating cultures were inoculated into 50 mls. SC or YPD medium containing 2% glucose in 250 ml. flasks at an initial density of 5×10⁵/ml. and continuously cultured at 30°C with rotary shaking for indicated times. To assess the effects of osmotic stress, sorbitol was added to these cultures to a final concentration of 1M. For experiments requiring ectopic expression of *CLN3*, SC medium was pre-depleted by growth of *sch9Δ* cells for 7 days followed by pelleting of cells and filter sterilization of medium. Exponential cultures of DBY746 cells transformed with pCLN3 or a vector control plasmid were seeded into standard SC medium. Beginning 24 hours later, cells were pelleted from medium and resuspended in pre-depleted medium every other day throughout the course of experiments. To assess the effects of caloric restriction on chronological lifespan, an equal number of cells from exponentially proliferating cultures were pelleted by centrifugation, washed twice with fresh medium containing 0.5% glucose and then seeded into this latter medium or into fresh medium containing 2% glucose. Caloric restriction experiments employed a slightly different formulation of SC medium compared to the formulation employed in other experiments. However, repetition of these experiments in the SC formulation employed in other experiments confirmed that the effects of caloric restriction were the same in both formulations (not shown). In all chronological lifespan measurements, aliquots removed from cultures at the indicated times were plated in triplicate on YPD agar to determine survival (as colony forming units). To determine budding indices and DNA content, cells in aliquots taken at each time point were pelleted by centrifugation and resuspended in water (for determining budding index) or 70% ethanol (for flow cytometry). The budding status of at least 500 cells from each aliquot was visually determined using a Nikon Eclipse E600 microscope with a 40× phase contrast objective. Just before examining cells, cell clumps were dissociated by sonication using a Model 60 Sonic Dismembrator sonicator (Fisher Scientific, Hampton, NH) for 10 seconds at power setting 5. To measure DNA content by flow cytometry, cells suspended in 70% ethanol were pelleted by centrifugation, washed with 50 mM sodium citrate (pH 7.5) and resuspended in 0.5 ml of this same buffer containing 0.5 mg/ml RNAse. After overnight incubation at 37°C, an additional 0.5 ml. of sodium citrate buffer containing 2 µM SYTOX Green (Invitrogen, Carlsbad, CA) was added to each sample. Stained cells were briefly sonicated as described above and DNA content was measured using a FacsCaliber flow cytometer (BD Biosciences, Woburn, MA) at a maximum flow rate of 500 cells/s. Flow cytometry data were processed using CellQuest (BD Biosciences) and Flojo (Tree Star Inc., Ashland, OR) software. Y axis scales indicating number of cells were maintained constant in all flow cytometry profiles for individual experiments. ## Measurements of mutation frequency Mutation frequency was assessed by determining frequency of resistance to the toxic amino-acid analogue canavanine conferred by mutations in the *CAN1* gene. An appropriate number of cells were plated on selective medium containing 60 mg/l L-canavanine instead of arginine and canavanine-resistant colonies were counted 3–5 days later. ## Cell Volume Measurements Analysis of cell volume was performed using the Z2 Coulter Particle Count and Size Analyzer. DBY746 and BY4741 strains were grown for 2 days at 30°C in 5 ml of either SC+2% glucose or SC+0.5% glucose to saturation (performed in triplicate for each condition). Cells from these cultures were inoculated into fresh media to an OD₆₀₀ of 0.1 in 10 ml of the same media. 1.0 OD of cells was collected at 6 hours, 24 hours, 72 hours, and 168 hours. The cells were spun down and resuspended in 100 µl sterile ddH₂O and immediately analyzed. The 100 µl volume was added to 10 ml of azide-free H₂O (Fisher Scientific) and analyzed on the Coulter counter (dilution factor = 100; range: 10 to 250 fL (femto-liters)). Prior to measurements, the cells were sonicated for ∼10s to break up any aggregates (duty cycle = 30%; continuous pulse). The mean cell volume in fL was manually calculated from the peak distribution. ## Measurements of mitochondrial DNA Cells were isolated at indicated times and after fixation in 70% ethanol, DNA was stained with DAPI as described previously to detect nuclear and cytoplasmic DNA. Photomicrographs of stained cells were obtained at 400× magnification with a Zeiss Axioskop microscope equipped with an Optronics Magnafire CCD camera. To determine relative amounts of mitochondrial compared to nuclear DNA, total yeast DNA was digested with the restriction enzyme XmnI, which produces a 6.3 kb fragment of nuclear DNA containing the *ACT1* gene and a 10.2 kb fragment of mitochondrial DNA containing sequences encoding 15S ribosomal RNA. XmnI-digested DNA was separated on 1% agarose gels and transferred to Duralon hybridization membranes. Membranes were probed with ³²Phosphate-labeled DNA produced by PCR amplification of sequences within each of these DNA fragments. Radioactive signals from membranes were captured using a Storm PhosphoImager (GE Healthcare). # Supporting Information We wish to thank Valter Longo, Paola Fabrizio, Stephen Elledge, Hannah Klein and Michael Breitenbach for yeast strains, Warren Heideman, Elizabeth Vallen, Paola Fabrizio, Valter Longo and Stephen Elledge for plasmids, Michael Breitenbach for critical reading of the manuscript, Benjamin Burhans (RPCI) for technical assistance, and Mitch Smith (UVA) for the use of a Coulter Counter. [^1]: Conceived and designed the experiments: JS JH WB MW KS. Performed the experiments: MW LF AP MV DS RH. Analyzed the data: JS JH WB MW KS. Contributed reagents/materials/analysis tools: JH. Wrote the paper: WB MW. [^2]: The authors have declared that no competing interests exist.

# Introduction Solid epithelial tumors evoke a reactive stromal response that is critical for growth and progression of the tumor. A reactive stroma is complex and consists of activated fibroblasts, newly formed vasculature, infiltrating immune cells, and extracellular matrix (ECM). Soluble signaling molecules such as cytokines and growth factors are well-documented mediators of interactions between cells in the tumor microenvironment. The tumor ECM also mediates communication between various cell types, in part by providing migration and adhesion signals. Although increased deposition and cross-linking of collagen in the tumor stroma is associated with increased tumor growth, little is known about how specific ECM adhesion signals are created and regulated. It has been suggested that proteolytic cleavage of collagen increases cancer growth, invasion, and angiogenesis. Thus is likely that signals induced by changes in the ECM act in conjunction with soluble cytokines produced by other stromal components to modulate tumor growth. Tumor associated fibroblasts (TAFs) are a key element of the reactive stroma. TAFs are fibroblasts that have undergone a major phenotypic change to an activated state characterized by increased proliferation, secretion of type I collagen, and expression of ECM-degrading proteases. TAF-derived proteases present in the tumor microenvironment play a pivotal role in remodeling the ECM to make it permissive for tumor cell invasion and infiltration by normal endothelial cells and immune cells such as macrophages. Fibroblast Activation Protein-α (FAP) is a membrane-bound, serine protease that is expressed by activated fibroblasts including TAFs, but is absent from normal healthy adult tissues. Although FAP reportedly cleaves type I collagen in vitro, the pathophysiological significance of FAP-generated collagen cleavage products is unclear. One possibility is that FAP participates in modifying ECM molecules to create and regulate cell adhesion in the tumor microenvironment. Macrophages are another major cellular component of the tumor stroma and their infiltration and accumulation in the tumor microenvironment is correlated with tumor progression, invasion, and poor patient prognosis. Tumor associated macrophages (TAMs) typically exhibit an M2 phenotype and are associated with secretion of a wide variety of growth factors, cytokines, and chemokines that suppress an anti-tumor immune response and promote tumor growth. Macrophages infiltrate and are retained in the tumor microenvironment through expression of specific adhesion proteins, such as integrins and scavenger receptors, that bind and mediate the adhesion of cells to ECM components of the tumor stroma. Interestingly, macrophages do not adhere well to native type I collagen, which is the most abundant ECM protein in a tumor stroma. This suggests that modifications of collagen are necessary for macrophage adhesion. Based on this background, we speculated that FAP, which is expressed by TAFs, cleaves collagen and converts it into an adhesive substrate for macrophages. # Materials and Methods ## Animals Animals used in this study include C57Bl/6 mice and SRA^-/- mice in a C57Bl/6 background purchased from Jackson Laboratories (Bar Harbor, ME, USA); and MMTV-PyMT obtained from Dr. S. Gendler (Mayo Clinic, Scottsdale, AZ). All animals were maintained as colonies at the University of Arkansas for Medical Sciences and housed on a 12-hour light-dark cycle. To generate a breast tumor model, male MMTV-PyMT mice and C57Bl6 female mice were bred, and the offspring genotyped for the presence of MMTV-PyMT transgene. Breast tumors from female offspring that were heterozygous for MMTV-PyMT were isolated and used for tumor immunohistochemistry. All animals were provided food (Teklan Global 16% protein rodent diet; Harlan Laboratories, Indianapolis, IN, USA) and water *ad libitum*. Animal care and use were performed according to protocols reviewed and approved by the Institutional Animal Care and Use Committee at the University for Arkansas for Medical Sciences. ## Protease activity assays Recombinant human FAP (R&D Systems, Minneapolis, MN, USA) was incubated with dye quenched (DQ) type I collagen or type IV collagen (100 μg/ml; Life Technologies, Carlsbad, CA, USA) in assay buffer according to manufacturer’s instructions. Proteolytic cleavage of each DQ substrate was assessed by measuring fluorescence in a Synergy-2 plate reader (Biotek, Winooski, VT, USA) using excitation/emission wavelengths of 485/528 nm. Baseline fluorescence was determined by incubation of DQ substrates in assay buffer in the absence of FAP. As a control for proteolytic activity, FAP was incubated with its synthetic substrate Z-Gly-Pro-AMC (Bachem, Bubendorf, Switzerland) according to manufacturer’s instructions. ## Cell isolation and treatment Mouse peritoneal macrophages (MPMs) were isolated from C57BL/6J and SR-A^-/- mice via peritoneal lavage with sterile saline from non- injected mice (for spreading assays) or from mice injected intraperitoneally with 4% thioglycollate 4 days prior to isolation (for attachment assays). For each assay, qualitatively similar results were obtained in preliminary experiments using non-elicited and elicited macrophages (data not shown). Isolated cells were immediately resuspended in DMEM GlutaMax (Life Technologies, Carlsbad, CA, USA) supplemented with FBS (10% vol/vol, Atlanta Biologicals, Flowery Branch, GA, USA), and penicillin/streptomycin (1%). Cell number and viability were assessed prior to use in experiments. ## Macrophage adhesion assays Macrophage attachment was assessed as described previously. Briefly, 12-well tissue culture dishes were coated for 5 h at 37°C with 10 μg/cm² of type I collagen (Stem Cell Technologies, Vancouver, Canada) or fibronectin (Sigma, St. Louis, MO, USA). Collagen coated wells were then treated (24 h; 37°C) with buffer (control), recombinant FAP (1 μg/ml), or FAP that was inhibited with 4 mM PMSF or heat inactivated at 80°C for 10 min. The treated wells were washed, and freshly isolated MPMs (10⁶ cells/well) plated for 30 min at 37°C. Non-adhered cells were removed by washing with PBS, and the number of cells remaining attached were quantified using a hemocytometer. Macrophage spreading was similarly examined. Tissue culture chamber slides (Nalge Nunc International, Naperville, IL, USA) were coated with type I collagen (10 μg/cm²; 4°C; 24 h), and then treated for 24 h at 37°C with buffer (control), active FAP, or heat inactivated FAP (1 μg/ml). Treated wells were washed, and freshly isolated MPMs (0.15 x 10⁶ cells/chamber) plated for 2 hr at 37°C. Non-adherent cells were removed, and then adherent cells were fixed with 4% paraformaldehyde and permeabilized with 0.1% Triton X-100. Cells were stained with Alexa-Fluor 647-conjugated phalloidin (Life Technologies) and nuclei were stained with DAPI. Images (40x) were digitally captured with Olympus CKX41 microscope and the surface area of cells quantified using AxioVision software (Carl Zeiss, Jena, Germany). Five independent images with at least 75 cells total were quantified in each experiment. ## Immunohistochemistry of MMTV-PyMT tumors Tumor tissues were isolated from MMTV-PyMT females before individual tumors reached 1 cm in any dimension. Isolated tumors were immediately embedded in OCT medium, rapidly frozen in liquid nitrogen, and stored at -80°C. Sections (8 μm) were cut onto glass slides and tissues fixed by incubation in ice-cold acetone for 20 min. Endogenous peroxidases were blocked with a dual endogenous enzyme block (Dako North America, Carpinteria, CA, USA). Additional blocking was performed with serum-free protein block (Dako) and 2.5% horse serum (Vector, Burlingame, CA, USA). To detect collagen, tissues were incubated with an anti- collagen antibody (Rabbit Anti-Mouse, EMD Millipore, Billerica, MA, USA), followed by Vector ImmPress Anti-Rabbit Reagent for alkaline phosphatase, and Vector ImmPact Red Substrate Solution. A dual-staining approach was used to identify SR-A-expressing macrophages and FAP-expressing fibroblasts. Following the blocking steps outlined above, tissues were incubated with primary anti-SR-A antibody (Goat Anti-Mouse, R&D Systems), then treated with Vector ImmPress Anti- Goat Reagent for peroxidase and Vector Immpact DAB Solution to visualize SR-A staining in brown. A second blocking step was then performed with 2.5% horse serum, and tissue sections incubated with primary anti-FAP antibody (Rabbit Anti-Mouse, Millipore), followed by Vector Immpress Anti-Rabbit Reagent for alkaline phosphatase and Vector Immpact Red Substrate Solution to visualize FAP staining in red. Tumor sections were counterstained with hematoxylin, dehydrated and coverslipped. ## Statistical analysis As indicated in individual figure legends, experiments were repeated at least three times and data analyzed with GraphPad Prism software using a *t*-test for comparing 2 groups, or ANOVA followed by the appropriate post-hoc statistical test to compare multiple groups. Differences with *p* \< 0.05 were considered statistically significant. # Results ## FAP selectively cleaves type I collagen FAP has been reported to cleave type I collagen, which is the most abundant ECM protein in a tumor stroma. This activity was confirmed using recombinant FAP and fluorescently quenched (DQ) collagen substrates that fluoresce when cleaved. As shown in, incubating FAP (1 μg/ml) with DQ type I collagen resulted in a time- dependent increase in fluorescence that was most evident during the first 24 h. Inhibiting FAP proteolytic activity with PMSF or by heat-inactivation (data not shown) prior to incubation with DQ type I collagen abolished the FAP dependent increase in fluorescence. In contrast to its effect on type I collagen, there was no increase in fluorescence when FAP was incubated with DQ type IV collagen. Together these results demonstrate that FAP selectively cleaves type I collagen, and establish conditions (1 μg/ml, 24 h) that were used to examine the consequence of FAP-mediated collagen cleavage in the macrophage adhesion assays described below. ## FAP-mediated cleavage of type I collagen increases macrophage adhesion Specific modifications of collagen have been shown to enhance macrophage adhesion. We therefore tested whether macrophage adhesion to type I collagen was increased following FAP-mediated cleavage. Primary MPMs were plated for 30 min (to assess attachment) or 2 h (to assess spreading) on type I collagen-coated tissue culture dishes that were untreated, treated with catalytically active FAP, or treated with FAP that was inhibited with PMSF (FAP + PMSF) or heat inactivation (inactive FAP). Macrophages attached poorly (≤ 1% of plated cells) to untreated (native) type I collagen. In contrast, the number of attached macrophages was significantly increased (\> 30% of plated cells) when plated on collagen treated with active FAP, but not when plated on collagen treated with PMSF-inhibited FAP. Similarly, relative to macrophages plated for 2 h on native type I collagen, macrophages that were plated on type I collagen that was treated with active FAP exhibited enhanced spreading as evidenced by a significant increase in surface area with a high percentage of cells having a surface area \> 100 μm². Macrophage spreading on collagen treated with inactive FAP was similar to that on native collagen. Overall, these results demonstrate that macrophage adhesion to type I collagen is substantially enhanced by FAP-mediated cleavage. ## Macrophage adhesion to FAP-cleaved collagen is integrin-independent Integrins are widely expressed surface proteins that mediate Ca²⁺-dependent cell adhesion to ECM, thus sequestering extracellular cations can disrupt this interaction. To determine whether increased macrophage adhesion to FAP-cleaved type I collagen is mediated by integrins, macrophage attachment was examined following chelation of divalent cations with EDTA. Primary MPMs were adhered to untreated (control) and FAP-cleaved type I collagen for 30 min in the presence or absence of EDTA (5 mM). Results depicted in show that macrophage attachment to type I collagen was substantially increased by FAP-mediated cleavage, and this increased attachment was not impaired by EDTA treatment. To confirm that integrin-mediated adhesion could be inhibited with this approach, MPM adhesion to fibronectin was tested in the presence and absence of EDTA. Over 80% of plated macrophages adhered to fibronectin in the absence of EDTA; however \<10% adhered in the presence of EDTA confirming disruption of integrin-mediated adhesion (data not shown). The divalent cation independence of the increased macrophage adhesion to FAP-cleaved type I collagen indicates that this adhesion is not mediated by integrins. ## Macrophage adhesion to FAP-cleaved collagen is mediated by SR-A Macrophages express a variety of receptors that recognize modified ECM components. In particular, SR-A mediates the divalent cation independent macrophage adhesion to a variety of substrates including glucose-modified collagen, denatured collagen, and collagenase-cleaved type I collagen. Therefore, we determined whether SR-A mediates macrophage adhesion to FAP- cleaved type I collagen. MPMs were adhered to tissue culture plates that were coated with type I collagen and left untreated (control) or treated with active FAP in the presence or absence of polyinosine, which is used to antagonize SR-A binding. As shown in, increased adhesion of MPMs to FAP-cleaved collagen was significantly inhibited in the presence of polyinosine. This result is consistent with a role for SR-A in mediating macrophage adhesion to FAP-cleaved collagen. Although polyinosine binds SR-A, it is not specific for SR-A. Therefore, we compared macrophage adhesion to FAP-cleaved type I collagen using MPMs isolated from wild type (SR-A^+/+) and SR-A knock out (SR-A^-/-) mice. Primary MPMs were plated for 30 min (to assess attachment) or 2 h (to assess spreading) on type I collagen-coated tissue culture dishes that were untreated, treated with active FAP, or treated with inhibited FAP. As shown in, both wild type and SR-A^-/- macrophages attached poorly to native type I collagen (buffer treated). Treating collagen with active FAP, but not with PMSF-inhibited FAP, significantly increased the attachment of wild-type, but not SR-A^-/- macrophages. In parallel, the majority of wild-type and SR-A^-/- macrophages remained rounded with a surface area \< 100 μm² when plated for 2 h on native type I collagen. In contrast, compared to SR-A^-/- macrophages, significantly more wild-type macrophages adopted a spread morphology with a surface area \>100 μm² when plated on type I collagen that was treated with active FAP, but not inactive FAP. These results identify SR-A as the macrophage receptor responsible for increased adhesion to FAP-cleaved type I collagen. ## SR-A expressing TAMs localize to the tumor stroma The finding that FAP-cleaved collagen enhances SR-A-mediated macrophage adhesion suggests that SR-A-expressing TAMs should be enriched in areas of tumor where TAFs are abundant. To assess this prediction, we isolated mammary tumors from female MMTV-PyMT mice and performed immunostaining for collagen, and dual staining for FAP, SR-A that would indicate co-localization of TAFs and TAMs, respectively. As shown in, the breast tumors are characterized by a distinct collagen-rich stromal compartment surrounding the acini of malignant epithelial cells. Importantly, dual staining for FAP-expressing TAFs and SR-A-expressing TAMs showed an abundant macrophage infiltrate into these tumors and that FAP- positive fibroblasts and SR-A-positive macrophages are localized to the same stromal areas in the tumors.. This co-localization of FAP-expressing TAFs and SR-A-expressing TAMs in the collagen-rich stroma supports the mechanism for localized stromal cell communication mediated by FAP and SR-A via the ECM modifications. # Discussion Our results demonstrate that FAP-mediated cleavage of collagen creates an adhesion substrate for macrophages that is recognized by SR-A. This conclusion is based on the increased attachment and spreading of macrophages on collagen treated with enzymatically active FAP, but not to collagen exposed to FAP that was enzymatically inactivated. The cation-independence of this adhesion and the disruption of this adhesive interaction by polyinosine, a compound that disrupts scavenger receptor interactions, suggested that SR-A mediates macrophage attachment to FAP-cleaved collagen. The specific importance of SR-A in mediating macrophage adhesion to FAP-cleaved collagen was confirmed by comparing the adhesion of wild type macrophages to those that lack SR-A. Together, these results identify FAP-cleaved type I collagen as a ligand for SR-A-mediated macrophage adhesion. FAP is an activated fibroblast-specific protease that is implicated in modifications of tumor stroma and correlated with higher histological grades of malignancies, metastasis, and poor patient prognosis. Importantly, FAP is not normally found in adult healthy tissues but occurs at sites of inflammation. Although the biological significance of specific FAP cleavage products is not known, there is increasing evidence that FAP activity plays an important role in modifying collagen matrices, particularly in tumors and atherosclerosis. For example, the expression and proteolytic activity of FAP was correlated with the presence of macrophages in human atheroma. In mouse tumor models, FAP^-/- and FAP enzymatic inhibition slowed tumor growth, increased stromal collagen content, and decreased collagen fibril organization in the tumor stroma compared to tumor tissues isolated from FAP^+/+ mice. It was further shown that FAP-targeted elimination of TAFs in the 4T1 mouse model of breast cancer reduced infiltration of TAMs and other immune cells to the tumor site. Our results demonstrating that FAP-cleaved collagen increases macrophage adhesion suggest a novel mechanism by which FAP regulates inflammatory responses within tumors by modulating the behavior of macrophages. Macrophages are key contributors to many chronic inflammatory conditions. Targeted influx and accumulation of macrophages are important determinants of macrophage function, and critical to the regulation of inflammatory, immune, and repair processes. In cancer, macrophage accumulation and interaction with other cells in the tumor microenvironment are linked to poor patient prognosis. Thus, an important implication of our findings is the possibility that increased FAP expression by TAFs in the tumor stroma results in the cleavage of type I collagen to promote the localized accumulation of macrophages. A key feature of the model described above is that the adhesion of macrophages in tumors and/or chronic inflammatory sites is modulated by ECM modification. Macrophages express multiple receptors that recognize ECM components, including integrins and pattern recognition receptors (PRRs) such as SR-A. SR-A mediates the cation-independent adhesion of macrophages to several modified ECM components including heat-denatured collagen and collagenase-treated type I collagen, but does not bind to native collagen. Our findings suggest that a novel biological function of FAP proteolysis of type I collagen is to expose previously masked adhesion sites for SR-A on the collagen. Although MMPs such as collagenase are known to aggressively degrade and convert native type I collagen into an adhesive substrate for SR-A, FAP is a serine protease that cleaves collagen differently than MMPs creating products that are not the same as those generated by collagenases. Thus, one cannot assume a priori that previous findings using collagenase to degrade collagen \[e.g.,\] apply to FAP-cleaved collagen. Nonetheless, our results showing that macrophage adhesion to FAP- cleaved collagen is cation-independent, inhibited by polyinosine, and absent in SR-A^-/- macrophages identify SR-A as a macrophage receptor that mediates adhesion to FAP-cleaved collagen. The role of SR-A in infection, atherosclerosis and Alzheimer’s disease are well studied \[reviewed in\]. However, little is known about the contributions of SR-A to the formation and progression of tumors. Recent reports suggest that SR-A may be a marker for macrophages present in aggressive tumors. Importantly, Neyen et al. demonstrated that SR-A deficiency in vivo protects mice from tumor progression and metastasis. In addition, we previously reported that SR-A- mediated adhesion activates multiple signaling cascades including Lyn-PI3K and PLA₂-12/15LOX pathways. We further showed that SR-A-dependent macrophage adhesion induces the release of PGE₂ which acts in an autocrine/paracrine mechanism to inhibit TNFα production and increase secretion of IL-10. Other studies examining the role of SR-A in mouse models of inflammatory disease demonstrated a similar association of SR-A with a tumor supportive M2 macrophage phenotype. These reports suggest that in addition to promoting macrophage retention, SR-A-mediated interaction with components of the tumor microenvironment may contribute to the regulation of TAM phenotype. ## Conclusion Our results show that FAP cleavage of type I collagen promotes macrophage adhesion (attachment and spreading), and that these interactions are specifically mediated by macrophage class A scavenger receptors (SR-A/CD204). We further show that FAP positive TAFs and SR-A positive TAMs co-localize in the stroma of breast tumors from MMTV-PyMT mice. Together our results implicate a novel biological role for FAP-cleaved type I collagen, and define a mechanism for dialog between fibroblasts and macrophages via a FAP-ECM-SR-A axis, that could increase macrophage retention and function in tumors and potentially other inflammatory sites. The authors would like to thank Beixiang He and Jessica Webber for technical assistance. We would also like to thank Dr. Ginell Post for critical reading of the manuscript. Support was provided by the Dept. of Pathology, the UAMS College of Medicine Research Council, and an NIH award to TK and SRP (R21CA185691). [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AM TK SRP. Performed the experiments: AM EH SV. Analyzed the data: AM EH SV TK SRP. Wrote the paper: AM EH TK SRP.

# Introduction A cluster of nitrate assimilation genes occurs in the ascomycetes *Aspergillus nidulans* and *Pichia angusta* and in the basidiomycetes *Hebeloma cylindrosporum*, which is mycorrhizal, and *Phanerochaete chrysosporium*, which is a wood-decayer. This cluster, abbreviated fHANT-AC, encodes a high affinity nitrate transporter (NRT2, TCDB 2.A.1.8.5) along with a nitrate reductase (EUKNR, EC1.7.1.3) and a ferredoxin-independent assimilatory nitrite reductase (NAD\[P\]H-NIR, EC1.7.1.4). The latter two proteins are required for the reduction of nitrate to ammonium. Ascomycetes and basidiomycetes are sister taxa that together form a clade called Dikarya, which contains the vast majority of described species of filamentous, terrestrial fungi. Without regard to clustering, *nrt2* is widely distributed in bacteria, plants, heterokonts and other groups, but not in opisthokonts outside Dikarya. *Euknr* is restricted to eukaryotes, and absent from opisthokonts outside Dikarya, and *NAD(P)H-nir* is known to occur in Dikarya, heterokonts and bacteria. The clustering of functionally related genes may be selectively favorable because it facilitates coordinated expression of the constituent genes. The “selfish operon” hypothesis puts forth that clusters can result from the selective maintenance of fitness-improving genes following horizontal gene transfer (HGT) events. The clustering of genes in selfish operons greatly increases success of HGT events when a phenotypic advantage or ability to exploit a new niche is conferred to the recipient. Gene clusters may facilitate the acquisition of entire metabolic pathways in the fungi. Examples mainly involve secondary metabolite clusters, which do not necessarily require HGT between fungal species to explain their distribution, and pathogenicity gene clusters. Transfers of supernumerary chromosomes and large fragments of chromatin remain the most compelling mechanisms, for the transfer of metabolic pathways. After discovering a potential horizontal transfer of *nrt2* previously, we conducted a series of phylogenetic analyses to reconstruct the history of the fHANT-AC and its constituent genes. We first searched available eukaryotic genome sequences for homologs of genes in fHANT-AC. We then estimated the organismal phylogeny of ascomycetes, basidiomycetes, and oomycetes using rRNA and RPB2 gene sequences. We contrasted the organismal phylogeny with results of independent and concatenated phylogenetic analyses of genes in the fHANT-AC, which we performed at two phylogenetic scales: within the fungi (rooted with the heterokont *Phytophthora*), and across diverse eukaryotes and eubacteria (rooted with selected prokaryotes). Phylogenetic analyses within the fungi suggest that the nitrate assimilation cluster in the ascomycete *T. reesei* was obtained by HGT from a basidiomycete and corresponds to a change in nutritional mode. This is the best evidence to date that horizontal transfer of a metabolic pathway can facilitate niche shift in fungi. Less conclusive phylogenetic and structural evidence from analyses across diverse eukaryotes and eubacteria cannot reject a heterokont origin of the nitrate assimilation cluster in Dikarya. # Results ## Distribution of Genes of the fHANT-AC in Eukaryotes is Patchy Homologs of each of the three genes of the fHANT-AC (SI) are limited to Viridiplantae, Rhodophyta, heterokonts (oomycetes+diatoms in this dataset) and Dikarya, with one exception in *Nematostella vectensis* (Metazoa), which possesses bacteria-like *nrt2* and *nitrite reductase* homologs (<http://genome.jgi-psf.org/Nemve1/Nemve1.home.html>). Viridiplantae and Rhodophyta possess genes homologous to the dikarya/heterokont *nrt2* and *euknr*, but not the fungal/heterokont *NAD(P)H-nir*. Instead, these groups maintain a ferredoxin-dependent nitrite reductase gene (*cpnir*, EC1.7.7.1) of probable plastid/cyanobacterial origin. Diatoms possess the three Dikarya/oomycete homologs in addition to *cpnir* (<http://genome.jgi- psf.org/Thaps3/Thaps3.home.html>). Diverse eubacteria possess *nrt2* and *NAD(P)H-nir* homologs, but not *euknr* which has an exclusively eukaryotic distribution. From the *Populus trichocarpa* genome, we recovered a β proteobacteria-like *NAD(P)H-nir*. Homologs of the dikarya/oomycete genes are absent from most Dikarya yeasts (with the exception of *Pichia angusta*), filamentous fungi outside Dikarya (*Rhizopus oryzae* and *Phycomyces blakesleeanus*), and the chytrid *Batrachochytrium dendrobatis*. ## Phylogenetic Analyses Suggest Horizontal Gene Transfer of the fHANT-AC A comparison of the organismal phylogeny and the phylogeny of fHANT-AC within fungi provides support for the hypothesis that fHANT-AC has been transferred horizontally from basidiomycetes to an ascomycete. In separate and combined analyses of nuclear ribosomal RNA genes (rDNA) and translated *rpb2* (RPB2) sequences using Bayesian, maximum likelihood (ML) and maximum parsimony (MP) methods (BPP = Bayesian Posterior Probabilities, MLB = Maximum Likelihood Bootstrap, MPB = Maximum Parsimony Bootstrap), we recovered species relationships that are congruent with previous molecular phylogenies. We received strong support (1.0 BPP, 100 MPB combined analysis and 100MLB rDNA and RPB2 analyses unless otherwise noted) for the monophyly of Basidiomycota (RPB2 = 97 MLB, rDNA = 63 MLB) including *Ustilago maydis*, the ascomycetes (rDNA = 81 MLB), Pezizomycotina (non-yeast ascomycete lineages), and Sordariomycetes including *Trichoderma reesei*. The phylogeny of the fHANT-AC genes in fungi is similar to the organismal phylogeny inferred with rRNA and RPB2 genes, with one striking exception: in independent and combined analyses, the three genes of the fHANT-AC of the ascomycete *T. reesei* were strongly supported (1.0 BPP, 100 MPB, 100 MLB) as sister to those of the basidiomycete *Ustilago maydis*. Additionally, a Shimodaira-Hasegawa test rejected the monophyly of ascomycete nitrate assimilation clusters when *T. reesei* sequences were included (p\<0.0001). Together, these analyses indicate strong conflict between the organismal and nitrate assimilation gene phylogenies that is best explained by HGT of the fHANT-AC from the Ustilaginales to *Trichoderma*. A search of the EST database at GenBank confirms that at least two of the horizontally transferred genes (*nrt2* and *euknr*) are expressed in *T. reesei*. We recovered no evidence of vertically derived homologs of these genes in *T. reesei* Higher-level analyses of HANT-AC proteins including diverse eubacteria sequences along with sequences from fungi, heterokonts and 25-29 (depending on gene distribution) eukaryotic lineages are not inconsistent with a heterokont origin of the fungal HANT-AC, although patchy distribution prevents a strong conclusion from phylogenetic analyses. NAD(P)H-NIR sequences from three species of *Phytophthora* (oomycetes) and *Phaeodactylum* and *Thalassiosira* (diatoms) do not form a heterokont clade. *Phytophthora* NAD(P)H-NIR sequences are sister to fungal sequences (1.0 BPP, 81 MPB, 57 MLB), and heterokonts plus fungi are supported as monophyletic by MP analyses (77 MPB), but not by ML or Bayesian analyses. Analyses of NRT2 suggest a red alga+heterokont+fungi clade (1.0 BPP, \<50 MPB, 74 MLB) that excludes green algae and plants in a strongly supported (1.0 BPP, 92 MPB, 99 MLB) eukaryote clade. These results are consistent with previous analyses extensively focused on fungi, which suggested that fungal NRT2 sequences form a monophyletic group that is nested within a paraphyletic assemblage of sequences from heterokonts. Analyses of EUKNR sequences, which were restricted to the eukaryotic distribution of the gene, revealed little phylogenetic resolution outside the fungi. We found no synapomorphic insertion or deletion events in any of these sequences that exclusively support a sister relationship between fungi and oomycetes. Future analyses that include eukaryotic and fungal lineages not represented in our survey could falsify the hypothesis of an ancient transfer from heterokonts to dikarya. ## Gene Clustering of fHANT-AC genes is Conserved, but Order is Not While fHANT-AC genes are usually clustered, gene orders within and flanking the cluster are not well conserved between lineages. A comparison of genes flanking the fHANT-AC in *Ustilago maydis* and *T. reesei* genomes did not reveal additional genes that appear to have been obtained by HGT. Most genes on the *T. reesei* scaffold 28 are highly similar to ascomycete genes (data not shown). In addition, the *Cip2* gene that occurs three genes downstream from the *nrt2* gene is a confirmed *T. reesei* sequence. These results suggest that the incongruous fHANT-AC was incorporated by chromosomal recombination involving a small number of genes rather than by acquisition of an additional chromosome. Genes flanking the fHANT-AC in *T. reesei* appear to be related to carbohydrate metabolism. We speculate that the occurrence of this cluster could influence carbohydrate metabolism by either upregulating or inhibiting transcription of these genes, and thereby contribute downstream to the superlative propensity of *T. reesei* to produce cellulases, *e.g.*in. # Discussion ## The fHANT-AC persists with genomic rearrangement The only groups found to maintain the three genes of the fHANT-AC in a cluster (as of June 2007) are the dikarya fungi and oomycete heterokonts (SI). *Euknr* and *nrt2* are widely distributed in dikarya fungi, chromalveolates (alveolates+heterokonts), and plants, whereas *NAD(P)H-nir* does not occur outside heterokonts and Dikarya. We cannot yet falsify the hypothesis that the fHANT-AC cluster originated in the lineage leading to heterokonts (SI) and was subsequently transferred to Dikarya. Because of the limited sample of non- dikarya fungi (SI) we cannot rule out the possibility of an earlier fungal origin. Nitrate assimilation clusters also occur in at least two green algae, but each of these clusters contains a unique complement of genes that are not all homologous to those in the fHANT-AC. The convergent evolution of a nitrate assimilation cluster in green algae could suggest a general selective advantage to clustering of nitrate assimilation genes. Clustering of nitrate assimilation genes is observed in varying degrees within Dikarya. In Basidiomycota where the genes are present, the clustered orientation of the three genes predominates with one known exception. A retroviral sequence interrupts the cluster between *nrt2* and *NAD(P)H-nir* in *Laccaria bicolor*, but the genes remain in the same locus. The cluster has disassembled in certain ascomycetes. At least two lineages have lost the clustering of the three genes while maintaining the genes themselves, most notably the Sordariomycetes, of which *T. reesei* is a member. *Botryotinia fuckeliana* maintains a cluster of two of the genes, while the closely related *Sclerotinia sclerotiorum* lacks clustering entirely. The orientation of the genes within the cluster is not phylogenetically conserved within Dikarya, suggesting that while rearrangement is common, these rearrangements do not necessarily disassemble the cluster. We propose three scenarios to explain the disassembly of the fHANT-AC in ascomycetes (SI). If cluster formation is assumed to be an unlikely event, then that would favor a scenario (SI) that only requires the assembly of the fHANT- AC to have occurred once. Metabolic gene clusters have been reported in several fungal lineages, and one occurrence of rapid assembly of a cluster has been suggested in yeast. However, there is currently no consensus regarding methods of measuring the relative rates of cluster assembly, disassembly and horizontal transfer in fungi. ## The Anomalous fHANT-AC in *Trichoderma reesei* coincides with Niche Shift *T. reesei,* an ascomycete adapted to wood decay occurs in a genus consisting largely of endophytes and parasites. *T. reesei* strain QM9414, the subject of the *T. reesei* genome project, was derived from QM6a, a strain originally defined by its inability to grow on nitrate as a sole nitrogen source, likely due to defects in nitrite reductase. Several sexually competent strains of *T. reesei,* however, are known to utilize nitrate. We know of no direct evidence of nitrate assimilation by other *Trichoderma* species, although the presence of nitrate has been linked to physiological responses, and suppression of growth in some species. Attempts to amplify *nrt2* from various *Trichoderma* species using PCR were not successful. A shift away from a tightly regulated nitrate assimilation pathway may have preceded the acquisition of a basidiomycete fHANT- AC by *T. reesei*. Analyses of gene synteny suggest that loss of the ascomycete nitrate assimilation genes in *T. reesei* was preceded by the disassembly of the cluster in Sordariomycetes. In one study, *Trichoderma* species (not including *T. reesei*) and *Sclerotinia sclerotiorum* (which also lacks an intact cluster) were shown to utilize ammonium preferentially, and possibly exclusively (in *Trichoderma)*, over nitrate in ammonium nitrate feeding experiments. The effect was also observed, but less pronounced, for other sordariomycetes (*Fusarium spp*.). These observations suggest that there was a change in the mode of nitrate assimilation in the evolution of sordariomycetes. This could have favored the acquisition of the basidiomycete fHANT-AC and subsequent loss of vertically inherited genes when *T. reesei* switched to a nitrogen-deprived, woody substrate. The horizontal transmission leading to the *T. reesei* cluster is a plausible scenario because some *Trichoderma* species are intracellular parasites of basidiomycetes. ## Analysis of the HANT-AC Suggests Widespread Loss, Convergence or a “Selfish Operon” The distribution of HANT-AC genes could suggest either widespread loss outside Dikarya, heterokonts and plants or an HGT origin of the fungal homologs. We can reject the null hypotheses that chromalveolate and plant NRT2 form a clade that excludes fungi (p = .038), that heterokont NAD(P)H-NIR are monophyletic exclusive of fungi (p\<.0001), and that an organismal phylogeny based on five nuclear protein coding genes (SI), restricted to taxa bearing HANT-AC genes, is consistent with the fungi+*Phytophthora* clade (p\<.0001) suggested by NAD(P)H-NIR analyses. Each of these results is consistent with an HGT origin of these genes in fungi. The eukaryotic *NAD(P)H-nir* could have been derived from a mitochondrial source, which would require multiple losses in different lineages of eukaryotes, or there could have been a more recent transfer from bacteria to a lineage leading to the heterokonts. The absence of *euknr* from other opisthokonts also suggests that this gene was secondarily acquired in Dikarya, although phylogenetic analyses do not clearly identify a source. The existence in *Phytophthora* of a cluster (oHANT-AC) homologous to the fHANT-AC, some of whose components appear to be vertically derived, is the strongest evidence of a heterokont origin of the fungal genes, implying that a cluster could have facilitated HGT. We present an hypothesis of the origin of the fHANT- AC in SI, which parsimoniously accounts for the observed distribution of HANT-AC genes in eukaryotes. There we show a single origin of the cluster in heterokonts from genes present in common ancestors of heterokonts and plants, followed by a transfer to the ancestor of Dikarya. Recent studies have suggested closer genetic links between oomycetes and fungi than would be predicted by organismal phylogeny. One suggested that a number of shared EST sequences relate to a mutual tendency toward pathogenicity. It was not clear if this is evidence of genomic convergence or genetic exchange. Another study suggested that at least four genes have been transferred from ascomycete fungi to oomycetes. A failure to detect homologous sequences in other heterokonts and non-dikarya fungi in that study leaves the origin of these sequences ambiguous, requiring the invocation of additional prokaryotic donors. In our study, the presence of homologous genes in other heterokonts and related lineages, but not in other fungi suggests that the direction of the hypothetical transfer of the fHANT-AC would be from a heterokont to a common ancestor of dikarya fungi. A limited sample of non-dikarya fungi leaves an earlier fungal origin also possible. In this case, the restriction of clustered HANT-AC genes to oomycetes suggests that the heterokonts could have produced the species that donated the fHANT-AC to dikarya fungi. While the current study alone is not conclusive as to genetic exchange between these groups, it contributes to a developing hypothesis of an ancient, intimate association involving Dikarya and heterokonts. The alternate hypothesis is that the HANT-AC arose convergently in fungi and oomycetes under similar environmental pressures. For example, in aerobic soils, reduced iron for denitrification is limited, leading to elevated nitrate. Iron- containing ferredoxin is required to donate electrons in plant nitrite reduction, but this function can be performed by NAD(P)H (derived from cellular respiration) in fungi and heterokonts; they could be partly freed from iron limitations through this mechanism. That certain diatoms are dominant in high- nitrate, low-iron environments seems to support this. This nutritional mode could benefit photosynthetic partners in ectomycorrhizal and lichen symbioses under aerobic conditions. In fact, ectomycorrhizal fungi both down-regulate plant nitrite reductase and increase iron uptake. In contrast, oomycete plant pathogens could out-compete their host for nitrate in iron-limited conditions. Alternatively, the absence of photosystems to efficiently replenish reduced ferredoxin in fungi and oomycetes could have necessitated convergent acquisition of a respiration-linked source of electrons for nitrite reduction. Co-regulation of the three genes in the fHANT-AC in *H. cylindrosporum* was recently demonstrated. This was earlier demonstrated in the ascomycetous yeast, *Pichia angusta*. The intergenic region between *NAD(P)H-nir* and *euknr* in *A. nidulans* serves as a bidirectional promoter for both genes. These findings collectively suggest that the fHANT-AC resembles a eukaryotic operon. The “selfish operon” is a coordinately regulated cluster of genes, which receive improved fitness because they are more readily maintained when transferred between organisms. While not truly a “selfish” element due to its likely positive influence on host fitness, evidence of HGT of the fHANT-AC makes it a candidate for a “selfish” eukaryotic operon. The current results suggest that acquisition of nutritional operons may promote ecological shifts in fungi. Differential utilization of the fHANT-AC in various strains of *T. reesei* could suggest this species is currently undergoing such an evolutionary event. ## A key acquisition in colonization of dry land? The Dikarya experienced a neoproterozoic diversification that was not paralleled in other fungal lineages, including some that also associate with plants. We speculate that the horizontally acquired ability to derive nitrogen from oxidized substrates, and to thus benefit photosynthetic hosts, may have been a key innovation that allowed the Dikarya to diversify in terrestrial habitats as free-living saprotrophs or mycorrhizal symbionts. # Materials and Methods ## Phylogenetic Analysis of Organisms Containing the High Affinity Nitrate Assimilation Gene Cluster (fHANT-AC) using rRNA and RPB2 gene sequences We used a combination of tblastn and searches for annotated genes to obtain phylogenetically relevant sequences for species noted with an \* in SI. Substitutions were made for *Pichia angusta* RNA polymerase II second largest subunit (RPB2, AAS67502, *Pichia guilliermondii*) and *Hebeloma cylindrosporum* (partial nuclear small ribosomal RNA subunit gene \[nssu\] AY745703 and *rpb2* DQ472718 from *H. velutipes* specimen PBM 2277) where genome projects were not available. We generated a partial nuclear large ribosomal RNA subunit gene (nlsu) sequence from *H.cylindrosporum* strain CBS558.96 (EF219136) using the primers LROR and LR5 (sequencing was performed on an ABI377 Automated DNA Sequencer using BigDye ver1.1; Applied Biosystems, Foster City, CA, USA). We concatenated nlsu and nssu nucleotide sequences and the inferred amino acid sequence of RPB2 for all available fungal genomes. Sequences were aligned with ClustalX, and adjusted manually. We created independent alignments (SI) for each gene by excluding two of the three genes, and a combined analysis that included all three genes. Ambiguously aligned characters were excluded from further analysis. Each alignment was analyzed by (A) 1000 maximum parsimony bootstrap replicates with ten random addition sequence replicates per bootstrap replicate, saving multiple trees at each replicate in PAUP\* 4.0b, (B) two Bayesian analyses using mixed protein models for amino acids and a GTR plus gamma model of evolution for nucleotides in MrBayes ver. 3.2, ; we ran the Bayesian analyses for one million generations, sampled trees every 100 generations and removed trees sampled before likelihoods for two runs converged and stabilized as the burnin, and (C) 100 Maximum Likelihood Bootstrap replicates of RPB2 with mixed protein models and of combined rDNA with mixed GTR models as implemented in RaxML-VI-HPC ver. 2.2.3. We also performed (D) maximum likelihood searches of a combined rDNA dataset and an RPB2 dataset in RaxML-VI- HPC ver. 2.2.3 (mixed GTR and protein models) for the optimal topologies and also for the best topology in which *T. reesei* sequences were constrained to form a clade with *U. maydis*. We then used a Shimodaira-Hasegawa test, as implemented in TreePuzzle ver. 5.2, to compare the likelihoods of resulting topologies. ## Analysis of the nitrate assimilation gene cluster in fungi We concatenated amino acid sequences inferred from three genes in the fHANT-AC (*nrt2, euknr* and *NAD(P)H-nir*) of *Hebeloma cylindrosporum*. We used tBlastn with this sequence to search genome projects shown in supporting materials (SI) for homologous sequences. We obtained *H. cylindrosporum* and *P. angusta* fHANT-AC sequences, and *T. reesei* EST sequences from GenBank. The identity of each sequence was verified by a blastp search of GenBank to determine the most similar sequences and conservation of functional domain architecture. We assembled an alignment consisting of concatenated amino acid sequences from the fHANT-AC for all available fungal genomes (as of May, 2006, shown with an \* in SI) and also *Phytophthora sojae* and *Phytophthora ramorum* genomes, which were included as the outgroup. Alignments and analyses for independent genes and combined sequences were conducted as described for amino acid sequences in the phylogenetic analysis. We performed maximum likelihood searches of the combined fHANT-AC dataset using mixed protein models in RaxML-VI-HPC (see above) for the optimal topology and also for the best topology in which ascomycete sequences were constrained to be monophyletic. We then compared the likelihoods of resulting topologies using a Shimodaira-Hasegawa test as described above for RPB2. ## Analyses of components of the fHANT-AC in diverse eukaryotes and prokaryotes We conducted higher level analyses of NRT2, EUKNR and NAD(P)H-NIR amino acid sequences retrieved from eukaryotic genome projects along with bacterial sequences from GenBank. For NRT2, approximately 200 bacterial sequences and for NAD(P)H-NIR approximately 500 bacterial sequences were initially obtained. These were combined with the eukaryotic genome sequences, aligned with mafft ver. 5.861 using the default settings. Long ends of the alignment were trimmed and a neighbor joining analysis was performed in PAUP\* 4.0b. Thirty-nine clades of NRT2 and eighty-three clades of NAD(P)H-NIR were then analyzed as described above for the fHANT-AC genes, except 500 maximum likelihood bootstraps were performed on each gene separately. We confirmed that major clades from previous NRT2 analyses were represented. Twenty-nine eukaryotic EUKNR sequences were also analyzed as described for the other protein sequences. Constraint analyses as described above forced the monophyly of heterokont NAD(P)H-NIR sequences. We performed analyses as described for a combined dataset of α-tubulin, β-tubulin, RNA polymerase II largest and second largest subunits, and elongation factor 1-alpha limited to taxa bearing homologs of genes in the fHANT-AC. We then performed constraint analyses as described above, which forced either fungi and *Phytophthora* or fungi and heterokonts to be monophyletic. ## Analysis of gene order and location in fungi We used the browser at the *T. reesei* and *Ustilago maydis* genome websites to catalog genes flanking the fHANT-AC in each species. We cataloged genes of up to four inferred or hypothetical proteins for each species. We used tBlastn with a concatenated amino acid sequence of the three proteins encoded by the fHANT-AC of *Coprinopsis cinerea* in order of nitrate metabolism (NRT2-EUKNR-NAD(P)H-NIR) to obtain the location and relative direction of transcription of each gene in the eukaryotic genomes and bacterial genomes shown in supporting materials. We also performed tBlastn using the individual genes, and used 45% amino acid similarity to determine the absence of homologs for each gene in eukaryotes and 40% amino acid similarity in bacteria. This was repeated using the *Arabidopsis thaliana* ferredoxin-dependent nitrite reductase (NP_179164.1) as the query. # Supporting Information We would like to thank Z. Wang and several anonymous reviewers for helpful suggestions, and P. Matheny, M. Binder and C. Khatchikian for editorial help. [^1]: Conceived and designed the experiments: DH JS. Performed the experiments: JS. Analyzed the data: JS. Wrote the paper: DH JS. [^2]: The authors have declared that no competing interests exist.