text
stringlengths 65
113k
|
|---|
Development of a task-free method for presurgical mapping of language function is important for use in young or cognitively impaired patients. Resting state connectivity fMRI (RS-fMRI) is a task-free method that may be used to identify cognitive networks. We developed a voxelwise RS-fMRI metric, Functional Connectivity Hemispheric Contrast (FC-HC), to map the language network and determine language laterality through comparison of within-hemispheric language network connections (Integration) to cross-hemispheric connections (Segregation). For the first time, we demonstrated robustness and efficacy of a RS-fMRI metric to map language networks across five groups (total N = 243) that differed in MRI scanning parameters, fMRI scanning protocols, age, and development (typical vs pediatric epilepsy). The resting state FC-HC maps for the healthy pediatric and adult groups showed higher values in the left hemisphere, and had high agreement with standard task language fMRI; in contrast, the epilepsy patient group map was bilateral. FC-HC has strong but not perfect agreement with task fMRI and thus, may reflect related and complementary information about language plasticity and compensation. |
Intracortical myelin is involved in speeding and synchronizing neural activity of the cerebral cortex and has been found to be disrupted in various psychiatric disorders. However, its role in obsessive-compulsive disorder (OCD) has remained unknown. In this study, we investigated the alterations in intracortical myelin and their association with white matter (WM) microstructural abnormalities in OCD. T1-weighted and diffusion-weighted brain images were obtained for 51 medication-naïve patients with OCD and 26 healthy controls (HCs). The grey/white matter contrast (GWC) was calculated from T1-weighted signal intensities to characterize the intracortical myelin profile in OCD. Diffusion parameters, including fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD) and radial diffusivity (RD), were extracted from diffusion-weighted images to examine the WM microstructure in OCD. Compared with HCs, patients with OCD showed increased GWC in the bilateral orbitofrontal, cuneus, lingual and fusiform gyrus, left anterior cingulate, left superior parietal, right inferior parietal, and right middle frontal cortices, suggesting reduced intracortical myelin. Patients with OCD also showed decreased FA in several WM regions, with a topology corresponding to the GWC alterations. In both groups, the mean GWC of the significant clusters in between-group GWC analysis was correlated negatively with the mean FA of the significant clusters in between-group FA analysis. In patients with OCD, the FA of a cluster in the right cerebellum correlated negatively with the Yale-Brown obsessive-compulsive scale scores. Our results suggest that abnormal intracortical and WM myelination could be the microstructural basis for the brain connectivity alterations and disrupted inhibitory control in OCD. |
We describe the USC Multimodal Connectivity Database (http://umcd.humanconnectomeproject.org), an interactive web-based platform for brain connectivity matrix sharing and analysis. The site enables users to download connectivity matrices shared by other users, upload matrices from their own published studies, or select a specific matrix and perform a real-time graph theory-based analysis and visualization of network properties. The data shared on the site span a broad spectrum of functional and structural brain connectivity information from humans across the entire age range (fetal to age 89), representing an array of different neuropsychiatric and neurodegenerative disease populations (autism spectrum disorder, ADHD, and APOE-4 carriers). An analysis combining 7 different datasets shared on the site illustrates the diversity of the data and the potential for yielding deeper insight by assessing new connectivity matrices with respect to population-wide network properties represented in the UMCD. |
There is a need for accurate quantitative non-invasive biomarkers to monitor myelin pathology in vivo and distinguish myelin changes from other pathological features including inflammation and axonal loss. Conventional MRI metrics such as T2, magnetization transfer ratio and radial diffusivity have proven sensitivity but not specificity. In highly coherent white matter bundles, compartment-specific white matter tract integrity (WMTI) metrics can be directly derived from the diffusion and kurtosis tensors: axonal water fraction, intra-axonal diffusivity, and extra-axonal radial and axial diffusivities. We evaluate the potential of WMTI to quantify demyelination by monitoring the effects of both acute (6weeks) and chronic (12weeks) cuprizone intoxication and subsequent recovery in the mouse corpus callosum, and compare its performance with that of conventional metrics (T2, magnetization transfer, and DTI parameters). The changes observed in vivo correlated with those obtained from quantitative electron microscopy image analysis. A 6-week intoxication produced a significant decrease in axonal water fraction (p<0.001), with only mild changes in extra-axonal radial diffusivity, consistent with patchy demyelination, while a 12-week intoxication caused a more marked decrease in extra-axonal radial diffusivity (p=0.0135), consistent with more severe demyelination and clearance of the extra-axonal space. Results thus revealed increased specificity of the axonal water fraction and extra-axonal radial diffusivity parameters to different degrees and patterns of demyelination. The specificities of these parameters were corroborated by their respective correlations with microstructural features: the axonal water fraction correlated significantly with the electron microscopy derived total axonal water fraction (ρ=0.66; p=0.0014) but not with the g-ratio, while the extra-axonal radial diffusivity correlated with the g-ratio (ρ=0.48; p=0.0342) but not with the electron microscopy derived axonal water fraction. These parameters represent promising candidates as clinically feasible biomarkers of demyelination and remyelination in the white matter. |
The problem of emotion recognition has been tackled by researchers in both affective computing and cognitive neuroscience. While affective computing relies on analyzing visual features from facial expressions, it has been proposed that humans recognize emotions by internally simulating the emotional states conveyed by others' expressions, in addition to perceptual analysis of facial features. Here we investigated whether and how our internal feelings contributed to the ability to decode facial expressions. In two independent large samples of participants, we observed that individuals who generally experienced richer internal feelings exhibited a higher ability to decode facial expressions, and the contribution of internal feelings was independent of face recognition ability. Further, using voxel-based morphometry, we found that the gray matter volume (GMV) of bilateral superior temporal sulcus (STS) and the right inferior parietal lobule was associated with facial expression decoding through the mediating effect of internal feelings, while the GMV of bilateral STS, precuneus, and the right central opercular cortex contributed to facial expression decoding through the mediating effect of face recognition ability. In addition, the clusters in bilateral STS involved in the two components were neighboring yet separate. Our results may provide clues about the mechanism by which internal feelings, in addition to face recognition ability, serve as an important instrument for humans in facial expression decoding. |
The maintenance of sensory information in working memory (WM) is mediated by the attentional activation of stimulus representations that are stored in perceptual brain regions. Using event-related potentials (ERPs), we measured tactile and visual contralateral delay activity (tCDA/CDA components) in a bimodal WM task to concurrently track the attention-based maintenance of information stored in anatomically segregated (somatosensory and visual) brain areas. Participants received tactile and visual sample stimuli on both sides, and in different blocks, memorized these samples on the same side or on opposite sides. After a retention delay, memory was unpredictably tested for touch or vision. In the same side blocks, tCDA and CDA components simultaneously emerged over the same hemisphere, contralateral to the memorized tactile/visual sample set. In opposite side blocks, these two components emerged over different hemispheres, but had the same sizes and onset latencies as in the same side condition. Our results reveal distinct foci of tactile and visual spatial attention that were concurrently maintained on task-relevant stimulus representations in WM. The independence of spatially-specific biasing mechanisms for tactile and visual WM content suggests that multimodal information is stored in distributed perceptual brain areas that are activated through modality-specific processes that can operate simultaneously and largely independently of each other. |
Recently, much attention has been focused on the definition and structure of the hippocampus and its subfields, while the projections from the hippocampus have been relatively understudied. Here, we derive a reliable protocol for manual segmentation of hippocampal white matter regions (alveus, fimbria, and fornix) using high-resolution magnetic resonance images that are complementary to our previous definitions of the hippocampal subfields, both of which are freely available at https://github.com/cobralab/atlases. Our segmentation methods demonstrated high inter- and intra-rater reliability, were validated as inputs in automated segmentation, and were used to analyze the trajectory of these regions in both healthy aging (OASIS), and Alzheimer's disease (AD) and mild cognitive impairment (MCI; using ADNI). We observed significant bilateral decreases in the fornix in healthy aging while the alveus and cornu ammonis (CA) 1 were well preserved (all p's<0.006). MCI and AD demonstrated significant decreases in fimbriae and fornices. Many hippocampal subfields exhibited decreased volume in both MCI and AD, yet no significant differences were found between MCI and AD cohorts themselves. Our results suggest a neuroprotective or compensatory role for the alveus and CA1 in healthy aging and suggest that an improved understanding of the volumetric trajectories of these structures is required. |
Olfactory adaptation, characterized by attenuation of response to repeated odor stimulations or continuous odor exposure, is an intrinsic feature of olfactory processing. Adaptation can be induced by either "synaptic depression" due to depletion of neurotransmitters, or "enhanced inhibition" onto principle neurons by local inhibitory interneurons in olfactory structures. It is not clear which mechanism plays a major role in olfactory adaptation. More importantly, molecular sources of enhanced inhibition have not been identified. In this study, olfactory responses to either repeated 40-s stimulations with interstimulus intervals (ISI) of 140-s or 30-min, or a single prolonged 200-s stimulus were measured by fMRI in different naïve rats. Olfactory adaptations in the olfactory bulb (OB), anterior olfactory nucleus (AON), and piriform cortex (PC) were observed only with repeated 40-s odor stimulations, and no olfactory adaptations were detected during the prolonged 200-s stimulation. Interestingly, in responses to repeated 40-s odor stimulations in the PC, the first odor stimulation induced positive activations, and odor stimulations under adapted condition induced negative activations. The negative activations suggest that "sparse coding" and "global inhibition" are the characteristics of olfactory processing in PC, and the global inhibition manifests only under an adapted condition, not a naïve condition. Further, we found that these adaptations were NMDA receptor dependent; an NMDA receptor antagonist (MK801) blocked the adaptations. Based on the mechanism that glutamate NMDA receptor plays a role in the inhibition onto principle neurons by interneurons, our data suggest that the olfactory adaptations are caused by enhanced inhibition from interneurons. Combined with the necessity of the interruption of odor stimulation to observe the adaptations, the molecular source for the enhanced inhibition is most likely an increased glutamate release from presynaptic terminals due to glutamate over-replenishment during the interruption of odor stimulation. Furthermore, with blockage of the adaptations, the data reveal that orbital, medial & prefrontal, and cingulate cortices (OmPFC) are involved in the olfactory processing. |
In the context of neurologic disorders, dynamic susceptibility contrast (DSC) and dynamic contrast enhanced (DCE) MRI provide valuable insights into cerebral vascular function, integrity, and architecture. Even after two decades of use, these modalities continue to evolve as their biophysical and kinetic basis is better understood, with improvements in pulse sequences and accelerated imaging techniques and through application of more robust and automated data analysis strategies. Here, we systematically review each of these elements, with a focus on how their integration improves kinetic parameter accuracy and the development of new hemodynamic biomarkers that provide sub-voxel sensitivity (e.g., capillary transit time and flow heterogeneity). Regarding contrast mechanisms, we discuss the dipole-dipole interactions and susceptibility effects that give rise to simultaneous T<sub>1</sub>, T<sub>2</sub> and T<sub>2</sub><sup>∗</sup> relaxation effects, including their quantification, influence on pulse sequence parameter optimization, and use in methods such as vessel size and vessel architectural imaging. The application of technologic advancements, such as parallel imaging, simultaneous multi-slice, undersampled k-space acquisitions, and sliding window strategies, enables improved spatial and/or temporal resolution of DSC and DCE acquisitions. Such acceleration techniques have also enabled the implementation of, clinically feasible, simultaneous multi-echo spin- and gradient echo acquisitions, providing more comprehensive and quantitative interrogation of T<sub>1</sub>, T<sub>2</sub> and T<sub>2</sub><sup>∗</sup> changes. Characterizing these relaxation rate changes through different post-processing options allows for the quantification of hemodynamics and vascular permeability. The application of different biophysical models provides insight into traditional hemodynamic parameters (e.g., cerebral blood volume) and more advanced parameters (e.g., capillary transit time heterogeneity). We provide insight into the appropriate selection of biophysical models and the necessary post-processing steps to ensure reliable measurements while minimizing potential sources of error. We show representative examples of advanced DSC- and DCE-MRI methods applied to pathologic conditions affecting the cerebral microcirculation, including brain tumors, stroke, aging, and multiple sclerosis. The maturation and standardization of conventional DSC- and DCE-MRI techniques has enabled their increased integration into clinical practice and use in clinical trials, which has, in turn, spurred renewed interest in their technological and biophysical development, paving the way towards a more comprehensive assessment of cerebral hemodynamics. |
Brain extraction from 3D medical images is a common pre-processing step. A variety of approaches exist, but they are frequently only designed to perform brain extraction from images without strong pathologies. Extracting the brain from images exhibiting strong pathologies, for example, the presence of a brain tumor or of a traumatic brain injury (TBI), is challenging. In such cases, tissue appearance may substantially deviate from normal tissue appearance and hence violates algorithmic assumptions for standard approaches to brain extraction; consequently, the brain may not be correctly extracted. This paper proposes a brain extraction approach which can explicitly account for pathologies by jointly modeling normal tissue appearance and pathologies. Specifically, our model uses a three-part image decomposition: (1) normal tissue appearance is captured by principal component analysis (PCA), (2) pathologies are captured via a total variation term, and (3) the skull and surrounding tissue is captured by a sparsity term. Due to its convexity, the resulting decomposition model allows for efficient optimization. Decomposition and image registration steps are alternated to allow statistical modeling of normal tissue appearance in a fixed atlas coordinate system. As a beneficial side effect, the decomposition model allows for the identification of potentially pathological areas and the reconstruction of a quasi-normal image in atlas space. We demonstrate the effectiveness of our approach on four datasets: the publicly available IBSR and LPBA40 datasets which show normal image appearance, the BRATS dataset containing images with brain tumors, and a dataset containing clinical TBI images. We compare the performance with other popular brain extraction models: ROBEX, BEaST, MASS, BET, BSE and a recently proposed deep learning approach. Our model performs better than these competing approaches on all four datasets. Specifically, our model achieves the best median (97.11) and mean (96.88) Dice scores over all datasets. The two best performing competitors, ROBEX and MASS, achieve scores of 96.23/95.62 and 96.67/94.25 respectively. Hence, our approach is an effective method for high quality brain extraction for a wide variety of images. |
There is a growing interest in neuroscience in assessing the continuous, endogenous, and nonstationary dynamics of brain network activity supporting the fluidity of human cognition and behavior. This non-stationarity may involve ever-changing formation and dissolution of active cortical sources and brain networks. However, unsupervised approaches to identify and model these changes in brain dynamics as continuous transitions between quasi-stable brain states using unlabeled, noninvasive recordings of brain activity have been limited. This study explores the use of adaptive mixture independent component analysis (AMICA) to model multichannel electroencephalographic (EEG) data with a set of ICA models, each of which decomposes an adaptively learned portion of the data into statistically independent sources. We first show that AMICA can segment simulated quasi-stationary EEG data and accurately identify ground-truth sources and source model transitions. Next, we demonstrate that AMICA decomposition, applied to 6-13 channel scalp recordings from the CAP Sleep Database, can characterize sleep stage dynamics, allowing 75% accuracy in identifying transitions between six sleep stages without use of EEG power spectra. Finally, applied to 30-channel data from subjects in a driving simulator, AMICA identifies models that account for EEG during faster and slower response to driving challenges, respectively. We show changes in relative probabilities of these models allow effective prediction of subject response speed and moment-by-moment characterization of state changes within single trials. AMICA thus provides a generic unsupervised approach to identifying and modeling changes in EEG dynamics. Applied to continuous, unlabeled multichannel data, AMICA may likely be used to detect and study any changes in cognitive states. |
It has been well established over the last two decades that walking is not merely an automatic, motoric activity; it also utilizes executive function circuits, which play an increasingly important role in walking for older people and those with mobility and cognitive deficits. Dual-task walking, such as walking while performing a cognitive task, is a necessary skill for everyday functioning, and has been shown to activate prefrontal lobe areas in healthy older people. Another well-established point in healthy aging is the loss of grey matter, and in particular loss of frontal lobe grey matter volume. However, the relationship between increased frontal lobe activity during dual-task walking and loss of frontal grey matter in healthy aging remains unknown. In the current study, we combined oxygenated hemoglobin (HbO<sub>2</sub>) data from functional near-infrared spectroscopy (fNIRS), taken during dual-task walking, with structural MRI volumetrics in a cohort of healthy older subjects to identify this relationship. We studied fifty-five relatively healthy, older participants (≥65 years) during two separate sessions: fNIRS to measure HbO<sub>2</sub> changes between single-task (i.e., normal walking) and dual-task walking-while-talking, and high-resolution, structural MRI to measure frontal lobe grey matter volumes. Linear mixed effects modeling was utilized to determine the moderation effect of grey matter volume on the change in prefrontal oxygenated hemoglobin between the two walking tasks, while controlling for covariates including task performance. We found a highly significant interaction effect between frontal grey matter volume and task on HbO<sub>2</sub> levels (p < 0.0001). Specifically, increased HbO<sub>2</sub> levels during dual-task compared to single-task walking were associated with reduced frontal grey matter volume. Regional analysis identified bilateral superior and rostral middle gyri as the primary areas driving these results. The findings provide support for the concept of neural inefficiency: in the absence of behavioral gains, grey matter loss in relatively healthy, older individuals leads to over-activation of frontal lobe during a cognitively demanding walking task with established clinical and predictive utility. |
Individuals with impulsive and addictive disorders, including drug addiction, binge eating/obesity, and problem gambling, exhibit both impaired control over behavior and heightened sensitivity to reward. However, it is not known whether such deviation in inhibitory and reward circuitry among clinical populations is a cause or consequence of the disorders. Recent evidence suggests that these constructs may be related at the neural level, and together, increase risk for engaging in maladaptive behaviors. The current study examined the degree to which brain function during inhibition relates to brain function during receipt of reward in healthy young adults who have not yet developed problem behaviors. Participants completed the stop signal task to assess inhibitory control and the doors task to assess reactivity to monetary reward (win vs loss) during functional magnetic resonance imaging (fMRI). Brain activation during response inhibition was negatively correlated with brain activation during reward. Specifically, less brain activation in right prefrontal regions during inhibition, including the right inferior frontal gyrus, middle frontal gyrus, and supplementary motor area, was associated with greater brain activation in left ventral striatum during receipt of monetary reward. Moreover, these associations were stronger in binge drinkers compared to non-binge drinkers. These findings suggest that the systems are related even before the onset of impulsive or addictive disorders. As such, it is possible that the association between inhibitory and reward circuitry may be a prospective marker of risk. |
Older adults experience difficulties in daily situations that require flexible information selection in the presence of multiple competing sensory inputs, like for instance multi-talker situations. Modulations of rhythmic neural activity in the alpha-beta (8-30 Hz) frequency range in posterior brain areas have been established as a cross-modal neural correlate of selective attention. However, research linking compromised auditory selective attention to changes in rhythmic neural activity in aging is sparse. We tested younger (n = 25; 22-35 years) and older adults (n = 26; 63-76 years) in an attention modulated dichotic listening task. In this, two streams of highly similar auditory input were simultaneously presented to participants' both ears (i.e., dichotically) while attention had to be focused on the input to only one ear (i.e. target) and the other, distracting information had to be ignored. We here demonstrate a link between severely compromised auditory selective attention in aging and a partial reorganization of attention-related rhythmic neural responses. In particular, in old age we observed a shift from a self-initiated, preparatory modulation of lateralized alpha rhythmic activity to an externally driven response in the alpha-beta range. Critically, moment-to-moment fluctuations in the age-specific patterns of self-initiated and externally driven lateralized rhythmic activity were associated with behavioral performance. We conclude that adult age differences in spatial selective attention likely derive from a functional reorganization of rhythmic neural activity within the aging brain. |
Due to the low temporal resolution of BOLD-fMRI, imaging studies on human brain function have almost exclusively focused on instantaneous correlations within the data. Developments in hardware and acquisition protocols, however, are offering data with higher sampling rates that allow investigating the latency structure of BOLD-fMRI data. In this study we describe a method for analyzing the latency structure within BOLD-fMRI data and apply it to resting-state data of 94 participants from the Human Connectome Project. The method shows that task-positive and task-negative networks are integrated through traveling BOLD waves within early visual cortex. The waves are initiated at the periphery of the visual field and propagate towards the fovea. This observation suggests a mechanism for the functional integration of task-positive and task-negative networks, argues for an eccentricity-based view on visual information processing, and contributes to the emerging view that resting-state BOLD-fMRI fluctuations are superpositions of inherently spatiotemporal patterns. |
Understanding how the anatomy of the human brain constrains and influences the formation of large-scale functional networks remains a fundamental question in neuroscience. Here, given measured brain activity in gray matter, we interpolate these functional signals into the white matter on a structurally-informed high-resolution voxel-level brain grid. The interpolated volumes reflect the underlying anatomical information, revealing white matter structures that mediate the interaction between temporally coherent gray matter regions. Functional connectivity analyses of the interpolated volumes reveal an enriched picture of the default mode network (DMN) and its subcomponents, including the different white matter bundles that are implicated in their formation, thus extending currently known spatial patterns that are limited within the gray matter only. These subcomponents have distinct structure-function patterns, each of which are differentially observed during tasks, demonstrating plausible structural mechanisms for functional switching between task-positive and -negative components. This work opens new avenues for the integration of brain structure and function, and demonstrates the collective mediation of white matter pathways across short and long-distance functional connections. |
Placebos can reduce pain by inducing beliefs in the effectiveness of an actually inert treatment. Such top-down effects on pain typically engage lateral and medial prefrontal regions, the insula, somatosensory cortex, as well as the thalamus and brainstem during pain anticipation or perception. Considering the level of large-scale brain networks, these regions spatially align with fronto-parietal/executive control, salience, and sensory-motor networks, but it is unclear if and how placebos alter interactions between them during rest. Here, we investigated how placebo analgesia affected intrinsic network coupling. Ninety-nine human participants were randomly assigned to a placebo or control group and underwent resting-state fMRI after pain processing. Results revealed inverse coupling between two resting-state networks in placebo but not control participants. Specifically, networks comprised the bilateral somatosensory cortex and posterior insula, as well as the brainstem, thalamus, striatal regions, dorsal and rostral anterior cingulate cortex, and the anterior insula, respectively. Across participants, more negative between-network coupling was associated with lower individual pain intensity as assessed during a preceding pain task, and there was no significant relation with expectations of medication effectiveness in the placebo group. Altogether, these findings provide initial evidence that placebo analgesia affects the intrinsic communication between large-scale brain networks, even in the absence of pain. We suggest a theoretical model where placebo analgesia might affect processing within a descending pain-modulatory network, potentially segregating it from somatosensory regions that may code for painful experiences. |
Deep brain stimulation (DBS) can be a very efficient treatment option for movement disorders and psychiatric diseases. To better understand DBS mechanisms, brain activity can be recorded using magnetoencephalography (MEG) with the stimulator turned on. However, DBS produces large artefacts compromising MEG data quality due to both the applied current and the movement of wires connecting the stimulator with the electrode. To filter out these artefacts, several methods to suppress the DBS artefact have been proposed in the literature. A comparative study evaluating each method's effectiveness, however, is missing so far. In this study, we evaluate the performance of four artefact rejection methods on MEG data from phantom recordings with DBS acquired with an Elekta Neuromag and a CTF system: (i) Hampel-filter, (ii) spectral signal space projection (S3P), (iii) independent component analysis with mutual information (ICA-MI), and (iv) temporal signal space separation (tSSS). In the sensor space, the largest increase in signal-to-noise (SNR) ratio was achieved by ICA-MI, while the best correspondence in terms of source activations was obtained by tSSS. LCMV beamforming alone was not sufficient to suppress the DBS-induced artefacts. |
Tau neurofibrillary tangles, a pathophysiological hallmark of Alzheimer's disease (AD), exhibit a stereotypical spatiotemporal trajectory that is strongly correlated with disease progression and cognitive decline. Personalized prediction of tau progression is, therefore, vital for the early diagnosis and prognosis of AD. Evidence from both animal and human studies is suggestive of tau transmission along the brains preexisting neural connectivity conduits. We present here an analytic graph diffusion framework for individualized predictive modeling of tau progression along the structural connectome. To account for physiological processes that lead to active generation and clearance of tau alongside passive diffusion, our model uses an inhomogenous graph diffusion equation with a source term and provides closed-form solutions to this equation for linear and exponential source functionals. Longitudinal imaging data from two cohorts, the Harvard Aging Brain Study (HABS) and the Alzheimer's Disease Neuroimaging Initiative (ADNI), were used to validate the model. The clinical data used for developing and validating the model include regional tau measures extracted from longitudinal positron emission tomography (PET) scans based on the <sup>18</sup>F-Flortaucipir radiotracer and individual structural connectivity maps computed from diffusion tensor imaging (DTI) by means of tractography and streamline counting. Two-timepoint tau PET scans were used to assess the goodness of model fit. Three-timepoint tau PET scans were used to assess predictive accuracy via comparison of predicted and observed tau measures at the third timepoint. Our results show high consistency between predicted and observed tau and differential tau from region-based analysis. While the prognostic value of this approach needs to be validated in a larger cohort, our preliminary results suggest that our longitudinal predictive model, which offers an in vivo macroscopic perspective on tau progression in the brain, is potentially promising as a personalizable predictive framework for AD. |
The functional organization of neural processes is constrained by the brain's intrinsic structural connectivity, i.e., the connectome. Here, we explore how structural connectivity can improve the representation of brain activity signals and their dynamics. Using a multi-modal imaging dataset (electroencephalography, structural MRI, and diffusion MRI), we represent electrical brain activity at the cortical surface as a time-varying composition of harmonic modes of structural connectivity. These harmonic modes are known as connectome harmonics. Here we describe brain activity signal as a time-varying combination of connectome harmonics. We term this description as the connectome spectrum of the signal. We found that: first, the brain activity signal is represented more compactly by the connectome spectrum than by the traditional area-based representation; second, the connectome spectrum characterizes fast brain dynamics in terms of signal broadcasting profile, revealing different temporal regimes of integration and segregation that are consistent across participants. And last, the connectome spectrum characterizes fast brain dynamics with fewer degrees of freedom than area-based signal representations. Specifically, we show that a smaller number of dimensions capture the differences between low-level and high-level visual processing in the connectome spectrum. Also, we demonstrate that connectome harmonics capture more sensitively the topological properties of brain activity. In summary, this work provides statistical, functional, and topological evidence indicating that the description of brain activity in terms of structural connectivity fosters a more comprehensive understanding of large-scale dynamic neural functioning. |
The "primary" or "first-order relay" nuclei of the thalamus feed the cerebral cortex with information about ongoing activity in the environment or the subcortical motor systems. Because of the small size of these nuclei and the high specificity of their input and output pathways, new imaging protocols are required to investigate thalamocortical interactions in human perception, cognition and language. The goal of the present study was twofold: I) to develop a reconstruction protocol based on in vivo diffusion MRI to extract and measure the axonal fiber tracts that originate or terminate specifically in individual first-order relay nuclei; and, II) to test the reliability of this reconstruction protocol. In left and right hemispheres, we investigated the thalamocortical/corticothalamic axon bundles linking each of the first-order relay nuclei and their main cortical target areas, namely, the lateral geniculate nucleus (optic radiation), the medial geniculate nucleus (acoustic radiation), the ventral posterior nucleus (somatosensory radiation) and the ventral lateral nucleus (motor radiation). In addition, we examined the main subcortical input pathway to the ventral lateral posterior nucleus, which originates in the dentate nucleus of the cerebellum. Our protocol comprised three components: defining regions-of-interest; preprocessing diffusion data; and modeling white-matter tracts and tractometry. We then used computation and test-retest methods to check whether our protocol could reliably reconstruct these tracts of interest and their profiles. Our results demonstrated that the protocol had nearly perfect computational reproducibility and good-to-excellent test-retest reproducibility. This new protocol may be of interest for both basic human brain neuroscience and clinical studies and has been made publicly available to the scientific community. |
Higher brain dysfunction, such as language delay, is a major concern among preterm infants. Cerebral substrates of cognitive development in preterm infants remain elusive, partly because of limited methods. The present study focuses on hemodynamic response patterns for brain function by using near-infrared spectroscopy. Specifically, the study investigates gestational differences in the hemodynamic response pattern evoked in response to phonetic changes of speech and cerebral hemispheric specialization of the auditory area in preterm infants (<i>n</i> = 60) and term infants (<i>n</i> = 20). Eighty neonates born between 26 and 41 weeks of gestational age (GA) were tested from 33 to 41 weeks of postmenstrual age (PMA). We analyzed the hemodynamic response pattern to phonemic and prosodic contrasts for multiple channels on temporal regions and the laterality index of the auditory area. Preterm infants younger than 39 weeks of PMA showed significantly atypical hemodynamic patterns, with an inverted response shape. Partial correlation analysis of the typicality score of hemodynamic response revealed a significant positive correlation with PMA. The laterality index of preterm infants from 39 weeks of PMA demonstrated a tendency rightward dominance for prosodic changes similar to term infants. We provide new evidence that alterations in hemodynamic regulation and the functional system for phonemic and prosodic processing in preterm infants catch up by their projected due dates. |
Elucidating the relationship between neuronal metabolism and the integrity of the cholinergic system is prerequisite for a profound understanding of cholinergic dysfunction in Alzheimer's disease. The cholinergic system can be investigated specifically using positron emission tomography (PET) with [<sup>11</sup>C]N-methyl-4-piperidyl-acetate (MP4A), while neuronal metabolism is often assessed with 2-deoxy-2-[<sup>18</sup>F]fluoro-d-glucose-(FDG) PET. We hypothesised a close correlation between MP4A-perfusion and FDG-uptake, permitting inferences about metabolism from MP4A-perfusion, and investigated the patterns of neuronal hypometabolism and cholinergic impairment in non-demented AD patients. MP4A-PET was performed in 18 cognitively normal adults and 19 patients with mild cognitive impairment (MCI) and positive AD biomarkers. In nine patients with additional FDG-PET, the sum images of every combination of consecutive early MP4A-frames were correlated with FDG-scans to determine the optimal time window for assessing MP4A-perfusion. Acetylcholinesterase (AChE) activity was estimated using a 3-compartmental model. Group comparisons of MP4A-perfusion and AChE-activity were performed using the entire sample. The highest correlation between MP4A-perfusion and FDG-uptake across the cerebral cortex was observed 60-450 s after injection (r = 0.867). The patterns of hypometabolism (FDG-PET) and hypoperfusion (MP4A-PET) in MCI covered areas known to be hypometabolic early in AD, while AChE activity was mainly reduced in the lateral temporal cortex and the occipital lobe, sparing posterior midline structures. Data indicate that patterns of cholinergic impairment and neuronal hypometabolism differ significantly at the stage of MCI in AD, implying distinct underlying pathologies, and suggesting potential predictors of the response to cholinergic pharmacotherapy. |
Research on neurophysiological impairments associated with binge drinking (BD), an excessive but episodic alcohol use pattern, has significantly increased over the last decade. This work is the first to systematically review -following PRISMA guidelines- the empirical evidence regarding the effects of BD on neural activity -assessed by electroencephalography- of adolescents and young adults. A systematic review was conducted in 34 studies (N = 1723). Results indicated that binge drinkers (BDs) showed similar behavioral performance as non/low drinkers. The most solid electrophysiological finding was an augmented P3 amplitude during attention, working memory and inhibition tasks. This increased neural activity suggests the recruitment of additional resources to perform the task at adequate/successful levels, which supports the neurocompensation hypothesis. Similar to alcoholics, BDs also displayed increased reactivity to alcohol-related cues, augmented resting-state electrophysiological signal and reduced activity during error detection -which gives support to the continuum hypothesis. Evidence does not seem to support greater vulnerability to BD in females. Replication and longitudinal studies are required to account for mixed results and to elucidate the extent/direction of the neural impairments associated with BD. |
A major caveat with investigations on schizophrenic patients is the difficulty to control for medication usage across samples as disease-related neural differences may be confounded by medication usage. Following a thorough literature search (632 records identified), we included 37 studies with a total of 740 medicated schizophrenia patients and 367 unmedicated schizophrenia patients. Here, we perform several meta-analyses to assess the neurofunctional differences between medicated and unmedicated schizophrenic patients across fMRI studies to determine systematic regions associated with medication usage. Several clusters identified by the meta-analysis on the medicated group include three right lateralized frontal clusters and a left lateralized parietal cluster, whereas the unmedicated group yielded concordant activity among right lateralized frontal-parietal regions. We further explored the prevalence of activity within these regions across illness duration and task type. These findings suggest a neural compensatory mechanism across these regions both spatially and chronically, offering new insight into the spatial and temporal dynamic neural differences among medicated and unmedicated schizophrenia patients. |
Effective learning from performance feedback is vital for adaptive behavior regulation necessary for successful cognitive performance. Yet, how this learning operates in clinical groups that experience cognitive dysfunction is not well understood. Multiple sclerosis (MS) is an autoimmune, degenerative disease of the central nervous system characterized by physical and cognitive dysfunction. A highly prevalent impairment in MS is cognitive fatigue (CF). CF is associated with altered functioning within cortico-striatal regions that also facilitate feedback-based learning in neurotypical (NT) individuals. Despite this cortico-striatal overlap, research about feedback-based learning in MS, its associated neural underpinnings, and its sensitivity to CF, are all lacking. The present study investigated feedback-based learning ability in MS, as well as associated cortico-striatal function and connectivity. MS and NT participants completed a functional magnetic resonance imaging (fMRI) paired-word association task during which they received trial-by-trial monetary, non-monetary, and uninformative performance feedback. Despite reporting greater CF throughout the task, MS participants displayed comparable task performance to NTs, suggesting preserved feedback-based learning ability in the MS group. Both groups recruited the ventral striatum (VS), caudate nucleus, and ventromedial prefrontal cortex in response to the receipt of performance feedback, suggesting that people with MS also recruit cortico-striatal regions during feedback-based learning. However, compared to NT participants, MS participants also displayed stronger functional connectivity between the VS and task-relevant regions, including the left angular gyrus and right superior temporal gyrus, in response to feedback receipt. Results indicate that CF may not interfere with feedback-based learning in MS. Nonetheless, people with MS may recruit alternative connections with the striatum to assist with this form of learning. These findings have implications for cognitive rehabilitation treatments that incorporate performance feedback to remediate cognitive dysfunction in clinical populations. |
High frequency oscillations (HFOs, 80–500 Hz) in invasive EEG are a biomarker for the epileptic focus. Ripples (80–250 Hz) have also been identified in non-invasive MEG, yet detection is impeded by noise, their low occurrence rates, and the workload of visual analysis. We propose a method that identifies ripples in MEG through noise reduction, beamforming and automatic detection with minimal user effort. We analysed 15 min of presurgical resting-state interictal MEG data of 25 patients with epilepsy. The MEG signal-to-noise was improved by using a cross-validation signal space separation method, and by calculating ~ 2400 beamformer-based virtual sensors in the grey matter. Ripples in these sensors were automatically detected by an algorithm optimized for MEG. A small subset of the identified ripples was visually checked. Ripple locations were compared with MEG spike dipole locations and the resection area if available. Running the automatic detection algorithm resulted in on average 905 ripples per patient, of which on average 148 ripples were visually reviewed. Reviewing took approximately 5 min per patient, and identified ripples in 16 out of 25 patients. In 14 patients the ripple locations showed good or moderate concordance with the MEG spikes. For six out of eight patients who had surgery, the ripple locations showed concordance with the resection area: 4/5 with good outcome and 2/3 with poor outcome. Automatic ripple detection in beamformer-based virtual sensors is a feasible non-invasive tool for the identification of ripples in MEG. Our method requires minimal user effort and is easily applicable in a clinical setting.
Highlights
Cross-validation signal space separation and beamformer increase the SNR in MEG.
Automatic detection of MEG ripples in the time domain is feasible.
Our method identifies ripples with minimal user effort and is clinically applicable.
Automatically detected ripples are concordant with MEG spikes in 14/16 patients.
Automatically detected ripples are concordant with resection area in 6/8 patients.
## Introduction
All investigations in the workup for epilepsy surgery aim to identify the epileptogenic zone sensitively and specifically. The trade-off between sensitivity and specificity often involves the invasiveness of the investigation. Interictal epileptiform discharges, also called spikes, in electroencephalography (EEG), electrocorticography (ECoG) and magnetoencephalography (MEG) are often used to estimate the location of the epileptogenic zone, but spikes might not be very specific ( , ). High frequency oscillations (HFOs, 80–500 Hz) are electrophysiological transients that are used as biomarkers for the epileptogenic zone in ECoG, and show a high sensitivity and specificity ( , , ). The use of HFOs as a biomarker in non-invasive investigations is a topic of current research. Ripples (80–250 Hz) have been found in both EEG and MEG ( , , , ). A specific and sensitive non-invasive biomarker would reduce the need for invasive investigations.
MEG is a promising recording technique for ripple analysis, because of its generally higher spatial resolution than clinical EEG. Analysis of ripples in MEG is a recent development. Few MEG studies have analysed high gamma or ripples in patients with epilepsy, either by looking at the spectral content ( , , , , ), or by searching for short lasting oscillations that stand out from the baseline ( , , ).
The large number of sensors in modern whole-head MEG systems is an advantage for localization, but makes visual analysis of ripples very time consuming. Automatic detection algorithms for invasive ripples have been developed, but direct application to MEG signals is difficult due to differences in signal characteristics. A recent study ( ) used a detection algorithm to find ripples in MEG based on an increase in root mean square amplitude in 10 narrow frequency bands between 40 and 160 Hz. After rejection of possible artefacts and visual validation by two reviewers, ripples were identified in 8 out of 17 patients. This algorithm was developed to detect ripples with a high sensitivity. Another algorithm, developed by Burnos and colleagues ( , ), identifies possible ripples by using the Stockwell entropy ( ) of the signal and detects ripples based on the presence of a high frequency component with well-defined characteristics in the time-frequency spectrum. This algorithm was designed to detect ripples with a high specificity for the seizure onset zone.
The low amplitude of the ripples, combined with high amplitude background noise, result in a low signal-to-noise ratio (SNR) and mean that it can be hard to (automatically) distinguish ripples from the baseline. In a previous study we have shown that the use of beamformer virtual sensors can increase the signal-to-noise ratio, and show ripples that were not visible in the physical sensors ( ). These ripples were marked visually for 70 virtual sensors placed in a priori defined areas of interest. Covering the whole head with virtual sensors would increase the sensitivity, but at the same time would hugely increase the number of channels, rendering visual analysis impractical.
The aim of this study was to generate beamformer virtual sensors throughout the cortex to increase the chance of finding ripples, and to detect these ripples with an automatic detection algorithm with as little manual reviewing as possible. To enable automatic detection, we further increased the SNR by pre-processing the data with the extended signal space separation (xSSS) method, which combines efficient interference elimination and reduction of sensor noise (manuscript in preparation). We adapted the ripple detector algorithm developed by to work with our MEG virtual sensor data. With this detector it was possible to automatically analyse the approximately 2400 beamformer virtual sensors for the presence of ripples, showing that the approach would be applicable in a clinical setting. We compared the identified ripple locations to the clinical information of each patient in order to determine the validity of the approach.
## Methods
### Patients
Patients with refractory epilepsy in the presurgical workup for epilepsy surgery at the University Medical Centre Utrecht, who had an MEG registration in 2012 or 2013 at the VU University Medical Centre in Amsterdam, were included. Patients without epileptic spikes in the MEG, according to the clinical report, were excluded, since patients with spikes have a higher chance of showing ripples ( ). Also MEG recordings with extensive high frequency artefacts were excluded. We determined the resected brain area in patients who had undergone surgery based on post-surgical MRI (if available) or based on a description of the surgery. Patients were considered seizure free if they had an Engel score of 1 at the longest available follow up. All patients or caretakers gave written informed consent for use of their data for research.
### MEG data acquisition
MEG recordings were performed with a 306-channel whole head Elekta Neuromag® system (Elekta Oy, Helsinki, Finland) in a magnetically shielded room (VacuumSchmelze GmbH, Hanau, Germany). The system consists of 102 sensor units, each with two gradiometers and one magnetometer. Four or five head localization coils continuously recorded the position of the head in the MEG helmet. The data were recorded with a 1250 Hz sampling frequency, a low-pass anti-aliasing filter of 410 Hz and a high-pass filter of 0.1 Hz. Recordings were made with closed eyes, and in a supine position, to minimize head movement. A fifteen-minute resting-state interictal recording was used for analysis. Other recordings included a motor task and somatosensory stimulation, but these data were not used in this study. The position of the head localization coils and the shape of the scalp were digitized using a 3D digitizer (Fastrak, Polhemus, Colchester, VT, USA).
### Anatomical MRI
Each MEG recording was co-registered with a T1-weighted structural magnetic resonance image (MRI) of the patient with surface matching software developed by one of the authors (AH). This resulted in a co-registration error of approximately 4 mm ( ). A single sphere, which fitted best to the outline of the scalp, was used as volume conductor model. This model was used for the beamformer analysis described below.
We used the same T1 MRI to reconstruct virtual sensors in the grey matter. This was done by segmenting the grey matter in SPM12 in Matlab (version 8.5.0; Mathworks Inc., Natick, MA, USA), down sampling the grey matter voxels to get a minimum inter-sensor distance of 5 mm, and excluding all voxels below the nose. Cerebellar grey matter voxels were excluded, but deep structures like the hippocampus and interhemispheric grey matter were maintained. The remaining voxels were used as virtual sensor locations. The coverage of virtual sensors was visually checked for each patient. Each patient had between 2060 and 2788 virtual sensor locations (average 2421, ).
Locations of beamformer virtual sensors for 4 different axial slices in patient 22. All grey matter voxels were segmented from the MRI and down sampled to a minimum inter-sensor distance of 5 mm. Cerebellar grey matter voxels were excluded, but deep structures like the hippocampus and interhemispheric grey matter were maintained.
Fig. 1
### Data processing
We removed the signal from the head localization coils with a band-stop filter and applied the new cross-validation signal space separation (xSSS) method implemented in a research software module (Elekta MaxFilter version 3.0, not commercially available). Compared to the spatial SSS ( ) and spatiotemporal tSSS ( ), the xSSS method has two important novelties: cross-validation for extracting and suppressing uncorrelated channel artefacts and noise, as well as covariance-based regularization of the SSS reconstruction for reducing the sensor noise. Details of the xSSS pre-processing are described in .
We used a scalar beamformer similar to Synthetic Aperture Magnetometry ( ) that is implemented in a research software module (Elekta Beamformer version 2.2, not commercially available). The 80 Hz high-pass-filtered, pre-processed signal was used for data covariance, and the first 10 s of the unfiltered pre-processed signal were used to estimate noise covariance. Both magnetometer and gradiometer data were used to calculate the beamformer solution, so that the relative advantages of the two sensor-types are combined (i.e. magnetometers for deeper sources; gradiometers with higher SNR for superficial sources). Normalized beamformer weights were calculated and used to reconstruct time series for the virtual sensor locations ( , , ).
### Ripple detection
Ripples were automatically detected in all virtual sensors by an adapted version of the HFO detector developed by , . The original detector has previously been optimized for use on intracranial grid and depth electrode signals, which have a higher SNR than non-invasive MEG signals. We adapted the parameters of the detector and added extra requirements for true ripples to deal with the increased noise levels. The detector filtered all channels with an elliptic band pass filter between 70 and 253 Hz (− 3 dB points) with a stop band attenuation of 60 dB on both sides, and a band pass attenuation of 0.5 dB. The algorithm was applied on filtered individual channels and has a two-step approach: first a baseline was identified by computing the Stockwell entropy for 120 random one second epochs; samples with entropy higher than the threshold (0.85 ∗ maximum entropy) were considered as baseline. In the second step the ripples were identified. An envelope for all baseline segments was calculated with the Hilbert transform, a cumulative distribution function (CDF) of all segments was constructed, and the 98th percentile of this CDF was used as a threshold for potential ripples for that channel. When the Hilbert envelope of a channel exceeded this channel threshold for at least 20 ms, a potential ripple was found. A true ripple was defined when for a potential ripple a) the Stockwell entropy during the event was stable; the maximum entropy was smaller than 125% of the minimum entropy, excluding the first and last sample, b) the absolute amplitude was higher than the absolute amplitude plus one standard deviation of 1000 samples before and 1000 samples after the potential ripple, and c) a distinct component was present in the time-frequency spectrum between 40 and 250 Hz, detected by a peak above 40 Hz preceded by a trough in the power spectral density (PSD, ).
Schematic overview of automatic ripple detection algorithm, with examples of a true ripple (left) and an event that is not in the final output of the detector (right), because the entropy is not stable over the length of the event. A) Unfiltered virtual sensor signal. B) 80 Hz high-pass filtered signal, showing the true ripple (left) and false detection (right). C) Stockwell entropy over the length of the event is stable for the true ripple (left), and irregular for the false detection (right). D) Time-frequency decomposition shows a high frequency component for the true ripple (~ 100 Hz) and the spike that can be seen in part A (12–20 Hz, left), and less distinct components and high frequency artefacts for the false detection (right). E) The power spectral density (PSD) also shows the high frequency component in the true ripple (60–100 Hz, left), and the irregular high frequency activity for the false detection (right).
Fig. 2
As automatic ripple detectors have the tendency to include a large number of false positives, we checked the performance of the detector in each patient by visually reviewing a selective set of detected (‘true’) ripples. All moments in time that at least one ripple was detected (ripple-times) were extracted, and the review set was comprised by a maximum of three randomly chosen virtual sensors with ripples at each ripple-time. The reviewer was presented a 10 s trace of the unfiltered virtual sensor at the time of the ripple, a 1 s trace of the unfiltered virtual sensor, and a 1 s trace of 80 Hz high pass filtered virtual sensor, with the marked event in all traces, in a custom-made graphical user interface. The reviewer determined if the automatically detected ripple was true or not. If more than half of the reviewed ripples at a ripple-time were considered true, all ripples at that ripple-time in all channels were considered true, also the ripples that were not included in the review set. As the review set consisted of maximum three ripples at a certain ripple-time, all ripples at that ripple-time were considered true if > 66% of the ripples in the review set were considered true ( ). This strategy minimized the number of ripples to be reviewed, while all ripple-times were evaluated. Potentially true ripples at the same time as artefact detections at other channels could be excluded with this approach. The reviewer was blinded for the clinical information and for the location of the channel that was reviewed. The location of the ripples in the analysis is the location of the virtual sensors in which ripples were detected. We did not systematically review the raw MEG data at the same time-points, because in an earlier study we found that at 78% of the ripple-times, the raw MEG only showed noise ( ). We did review the unfiltered virtual sensor data to decrease the chance of marking artefacts. shows examples of events that were considered true ripples, and events that were considered artefacts, together with the physical sensor channels.
Schematic overview of the review process. The automatic detector has detected ripples in all ∼ 2400 virtual sensor channels. All moments in time where at least one ripple was detected (ripple-times) were extracted, and a review set was comprised by a maximum of three randomly chosen virtual sensors with ripples at each ripple-time. The reviewer was presented a 10 s trace of the unfiltered virtual sensor at the time of the ripple, a 1 s trace of the unfiltered virtual sensor and a 1 s trace of the 80 Hz high pass filtered virtual sensor, with the marked event in all traces. The reviewer determined if the automatically detected ripple was true or not. If > 2/3 of the reviewed ripples at a ripple-time were considered true, all ripples at that ripple-time in all channels were considered true, also the ripples that were not included in the review set.
Fig. 3
Examples of ripples that were approved (A + B) and not approved (C + D) during the visual check. On the left side we show the physical sensors after xSSS preprocessing closest to the virtual sensors that are shown on the right. The left part of each sensor set shows unfiltered data. The grey area is 80 Hz high pass filtered and shown on the right. Vertical lines indicate the same moment in time. In part A and B the true ripples are underlined, and a time frequency spectrum of each signal is shown below. Some sign of the ripple can be found in the physical channels, but only the virtual channels show a clear ripple. In part C and D the falsely marked ripples by the detector are underlined. These were discarded by the reviewer and not used for further analysis.
Fig. 4
### Spike dipole analysis
The primary, non-propagated, and therefore clinically most important epileptic spikes in the physical sensor channels were marked and evaluated for a clinical report by a team of clinicians, MEG/EEG technicians and physicists. These primary spikes were localized with a dipole fit at every sample from half-way of the flank preceding the top, to the top of the spike, with a single moving equivalent current dipole (using the Elekta Source Modelling software version 5.5). The locations of the fitted dipoles were used to compare with the locations of the ripples.
### Analysis
The results of the ripples after automatic detection and review were visualized on axial slices of the patient's MRI, and in a 3D figure. The concordance between the area(s) with ripples and the area(s) with spikes in the MEG was assessed visually and was classified as good (+) if all ripples were located in the same lobe as the spike dipoles, moderate (=) if any ripple was located in the same lobe as the spike dipoles, and bad (−) for discordance. A similar classification strategy was used to assess the concordance between the area with ripples and the resected brain area for those patients who had undergone surgery. Concordance was good (+) if > 50% of the ripple locations were included in the resection, at lobar level, moderate (=) if < 50% ripple locations were included in the resection, and bad (−) for discordance. We classified the concordance between the MEG spike dipole locations and the resected brain area by using the same criteria as for ripples.
Twelve patients (14–25) had already been included in a previous study in which we visually marked ripples in a predefined area of interest using the same MEG recordings ( ). Here, we were therefore able to compare the number of automatically identified ripple-times to the number of visually marked ripple-times in these patients.
Statistical analyses were performed using IBM SPSS Statistics 23 (IBM Corp., Armonk, NY, USA); a p -value < 0.05 was considered significant.
## Results
### Patients
Fifty-eight patients had an MEG recording in 2012 or 2013, of whom 32 did not show epileptic spikes in the clinical analysis. The MEG of one patient showed such artefacts that the patient had to be excluded from the analysis. The other 25 patients were included: they had a mean age of 12 years (range: 4–29) and 19 were male. Fifteen patients had undergone epilepsy surgery, for which the resection area was determined based on all available presurgical investigations, including MEG spikes ( ). Ten patients were seizure free after surgery (Engel 1 outcome). The average follow up time for all patients was 2.2 years (range: 0.5–4 years).
Patient characteristics, showing the location of MEG spikes, interictal EEG abnormalities, ictal EEG onset, PET abnormalities, SPECT abnormalities, pathology and/or MRI findings, and surgery with Engel outcome and duration of follow-up.
Table 1
### MEG pre-processing
The previous study ( ) utilized the standard SSS methods for suppressing magnetic interference ( , ). The cross-validation SSS method in the present study required more computing steps (see for details). Altogether, the xSSS pre-processing time of a 15-min long recording was about 20 min on a 16GB RAM four-core Linux workstation (HP Z600). Creating the approximately 2400 beamformer virtual sensors took about 3 h on the same workstation.
### Ripple detection
The ripple detection algorithm processed batches of 100 virtual sensors with 15 min of signal in 45 min on an 8GB RAM, 2.6 GHz CPU laptop. Detecting ripples in all 2400 channels per patient took about 18 h. It identified ripples in all patients before visual review, on average 905 ripples per patient (range: 79–3924). The review set consisted of 11 to 546 ripples per patient (average 148), and it took approximately 5 min per patient to review this set. The number of ripples excluded after visual review varied from 67 to 2950 per patient (average 737). The ripple detection algorithm thus had a false positive rate of 81.5%. This high false positive rate was accepted to ensure a good sensitivity. The majority of false positive detections were movement artefacts or EMG-like activity.
### Ripple rates
After reviewing, 16 of the 25 patients (64%) showed ripples. In these 16 patients, on average 18 ripple-times were identified, which were on average 261 ripples per patient, with an average rate of 1.31 per minute ( ). Ripples were found on 165 virtual sensors on average, and this number was not correlated to the total number of virtual sensors in a patient (Spearman's rho(23) = 0.36, p = 0.08).
MEG results: number of virtual sensors (VS) in each patient, duration of the recording, location of the identified ripples, concordance between ripples and MEG spikes, concordance between ripples and the resection area, the number of moments that a ripple was found in at least one channel (ripple-times), the total number of VS that showed ripples, the ripple-times per minute, and the concordance between MEG spikes and the resection area. Concordance is classified as good (+), moderate (=) or bad (−).
Table 2
### Ripple locations
Visual analysis showed good concordance of the location of the ripples at the lobar level with the location of the MEG spikes in 10/16 patients with ripples. Four patients showed moderate concordance, because some ripple locations were outside of spike locations. Of these four patients, the main focus of ripples in two patients (1 and 3) was also a spike location. Bad concordance was seen in two patients (14 and 17), both with only few ripple-times (3 and 1) and few channels with ripples (4 channels both, ). Examples for individual patients are shown in .
Ripple results for three patients. Ripples are visualized in a 3D figure (top), as well as axial MRI slices (bottom right). The ripple locations are compared to the spike dipoles from the clinical report (bottom left). All three patients show good concordance with MEG spike dipoles, as all ripple locations are also spike locations at lobar level. For patient 6, ripples were found unilaterally right centro-temporal, and spike dipoles were fitted bilateral centro-temporal. This was classified as good concordance, as the ripple location was also a spike location. Patient 6 did not undergo surgery because the number of seizures was too low. Patient 13 underwent surgery where a cortical tuber right frontal and a tuber right temporal were removed, but the seizure frequency did not change (Engel 4B). Patient 15 underwent a right temporo-lobectomy with amygdalohippocampectomy and was seizure free (Engel 1A). Postoperative MRI was not available.
Fig. 5
Eight patients with ripples underwent surgery, of whom five were seizure free after resection (Engel score 1). Patient 4 and 15 were seizure free, and the MEG ripples showed good concordance with the resection site. The other three patients who were seizure free showed a moderate (patient 16 and 22) or bad (patient 24) concordance between MEG ripples and the resection site. In all three a temporo-lobectomy with amygdalohippocampectomy was part of the surgery. The three patients with ripples who did not become seizure free showed good (patient 9), moderate (patient 13) and bad (patient 11) concordance with the resection site. Patient 9 had an incomplete resection of the lesion. The MEG spikes in patients 11 and 13 were multifocal, and did not perform better than ripples in identification of the resection site.
We also determined the concordance between the MEG spike dipole locations and the resection area. For the eight patients with ripples who underwent surgery the spike and the ripple concordance were the same in six patients, and the spikes performed better than the ripples in the other two patients. For all ten patients who underwent surgery with good outcome, the spikes showed good concordance with the resection site in six patients, moderate concordance in 2 patients and bad concordance in 2 patients ( ).
### Comparison with visual analysis
The number of ripple-times identified by automatic and visual analysis were comparable and not significantly different (Wilcoxon Signed Rank Test, Z = − 0.28, p = 0.78, ). Only for patient 21 the difference was striking, as we found 109 ripple-times automatically, and only 19 by visual marking. This is probably due to the limited spatial sampling of the visually marked sensors.
Number of automatically identified ripple-times compared to the number of visually marked ripples in ( ). The numbers are comparable and not significantly different ( p = 0.78).
Fig. 6
Ripples were marked visually in 8/12 patients; in 6 of whom ripples were also found automatically. The two patients in whom visually marked ripples were not detected automatically had only 1 and 2 visual ripple-times. Two patients in whom we did not find ripples visually, showed ripples after automatic detection.
## Discussion
We show the feasibility of automatic detection and visualization of ripples in clinical MEG recordings. We used cross-validation SSS pre-processing and beamformer virtual sensors to increase the SNR and therefore were able to find ripples in 16 of the 25 patients in this study (64%). We validated these ripples by comparison with MEG spike dipole findings, which showed good or moderate concordance in 14 of the 16 patients with ripples. For six out of eight patients who had surgery, the ripple locations showed good or moderate concordance with the resection area: 4/5 with good outcome and 2/3 with poor outcome. Performing this analysis required only minimal review of the detected ripples, allowing for application in clinical practice.
The large amount of data of MEG routinely acquired in pre-surgical assessments requires a good data analysis strategy. The approach has to be accurate, as well as fast and easy to use for non-specialists, to be useful in clinical practice. Automatic detection algorithms for ripples usually have a high false positive rate, to ensure all true ripples are caught ( ). This is especially crucial in MEG, where the ripple-rates are very low compared to intracranial recordings ( , ). Visual review of the automatically detected events is usually the solution, but even this is a cumbersome job when > 300 channels with 80 Hz high pass filtered signal have to be reviewed. Our proposed algorithm takes time to run – approximately 3 h to create 2400 virtual sensor signals and 18 h to run the ripple detector on all these channels – but these steps are unsupervised. Determining the virtual sensor locations can also be automated. By creating a smart subset of detected ripples to review visually, the time a reviewer needs to spend on ripple analysis in one patient is reduced to 5 min to check the subset of detected ripples and exclude the false detections. The complete procedure, from raw MEG data to detected ripples, took approximately 21.5 h per patient, in which maximum half an hour of human work is involved, to initially check the quality of the recording and to check the subset of detected ripples.
The fact that ripples can be found in non-invasive MEG and EEG was long considered impossible, because the generators would be too small ( ). The number of studies disproving this statement is growing, especially in EEG. The high density of MEG sensors and the ease to create a forward model for MEG would suggest that MEG is more suitable for HFO analysis than clinical EEG. However the magnitude of the background noise in MEG, and the interference induced by electrical power lines, vehicles, or heart beats, for example, might deem this untrue ( ). Passive or active shielding, smart geometry of gradiometers and magnetometers, synthetic higher order gradiometers ( ), signal space separation ( , ), and beamforming ( , ) can be used to improve the SNR. In a previous study we have shown that it is possible to identify epileptic ripples in the time domain in MEG data that was pre-processed ( ). In that study we only sampled a small area of interest, and calculated beamformer virtual sensors based on spike markings. In this study we used the whole 80 Hz filtered 15 min signal as data covariance, thus minimizing the effect of a small covariance matrix on the quality of reconstructed sources and power estimation ( ). We further improved the SNR by using the cross-validation signal space separation (xSSS; manuscript in preparation) that reduced both magnetic interference and sensor noise. The resulting signals were of such quality that automatic detection of ripples was possible.
The rate of true ripples was 1.3/minute in patients with ripples, which is low, but comparable to visually marked ripples ( ). One other study that automatically detected ripples in the time domain found ripples in 8 out of 17 patients (47%), without using beamformer virtual sensors, and found similar ripple rates ( ).
Analysis of ripples can be difficult, as filtering of sharp transients can result in ripple-like oscillations ( ). Filter artefacts of sharp transients show activity over all frequencies, low to high. Our automatic detector discards such high frequency activity that is part of broadband activity, because the power spectral density will not show a distinct high frequency peak ( E). These ripples are considered artefacts. However, we often see ripples at the same time as epileptogenic spikes, which are not connected in the frequency spectrum, and therefore no filter artefacts. These were events that we considered as true ripples. Ripples in invasive recordings can also be a physiological phenomenon, and distinction between physiological and pathological ripples is difficult when no tasks are performed. Physiological ripples have not (yet) been found in spontaneous non-invasive recordings. As most ripples in our study seemed to relate to the epileptic focus, we assumed they were pathological.
With our proposed method, all moments a ripple was present in at least one channel (ripple-times) were considered. This resulted in a large area that seems involved in ripple generation. This is in contrast with the idea that ripples in intracranial ECoG are thought to be generated by only a small brain area ( ). Reasons for the relatively widespread ripples in MEG can be found in the spatial smoothness of the measured magnetic fields ( ), and the blurring effect, or leakage, of the spatial filtering beamformer algorithm ( , ). The diameter of the ripple cluster seemed also larger than the spike clouds, but dipole clouds were the result of analysis of several selected spikes, fitted over multiple latencies, and presented here without confidence volumes, while ripples were detected on each virtual channel independently. For these reasons, determining the actual size of the ripple generating area and comparison with spikes is difficult. To be able to draw conclusions on the ripple generation area, we would at least need to reconstruct the sources of spikes and ripples in a similar fashion.
The results for ripples as marker for the resection area were not better than those for spikes, and spikes were a good marker. Ripples should not replace spikes as a biomarker, but they can be an addition to the spike information, to strengthen the result of the MEG. Of course the added value of ripples would be larger if we can also find them in patients without spikes in MEG.
We used these spikes as gold standard to determine the reliability of the detected ripples, which is not the best gold standard, but one that was available for all patients. The best gold standard for identification of the epileptogenic zone is seizure freedom after resective surgery. As MEG in our centre is used mainly for patients without a clear hypothesis about the epileptogenic zone, i.e. the most difficult cases, unfortunately only eight patients with ripples were considered eligible for epilepsy surgery, of which 5 were successful. The resection area was concordant with the MEG ripples in two of them. Interestingly, the other three patients had a temporo-lobectomy with amygdalohippocampectomy, which suggests that detection of deep mesiotemporal sources is difficult. The insensitivity of MEG to sources in the mesiotemporal lobe has been stated before ( ), also for ripples ( ). However, even for two of these difficult cases we still found moderate concordance, suggesting that the improved SNR offered by beamforming (and xSSS in this study) may come to our aid, as shown previously for interictal spikes ( ). In the patients with poor outcome after surgery, ripples showed concordance with the resection area in two of the three patients, which can indicate that the resection was incomplete.
We included patients with spikes in the unfiltered MEG. The presence of spikes was not required to perform the analysis, but it increased the chance of finding ripples ( ). We found ripples in 64% of the patients with spikes in the MEG, which is in line with 61–88% of focal epilepsy patients with ripples that are reported in scalp EEG ( , , ). Our results also suggest that the chance of good localization is higher when the number of identified ripples is higher. Therefore the performance of the method might improve when longer epochs are analysed.
We used a method to facilitate ripple detection in the time domain, where the virtual electrode time series were constructed on the basis of the whole recording. The localization of the detected ripples could be improved by applying source localization in the ripple band at the ripple-times, combining information from all channels at all ripple-times, and thereby increases the SNR in the spatial and temporal domain. This would also give more insight in the true size of the ripple generating area. Methods such as beamforming ( ) or the wavelet maximum entropy of the mean approach (wMEM, ( )) could be used for such a next step to improve localization accuracy of the automatically identified ripples.
## Conclusion
We generated beamformer virtual sensors throughout the brain to increase the chance of finding ripples, and detected these ripples with an automatic detection algorithm with minimum human intervention. We have shown that this approach is feasible and that the identified ripples correlated with the MEG spike dipoles and with the resected area in the subset of patients who were successfully operated. Further validation of the MEG ripples as a biomarker for the epileptogenic zone has to be performed in a larger cohort of patients who underwent surgery. This automatic analysis method paves the way for such studies.
## Funding
N. van Klink is supported by the Dutch Brain Foundation fund (number 2013-139) and the Dutch Epilepsy Foundation fund 15-09. S. Burnos is supported by Vontobel Stiftung, EMDO Stiftung, Herzog-Egli Stiftung and Swiss National Science Foundation (SNSF 320030_156029). S. Taulu is supported by a grant from the Washington State Life Sciences Discovery Fund (LSDF). The Wellcome Trust Laboratory for MEG Studies at the Aston Brain Centre, Aston University, UK, is supported by the Wellcome Trust and the Dr Hadwen Trust for Humane Research. M. Zijlmans is supported by the Rudolf Magnus Institute Talent Fellowship 2012 and ZonMW Veni 91615149. J. Nenonen, L. Helle and S. Taulu are or were employed by Elekta Oy. There are no further potential conflicts of interest to be disclosed.
|
## Objective
We combined diffusion MRI (dMRI) with quantitative T1 (qT1) relaxometry in a sample of school-aged children born preterm and full term to determine whether reduced fractional anisotropy (FA) within the corpus callosum of the preterm group could be explained by a reduction in myelin content, as indexed by R1 (1/T1) from qT1 scans.
## Methods
8-year-old children born preterm ( n = 29; GA 22–32 weeks) and full term ( n = 24) underwent dMRI and qT1 scans. Four subdivisions of the corpus callosum were segmented in individual native space according to cortical projection zones (occipital, temporal, motor and anterior-frontal). Fractional anisotropy (FA) and R1 were quantified along the tract trajectory of each subdivision and compared across two birth groups.
## Results
Compared to controls, preterm children demonstrated significantly decreased FA in 3 of 4 analyzed corpus callosum subdivisions (temporal, motor, and anterior frontal segments) and decreased R1 in only 2 of 4 corpus callosum subdivisions (temporal and motor segments). FA and RD were significantly associated with R1 within temporal but not anterior frontal subdivisions of the corpus callosum in the term group; RD correlated with R1 in the anterior subdivision in the preterm group only.
## Conclusions
Myelin content, as indexed by R1, drives some but not all of the differences in white matter between preterm and term born children. Other factors, such as axonal diameter and directional coherence, likely contributed to FA differences in the anterior frontal segment of the corpus callosum that were not well explained by R1.
Highlights
Reduced FA in 3 corpus callosum segments may relate to myelin or axonal factors.
Reduced R1 in PT children in 2 corpus callosum segments suggests reduced myelin.
Reduced R1 explains reduced FA in the posterior but not anterior corpus callosum.
Changes in myelin content explain some but not all white matter differences in PT.
## Introduction
White matter is a dynamic neural structure that changes during development and learning, and is associated with sensorimotor and cognitive functions ( ; ; ). White matter disturbances are linked to neurological and psychological disorders in adults ( ). Disturbances in the typical maturation of white matter have also been reported in several clinical pediatric populations, such as children born preterm. Diffuse white matter injuries observed in the aftermath of preterm birth (PT) have been attributed to the susceptibility of pre-oligodendrocytes, precursors of myelin forming glia, to destruction from hypoxia-ischemia and neuroinflammation during the perinatal period ( ; ) and are associated with poor neurodevelopmental outcomes ( ; ).
Abnormalities in white matter structure can be assessed using diffusion MRI (dMRI). In the newborn period, reduced FA and/or increased mean diffusivity in preterm compared to full term infants (FT) is consistent with the putative pathogenesis and resulting reductions in myelin content ( ; ). In older preterm children, FA differences have also been observed ( ; ; ; ; ). Recent meta-analyses of dMRI in preterm samples found that the genu, body and splenium of the corpus callosum consistently demonstrate significantly reduced FA in preterm as compared to full term children ( ). Pathways of the corpus callosum traverse periventricular white matter regions that show vulnerability to white matter injury in children born preterm ( ; ; ). However, studies have yet to establish whether differences seen in the corpus callosum reflect reductions in myelin content or other downstream effects on additional tissue properties that are indexed by dMRI metrics (e.g., crossing fibers, axonal diameter and density) and that may change over development.
In the present study, we combined dMRI with qT1 relaxometry in school-aged children born preterm or term to determine whether reduced FA in corpus callosum subdivisions of the preterm group reflected reduced myelin content. R1, derived from qT1 scans (R1 = 1/T1), is a measure of the longitudinal relaxation rate of water protons in a magnetic field. Rates of R1 (1/s) are most affected by the water content of an image voxel; voxels containing mostly water (e.g., cerebral spinal fluid) exhibit slower (~0.25/s) R1 rates than voxels that are primarily comprised of tissue (e.g., white matter), which exhibit higher (~1.2/s) rates. Rates of R1 are also sensitive to the biophysical composition of tissues within an image voxel, particularly to the tissue composition of myelin and of iron ( ; ; ). However, within white matter voxels specifically, up to 90% of the R1 signal is explained by variations in myelin content ( ). For these reasons, R1 is generally considered to be a useful proxy for tissue myeloarchitecture, particularly in white matter ( ).
Recent studies involving clinical populations have begun to use dMRI combined with R1 mapping techniques to dissociate the tissue properties underlying white matter changes observed in neurological disorders, such as multiple sclerosis ( ), and psychiatric illnesses, such as anorexia nervosa ( ). To our knowledge, the present study is the first to combine qT1 relaxometry with dMRI to interrogate the tissue properties underlying white matter changes observed in the aftermath of preterm birth. Based on the evidence described above, we used these two methods to test our specific hypotheses which were that compared to term children, (1) the current preterm group would replicate evidence for decreased FA in the posterior, body and anterior subdivisions of the corpus callosum as found in many previous studies and summarized in meta-analyses ( ), and (2) that these differences would be explained by decreased R1.
## Methods
### Participants
Participants were 8 year old preterm ( n = 43) and full term ( n = 37) children, recruited between 2012 and 2015 as part of a longitudinal study. The qT1 imaging sequence used here became available during the study and was added to the protocol ( ). To be included in the present study, participants were required to have useable dMRI and qT1 data and have complete general intelligence measures. Of all the preterm participants in our sample, 14 were excluded from the present analyses because the children underwent a different MRI protocol ( n = 2), did not undergo the qT1 imaging protocol ( n = 9), or moved too much during scanning ( n = 3). Of all term participants in our sample, 13 were not included in the present analyses because they did not undergo the qT1 imaging protocol ( n = 4), or moved too much during scanning ( n = 9). The final sample thus consisted of 29 preterm (17 males, mean age = 8 years 2.6 months) and 24 full term (13 males, mean age = 8 years 1.7 months) participants. Preterm birth was defined as gestational age (GA) ≤ 32 weeks, the gestational age at which neonates are at greatest risk for white matter injury ( ). Full term birth was defined as GA ≥ 36 weeks or birth weight ≥ 2500 g. Preterm children were recruited from the High-Risk Infant Follow-Up Clinic at Lucile Packard Children's Hospital Stanford, local parent groups, and surrounding communities in the San Francisco Bay Area. Full term children were recruited through online parent groups, postings in local school newsletters and letters to families who had participated in past research studies in affiliated research laboratories at Stanford University. Exclusion criteria for all participants included neurological factors unrelated to preterm birth that would account for white matter differences amongst participants, including congenital anomalies, active seizure disorder, hydrocephalus or sensorineural hearing loss. Diagnosis of cerebral palsy (CP) was not an exclusion criterion since CP is associated with preterm birth. One preterm participant had mild CP. The experimental protocol was approved by the Stanford University Institutional Review Board #IRB-22233. A parent or legal guardian provided informed written consent and participants were compensated for participation.
Demographic characteristics assessed in this sample included socio-economic status (SES), as measured using a modified 4-Factor Hollingshead Index ( ). General intelligence was also assessed in all participants using the Wechsler Abbreviated Scale of Intelligence (WASI-II), a nationally standardized test of general intellectual abilities ( ).
Detailed medical information was available for 26 of the 29 preterm participants. Medical complications at birth in the preterm group were: 21 had respiratory distress syndrome; 4 developed bronchopulmonary dysplasia or chronic lung disease; 20 had hyperbilirubinemia; 11 had patent ductus arteriosus; 15 had retinopathy of prematurity or immature retinae; 4 developed necrotizing enterocolitis; 2 were small for GA (≤ 3rd percentile birth weight for GA). In terms of neuroimaging findings during the initial hospitalization, 15 had one or more mildly abnormal findings on head ultrasound or MRI (5 with grade I intraventricular hemorrhage (IVH); 1 with grade II IVH; 1 with small periventricular lesions; 3 with mild white matter injury; 1 with a transient vascular malformation; 3 had enlarged ventricles) and 1 had abnormal findings on head ultrasound or MRI (bilateral grade III IVH). To follow up on the early abnormalities seen in preterm participants on head ultrasound or MRI, a neuroradiologist assessed T1-weighted MRI scans collected as part of the current longitudinal study for 5 features associated with white matter injury ( ; for description ( ; ). Of the 29 preterm subjects, 4 had abnormal T1-weighted scans, including the participant with mild CP.
### MRI acquisition
MRI data were acquired on a 3T Discovery MR750 scanner (General Electric Healthcare, Milwaukee, WI, USA) equipped with a 32-channel head coil (Nova Medical, Wilmington, MA, USA) at the Center for Cognitive and Neurobiological Imaging at Stanford University ( ). All subjects were scanned for research purposes without the use of sedation.
High-resolution T1-weighted (T1w) anatomical images were collected for each subject using an inversion recovery (IR)-prep 3D fast-spoiled gradient (FSPGR) sequence collected in the sagittal plane (0.9 × 0.9 × 0. 9 mm voxel size). The T1-weighted imaged was used as a common anatomical reference for the alignment of the diffusion tensor image (DTI) maps and qT1 maps.
dMRI data were acquired with a diffusion-weighted, dual-spin echo, echo-planar imaging sequence with full brain coverage. Diffusion weighted gradients were applied at 30 non-collinear directions with a b-value of 1000 s/mm . In addition, three volumes were acquired at b = 0 at the beginning of the scan. We collected 70 axial slices in each participant (TR = 8300 ms; TE = 83.1 ms; FOV = 220 mm; matrix size of 256 × 256, voxel size of 0.8549 × 0.8549 × 2 mm ).
qT1 relaxometry was acquired with SPGR echo images acquired at a four different flip angles (α = 4°, 10°, 20°, and 30°; TR = 14 ms; TE = 2 ms) ( ). A voxel size of 0.9375 × 0.9375 × 1.5 mm was chosen in order to acquire the four SPGR scans in a reasonably short time frame, appropriate for a pediatric clinical sample. To ensure that the scan parameters were similar between flip angles, SPGR scans were acquired sequentially without breaks or pre-scan normalization. To obtain an unbiased T1 map, we acquired a spin-echo inversion-recovery (SEIR) sequence with echo-planar imaging (EPI) read-out ( ) measured at lower resolution. Four inversion times were measured (2400, 1200, 400 and 50 ms; TR = 3 s; TE = 47 ms) with a voxel size of 1.875 × 1.875 × 4 mm . Unbiased T1 maps obtained from SEIR-EPI scans were used to correct for RF transmit bias in the higher-resolution SPGR scans as described and validated by ( ; ).
### Image preprocessing
The T1w images were first aligned to the canonical ac-pc orientation.
Diffusion weighted images were pre-processed with Vistasoft ( ), an open-source software package implemented in MATLAB R2012a (Mathworks, Natick, MA). The dual-spin echo dMRI sequence used here greatly reduces eddy-current distortions ( ) obviating the need for eddy-current correction. Subject motion during the diffusion weighted scan was corrected using a rigid body alignment algorithm ( ). Each diffusion weighted image was registered to the mean of the b = 0 images and the mean b = 0 image was registered automatically to the participant's T1w image, using a rigid body transformation (implemented in SPM8, no warping was applied). The combined transform that resulted from motion correction and alignment to the T1w anatomy was applied to the raw data once, and the transformed dMRI images were resampled to 2 × 2 × 2mm isotropic voxels. This step was performed because non-isotropic voxels may bias the tensor fit and distort both tracking and measurements of diffusion properties ( ). Diffusion gradient directions were then adjusted to fit the resampled diffusion data ( ).
For each voxel in the aligned and resampled volume, tensors were fit to the diffusion measurements using a robust least-squares algorithm, Robust Estimation of Tensors by Outlier Rejection (RESTORE), which is designed to remove outliers at the tensor estimation step ( ). A continuous tensor field was estimated using trilinear interpolation of the tensor elements. The eigenvalue decomposition of the diffusion tensor was calculated and the resulting three eigenvalues (λ1, λ2, λ3) were used to compute fractional anisotropy (FA), radial diffusivity (RD, i.e., the mean of λ2 and λ3) and axial diffusivity (AD, i.e., λ1) ( ). In the analyses presented here, we focused on FA because it was the metric used in the meta-analysis ( ).
Quantitative T1 maps were calculated using mrQ , ( ), an open-source software package implemented in MATLAB R2012a (Mathworks, Natick, MA). T1 fitting and bias correction was calculated using methods described in ( ; ). Briefly, local transmit-coil inhomogeneities are calculated by minimizing the difference between the unbiased T1 map calculated from low-resolution SEIR- epi images ( ) and the T1 map fit from the multiple flip angle SPGR images. Minimization was achieved using a nonlinear least-squares (NLS) solution that assumes transmit-coil inhomogeneities to be smooth in space ( ). These estimates were then interpolated the to the high-resolution SPGR to generate T1 (1/T1 = R1) maps.
To co-register a subject's quantitative T1 images to their dMRI data, we used the Advanced Normalization Tools (ANTs) software package ( ). This tool was used to warp the qT1 images to the non-diffusion weighted b = 0 images, as these images have relatively similar contrast ( ). This warping procedure is used to minimize mis-registration errors due to EPI distortions in the dMRI data. EPI distortions were minimal due to the 2× ASSET acceleration used for the readout of the diffusion-weighted images. After applying the diffeomorphic warp, image registration was manually inspected using the mrView software ( ). Manual inspection of the aligned images confirmed that the registration was accurate in all subjects. From the qT1 maps, we then derived R1, the inverse of T1 (R1 = 1/T1). We chose to analyze R1 for ease of interpretation: higher values of R1 would indicate increases in myelin content just as higher values of FA would be associated with increased myelin content. The validity of the current method has been demonstrated in clinical populations through findings of reduced R1 in patients with multiple sclerosis ( ) and adolescent girls with anorexia ( ), conditions in which myelin loss is expected.
### Quantification of white matter tissue properties
Automated Fiber Quantification (AFQ; ), a software package implemented in MATLAB R2012a (Mathworks, Natick, MA), was used to isolate and characterize the corpus callosum in individual native space ( ; ). In meta-analyses, the genu, body and splenium of the corpus callosum consistently demonstrate significantly decreased FA in preterm compared to full term children ( ). Further, the occipital and temporal segments of the corpus callosum traverse periventricular white matter regions that show vulnerability to injury in preterm children ( ; ; ). Therefore, we segmented four subdivisions of the corpus callosum based on the cortical projection zone of the fibers (occipital, temporal, motor, and anterior frontal) and quantified tract profiles of white matter tissue properties along the trajectory of each subdivision. The temporal (CC-temporal) and occipital (CC-occipital) subdivisions of the corpus callosum pass through the splenium and posterior periventricular white matter; the motor subdivision (CC-motor) passes through the body of the corpus callosum; and the anterior frontal (CC-ant-frontal) subdivision passes through the genu of the corpus callosum.
Methods for deterministic tractography, fiber tract identification, segmentation and quantification are described in detail in several previous publications ( ; ; ; ; ; ). shows the tracts in a representative preterm participant, including the defining ROIs used to segment each tract from the whole brain tractogram. Tract profiles of tissue parameters (FA and R1 = 1/T1) were calculated at 30 equidistant locations (nodes) along the central portion of each fiber tract bounded by the same two ROIs used for tract segmentation. This procedure generated an FA or R1 tract profile that described the variations in either FA or R1 along the central portion of the tract. Tract profiles were also averaged to produce a single mean value for each tract.
Tractography of 4 subdivisions of the corpus callosum. (a) Illustrates the following 4 subdivisions of the corpus callosum (CC), in the Splenium: CC-occipital (red), CC-temporal (yellow); Body: CC-motor (green), and Genu: CC-anterior frontal (blue). Tract renderings are displayed on a mid-sagittal and axial T1-weighted (T1w) image from a representative preterm child. Dashed lines represent the location of the regions of interest (ROIs) used to segment each callosal subdivision from the whole-brain fiber group. Colored arrows in panels b and c correspond to tract renderings presented in (a) to indicate the location of each tract within the on a mid-sagittal cross-section of the corpus callosum. Panel (b) displays the location of each subdivision on an FA map of a mid-sagittal cross section of the CC in the same representative preterm child. Panel (c) displays each subdivision on a quantitative T1 (R1 = 1/T1) map of a mid-sagittal cross-section of the same representative preterm child presented in (a).
Fig. 1
### Statistical approach
#### Group comparisons: demographic and clinical variables
All statistical analyses were performed using SPSS (version 24.0, IBM Corporation, 2014). We computed two-tailed t -tests for independent samples for each of the demographic and behavioral measures of age, gestational age, birth weight, SES and general intelligence. We computed a separate chi-square analysis for the one categorical demographic variable of sex. Threshold for significance was set to p < .05.
#### Group comparisons: FA and R1
Group differences in mean tract properties (FA or R1) were first assessed for each subdivision of the corpus callosum by calculating a two-tailed t-test for independent samples. Threshold for significance was set to p < .05 and a trend for significance at p < .1. We considered the probability of Type I error to be low given that several studies observe reduced FA within the corpus callosum in preterm compared to full term children ( ). For this reason, we did not correct for the number of comparisons made across the four callosal subdivisions. Effect size of group differences were reported using Cohen's d. Analyses were repeated removing the one participant with cerebral palsy to confirm that group differences were not driven by this child.
We then calculated two-tailed t-tests for independent samples, comparing FA or R1 values of the preterm and full term groups at each location along the tract profile. This analysis was conducted in order to (1) examine whether group differences in FA were accompanied by group differences in R1 in overlapping or non-overlapping locations along the tract profile and (2) confirm that group differences in either FA or R1 were not obscured due to the use of mean tract measures. The analysis of tract profiles utilized a nonparametric permutation-based method to control for 30 comparisons along the tract ( ). This procedure produced a critical cluster size for each of the candidate tracts (significant cluster size for all tracts was ≥ 5 locations/nodes). Differences along a tract were considered significant after correction for within tract comparisons if they occurred in a cluster larger than the critical cluster size.
We next performed a series of secondary analyses to further interpret group differences identified in along tract analyses. All secondary analyses described below were performed only for those tracts in which significant group differences identified in along tract analyses survived corrections for multiple comparisons. First, the contributions of axial diffusivity (AD) and radial diffusivity (RD) to group differences observed in FA were examined using a one-way multivariate analysis of variance (MANOVA; ( ; ; ). For these analyses group (preterm vs full term) served as the between-subjects variable and mean AD and mean RD computed from significant tract locations served as dependent variables. Significant group effects identified in the MANOVA were then examined with post-hoc univariate ANOVAs to establish whether group differences were driven by AD, RD or both.
To explore whether the underlying tissue properties that contributed to individual differences in FA were related to the underlying tissue properties that contributed to individual differences in R1, we performed Pearson correlations between mean FA and mean R1. Pearson correlations were performed based on exploratory analyses that demonstrated normal distributions for the majority of pathways in both FA and mean R1 in the both groups. We restricted the analyses to tract locations where group differences in FA identified in along tract analyses remained significant after correcting for multiple comparisons. We also explored associations of RD and R1 because RD has been shown to be sensitive to variations myelin content ( ). We anticipated a negative association. We again used Pearson correlations because of the normal distributions of RD and R1 for most tracts. We restricted the analyses to the same tract locations as we used for FA-R1 associations. These analyses were performed in each group separately to avoid pseudo-correlations, stemming from group differences.
Follow-up analyses were performed to assess whether group differences in FA or R1 remained significant after controlling variation in Full Scale IQ, which was found to significantly differ between children born preterm and full term in the current sample ( ; ; ). These analyses were computed using mean FA or mean R1 computed from significant tract locations entered into a univariate mixed analysis of covariance (mixed-ANCOVA) in which group (preterm vs full term) served as the between subjects factor and Full Scale IQ scores served as the covariate.
## Results
### Demographic and clinical variables
Results of the group comparisons of demographic and behavioral measures are presented in . By design, children in the preterm group had significantly decreased GA and birth weight than the full term sample. Children in the preterm group did not differ significantly from the full term group in age, sex, or SES. Children in the preterm group had mean general intelligence standard scores within the normal range but significantly lower than children in the term group.
Demographic measures for the preterm and full term sample.
Table 1
### Group comparisons: FA and R1
Compared to the full term group, children born preterm demonstrated significantly decreased mean tract FA in 3 of the 4 analyzed subdivisions of the corpus callosum (CC-temporal; CC-motor; CC-anterior frontal; ). Cohen's effect sizes for group FA differences in these three callosal subdivisions were moderate (0.4–0.5; ). Compared to the full term group, children born preterm also demonstrated significantly decreased mean tract R1 in 2 of the 4 analyzed subdivisions of the corpus callosum (CC-temporal; CC-motor; ). Cohen's effect sizes for group R1 differences in these two callosal subdivisions were moderate to large (0.6–0.7; ). No significant birth group differences were observed for mean tract FA and R1 of the CC-occipital ( ) or mean tract R1 of the CC-anterior frontal (Table 2). These findings remained the same after removing the one preterm participant with cerebral palsy.
Birth group differences in mean tract FA and R1 in pathways of the corpus callosum in preterm and full term children.
Table 2
Along tract analyses of tract profiles confirmed that significant decreases in FA observed in the preterm group for the CC-temporal and the CC-motor were accompanied by significant decreases in R1 at several overlapping tract locations (CC-temporal: b,f; CC-motor, c,g). By contrast, significant decreases in FA observed in the preterm group for the CC-anterior frontal were not accompanied by decreased R1 in the preterm group at any tract location ( d,h). Finally, in the CC-occipital subdivision, modest decreases were found in FA and R1 values of the preterm group in overlapping and non-overlapping tract regions that did not show significant group differences in mean-tract analyses ( a,e).
Preterm children demonstrate significantly decreased FA and decreased R1 in overlapping and non-overlapping regions of 3 subdivisions of the corpus callosum. FA tract profiles (a–d) and R1 tract profiles (e–h) for the preterm (solid green line) and full term (solid blue line) groups are shown for the CC-occipital (a,e), CC-temporal (b,f), CC-motor (c,g) and CC-anterior frontal (d,h). Dashed lines indicate ±1 standard error of the mean. FA and R1 values are plotted for 30 equidistant locations (nodes) between the two ROIs used to isolate the core of each tract. Shaded gray background indicates tract locations where preterm children demonstrated significantly decreased FA or R1 compared to full term children ( p < .05, corrected for 30 comparisons by controlling the family-wise error). Shaded tan background indicates tract locations where preterm children demonstrated significantly decreased FA or R1 compared to full term children (p < .05, uncorrected). Tract renderings from the same representative preterm subject presented in are displayed on an axial T1-weighted image next to each tract profile. Magnitude of t- tests for independent samples computed to visualize the location of group differences identified in along tract tests are displayed as a colored heat map on a cylinder surrounding tract renderings. Color bar represents magnitude of t statistics. (CC = corpus callosum; Ant = Anterior; PT = preterm; FT = full term; FA = fractional anisotropy).
Fig. 2
Secondary analyses demonstrated that decreased FA in the preterm group was driven primarily by significantly increased RD in both the CC-temporal and the CC-anterior frontal (Supplemental Table S1). No birth group differences in AD were observed for either tract.
### Associations between dMRI metrics (FA or RD) and R1
Correlation analyses revealed significant positive associations between mean FA and mean R1 of the CC-temporal in both the preterm (r = 0.73, p < .001; a,b) and full term group (r = 0.65, p = .001; a,c). Significant negative associations were also found between mean RD and mean R1 of the CC-temporal in both the preterm (r = −0.83, p < .001;) and full term group (r = −0.83, p < .001). By contrast, no association was observed between mean FA and mean R1 of the CC-anterior frontal in either the preterm (r = 0.18, p = .35; d,e) or the full term group (r = −0.26, p = .22; d,f). A significant negative association was found between mean RD and mean R1 of the CC-anterior frontal in the preterm (r = −0.48, p = .009) but not the full term group (r = 0.21, p = .34).
Decreased FA is associated with decreased R1 in posterior but not anterior tracts of the corpus callosum in preterm and full term children. Scatter plots represent the association between mean FA and mean R1 from significant locations along tract profiles of the CC-temporal (a) and CC-anterior frontal (d). Tract renderings from the same representative preterm subject (b,e) presented in and a representative full term subject (c,f) are displayed on an axial T1-weighted image next to scatter plots. Strength of correlations between FA and R1 at 30 equidistant locations (nodes) are displayed on a colored cylinder surround tract renderings for the CC-temporal (b,c) and CC-anterior frontal (e,f). White arrows indicate tract location where significant group differences in FA were observed for the CC-temporal (nodes 25–29; b,c) and CC-anterior frontal (nodes 18–23; e,f). Color bar represents Pearson correlation coefficients. (CC = corpus callosum; PT = preterm; FT = full term; FA = fractional anisotropy).
Fig. 3
### IQ
Secondary analyses also demonstrated that birth group differences in the CC-temporal remained significant for both FA ( F (1,50)) = 3.81, p = .03) and R1 ( F( 1,50) ) = 4.60, p = .02) after co-varying for group differences in IQ. Birth group differences in the CC-anterior frontal for FA also remained significant after co-varying for group differences in IQ ( F( 1,50 ) = 4.17, p = .02).
## Discussion
In this study of school-aged children born preterm and at term, we employed complementary methods of dMRI and qT1 relaxometry to examine whether birth group differences within the corpus callosum may be explained by reductions in myelin content. We hypothesized that, compared to children born at term, preterm children would show decreased FA within pathways of the corpus callosum consistent with previous research ( ). Group differences in FA were driven by differences in RD, not AD. We also hypothesized that reduced FA would be accompanied by reduced R1, suggestive of reduced myelin content. The first hypothesis was confirmed for the occipital, temporal, motor and anterior frontal subdivisions of the corpus callosum. The second hypothesis was confirmed for the temporal and motor subdivisions, but not for the anterior-frontal subdivision. Further analyses revealed that FA and RD were significantly associated with R1 within posterior but not anterior subdivisions of the corpus callosum, except for RD which correlated with R1 in the preterm group only. Secondary analyses confirmed that group differences in FA or R1 for the temporal and anterior frontal subdivisions were unlikely to be explained by birth group differences in IQ. Taken together, the present findings suggest that white matter abnormalities observed in school-aged children born preterm are likely to involve multiple tissue properties, including, but not limited to myelin. The present findings also suggest that the balance of tissue properties indexed by diffusion metrics may vary across different white matter tracts of the brain and between children born preterm or term.
### Group differences in FA reflect differences in myelin content in posterior CC
Birth group differences in FA observed here for tracts that traverse the posterior corpus callosum (CC-temporal, CC-occipital) are consistent with several previous dMRI studies that have reported decreased FA within regions of the splenium in neonates and children born preterm compared to their term born peers ( ; ). Effect sizes for group differences in FA were generally comparable to group differences observed in anterior and posterior regions the corpus callosum reported in a previous study of 9–17 year old children and adolescents born preterm and full term that employed similar tractography approaches ( ). Here, novel evidence for decreased R1 within the CC-temporal and CC-motor segments is consistent with evidence demonstrating the susceptibility of posterior and periventricular white matter to injuries from hypoxia-ischemia and inflammation in non-human animal models of prematurity ( ; ; ). The present findings suggest dysmaturity of pre-oligodendrocytes beyond the initial perinatal insult may lead to permanent reductions in the amount of myelin in pathways captured by CC-temporal and CC-motor tracts ( ; ; ). Future studies of preterm children that combine dMRI with qT1 methods may clarify the white matter tissue properties directly impacted by specific complications of preterm birth (eg., hypoxia-ischemia, inflammation). Longitudinal studies employing these techniques will be important for specifying the white matter tissue properties that may explain variability observed neurodevelopmental outcomes ( ; ) or that may change during development in response to specific environmental experiences or medical treatments.
### Group differences in FA reflect tissue properties other than myelin in anterior CC
Contrary to our initial hypothesis, we observed significantly decreased FA in the preterm compared to the full term group that was not explained by a parallel decrease in R1 within the CC-anterior frontal. Although the present findings are consistent with several previous dMRI studies demonstrating decreased FA within the genu in preterm compared to full term children ( ), we did not find evidence to suggest that decreased levels of FA could be explained by reduced myelin content. Instead, the current findings suggest that additional tissues properties indexed by dMRI metrics that reflect axonal microstructure and fiber organization may account for the present decreases in FA and increased RD, including decreased levels of axonal packing, decreased fiber coherence from increased amounts of fiber crossings, or larger axonal diameters ( ). Anterior regions belonging to the genu of the corpus callosum contain higher proportions of unmyelinated axons and denser packing of small diameter fibers compared to medial and posterior regions of the corpus callosum ( ; ). Myelination is also known to proceed slower and peak later in frontal white matter regions compared to medial and posterior brain areas ( ; ). Such evidence may account for why periventricular and posterior white matter areas demonstrate increased susceptibility to white matter injuries from complications of preterm birth ( ; ). This evidence may also explain why associations between dMRI metrics (FA or RD) and R1 were most evident across birth groups in posterior but not anterior callosal subdivisions.
Deciphering whether decreased FA within anterior corpus callosum pathways reflects decreases in axonal packing or increases in the amount of crossing fibers or axonal diameter will require the application of qT1 imaging and additional sophisticated dMRI analyses for quantifying the amount of fiber crossing and axonal density (e.g., NODDI ( )) and additional MRI imaging methods for quantifying the axonal diameter (e.g., AxCaliber ( )). The combination of these techniques along with qT1 can be used to obtain greater specificity for tissue properties contributing to variations in dMRI that are observed across preterm and full term groups and across and within white matter tracts of the brain. Application of these methods in longitudinal studies will also help to establish how myelination may differ across preterm and full term children.
### Limitations and future directions
Although R1 is strongly related to the amount of myelin within white matter voxels, we cannot rule out the possibility that reduced R1 within the preterm sample may also reflect other tissue properties, such as decreased iron deposition or increased in water content, both possibly arising from axonal loss. However, evidence suggests that the cellular processes that may affect tissue concentrations of water content and of iron deposition may be related to myelination ( ; ; ). The present study also cannot determine whether the observed changes in R1 are the consequence of reduced myelin content beginning in the neonatal period, or secondary changes based on the dysregulation of myelination in the aftermath of preterm birth (eg., axonal loss). Understanding how such white matter tissue properties are impacted in the long term aftermath of preterm birth is likely to require the use of animal models and additional imaging modalities in developing human children.
The present study was based on a convenience sample that was of modest size. We also made no corrections for the four comparisons performed across different callosal segments. However, given that the present findings replicated previous evidence for reduced FA within corpus callosum in preterm compared to full term groups ( ) we expected the likelihood for type I error to be low. Birth group differences observed in FA and R1 in the CC-occipital were modest and would be better tested in larger samples of preterm children than in the current study. Without longitudinal data, we cannot establish whether the observed group differences reflect differences resulting from initial injuries, compensation for injury, or differences in developmental changes. Nevertheless, the present study makes important contributions to understanding white matter tissue properties contributing to altered white matter microstructure observed in older children born preterm.
On-going analyses will examine relationships between dMRI and qT1 metrics in relation to functional outcomes, including individual variation in language and reading outcomes in both children born preterm and at term. The present findings emphasize the importance of including additional metrics based on qT1 measurements in studies of dMRI, in order to aid in the interpretation of group differences identified with dMRI and to understand structure-function associations. The combination of methods is critical to deepening our understanding of the contributions of white matter in development, in health, and in disease.
|
Highlights
Electroconvulsive therapy (ECT) induced enlargement in hippocampus volume contrast to pharmaceutical therapy in schizophrenia.
Both ECT responders and non-responders show hippocampal volume expansion.
Increased FC between hippocampus and brain cognitive networks only in ECT responders.
Electroconvulsive therapy (ECT) is considered a treatment option in patients with drug-resistant schizophrenia (SZ). However, approximately one-third of patients do not benefit from ECT in the clinic. Thus, it is critical to investigate differences between ECT responders and non-responders. Accumulated evidence has indicated that one region of ECT action is the hippocampus, which also plays an important role in SZ pathophysiology. To date, no studies have investigated differences in ECT effects in the hippocampus between treatment responders and non-responders. This study recruited twenty-one SZ patients treated for four weeks with ECT (MSZ, n = 21) and twenty-one SZ patients who received pharmaceutical therapy (DSZ, n = 21). The MSZ group was further categorized into responders (MSR, n = 10) or non-responders (MNR, n = 11) based on treatment outcomes by the criterion of a 50% reduction in the Positive and Negative Syndrome Scale total scores. Using structural and resting-state functional MRI, we measured the hippocampal volume and functional connectivity (FC) in all SZ patients (before and after treatment) and 23 healthy controls. In contrast to pharmaceutical therapy, ECT induced bilateral hippocampal volume increases in the MSZ. Both the MSR and MNR exhibited hippocampal expansion after ECT, whereas a lower baseline volume in one of hippocampal subfield (hippocampus-amygdala transition area) was found in the MNR. After ECT, increased FC between the hippocampus and brain networks associated with cognitive function was only observed in the MSR. The mechanism of action of ECT in schizophrenia is complex. A combination of baseline impairment level, ECT-introduced morphological changes and post-ECT FC increases in the hippocampus may jointly contribute to the post-ECT symptom improvements in patients with SZ.
## Introduction
Schizophrenia (SZ) is a serious mental disorder mainly characterized by multidimensional psychotic syndrome, such as positive and negative symptoms, as well as cognitive and affective impairments ( ). In SZ, electroconvulsive therapy (ECT) is considered a treatment option, particularly in patients with drug-resistant symptoms or to resolve acute symptoms ( ). Although ECT has been a conventional technique in clinical treatment, its mechanisms of action on the brain have not been fully clarified.
Since ECT was introduced to clinical practice, several hypotheses, including monoamine neurotransmitter, neuroendocrine and anticonvulsant theories ( ), have been provided to interpret possible mechanisms of action. ECT-related neuroplasticity is one of the potential mechanisms, especially neuroplasticity occurring in the hippocampus ( ; ; ). Several studies found reproducible results that electroconvulsive shock induced neurogenesis in the dentate gyrus of the hippocampus in an animal model ( ; ). In humans, accumulated evidence has also indicated that ECT induced volume increases in the hippocampus ( ; ; ). More recently, Takamiy and colleagues systematically reviewed MRI studies investigating structural changes due to ECT in patients with depression and quantitatively analysed whether ECT induced hippocampal and other brain region structural changes through a meta-analytic approach ( ). They found that both right and left hippocampal and amygdalar volumes increased after ECT. In addition, our previous study applied a data-driven method (voxel-based morphometry, VBM) to detect ECT-induced grey-matter alterations across the whole brain and found significant grey matter increases within the hippocampus ( ). Although some studies have also reported ECT-induced changes in other regions, such as the striatum ( ), cingulate cortex ( ) and insula ( ), the current hypothesis-driven study focused on ECT-induced hippocampal plasticity, compared hippocampal volume and FC between ECT responders and non-responders to link ECT-related hippocampal plasticity with clinical outcomes in SZ.
The hippocampus anatomically connects brain regions that mediate emotional and cognitive regulation, which supports the hippocampus as having a key role in SZ-related circuitry ( ; ). In addition, structural and functional disruptions in the hippocampus have been implicated in the pathophysiology of SZ. On the one hand, hippocampal volume reduction has been widely reported in previous SZ studies ( ; ; ). However, several structural MRI studies have reported contrasting results with no differences in hippocampal volume between SZ and healthy subjects ( ; ). These inconsistencies might be explained by illness duration or treatment strategy ( ). As there are these inconsistent results regarding the hippocampal volume reductions in SZ, it is meaningful for clinical research to clarify the relationship between the levels of hippocampal volume reduction and clinical responses to ECT. A substantial number of studies have indicated that there are FC alterations in the hippocampus in patients with SZ ( ; ). Moreover, studies have found associations between hippocampal FC changes and clinical symptoms in SZ patients ( ; ). These findings not only suggest that hippocampal FC is highly relevant to the symptoms and pathobiology of SZ but also imply that clinical treatment outcomes (e.g., symptom remission) may be directly reflected in FC changes. However, to date, no study has investigated whether ECT causes differential FC changes in the hippocampus between responders and non-responders and assessed the relationship between FC changes and symptom remission.
Clinically, approximately one-third of patients do not benefit from ECT ( ); thus, it is of high clinical significance to investigate differences between ECT responders and non-responders. Taking into account these issues, in this observational study, which was designed in a similar manner as a previous study ( ), we systematically assessed ECT effects on hippocampal structure and function in patients with SZ. Our specific aims were to clarify (1) whether the hippocampal volume changes induced by ECT were different than those treated with drug treatment alone. We hypothesized that the hippocampal volume would be increased after ECT treatment but would not change after drug treatment alone; (2) whether the hippocampal volume increase is an ECT effect common to both clinical responders and non-responders. We hypothesized that both responders and non-responders would show increased hippocampal volume after ECT; (3) whether the hippocampal FC changes are ECT effects that are specific to either the responders or non-responders. We hypothesized that ECT would increase hippocampal FC in the responders but decreased FC in the non-responders; and (4) whether there were hippocampal volume and FC differences between ECT responders and non-responders at baseline.
## Materials and methods
### Participants
In the present study, forty-two patients with acute SZ were divided into two groups according to treatment strategy. One group received a four-week ECT series in addition to antipsychotic drugs (MSZ group, n = 21); the other group received only antipsychotic drugs (DSZ group, n = 21). All inpatients were recruited from the Shanghai Mental Health Centre (SMHC) from October 2013 to January 2015. The patients were diagnosed with SZ by trained clinical psychiatrists using the SCID-I/P (Structural Clinical Interview for DSM-IV-TR, Patient edition) and met the indications for ECT. In addition, the patients had no history of ECT within the previous six months. Psychiatric symptom severity was assessed by the Positive and Negative Syndrome Scale (PANSS), and the total PANSS scores of all patients were greater than 60. All patients received antipsychotic medications, and the daily antipsychotic medication dosage was converted to chlorpromazine equivalents (mg/d) ( ). Additional details regarding the antipsychotic medication for each patient are provided in Supplementary Information 1. A sample of healthy controls (HC, n = 23), which was matched to the patient groups by age, sex and education level ( ), was also recruited from the faculty in SMHC. All healthy controls did not have a lifetime psychiatric disorder or family history of psychosis in their first-degree relatives. Potential participants were excluded if they had brain injuries, organic mental disorders, neurological abnormalities, other serious physical illnesses, dementia, substance abuse or dependence, or contraindications to MRI. The Ethics Committee of SMHC approved the study protocol. Written informed consent was obtained from all subjects prior to study participation.
Demographic and clinical data of participants.
Table 1
### Electroconvulsive therapy
Using a therapeutic apparatus Thymatron System IV (Somatics, Lake Bluff, IL, USA), ECT was administered 3 times weekly for 4 weeks and 12 sessions. Before ECT, succinylcholine chloride (1.0 mg/kg) was used to relax muscles, atropine (0.5 mg) was applied to reduce airway secretion, etomidate (0.21–0.3 mg/kg) and propofol (1.82–2.44 mg/kg) were conducted to keep anaesthesia. Two electrodes were placed at bilateral temporal scalps. The main parameters of ECT were similar for all patients (frequency, 10–70 Hz; maximum charge delivered, 504 mC; output current, 0.9 A; pulse width, 1.0 ms; maximum stimulus duration, 8 s). During ECT, we monitored motor convulsions and induced tachycardia, and also recorded electroencephalogram and electromyogram (when necessary). The antipsychotic therapy remained stable during ECT period except for the discontinuation of pharmacotherapy in the morning before ECT.
### Symptom assessments
At pre- and post-treatment, we measured the positive (PANSS-P), negative (PANSS-N), general psychopathology (PANSS-G) subscales and total scores (PANSS-T). The PANSS reductive ratio was defined as percentage PANSS changes as: △PANSS-T% = (PANSS-T −PANSS-T ) × 100 /(PANSS-T −30). PANSS-T −30 was used as baseline value instead of PANSS-T as 30 was the lowest possible value for PANSS total score ( ). Similarly, △PANSS-P%, △PANSS-N%, △PANSS-G% were also calculated. After ECT, the MSZ group was further classified as the ECT responder (MSR group, n = 10) and non-responder (MNR group, n = 11) groups, with the criterion of less than 50% individual symptom relief for non-response according to the PANSS total reductive ratio ( ). Similarly, the DSZ group was also grouped to the drug responder (DR, n = 12) and non-responder groups (DNR, n = 9). The responders and non-responders did not differ in terms of gender, age, education, illness duration and baseline PANSS scores ( ).
### Data acquisition
Whole-brain imaging data were acquired using a 3-T Siemens Magnetom veriosyngo MR B17 scanner. Functional MRI data were obtained by a gradient echo planar imaging (EPI) sequence (TR 2000 ms; TE, 30 ms; flip angle, 90°; FOV, 220 mm × 220 mm; matrix, 64 × 64; slice thickness, 4 mm; 30 slices; voxel size, 3.4 × 3.4 × 3.4 mm; 180 vol). In addition, high-resolution T1-weighted structural images were collected using a magnetization-prepared rapid acquisition gradient echo (MPRAGE) sequence (TR, 2530 ms, TE, 2.56 ms, flip angle, 7°, inversion time, 1100 ms, FOV, 256 mm × 256 mm, matrix, 256 × 256, 224 slices, slice thickness, 1 mm; voxel size, 1.0 × 1.0 × 1.0 mm). The patients were scanned twice before (baseline) and after 4-week therapy. The HC underwent scanning only at baseline. The first MRI scan of patients was obtained within 24 h before the first ECT, and the final MRI scan was collected 24–48 h after the last session of ECT. Participants were instructed to keep their eyes closed, relax, but not to fall sleep.
### Data processing
#### 2.5.1. Hippocampal structural analysis
A longitudinal segmentation of hippocampal substructures was performed using FreeSurfer (version 6.0, ). Specifically, a longitudinal processing stream ( ) including skull stripping, tissue segmentation, surface reconstruction, registration and parcellation, was applied to all subjects' T1 images in Freesurfer. This stream created an unbiased within-subject template space by using robust, inverse consistent registration across scanning images ( ). A longitudinal segmentation of hippocampal subfields tool in Freesurfer was also applied to segmented hippocampal regions to obtain twelve hippocampal subfields for each hemisphere ( ). The longitudinal algorithm uses a subject-specific atlas and treats all scanning time points the same way to avoid processing bias, and thus increases the robustness of segmentation ( ). Whole hippocampal volume and twelve subfields volumes were measured for each hemisphere, including the hippocampal tail, subiculum, presubiculum, parasubiculum, cornu ammonis area 1 (CA1), CA3, CA4, hippocampal fissure, granular cells layer of the dentate gyrus (GC-ML-DG), molecular layer, hippocampus-amygdala transition area (HATA) and fimbria. Illustrations of hippocampal subfield segmentation for each subject are provided in Supplementary Information 2.
#### 2.5.2. Hippocampal resting-state functional connectivity analysis
Resting-state fMRI processing was performed in SPM12 ( ) and DPABI ( ) software. The fMRI data preprocessing pipeline was similar with previous study ( ; ; ) and was only briefly described here. Firstly, the first 10 time points were removed for signal equilibrium and to allow the subjects’ adaptation to the scanning noise; secondly, we did the slice-timing correction, realignment correction, normalization and resampling to 3 × 3 × 3 mm ; thirdly, the nuisance covariates including 24 motion parameters, white matter and cerebrospinal fluid signals and linear trending were regressed out; then, we performed temporally scrubbing ( ) and temporal filtering (0.01–0.1 Hz); finally, the data was smoothed (FWHM= 6 mm). Following the preprocessing, seed-based FC analysis was used to evaluate the hippocampal FC. Four regions of bilateral rostral hippocampus (RosHIP) and caudal hippocampus (CauHIP) were defined as the seed ROIs according to the Human Brainnetome Atlas ( ). Pearson's correlation coefficients between the time series of each seed and that of the other voxels in the whole brain were calculated and then Fisher's z-transformed to z-scores. For each subject and each seed, a FC z-score map was obtained and used for following statistical analysis.
### 2.6. Statistical analysis
#### 2.6.1. Hippocampal volume comparisons
To determine whether the hippocampal volume changes were a specific effect induced by ECT rather than drug, repeated measured two-way ANOVA was performed on the volumes of the whole hippocampus and each subfield with the between-subject factor (treatment strategy: MSZ vs. DSZ) and the within-subject factor (time: t vs. t ). Post-hoc paired t-tests were performed in the MSZ and DSZ groups to indentify longitudinal changes after controlling for the overall intracranial volume.
To further examine whether hippocampal volume changes were different between the ECT responders and non-responders, repeated measured two-way ANOVA was conducted with the between-subject factor (MSR vs. MNR) and the within-subject factor (time: t vs. t ). Two post-hoc paired t-tests were separately performed in the MSR and MNR groups to characterize the longitudinal changes in hippocampal volume after the ECT when controlling the effect of the overall intracranial volume.
Additionally, ANOVA and post-hoc tests were used to compare the baseline differences among the MSR, MNR and HC groups.
#### 2.6.2. Hippocampal FC comparisons
To verify the hypothesis that ECT induced specific changes in hippocampal FC in the ECT responders, repeated measured two-way ANOVA was conducted with the between-subject factor (outcome: MSR vs. MNR) and the within-subject factor (time: t vs. t ); the interaction effect of outcome and time was used to investigate the specific changes observed in the ECT responders. Two post-hoc paired t-tests were separately performed in the MSR and MNR groups to detect the longitudinal changes in hippocampal FC after the ECT. In addition, ANOVA and post-hoc tests were applied to the comparisons amongststststst the MSR, MNR and HC groups at the baseline. For these analyses on the voxel-wise FC z-score maps, a multiple comparison correction was performed using a height threshold ( z > 2.7) for individual voxels and a cluster size based on the Gaussian random field theory, which corresponds to p < 0.05 after correction ( ).
### . Relationship between hippocampal changes and symptom improvements
The hippocampal volume and FC values that exhibited significant longitudinal changes were extracted. The hippocampal volume change was defined as the difference in volume (Volume - Volume ). Similarly, the FC changes were computed as difference values (FC - FC ). As these values did not conform to a normal distribution according to the Kolmogorov-Smirnov tests, Spearman rank correlations were used to assess the relationships between the hippocampal volume or FC changes and the reductive ratios of the symptoms (△PANSS-P%, △PANSS-N%, △PANSS-G%, and △PANSS-T%) in the MSR group.
### Power analysis on the longitudinal changes in hippocampal volume and FC
Finally, to estimate the statistical power for these longitudinal changes in hippocampal volume and functional connectivity in the MSR and MNR groups, power analyses were conducted using the G*Power 3.1.9.2 ( ).
### Additional analysis
#### Baseline comparisons between MSZ and DSZ
To investigate the differences in hippocampal volume and FC between the MSZ and DSZ groups at baseline, we performed two-sample t-tests to compare the differences between the MSZ and DSZ in the volume of each hippocampal subfield and FC.
#### Comparisons between responders and non-responders
In addition, we divided all the SZ patients into responders (SZR group: N = 22; 14 female; 30.72±7.38 years of age) and non-responders (SZNR group: N = 20; 9 female; 29.05±7.24 years of age) to further compare the baseline volume and FC across groups and the group-by-time interactions using repeated measured ANOVA.
#### FC changes in the DR and DNR groups
To further investigate the FC changes due to pharmacological treatment, paired t-tests were used to separately compare the differences between t and t in the DR and DNR groups.
## Results
### Hippocampal volume changes
Repeated measured ANOVA showed a significant interaction effect between the treatment strategy (MSZ vs. DSZ) and the time (t vs. t ) in bilateral hippocampal volumes (left hippocampus, F = 12.76, p <0.001; right hippocampus, F = 23.70, p <0.001) and certain subfields ( ). Post-hoc paired t -test analysis showed significant increases in bilateral whole hippocampal volumes in the MSZ group when controlling for the effect of overall intracranial volume (left, t = 3.97, p <0.001; right, t = 4.53, p <0.001). However, a volume increase was not observed in the DSZ group. Details of changes in all hippocampal subfields are shown in Supplementary Information 3.
Different changes of hippocampus volume between MSZ group and DSZ group. * represents that repeated measured ANOVA showed a significant interaction effect between the treatment strategy (MSZ vs. DSZ) and the time (t vs. t ) in bilateral hippocampal whole volumes and certain subfields. Abbreviation: MSZ, schizophrenia patients treated by ECT; DSZ, schizophrenia patients treated by antipsychotic drugs; CA, cornu ammonis area; GC-ML-DG, granular cells layer of the dentate gyrus; HATA, hippocampus-amygdala transition area.
Fig. 1
According to their remission status, patients in the MSZ group were divided into MSR and MNR groups. Repeated measured ANOVA (Group effect: MSR vs. MNR; Time effect: t vs. t ) showed a significant time effect in bilateral hippocampal volumes (left, F = 17.65, p <0.001; right, F = 29.57, p <0.001). Further paired t-tests showed that ECT induced significant volume increases in the bilateral hippocampus and certain subfields for both the MSR group and MNR group even after controlling for the overall intracranial volume ( and Supplementary Information 4). Moreover, a power analysis exhibited a high statistical power (> 0.8) for these longitudinal changes (Supplementary Information 5).
Longitudinal changes of hippocampus and subfields volumes in MSR and MNR after ECT. * represents that paired t-tests indicate a significant increased volume in bilateral hippocampus and certain subfields in the MSR group and MNR group. Abbreviation: MSR, schizophrenia patients with symptom remission after ECT; MNR, schizophrenia patients without symptom remission after ECT; CA, cornu ammonis area; GC-ML-DG, granular cells layer of the dentate gyrus; HATA, hippocampus-amygdala transition area. The represents that outliers (>mean±2*SD) dots were removed when performing the statistical analyses.
Fig. 2
Baseline comparisons showed significant group differences amongststststst the MSR, MNR and HC groups in the left CA1, left CA3, left molecular_layer, bilateral fissure, GC-ML-DG, CA4 and HATA ( and Supplementary Information 6). However, after multiple comparison corrections by Bonferroni correction, only the bilateral HATA (left, F = 8.26, p <0.001; right, F = 10.88, p <0.001) and CA4 (left, F = 9.19, p <0.001; right, F = 8.23, p = 0.001) remained significant. Post-hoc tests showed that in the left HATA, the MNR group had lower volume than the MSR and HC groups; in the right HATA, both the MNR and MSR groups showed lower volumes than the HC group; in the bilateral CA4, the MSR group showed higher volumes than the HC group ( and Supplementary Information 6).
Baseline comparisons in hippocampal and subfields volume among MSR, MNR and HC. * represents that ANOVA and post-hoc tests indicate a significant difference among the three groups. Abbreviation: MSR, schizophrenia patients with symptom remission after ECT; MNR, schizophrenia patients without symptom remission after ECT; CA, cornu ammonis area; GC-ML-DG, granular cells layer of the dentate gyrus; HATA, hippocampus-amygdala transition area. The represents that outliers (>mean±2*SD) dots were removed when performing the statistical analyses.
Fig. 3
### Hippocampal FC changes
Paired t-tests showed that after ECT, the MSR exhibited significantly increased FC between the hippocampus and prefrontal cortex as well as between the hippocampus and regions in default mode network (DMN). No decreases in FC were observed in the MSR group. However, in the MNR group, the post-ECT patients showed decreased FC in the hippocampus and primary sensory network. No increases in FC were observed in the MNR group. Detailed information is provided in and Supplementary Information 7. Moreover, a power analysis exhibited a high statistical power (> 0.8) for these longitudinal changes (Supplementary Information 5).
Longitudinal FC changes between pre-ECT and post-ECT in MSR group and MNR group.
Abbreviations: MSR, schizophrenia patients with symptom remission after ECT; MNR, schizophrenia patients without symptom remission after ECT; RosHIP, rostral hippocampus; CauHIP, caudal hippocampus; MTG, middle temporal gyrus; MFG, middle frontal gyrus; AG, angular gyrus; ITG, inferior temporal gyrus; STG, superior temporal gyrus; PoC, postcentral gyrus; MOS, middle occipital cortex.
Fig. 4
Consistent with the above findings, significant interaction effects between the outcomes (MSR vs. MNR) and time (t vs. t ) were observed for FC between the hippocampus and DMN as well as between the hippocampus and primary sensory network. Detailed information is shown in Supplementary Information 8.
In addition, baseline comparisons among the MSR, MNR and HC groups found significant group differences in FC between the left cauHIP and the right middle frontal gyrus (MFG), between the right cauHIP and the right MFG, and between the left rosHIP and right putamen (Supplementary Information 9).
### Relationship between hippocampal changes and symptom improvements
In the MSR group, a significant association was observed between the left CA4 increase and the general psychopathology reduction ratio (Rho = 0.697, p = 0.025). In addition, a significant correlation was observed between the change in FC (left cauHIP and right angular gyrus) and the general psychopathology reduction ratio (Rho = 0.721, p = 0.019).
### Additional analysis results
#### Baseline comparisons between MSZ and DSZ
The baseline comparisons between the MSZ and DSZ groups showed that there was no significant difference between the two patient groups in bilateral hippocampal volumes. The MSZ group exhibited higher FC between the bilateral superior temporal gyrus and bilateral cauHIP than the DSZ group. Details of baseline comparisons between the MSZ and DSZ groups are shown in Supplementary Information 10.
#### Comparisons between responders and non-responders
By dividing all the SZ patients into SZR and SZNR, we found that at baseline, the SZR group had higher volumes in the left CA1, left molecular layer, left GC-ML-DG, left CA3 and left CA4 than the SZNR group. Post-hoc paired t-tests showed that in the SZR group, there were increased volumes in the right hippocampus and subfields after treatment. Details of baseline comparisons between the SZR and SZNR groups and longitudinal changes between t and t are shown in Supplementary Information 11.
#### FC changes in the DR and DNR groups
Paired t-tests showed that the DR group had increased FC between the left rosHIP and right insula as well as between the left rosHIP and right inferior frontal gyrus (IFG) after pharmacological treatment. However, the DNR group exhibited decreased FC between the left rosHIP and right middle frontal gyrus (MFG) and between the left cauHIP and right MFG after pharmacological treatment. Detailed information on FC changes in the DR and DNR groups is shown in Supplementary Information 12.
## Discussion
To our knowledge, this is the first study investigating hippocampal volume and FC changes between ECT responders and non-responders in SZ. As expected, four main findings are as follows: (1) Hippocampal volume increases were only observed in patients with SZ treated by ECT. This suggested that the hippocampal volume increase was a specific ECT effect rather than a drug effect. (2) Both the MSR and MNR groups exhibited increased hippocampal volume following ECT, which further indicated that hippocampal volume increases were an inherent effect of ECT. (3) Interestingly, in the MSR group, we found increased FC between the hippocampus and higher-order cognitive networks; however, in the MNR group, we observed reduced FC between the hippocampus and primary sensory networks, including visual, sensorimotor and auditory networks. These findings suggested that ECT-induced changes in hippocampal FC with higher-order cognitive networks might be associated with clinical symptoms. (4) Finally, baseline comparisons showed that compared with the HC group, the MNR group had lower baseline volume in the HATA and higher FC between the hippocampus and putamen, which contributed to the prediction for ECT treatment outcomes.
Our investigation corroborated previous findings of hippocampal volume increases following ECT in psychiatric disorders ( ; , ; , ; ) and further compared the differences between ECT and drug-only samples. Our study revealed increased volumes in the bilateral hippocampus only in the ECT sample, whereas evidence for such structural changes was absent in the drug-only sample. Furthermore, the current study divided the ECT samples into MSR and MNR groups and observed a hippocampal volume increase common to both groups. This finding indicated that these changes in hippocampal volume induced by ECT were not unique to patients with improved symptoms. In this study, the ECT-related brain changes in the hippocampus observed in patients with SZ were similar to those observed in patients with MDD, which is consistent with previous studies ( ; ). Thomann and colleagues reported a similar pattern of brain volume changes in the hippocampus of individuals with SZ and MDD ( ; ). In addition, hippocampal deficits have frequently been reported to be robust in patients with SZ ( ; ), and previous studies provided evidence that such abnormalities also exist in those with MDD ( ; ). This suggests that ECT modulates neural effects that are not diagnosis-specific but are critical for both affective and non-affective psychoses.
Accumulated evidence has indicated that SZ is related to aberrant functional interactions between large-scale brain networks and cortical-subcortical pathways ( ; ; ; ; ). To date, several neuroimaging studies have investigated FC changes induced by ECT using resting-state fMRI ( ; ; ). Abbott reported that hippocampal FC increased after ECT in MDD patients and correlated with depressive symptom reduction ( ). Our recent study found increased functional integration in the DMN in patients with SZ following ECT using a data-driven FC density analysis ( ). However, very few neuroimaging studies have addressed whether the brain FC affected by ECT differs between responders and non-responders. To our knowledge, only one study used baseline resting-state FC networks to predict ECT clinical responsiveness, although they did not investigate the longitudinal alterations that may have contributed to symptom improvement ( ). These findings supported the hypothesis that FC networks are specifically altered in patients who respond to ECT. Interestingly, in the current study, we found increased FC between the hippocampus and higher-order cognitive networks, especially the default mode network (DMN), in the MSR group. The hippocampus is a key region for memory encoding and retrieval functions ( ). The DMN has been widely implicated in self-referential processes related to internal mental states ( ; ). The dorsolateral prefrontal cortex, as an important region of the central executive network, is responsible for higher-order cognitive function and is crucial for interfacing with the external environment ( ; ). A possible ECT mechanism could be reinforcing the connectivity between the hippocampus and higher-order cognitive networks to manage the information integration between the internal- and external-based mental landscapes and thus influence the clinical symptoms. In addition, an association between FC changes and symptom reductions was observed, which further demonstrated the relationship between ECT effects and clinical treatment outcomes. In addition, we found reduced FC between the hippocampus and primary sensory networks in the non-responders. Notably, after ECT, these so-called "non-responders" exhibited an incomplete remission relative to the "responders", rather than a worsening of clinical symptoms. These FC reductions may be another ECT mechanism that weakens the connectivity of the hippocampus and primary sensory networks to block access of the primary information and achieve mild symptom remission. In this manner, the lack of bottom-up primary information may give rise to abnormalities in higher-order information processing. Therefore, these patients failed to attain complete symptom remission. However, these interpretations are speculative; thus, future studies should test this hypothesis by task-based fMRI.
Baseline comparisons indicated that compared with the HCs, the SZ patients in the MSR and MNR groups showed lower hippocampal subfield volumes in the HATA. This finding is also consistent with previous meta-analysis studies ( ). Furthermore, our study found a lower baseline volume in the HATA in the MNR group. To our knowledge, this is the first study reporting that a lower baseline hippocampal subfield volume correlated with poorer ECT outcomes. In addition, FC baseline comparisons also found higher hippocampus-putamen FC in the MNR group than in the MSR and HC groups. However, the baseline differences between responders and non-responders may have also been influenced by age, disease duration, episode duration, medication and other random factors. Thus, the baseline differences should be interpreted with caution. Despite these potential confounds, the results from the current study suggest that pre-existing or more serious reductions in volume in the hippocampus may have a negative impact on clinical outcomes to ECT, which could help psychiatrists, clinicians and patients make better treatment decisions.
Despite these encouraging findings, several limitations must be acknowledged. First, the current sample size is relatively modest. Although sufficient statistical power was provided by the power analysis in this study, a larger sample size would be more useful to increase the reliability and sensitivity. Second, the SZ participants included medicated and chronic patients. Antipsychotic medication and illness duration may have confounding effects on brain structure and function ( ). Although we included a matched pharmacotherapy group as the treatment control, the effects of concomitant antipsychotics during the ECT course cannot be completely ruled out. Third, neuropsychological assessments were not evaluated; thus, we did not assess the associations between the changes in cognitive function and brain changes. Fourth, we assumed that the volume and function of the normal brain would not significantly change over the course of one month; thus, healthy controls were only scanned at baseline. Finally, as patients were not randomized to each group, some potential bias may have been introduced into the analyses.
In conclusion, this study demonstrated that ECT induced a hippocampal volume increase in patients with SZ that was common to both responders and non-responders. However, we observed hippocampal FC increases in the clinical responders, while we observed decreases in non-responders. Furthermore, the FC changes correlated with symptom improvements. These findings identified ECT-induced effects in the hippocampus. In addition, the ECT-induced improvements in both structure and function in the hippocampus might imply an important mechanism of action of ECT in patients with SZ.
## Declaration of Competing Interest
There is no conflict of interest.
|
## Background
The basal forebrain is a subcortical structure that plays an important role in learning, attention, and memory. Despite the known subcortical involvement in frontotemporal dementia (FTD), there is little research into the role of the basal forebrain in this disease. We aimed to investigate differences in basal forebrain volumes between clinical, genetic, and pathological diagnoses of FTD.
## Methods
356 patients with FTD were recruited from the UCL Dementia Research Centre and matched on age and gender with 83 cognitively normal controls. All subjects had a T1-weighted MR scan suitable for analysis. Basal forebrain volumes were calculated using the Geodesic Information Flow (GIF) parcellation method and were compared between clinical (148 bvFTD, 82 svPPA, 103 nfvPPA, 14 PPA–NOS, 9 FTD–MND), genetic (24 MAPT, 15 GRN, 26 C9orf72 ) and pathological groups (28 tau, 3 FUS, 35 TDP-43) and controls. A subanalysis was also performed comparing pathological subgroups of tau (11 Pick's disease, 6 FTDP-17, 7 CBD, 4 PSP) and TDP-43 (12 type A, 2 type B, 21 type C).
## Results
All clinical subtypes of FTD showed significantly smaller volumes than controls ( p ≤ 0.010, ANCOVA), with svPPA (10% volumetric difference) and bvFTD (9%) displaying the smallest volumes. Reduced basal forebrain volumes were also seen in MAPT mutations (18%, p < 0.0005) and in individuals with pathologically confirmed FTDP-17 (17%), Pick's disease (12%), and TDP-43 type C (8%) ( p < 0.001).
## Conclusion
Involvement of the basal forebrain is a common feature in FTD, although the extent of volume reduction differs between clinical, genetic, and pathological diagnoses. Tauopathies, particularly those with MAPT mutations, had the smallest volumes. However, atrophy was also seen in those with TDP-43 type C pathology (most of whom have svPPA clinically). This suggests that the basal forebrain is vulnerable to multiple types of FTD-associated protein inclusions.
## Introduction
Frontotemporal dementia (FTD) is a clinically, genetically, and pathologically heterogeneous disease. FTD can be characterised clinically by personality change and cognitive dysfunction, known as behavioural variant FTD (bvFTD) ( ) or by language difficulties, termed primary progressive aphasia (PPA) ( ). In around a third of cases, FTD is caused by a genetic mutation, usually in one of three genes: microtubule associated protein tau ( MAPT ), chromosome 9 open reading frame 72 ( C9orf72) or progranulin ( GRN ) ( ). Neuropathologically, tau, transactive response DNA binding protein 43 kDa (TDP-43), and fused in sarcoma (FUS) inclusions are the most common cause of brain abnormalities in FTD ( ; ). FTD is classically associated with a pattern of atrophy centred around the frontal and temporal lobes, however, a number of studies have also shown early subcortical involvement ( ; ; ).
The basal forebrain is a subcortical structure located on the medio-ventral portion of the brain, ventral to the striatum ( ). It consists of a number of different structures, including the medial septal nucleus, the diagonal band of Broca, and the nucleus basalis of Meynert ( ), with the major cholinergic pathways arising from these nuclei and innervating large portions of the neocortex and limbic system ( ). Cholinergic activity linked to the basal forebrain has been shown to be crucial for several cognitive processes including attention ( ; ; ) learning, and memory ( ).
Representative figure of the basal forebrain and its anatomy, based on . The basal forebrain segmentation is mapped on to the T1-weighted ICBM152 2009c nonlinear symmetric - 1 × 1 × 1 mm template (McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University). PO = preoptic area, BN = bed nucleus, DB = nucleus of the diagonal band, nbM = basal nucleus of Meynert, OlfA = olfactory area, OlfT = olfactory tract, SI = substantia innominata. The coronal slice corresponds to MNI y = 131.
Fig. 1
To date, the vast majority of research into the basal forebrain has been carried out in Alzheimer's disease (AD). Structural MRI ( ) and post-mortem studies ( ; ; ) have shown severe degeneration of the basal forebrain cholinergic nuclei in AD the extent of which is associated with the severity of cognitive impairment ( ; ; ; ). So far, structural MRI studies of the basal forebrain in FTD have shown a reduction in volume in PPA ( ; , ) and suggested that basal forebrain nuclei are crucial for language function ( ; ). Research in FTD-associated tauopathies, including progressive supranuclear palsy (PSP) ( ; ) and corticobasal degeneration (CBD) ( ) has also shown a reduction in basal forebrain volumes at post-mortem. However, comprehensive research into basal forebrain involvement in other forms of FTD is limited. The primary objective of this study was to investigate the pattern of basal forebrain involvement in the different clinical, genetic, and pathological forms of FTD.
## Methods
### Participants
We identified 356 patients from the UCL Dementia Research Centre FTD cohort who had a good quality T1-weighted magnetic resonance (MR) scan. Of these patients 148 were diagnosed with bvFTD, 82 with semantic variant PPA (svPPA), 103 with nonfluent variant PPA (nfvPPA), 14 with PPA not otherwise specified (PPA–NOS), and 9 with a diagnosis of FTD with motor neurone disease (FTD–MND), see . We did not include patients with logopenic variant PPA who are likely to have underlying AD pathology. These patients were matched on age and gender with 83 cognitively normal controls who also had a suitable T1–weighted MR scan that passed standard quality control checking. Informed consent was obtained from all participants, and the study was approved by the local ethics committee.
Overview of the study cohort, showing how patients were stratified in to clinical, genetic, and pathological groups and further divided in to pathological subtypes.
Fig. 2
### MR acquisition
All participants had a 3D T1-weighted MR scan on one of three scanners: 215 on a 1.5T GE MRI scanner, 176 on a 3T Siemens Trio MRI scanner, and 48 on a 3T Siemens Prisma MRI scanner. MR scans underwent standard methods for quality control and were excluded if scans showed artefacts, motion, any vascular lesions or any other brain lesions not related to FTD.
### Patient groups
We also stratified participants by their genetic and pathological cause of FTD. Within the FTD patient group, 65 patients were carriers of a mutation in one of the FTD causing genes: 24 in the MAPT ( ), 15 in the GRN ( ; ), and 26 in the C9orf72 ( ; ) group (see Supplementary Table 1 for the individual mutations). Post-mortem pathological confirmation was available for 66 FTD patients. Neuropathological examination of brain tissue was carried out at the Queen Square Brain Bank for Neurological Disorders at UCL following standard histopathological methods (see Supplementary Material) where underlying pathology was identified for these patients: 28 with a tauopathy, 35 with a TDP-43 proteinopathy and 3 FUSopathy. None of the patients had significant secondary pathology. We also performed a secondary analysis to investigate whether the specific type of pathology had an impact on basal forebrain volumes: in the tauopathies, 11 with Pick's disease, 7 with CBD, 4 with PSP and 6 with FTDP-17 (i.e. pathology associated with MAPT mutations); and in the TDP-43 proteinopathies, 12 with type A, 2 with type B, and 21 with type C). In the FUS pathology group, there were 2 patients with atypical frontotemporal lobar degeneration (FTLD) with ubiquitin inclusions (aFTLDU) and 1 with neuronal intermediate filament inclusion disease (NIFID) so no subanalysis was performed.
### 2.4. Statistical analysis
Basal forebrain volumes were automatically extracted using the Geodesic Information Flow algorithm (GIF) ( ). GIF is an automated technique for the segmentation and volume extraction of 162 grey matter and white matter regions on 3D T1-weighted MR images. It is a multi-atlas propagation approach, which uses the neuroanatomical annotations from the Neuromorphometric atlas [ ] to parcellate in a step-wise way MR images in their native space, after bias-correction of field inhomogeneities. We focused on comparing the whole basal forebrain structure, calculated by summing left and right basal forebrain volumes. The basal forebrain volume was expressed as a percentage of total intracranial volume (TIV), computed with statistical parametric mapping (SPM) version 12 software ( ) in MATLAB ( ). All segmentations underwent quality control checking. Statistical analyses were performed in SPSS software (SPSS Inc, Chicago, IL, USA) v22.0, between the patient groups and controls, tested using an ANCOVA and adjusting for scanner type, gender, age, and disease duration (for patient group comparisons). Results were corrected for multiple comparisons (Bonferroni's correction) with a threshold of p < 0.01.
## Results
Demographic data as well as basal forebrain volumes for FTD patient groups and controls are reported in . The mean disease duration for the FTD group, at the time of scan, was 3.8 years (SD = 3.1). There was no significant difference in age between FTD patients (64.2 years, SD = 8.4) and controls (61.5 years, SD = 11.8, p = 0.454, t -test), and there was no difference in scanner type ( p = 0.151, Chi-square test), or gender ( p = 0.349, Chi-square test). No significant differences in scanner type were found across clinical, genetic, and pathological diagnoses ( p = 0.175, p = 0.418, p = 0.604, Chi-square test). There was also no significant difference in disease duration across clinical and pathological diagnoses ( p = 0.057, p = 0.556, ANOVA). However, compared to C9orf72 (5.5 years) and MAPT (5.7 years), GRN carriers did have a significantly shorter disease duration (2.9 years, p = 0.016).
Demographics and clinical variables for the FTD patient groups and controls, together with basal forebrain volumes. Values denote means (standard deviation) or n (%). Mean volumes in bold denote significant p -values in comparison to controls once adjusting for multiple comparisons.
Table 1
Basal forebrain volumetric differences between controls and clinical, genetic, and pathological groups are displayed in . Overall, FTD patients had significantly smaller basal forebrain volumes than controls (8% difference overall, p < 0.0005, ANCOVA) regardless of the clinical, genetic, or pathological diagnosis. All individual clinical subtypes showed significantly smaller volumes than controls ( p ≤ 0.010, ANCOVA). The svPPA group had the smallest volumes (10% difference from controls), followed by bvFTD (9%), FTD–MND (9%), PPA–NOS (8%), and lastly nfvPPA (5%). Both the bvFTD (4% difference, p = 0.016) and svPPA (5%, p = 0.003) groups had reduced basal forebrain volumes in comparison to nfvPPA patients ( ). No other significant differences were found when comparing within clinical diagnosis.
Volume of the basal forebrain, as a percentage of total intracranial volume, between FTD patients and controls, by clinical, genetic, and pathological diagnosis.
Fig. 3
Basal forebrain volumetric comparisons between controls and the different clinical, genetic, and pathological groups. Volumes are adjusted for age, scanner type and gender, as well as disease duration for within patient group comparisons.
Table 2
When stratifying by genetic diagnosis, only the MAPT group had significantly smaller basal forebrain volumes compared with controls (18% difference, p < 0.0005) ( and ). Patients in the MAPT group also displayed reduced volumes compared to both the C9orf72 (17%, p < 0.0005), and GRN (14%, p < 0.0005) groups. No volumetric differences in basal forebrain volumes were found between the C9orf72 and GRN genetic groups.
When stratifying by pathology, the smallest volumes were seen in the tauopathies (10% difference from controls, p < 0.0005). However, there was also a significant difference from controls for the TDP-43 proteinopathies (6% difference, p = 0.003), with a non-significant difference in the FUSopathies (10% difference, p = 0.172). Within the specific tauopathies, we found that only subjects with FTDP-17 and Pick's disease (17% and 12%, p < 0.0005) had smaller volumes than controls, although there was a trend to lower volumes in those with PSP ( p = 0.030). In the TDP-43 proteinopathies, only those with type C had significantly smaller basal forebrain volumes (8%, p = 0.001) than controls.
## Discussion
Using the robust GIF parcellation method, this study has demonstrated that the basal forebrain shows significant volume reduction in FTD. Three major findings emerged from this study. First, the basal forebrain is reduced across the different clinical presentations of FTD. Second, individuals with MAPT mutations have significantly smaller basal forebrain volumes than the other genetic groups. Thirdly, tauopathies have the smallest basal forebrain volumes, driven by both patients with FTDP-17 (i.e. MAPT mutations) and Pick's disease, whilst in the TDP-43 proteinopathies, lower volumes are driven by a decrease in those with type C pathology.
We show similar reductions of basal forebrain volumes in all clinical diagnoses of FTD within the study. Individuals with svPPA and bvFTD had the lowest volumes, with lower basal forebrain volumes in comparison to nfvPPA in both of these groups. This finding is in line with previous studies showing significant reductions in basal forebrain volumes particularly in svPPA, and to a lesser extent in nfvPPA ( ; ). However, to our knowledge this is the first study to demonstrate the differences in basal forebrain volumes across all of the FTD diagnoses, including bvFTD.
Our results also indicate that, within the genetic and pathological diagnoses, the basal forebrain is mainly involved in individuals with tauopathies, specifically those with MAPT mutations (FTDP-17), and also those with Pick's disease pathologically. Previous research has demonstrated the role of tau in basal forebrain degeneration. Early deposition of tau in the basal forebrain is seen in AD, which correlates with cognitive decline ( ). Other research has suggested that tau pathology in the basal forebrain is an early event in the transition from mild cognitive impairment to AD ( ). Although FTD-associated tau differs from the tau neurofibrillary tangle pathology seen in AD, it appears that different structural conformations of tau can similarly affect the basal forebrain. The fact that different types of FTD-associated tau, including FTDP-17 and Pick's disease, both exhibited reduced basal forebrain volumes, supports this notion. Research has demonstrated that cholinergic nuclei are differently affected by pathogenetic mechanisms underlying neurodegenerative disease. For example, in AD the basal forebrain cholinergic nuclei are more affected than the midpontine cholinergic nuclei in the brainstem, whereas the reverse is seen in Parkinson's disease ( ). Whilst one previous study has shown reduced basal forebrain volumes in patients with both CBD and PSP ( ), our results demonstrated significant differences only in cases with PSP pathology. This is in line with other research that has shown relative preservation of basal forebrain nuclei in CBD compared to PSP ( ). Taken together, this research suggests there is a greater vulnerability of the basal forebrain cholinergic nuclei to certain forms of tau pathology. Importantly, it has been suggested previously that tau only causes neurodegeneration in the basal forebrain in tauopathies where there is also concurrent amyloid deposition, for example in AD ( ). However, our research has shown this is not the case, as individuals with primary tauopathies do not have co-occurring amyloid pathology.
We also show that individuals with TDP-43 type C pathology had significantly smaller basal forebrain volumes than controls. This is consistent with the finding in the clinical groups of decreased volume in svPPA, as the majority of patients with this phenotype have TDP-43 type C pathology ( ). TDP-43 pathology is seen in the basal forebrain in amyotrophic lateral sclerosis ( ), however the role of TDP-43 in the basal forebrain in FTD has not been widely investigated. One study has shown the greatest severity of TDP-43 inclusions in the whole brain were in the basal forebrain ( ). Furthermore, within this population the most prominent TDP-43 pathology was seen in the diagonal band of Broca in a subject with FTD ( ). Therefore, there is some evidence to suggest that the basal forebrain is vulnerable to TDP-43 pathology in FTD patients, although more research into the type of TDP-43 pathology present in the basal forebrain in distinct clinical subtypes of FTD is required.
The brain cholinergic system is an extensive network of projection neurons that innervate several brain areas. Neurons arise from the peduncolo-pontine nucleus (PPN) and dorsolateral tegmental nucleus (DLN) in the brainstem and project to the thalamus, hypothalamus, globus pallidus, striatum, and the basal forebrain ( ; ; ). Subcortical atrophy in FTD has previously been demonstrated in the thalamus, ( ; ), hypothalamus ( ), and striatum ( ). These regions are highly connected by cholinergic neurons to the basal forebrain, therefore it is perhaps unsurprising that this region is similarly affected. One study has shown that hypometabolism in the septal region of the basal forebrain is associated with reduced performance on a prosocial sentiment task in individuals with bvFTD ( ), suggesting this region may play a role in social cognition. Furthermore, other research has suggested the basal forebrain is important for language function ( ; , ), which is of particular interest considering individuals with svPPA had significantly reduced volumes in this study. Due to the extensive cholinergic network involved, it is likely that basal forebrain volume reductions play a prominent role in cognitive impairment and language function in FTD. However, more research is needed to investigate the role of the basal forebrain and resulting impairment in FTD.
Whether atrophy of the basal forebrain disrupts the entire cholinergic system in FTD remains unclear. However, there is no evidence that anticholinesterase therapies are helpful in bvFTD or PPA ( ), and in fact some evidence that cholinesterase inhibitors can worsen behaviour in bvFTD ( ; ). However, trials will have included patients with multiple different types of FTD pathology, and the work here suggests that such therapy may be helpful in specific subsets of patients.
To our knowledge, this is the first study to extensively investigate the role of the basal forebrain in FTD, particularly in those with confirmed genetic and pathological forms. This was a retrospective study, and as such, accompanying clinical and neuropsychological data were not uniformly collected and therefore not available for comparative analysis. Furthermore, a number of our patient groups had small sample sizes, and therefore results should be interpreted with caution, and replication in larger datasets will be important. It will also be important to use higher resolution MR imaging to better understand the differential involvement of specific basal forebrain subnuclei within the different clinical, genetic, and pathological subtypes of FTD.
## CRediT authorship contribution statement
Rhian S. Convery: Writing - original draft, Formal analysis, Writing - review & editing. Mollie R. Neason: Data curation. David M. Cash: Data curation. M. Jorge Cardoso: Data curation. Marc Modat: Data curation. Sebastien Ourselin: Data curation. Jason D. Warren: Data curation. Jonathan D. Rohrer: Conceptualization, Data curation. Martina Bocchetta: Conceptualization, Data curation.
|
Highlights
ico brain dm is an automated brain MRI segmentation faster than Freesurfer.
Significantly higher accuracy was obtained for several brain structures, including hippocampus.
ico brain dm volumes had a test-retest error below normal annual atrophy rates.
ico brain dm temporal lobe volume had highest sensitivity in discriminating Alzheimer's.
Brain volumes computed from magnetic resonance images have potential for assisting with the diagnosis of individual dementia patients, provided that they have low measurement error and high reliability. In this paper we describe and validate ico brain dm , an automatic tool that segments brain structures that are relevant for differential diagnosis of dementia, such as the hippocampi and cerebral lobes. Experiments were conducted in comparison to the widely used FreeSurfer software. The hippocampus segmentations were compared against manual segmentations, with significantly higher Dice coefficients obtained with ico brain dm (25–75th quantiles: 0.86–0.88) than with FreeSurfer (25–75th quantiles: 0.80–0.83). Other brain structures were also compared against manual delineations, with ico brain dm showing lower volumetric errors overall. Test-retest experiments show that the precision of all measurements is higher for ico brain dm than for FreeSurfer except for the parietal cortex volume. Finally, when comparing volumes obtained from Alzheimer's disease patients against age-matched healthy controls, all measures achieved high diagnostic performance levels when discriminating patients from cognitively healthy controls, with the temporal cortex volume measured by ico brain dm reaching the highest diagnostic performance level (area under the receiver operating characteristic curve = 0.99) in this dataset.
## Introduction
Structural neuroimaging with magnetic resonance imaging (MRI) (or computed tomography (CT)) plays a key role in the diagnostic work-up of dementia. It allows to rule out structural lesions of the brain that might cause cognitive problems. In addition, structural neuroimaging may contribute to the early and differential diagnosis of the neurodegenerative disease underlying the dementia syndrome ( ; ; ; , ; ). Indeed, neurodegenerative disorders that cause dementia are often associated with typical brain atrophy patterns. Alzheimer's disease (AD), for instance, is characterized by medial temporal lobe atrophy, including the hippocampus, and parietal atrophy. Frontotemporal dementia, on the other hand, mainly presents with atrophy of the frontal and (anterior and / or lateral parts of the) temporal lobes. Dementia with Lewy bodies usually does not show specific structural abnormalities, while vascular dementia is mainly characterized by global atrophy and diffuse white matter lesions, lacunes and/or strategic infarcts. As such, global and focal atrophy together with vascular disease are important factors to consider when establishing a differential dementia diagnosis. Gradually, these factors are being included into diagnostic clinical criteria for dementia ( ; ; ; ; ).
Besides contributing to differential diagnosis of prevalent dementia, structural neuroimaging may also aid in predicting progression to dementia in subjects who have not reached the dementia stage yet. MRI studies have shown hippocampal atrophy to be associated with increased risk of progression to dementia due to AD ( ). Hippocampal atrophy is included as a biomarker for early AD diagnosis in the revised diagnostic criteria of the National Institute on Aging – Alzheimer Association working group ( ; ).
In order to segment brain regions-of-interest and measure brain atrophy, fully automated processing techniques have been developed. These can be used in large study cohorts, saving both time and costs, and are easily reproducible, as opposed to manual segmentation by neuroanatomical experts or semi-automated measures that still require a priori information on the region-of-interest ( ; ; ; ; ; ; ). FreeSurfer is a very frequently used automatic tool ( ); depending on hardware, may require a long computation time of up to several tens of hours per scan ( ).
Applying automated measures of brain volumes on individual dementia patients requires a low measurement error and high reliability. For instance, a meta-analysis pointed to an annual atrophy rate of the hippocampus of 4.66% in AD patients compared with 1.41% in controls ( ). Hence, the measurement error of the brain volumetric measures should be minimal, in order to draw meaningful conclusions in individual patients.
In this study we validate an automated method to measure volumes of the whole brain (WB), total gray matter (GM), frontal, parietal and temporal cortex, hippocampi, and lateral ventricles. In order to evaluate the applicability of the method for brain volume quantification of individual dementia patients, this paper focuses on the accuracy, reliability and diagnostic performance of these volumetric measures.
## Materials and methods
### Dataset 1.a (accuracy)
Dataset 1.a was acquired from 35 healthy subjects (mean age 34 (±20 SD) years, 67% females,) as part of the OASIS project ( ). Manual brain segmentations were produced by Neuromorphometrics, Inc. (neuromorphometrics.com) using the brainCOLOR labeling protocol. The data were part of the 2012 MICCAI Multi-Atlas Labeling Challenge, where 15 subjects were used as training and the remaining 20 images were used for testing. Since all 35 manual segmentations were made available, we do not make this distinction and, thus, we report results on all 35 images. The 3D magnetization-prepared rapid gradient-echo (MP-RAGE) T1-weighted MRIs were acquired using a 1.5T Siemens Vision MR scanner, voxel size of 1 × 1 × 1 mm and dimensions up to 256 × 334 × 256 mm.
### Dataset 1.b (accuracy)
Dataset 1.b was acquired from 46 subjects of a memory clinic-based research population who participated in a study at the University of Antwerp, Belgium (mean age 72.0 (±7.8 SD) years, 50.0% females, Mini–Mental State Examination (MMSE) score 25.8 ± 3.1). This population consisted of 6 cognitively healthy controls as well as patients with subjective cognitive decline ( n = 3), mild cognitive impairment ( n = 28) and dementia due to AD ( n = 9). Local ethics committees (Hospital Network Antwerp and University of Antwerp / Antwerp University Hospital) approved the study and all patients signed informed consent forms. MR imaging was performed on each subject on a 3T whole body scanner with a 32-channel head coil (Siemens Trio/PrismaFit, Erlangen, Germany). The 3D MP-RAGE (TR/TE = 2200/2.45 ms) was used to obtain 176 axial slices without slice gap and 1.0 mm nominal isotropic resolution (FOV = 192 × 256 mm).
An expert (LC) performed bilateral manual hippocampus segmentation on all subjects according to the EADC-ADNI harmonized hippocampus segmentation guidelines ( ). These manual segmentations were further used as ground truth references.
### Dataset 2 (reproducibility)
Dataset 2 consisted of 42 cognitively healthy subjects (i.e., having score 0 on the Clinical Dementia Rating scale) who received longitudinal scans up to 10 days apart (mean age 61.4 (±8.6 SD) years, 59.5% females), provided by the publicly available database OASIS-3 ( ). MR imaging was performed on each subject on a 3T whole body scanner with a 16-channel head coil (Siemens TIM Trio or BioGraph mMR PET-MR, Erlangen, Germany). The baseline and follow-up scans of three subjects were done on the same scanner, while all other 39 subjects had different scanner types for their baseline and follow-up scans.
The MP-RAGE protocol of TIM Trio scanner was as follows: TR/TE = 2400/3.16 ms, ±176 axial slices without slice gap and 1.0 mm nominal isotropic resolution (FOV = 256 × 256 mm). The MP-RAGE protocol of BioGraph mMR PET-MR scanner was as follows: TR/TE = 2300/2.95 ms, ±176 axial slices without slice gap and 1.2 mm nominal isotropic resolution (FOV = 256 × 256 mm).
### Dataset 3 (diagnostic performance)
Dataset 3 consisted of 46 AD patients (age 71.5 ± 7.2, 60.9% females, Mini–Mental State Examination (MMSE) 19.2 ± 4) and 23 cognitively healthy subjects (age 70.4 ± 7.1, 47.8% females, MMSE 29.4 ± 0.8) of the publicly available MIRIAD database (miriad.drc.ion.ucl.ac.uk). An overview of the MIRIAD demographics, diagnostic procedures, and imaging protocol was published previously ( ). In brief, AD patients were diagnosed with mild–moderate probable AD according to the NINCDS–ADRDA clinical criteria ( ), while the control subjects did not have subjective cognitive complaints, nor evidence of cognitive impairment. All scans were conducted on a 1.5T whole body scanner (GE Medical systems Signa, Milwaukee, Wisconsin, USA). Three-dimensional T1-weighted (T1w) images were acquired with an IR-FSPGR (inversion recovery prepared fast spoiled gradient recalled) sequence, FOV 240 mm, 256 × 256 matrix, 124 1.5 mm coronal partitions, TR/TE = 15/5.4 ms.
A summary of the 3 datasets can be found in .
Short overview of datasets used for method validation.
Table 1
### MRI analysis
#### ico brain dm
ico brain dm (version 4.3) is a medical device software that measures relevant volumes of brain structures to assist radiologic assessment of dementia patients. The general ico brain pipeline segments a T1w image into white matter, gray matter and cerebrospinal fluid. When a FLAIR image is available, white matter FLAIR hyper-intensities are also identified and included in the white matter segmentation. The main blocks of the ico brain pipeline have been described previously ( ); in short, after skull stripping and bias correction, the T1w image is segmented using a probabilistic image intensity model and non-rigidly propagated tissue priors from an MNI atlas ( ). Lesion segmentation is obtained as intensity outliers to a probabilistic FLAIR image segmentation, and the tissue segmentation is improved iteratively by re-segmenting the lesion-filled T1w image. Volumes are normalized for head size, using the determinant of the affine transformation to MNI atlas as a scaling factor. ico brain dm further refines this main tissue segmentation in order to obtain sub-segmentations of cortical gray matter lobes and of the hippocampi.
Sub-segmentations of cortical lobes are obtained from the ico brain cortical gray matter segmentation, annotated according to a set of cortical labels available in MNI space ( ). Initial non-rigid registration ( ) between the patient's T1w image and the MNI template is used to obtain a first propagation of the cortical labels from atlas space (“CGM labels”) to the patient's T1w image space. This label propagation is further refined through a second non-rigid registration between the skeleton of the patient's binarized cortical gray matter segmentation and the skeleton of the binarized propagated CGM labels. Finally, each cortical gray matter voxel ias assigned as the cortical label matching the closest voxel in the skeleton of the non-rigidly propagated CGM labels.
Segmentation of the hippocampi starts from the T1w scans preprocessed by the ico brain pipeline, including bias field correction, brain orientation and skull stripping. After preprocessing, a multi-atlas segmentation approach registers binary anatomical priors (i.e., a set of manually annotated hippocampi corresponding to the guidelines of the EADC-ADNI harmonized protocol - ( )) for left and right hippocampi to the T1w image space using an affine and a non-rigid image registration algorithm. The propagated segmentations are then combined into one probabilistic segmentation for each hippocampus. This label fusion is based on a local ranking using the locally normalized cross correlation as a similarity metric ( ). Subsequently, the probabilistic segmentation of each hippocampus is used as a prior in an intensity-based 2-step maximum likelihood expectation-maximization algorithm ( ) to obtain the final hippocampus segmentation. As a post-processing step, voxels mainly considered as CSF by the main tissue segmentation are excluded from the hippocampus segmentation, to keep in line with the EADC-ADNI harmonized protocol, which agreed on excluding internal CSF pools from manual hippocampus segmentation. ico brain dm was executed on a Linux server with 8 CPU cores (Intel Xeon Platinum 8000) and 16 GB RAM, and required between 15 and 30 min per scan to complete.
#### FreeSurfer
The Freesurfer image analysis suite (version 6.0) is documented and freely available for download online ( ) and has been thoroughly described elsewhere ( ; ). In this paper, we used the recon-all stream with fully-automated directive -all , in order to reconstruct all brain volumes, including cortical and subcortical parcellations. Since we used very diverse datasets, they were all processed with identical command and default parameters, without optimizing for a specific dataset (e.g., without −3T or -mprage options).
Cortical labels corresponding to the frontal, temporal and parietal gray matter regions were grouped in order to obtain volumes of the same three cortical lobe regions as for ico brain dm .
When reporting volumes normalized for head size, in order to obtain brain volumes in the same range as ico brain , we performed a scaling of the FreeSurfer volumes using the formula below, where 1985.026 ml is the intracranial volume of the MNI template used in ico brain and ‘Estimated Total Intracranial Volume’ is the total intracranial volume reported by FreeSurfer:
FreeSurfer's more recent functionality for segmentation of hippocampal subfields and nuclei of the amygdala ( ) was also applied on the accuracy datasets 1.a and 1.b, from which volumes of the whole left and right hippocampi were extracted.
FreeSurfer was executed on a Linux server with 16 CPU cores (Intel Xeon Platinum 8000) and 64GB RAM, and required between 9 and 13 h per scan to complete.
Both ico brain and FreeSurfer used only the T1w images as input.
### Validation
ico brain dm and FreeSurfer were validated in terms of accuracy, reproducibility and diagnostic performance of all measures. Accuracy of the hippocampal segmentation received special attention, as it was compared against two different approaches implemented in FreeSurfer. Statistical analyses were performed using the integrated development environment for R programming language, RStudio (version 1.0.136) ( ). Per experiment, significant differences between ico brain dm and FreeSurfer were evaluated using the nonparametric Wilcoxon signed-rank test, using R package ‘MASS’ ( ), at significance level 0.01.
First, we quantified measurement error of all structures and in particular of the hippocampus segmentation with respect to manual ground truth segmentation (datasets 1.a and 1.b). The measurement error was computed as the (absolute) volume difference between ground truth volume and ico brain dm or FreeSurfer volume. In addition, accuracy of the hippocampal segmentation was assessed by the Dice similarity coefficient (DSC). DSC was used to measure the similarity between the ground truth and the automatic segmentation results separately for left and right hippocampus and for total hippocampal volume for each method. According to ( ) a DSC of 0.80 can be considered a good accuracy value, since it was measured by previous studies as the average rate of similarity between two manual hippocampus segmentations performed by experienced operators.
Subsequently, we assessed reproducibility of all measures on test–retest images from cognitively healthy subjects (dataset 2), based on the absolute volume difference between these pairs of images.
Finally, the diagnostic performance of the measures to distinguish AD patients and cognitively healthy subjects was evaluated (dataset 3) by means of a receiver operating characteristic curve (ROC) analysis with DeLong tests at significance level 0.05, using the ‘pROC’ package ( ).
## Results
### Accuracy of brain (sub)structures segmentation
illustrates the accuracy results for the brain segmentation obtained by ico brain dm and FreeSurfer on dataset 1.a (MICCAI 2012 challenge). These results are also summarized in . It is obvious that several volumes are biased with respect to the ground truth volumes obtained from manual segmentation, and ico brain dm and FreeSurfer typically have the same bias direction (i.e. underestimation for WB, GM and the cortical lobes), with the exception of the hippocampi, where FreeSurfer's default hippocampus segmentation overestimates most of the volumes. On the other hand, FreeSurfer's hippocampal subfield functionality underestimates them. For all measurements, ico brain dm has lower bias and lower absolute error. Moreover, there are fewer outliers.
Scatter plots illustrating the brain volumes segmentations by ico brain dm and FreeSurfer (including FreeSurfer's hippocampal subfield functionality, denoted “FS subfields”) compared to expert manual segmentation on dataset 1.a.
Fig. 1
Accuracy of volumes obtained by ico brain dm and FreeSurfer when compared with expert manual segmentation on dataset 1.a (MICCAI 2012 challenge), where volume differences are computed as ground truth segmentation volume minus volume computed automatically by ico brain dm , FreeSurfer or FreeSurfer's hippocampal subfield functionality, “FS subfields”.
Table 2
### Accuracy of hippocampus segmentation
Continuing with the dataset 1.a, we report the DSC for hippocampus segmentations for ico brain dm at 0.8223 (0.8142; 0.8321) (median and interquartile range), while FreeSurfer's default hippocampus segmentation scores a DSC of 0.7988 (0.7867; 0.8158). FreeSurfer's newer hippocampal subfield functionality ( ) scores a slightly lower DSC of 0.7953 (0.7867; 0.8092).
illustrates the accuracy of the hippocampus segmentation obtained by ico brain dm and FreeSurfer (dataset 1.b), with panel A showing the absolute volume difference from ground truth, panel B the DSC, and panel C scatter plots of automated measurements versus manual ground truth. These results are also summarized in . The median absolute volume difference of ico brain dm was significantly lower than that of FreeSurfer's default stream and FreeSurfer's hippocampal subfield functionality, which is also supported by a significantly higher DSC for ico brain dm compared with FreeSurfer methods. It should be noted that 44/46 subjects had a DSC above 0.80 when segmented by ico brain dm compared with 35/46 subjects for FreeSurfer and 42/46 for FreeSurfer's hippocampal subfield functionality.
Accuracy of hippocampus segmentation by ico brain dm and FreeSurfer, including FreeSurfer's hippocampal subfield functionality, denoted “FS subfields”, when compared with expert manual segmentation on dataset 1.b. A. Absolute volume difference between manual and automated segmentation. B. Dice similarity coefficient between manual and automated segmentation. C. Scatterplots comparing ground truth volumes to those obtained from ico brain dm and FreeSurfer. Note: p- values are obtained from Wilcoxon signed-rank tests.
Fig. 2
Accuracy of hippocampus segmentation by ico brain dm and FreeSurfer, including FreeSurfer's hippocampal subfield functionality, denoted “FS subfields”, when compared with expert manual segmentation on dataset 1.b (only hippocampal segmentations), where volume differences are computed as ground truth segmentation volume minus volume computed automatically by ico brain dm , FreeSurfer or “FS subfields” software.
Table 3
shows two illustrations of hippocampus segmentations by ico brain dm and FreeSurfer with high and low DSCs, respectively.
Illustrations of hippocampus segmentation by an expert (ground truth), ico brain dm, and FreeSurfer from dataset 1.b. The top panel shows segmentations with high Dice similarity coefficient (0.90 for ico brain dm , 0.84 for FreeSurfer and 0.85 FreeSurfer's hippocampal subfield functionality), while segmentations with lower Dice similarity coefficients are presented in the bottom panel (0.79 for ico brain dm , 0.77 for FreeSurfer and 0.75 for FreeSurfer's hippocampal subfield functionality).
Fig. 3
### Reproducibility
illustrates the absolute volume differences between test and retest scans (dataset 2) for all measures. Detailed volume differences are presented in . The segmentations obtained by ico brain dm systematically tended to have lower volume differences than FreeSurfer, except for parietal lobe volume, with significant differences for whole brain, total gray matter, and hippocampal volumes.
Reproducibility of segmentations by ico brain dm and FreeSurfer on dataset 2, measured by the absolute volume difference between test-retest segmentations. Note: P values are obtained from Wilcoxon signed-rank tests.
Fig. 4
Reproducibility of segmentations by ico brain dm and FreeSurfer on dataset 2, measured by the absolute volume difference in millilitres between test and retest quantifications.
Table 4
### Diagnostic performance
As shown in , all measures from both ico brain dm and FreeSurfer have high area under the curve (AUC) levels to distinguish AD patients from cognitively healthy controls (dataset 3). Temporal lobe volume measured by ico brain dm produced the highest AUC (0.9896), which was significantly higher than the temporal lobe AUC produced by FreeSurfer (0.9565, P = 0.04646).
Diagnostic performance to differentiate AD patients from age-matched controls on dataset 3.
Table 5
## Discussion
In this paper, the automated method ico brain dm for measuring brain volumes is presented and compared to the widely used FreeSurfer. In order to assess the use of this method in clinical practice on MRI scans of individual dementia patients, the reliability of the method is evaluated in terms of accuracy, reliability and diagnostic performance of all measures. Results are compared to FreeSurfer, a well-validated and extensively used method for measuring brain volumes in clinical studies and trials. ico brain dm and FreeSurfer results on dataset 1.a demonstrated bias in most volumes compared to manual delineations. A systematic bias is not dangerous as such, because volumes obtained with a certain automated software would typically only be compared with the same software between patient groups or between patients and healthy controls. A reason for bias to manual delineations could be the absence of partial volume effect in the manual ground truth. Both ico brain dm and FreeSurfer compute theirs volumes from probability maps, where the voxels close to the brain contour are partly brain tissue, partly CSF, without sharp edges.
Hippocampus segmentation showed however a divergent trend between the 2 automated methods, with FreeSurfer's default stream overestimating most volumes, and ico brain dm slightly underestimating them. On the other hand, FreeSurfer's hippocampal subfields segmentation module ( ), which is currently included in FreeSurfer's development version and thus is not yet the default algorithm, underestimates the considered manual segmentations slightly more than ico brain dm . A recent paper ( ) reported state-of-the-art hippocampus segmentation results using deep convolutional neural network (CNN) ensembles, reaching a Dice score of 0.88 on the same MICCAI 2012 challenge dataset. However, the authors had to tune their CNN with transfer learning on a training subset of the MICCAI 2012 challenge dataset in order to reach these maximal performance results. Deep learning is increasingly superior to classical brain segmentation approaches, but it is limited by the amount, the diversity and the quality of the data used for training. ico brain dm results on dataset 1.b demonstrated a small measurement error for hippocampus segmentation, with a median absolute volume difference from ground truth of 0.230 ml. The similarity with ground truth was generally high, with a median DSC of 0.87 and 44/46 segmentations with a DSC above 0.80. The accuracy of ico brain dm was significantly higher than that of both the default hippocampal segmentation in FreeSurfer 6.0 recon-all stream and FreeSurfer's hippocampal subfields segmentation module ( ), confirming the same trends observed in dataset 1.a.
Bias in hippocampal volumes between automated methods and manual annotations is not surprising, since not all methods and all manual raters use the same definition of the hippocampal borders seen on MRI. The recent EADC-ADNI harmonized protocol ( ), which is used for the multi-atlas approach of ico brain dm , is more clearly defined compared to prior protocols, but it differs from the Center for Morphometric Analysis (CMA) guidelines ( ) underlying FreeSurfer's probabilistic atlas used by the default recon-all stream. Other recent studies such as ( ) found that FreeSurfer 6.0 overestimates the hippocampal volume by 20% compared to manual raters, which is explained by the fact that FreeSurfer includes further caudal regions, resulting in larger tails, as well as some voxels between hippocampus and lateral ventricles. On the other hand, the newer FreeSurfer hippocampal subfields segmentation module ( ) is based on a quite different definition of the hippocampal formation at the subregion level, using ultra-high resolution ex vivo MRIs. The total hippocampal volume obtained with this approach underestimates the volumes obtained from manual segmentations in both accuracy datasets considered in this paper. A potential explanation for this bias towards smaller volumes is that the hippocampus subfield atlas was built using elderly subjects, and was based on a detailed ex vivo MRI delineation protocol that cannot be performed on in vivo brain scans.
The test-retest error on dataset 2 was lower for ico brain dm for all measures except parietal lobe volume, although these differences were significant only for whole brain, total gray matter, and hippocampal volumes. Regarding hippocampal volume, the average test-retest absolute volume difference of the hippocampus is 0.111 ml, which represents 1.20% of the average ico brain dm hippocampal volume (measured by ico brain dm ; test and retest combined). As such, the measurement error is below the average annual hippocampal atrophy rates of 1.41% in healthy individuals ( ). For FreeSurfer's hippocampal subfields segmentation, which we explored in the accuracy experiments ( ), reported test-retest reliability of around 2.5% for the whole left and right hippocampus.
It should also be noted that test-retest exercises are usually performed with datasets on the same scanner. In this manuscript we evaluated test-retest reliability on different scanner types. This increases variability and is better in line with clinical practice.
Finally, when using dataset 3, we found that all measures achieve high diagnostic performance levels when discriminating AD patients from cognitively healthy controls. The temporal lobe volume measured by ico brain dm reached the highest diagnostic performance level (AUC = 0.9896). Although hippocampal atrophy is considered the most disease-specific for Alzheimer's disease, it is not surprising that this structure has slightly lower diagnostic performance compared to the temporal lobe volume, since lower volumes (such as hippocampus) are likely affected by proportionally higher measurement errors. Moreover, not all subjects had severe dementia, as dataset 3 consisted of mild-moderate probable AD.
Of note, the frontal lobe produced the lowest diagnostic performance levels, with FreeSurfer showing stronger differences compared to ico brain dm . In fact ico brain dm finds the frontal cortex volumes in this particular dataset as being close to normal values for that age. As this region is the least of all included measures affected in AD ( ), this result is in line with our expectations.
We conclude that due to its low measurement error, ico brain dm could be of added value to the clinical diagnostic practice of AD patients. In future studies the performance of the measures to diagnose (very) early stages of AD as well as to distinguish between different dementia illnesses should be further investigated.
## CRediT authorship contribution statement
Hanne Struyfs: Formal analysis, Investigation, Writing - original draft. Diana Maria Sima: Methodology, Software, Validation, Formal analysis, Writing - original draft, Writing - review & editing, Supervision. Melissa Wittens: Resources, Data curation, Validation, Writing - review & editing. Annemie Ribbens: Methodology, Project administration, Funding acquisition. Nuno Pedrosa de Barros: Methodology, Software, Validation, Writing - review & editing. Thanh Vân Phan: Methodology, Software, Validation, Writing - review & editing. Maria Ines Ferraz Meyer: Resources, Data curation, Software. Lene Claes: Resources, Data curation. Ellis Niemantsverdriet: Resources, Data curation. Sebastiaan Engelborghs: Writing - review & editing, Supervision, Funding acquisition. Wim Van Hecke: Conceptualization, Funding acquisition. Dirk Smeets: Conceptualization, Supervision, Project administration.
## Declaration of Competing Interest
The following authors are employed (or have been employed at the time of performing the work relevant for this paper) by ico metrix : Hanne Struyfs, Diana M. Sima, Annemie Ribbens, Nuno Pedrosa de Barros, Thanh Vân Phan, Lene Claes, Maria Ines Ferraz Meyer, Wim Van Hecke, Dirk Smeets. Melissa Wittens and Ellis Niemantsverdriet have no competing interests. Sebastiaan Engelborghs has received unrestricted research grants from Janssen Pharmaceutica NV and ADx Neurosciences (paid to institution).
|
Highlights
Differential tractography allows for quantification of neurodegeneration in HD.
Neurodegeneration can be detected before the onset of clinical manifestations.
UHDRS scores correlated with Differential Tractography in longitudinal data.
Huntington’s disease (HD) is a neurodegenerative disorder characterized by motor, psychiatric, and cognitive symptoms. Due to its diverse manifestations, the scientific community has long recognized the need for sensitive, objective, individualized, and dynamic disease assessment tools. We examined the feasibility of Differential Tractography as a biomarker to evaluate correlation of symptom severity and of HD progression at the individual level. Differential tractography is a novel tractography modality that maps pathways with axonal injury characterized by a decrease of anisotropic diffusion pattern. We recruited sixteen patients scanned at 0-, 6-, and 12-month intervals by diffusion MRI scans for differential tractography assessment and correlated its volumetric findings with the Unified Huntington’s Disease Rating Scale (UHDRS). Deterministic fiber tracking algorithm was applied. Longitudinal data was modeled using the generalized estimating equation (GEE) model and correlated with UHDRS scores, in addition to Spearman correlation for cross-sectional data. Our results show that volumes of affected pathways revealed by differential tractography significantly correlated with UHDRS scores in longitudinal data ( p-value < 0.001), and chronological changes in differential tractography also correlated with the changes in UHDRS ( p-value < 0.001). This technique opens new clinical avenues as a clinical translational tool to evaluate presymptomatic and symptomatic gene positive individuals. Our results provide support that differential tractography has the potential to be used as a dynamic imaging biomarker to assess at the individual level in a non-invasive manner, disease progression in HD. Critically important, differential tractography proves to be a quantitative tool for following degeneration in presymptomatic patients, with potential applications in clinical trials.
## Introduction
Huntington’s disease (HD) is a progressive chronic neurodegenerative disorder, resulting from a mutation in the huntingtin gene consisting of a CAG repeat expansion. The resulting protein has an expanded glutamine repeat near the N-terminus, resulting in a toxic gain of function. No effective treatment is available for HD, and the disease is universally fatal. Hallmarks of HD include choreic movements, extrapyramidal motor abnormalities, and cognitive impairment. HD patients may also present with behavioral abnormalities including anxiety, depression and compulsive behaviors ( ). A reliable approach to evaluate disease severity and progression has been challenging in HD. The assessment of the severity of clinical symptoms relies mostly on the Unified Huntington's Disease Rating Scale (UHDRS) for disease stage stratification ( ). UHDRS evaluates the motor, cognitive, behavioral, and functional capacity allowing for a quantitative assessment based on clinical presentation. Despite the usefulness of UHDRS, there is still an ongoing need for an objective imaging biomarker to assess disease onset, progression, and severity.
Studies such as PREDICT-HD ( ) and TRACK-HD ( ) have used standard MRI to quantify gross structural findings and investigate the correlation between neuroimaging findings with cognitive and biological imaging and motor outcome measures. TRACK-HD correlated volumetric MRI with UHDRS in premanifest and manifest patients. Both PREDICT-HD and TRACK-HD confirmed previous reports supporting the value of imaging markers, especially of striatal and whole-brain atrophy during the premanifest stage ( , ). A recent study ( ) that applied fixel-based analysis in a large sample of premanifest individuals suggests that white matter structures such as the cortico-basal ganglia display signs of degeneration or “vulnerability” at around 11–25 years after diagnosis, with preserved integrity as early as 25 years before diagnosis has been established. In addition, the aforementioned study observed that the sensory and motor components of the thalamus and the limbic and motor striatum have demonstrated to be at risk in this population, suggesting that clear observable white matter changes at the voxel level can be demonstrated years after diagnosis and not during the premanifest phase. These interesting findings allow the opportunity for the emergence of biomarkers capable of detecting onset of neurodegeneration before clinical manifestations, which in turn, with early initiation of disease modifying therapies, can potentially represent a better quality of life for these patients. Other studies have shown that white matter atrophy is evident in T1-weighted MRI with posterior-frontal white matter degeneration evident in at-risk individuals far from disease onset ( ). Structural MRI allows for the examination of gradual changes that occur in premanifest HD with MRI studies showing that these subjects have brain atrophy years before disease manifestation in pyramidal projection neurons in the motor and prefrontal cortices, and cingulate and angular gyri ( , ). However, volumetric findings in the above-mentioned studies applied a group-based approach and individual difference are of the utmost importance for clinical applications. Although it is recognized that structural MRI has been sensitive to measure for neuronal loss and total volume measurement in grey and white matter cortical areas ( ), there is still ongoing efforts to increase the search for specific markers for localization of volume loss and atrophy ( ).
White matter changes have been studied by implementing diffusion MRI to explore its clinical value in neurodegeneration. Techniques such as Diffusion Tensor Imaging (DTI) ( ) are capable of detecting structural changes in axonal pathways in HD patients ( , , , , , , , , ). Disruption of several white matter pathways including cortico-striatal motor projections, cingulum, uncinate fasciculus, thalamocortical projections, corpus callosum, and corticospinal tract, have been found in HD ( , , , , , , ). Furthermore, cognitive and motor parameters correlated with white matter DTI alterations in several studies ( , , , , , , , ). DTI remains a commonly used technique to study structural white matter changes in neurodegeneration, however, its clinical applications are limited due to its inability to resolve complex fiber orientations in the presence of free water (i.e. CSF volume acting as an artifact) ( , , ), while still only demonstrating a difference in HD patients at a group level when compared to a control population. Recent studies have applied beyond-DTI methods such as fixel-based analyses using constraint spherical deconvolution (CSD) in early HD ( , ) and in premanifest HD ( ) to identify neurodegeneration in HD. However, although the fixel-based approach provides a high angular resolution advantage ( ), several technical considerations need to be taken to avoid a critical flaw in tractography clinical studies ( ). Furthermore, the acceptance that DTI-based metrics are non-specific for neurodegeneration and disease progression assessment, warrants the opportunity to move beyond DTI-based approaches ( , , ).
Recently advanced diffusion MRI has acquired a more sophisticated diffusion model by resolving multiple diffusion sensitization and hundreds of diffusion sampling directions ( ). This significant improvement has allowed to resolve complex fiber orientation by using or resorting to a nonparametric approach ( ). This has led to the development of beyond-DTI tractography that can handle crossing-fibers ( ) and cope with the partial volume of free water ( ). Beyond-DTI tractography has been used in patients with aphasia to demonstrate a clear functional correlation of tractography white matter fiber bundles such as the arcuate fasciculus (AF), inferior longitudinal fasciculus (IFOF), uncinate fasciculus (UF), and middle longitudinal fasciculus (MdLF) with semantic and phonological abilities involved in language production ( ). Conventional tractography is not sensitive during early neuronal degeneration as tractography differences can only be demonstrated if anisotropy drops substantially below the tracking threshold, and although diffusion MRI has been explored as a potential biomarker for early onset neurodegeneration, anisotropy as a measurement at the voxel level is susceptible to local variability including but not limited to partial volume effect ( , , ) restricting its potential in the clinical setting ( , ). Our recent study demonstrated that differential tractography ( ) addressed these limitations by focusing on differences in anisotropy to track only the segment of the pathway with neuronal degeneration. In the aforementioned study, the analysis required two longitudinal scans of the same subjects to derive differences, but in the present study we implemented an advanced protocol that compared one patient’s scan with a cohort of control subjects. Differential tractography accomplishes a substantial improvement when compared to conventional tractography. The method performs a comparison of voxel-wise differences of diffusion properties, such as quantitative anisotropy (QA), allowing to only track changes resulting in highlighted tractograms with segments of degeneration. Volumes extracted from obtained tractograms result in a simple measurement of the amount of neurodegeneration. The volume of specific pathways with a decrease in anisotropy was used as a quantitative biomarker to correlate with clinical UHDRS scores. This novel modification allowed us to derive a numeric value of altered pathways for individual patients, hence enabling the opportunity to study the advantages of a true diffusion-based analysis technique as a clinical translational biomarker for early neuronal injury, in contrast to other technique such as Tractwise Fractional Anisotropy Statistics (TFAS) which applies fiber tracks as a skeleton to obtain underlying voxel Fractional Anisotropy (FA) for statistical analysis ( ). Furthermore, since differential tractography tracks neuronal injury along a fiber pathway, this provides the ability to differentiate true findings from errors occurring at the local voxel level, as errors generated locally stay withing the local limits, versus true neuronal injury that disseminates along axons ( ).
In the present study we applied differential tractography in pre-manifest and manifest HD to localize differences in anisotropy between base and repeat scans, along with statistical correlation of anisotropic differences with UHDRS clinical scores, including total motor score (UHDRS TMS), dystonia total, chorea total, rapid alternating movements (RAM), stroop color word, behavior, and total functional capacity (UHDRS TFC). In addition, compromised fiber pathways in HD patients were identified by comparing them with healthy controls and quantifying the volume of each affected pathways as a biomarker. Although we have not taken in consideration any hypothesis to specific white matter bundles, our study is exploratory, and we hypothesized that pathway alterations would correlate with the UHDRS in both cross-sectional and longitudinal settings. We demonstrate progressive degeneration as subjects were imaged at several time points and provide evidence that differential tractography can be used as a dynamic biomarker for progressive structural damage, which correlates with disease progression in HD patients.
## Materials and methods
### Patient characteristics and demographics
We recruited sixteen patients, including twelve manifest HD patients and four pre-manifest patients ( ). All patients gave their informed consent prior to their inclusion in the study. Manifest were symptomatic and pre-manifest were asymptomatic (all confirmed gene positive). Patients had three scans over a period of two years. Twelve patients had three scans, one patient had two scans, and three patients had one scan. The average scan interval from the first to the second scan was 6 ± 0.4 months (range 5 to 10 months) and the average scan interval from the first to the third scan was 12 ± 1 months (range 11 to 24 months). Patients underwent a comprehensive clinical evaluation on the day of the scan conducted by a neurologist specializing in movement disorders. Previous to each MRI, subjects were evaluated to assess their Unified Huntington Disease Rating Scale (UHDRS) ( ) scores, including motor, behavior, cognitive and functional assessments. A reconstructed averaged template was included from the CMU-60 database, a compiled diffusion MRI dataset of 60 healthy individuals acquired with a 257-diffusion sampling direction that served as control for our study.
Patient demographics.
### MRI acquisition
Diffusion spectrum imaging data were acquired on a 3 T Tim Trio System (Siemens, Erlangen, Germany) using a 32-channel coil. A head stabilizer was utilized to prevent head motion. A 25 min, 257-direction DSI scan with a twice-refocused spin-echo planar imaging sequence and multiple b values (repetition time = 9916 ms, echo time = 157 ms, voxel size = 2.4 mm × 2.4 mm × 2.4 mm, field of view = 231 mm × 231 mm, maximum b -value = 7000 s/mm ) was performed. For anatomical comparison, we included a high-resolution anatomical image using a 9-min T1-weighted axial MPRAGE sequence (repetition time = 2110 ms, echo time = 2.63 ms, flip angle = 8°, number of slices = 176, field of view = 256 mm × 256 mm, voxel size = 0.35 mm × 0.5 mm × 1.0 mm). We have used the same scanner as in the control population (CMU 60). The potential impact of the use of different scanners is addressed in the discussion section.
#### Differential tractography for individuals
The flowchart of our revised differential tractography ( ) analysis is demonstrated in . Diffusion imaging data of each patient ( A) was reconstructed to a common stereotaxic space using q-space diffeomorphic reconstruction (QSDR) ( , ), which is a method that satisfies the conservation of diffusible spins and reconstructs diffusion MRI data in a common standard space. QSDR was applied to generate the density distribution of anisotropic diffusion ( B). The red–greenblue colors represents the orientation of diffusion (red: left–right, green: anterior-posterior, blue: superior-inferior). QSDR allowed us to calculate the differences in anisotropic diffusion by comparing it with a normal population database (CMU-60 database, C) ( , ) to show the locations of local fibers with a decrease of anisotropic diffusion in study subjects, indicating changes in fiber integrity. We used a percentile rank lower than 5 of the decrease in anisotropy as the threshold to filter the results. D shows the piecewise fibers (color-coded by orientation) with substantial decreases, which were connected to guide the fiber tracking algorithm to map the exact segment of affected fiber bundles ( E). Fiber bundles were segmented based on a recent tractography atlas ( ) by using DSI Studio’s interface for manual fiber tracking and we cross referenced with the average population atlas. The tracking was determined using a deterministic fiber-tracking algorithm ( ) in our proprietary developed and open-source software DSI Studio ( ). Deterministic tractography applies quantitative anisotropy (QA) which relies on generalized q-sampling imaging (GQI) to estimate the orientation of individual fibers ( ), and spin distribution function (SDF) to provide the amount or density of diffusing water in any direction within a single voxel ( ), therefore posing a great advantage over widely used diffusivity-based estimations such as FA. The tracking begins from each local fiber orientation as seeds and propagates until no orientation is found in the propagation direction. A maximum turning angle of 60° was used with a step size of 1 mm. The determined trajectories, termed the affected tracts, are used to identify pathways with decreased connectivity.
Flowchart of the differential tractography analysis. The red–greenblue colors represents the orientation of diffusion (red: left–right, green: anterior-posterior, blue: superior-inferior). (A) The diffusion data of each subject was reconstructed in a common stereotaxic space using q-space diffeomorphic reconstruction (QSDR) to calculate the diffusible spin distribution function. (B) The reconstruction allows the visualization of the density distribution of the anisotropic diffusion of the subject in the standard space. (C) Values from a normal population are compared to the study subject to calculate the percentile rank. (D) Comparison is then used to map locations with anisotropic diffusion of subject sufficiently smaller than normal population (<5 percentile rank). (E) Fiber tracking algorithm is then associated with the data obtained by substantial decrease in anisotropic diffusion to map exact fiber pathways affected by the disease.
### Statistical methods
We conducted a statistical analysis to determine the correlation of the UHDRS scores with quantitative data of each region of interest obtained by differential tractography. Data was evaluated using a one-sided t -test and was then organized by longitudinal and cross-sectional analyses to determine the efficacy of the dynamic biomarker tested and have more control over brain regions tested and their correlation with clinical scores.
Longitudinal measures of subjects were modeled using the generalized estimating equation (GEE) model, a linear model similar to the mixed effect model that can investigate the correlation between tract volume and the clinical scores that evaluated the cognitive levels and severity of the disease. Sandwich estimate of the variance was used to avoid violations of normality.
Using the GEE model, we correlated differential tractography findings and UHDRS total scores for motor, cognitive, behavior, and functional capacity. Since the motor scores include assessments to evaluate the motor dysfunction in detail, we further correlated differential tractography with subscores under the motor assessment, including Total Motor Score (TMS), Dystonia Total, Chorea Total and Rapid Alternating Movements (RAM), to see whether there are meaningful findings specific to these subscores. The same setting was applied to the cognitive component represented by the subscore Stroop Color-Word. Lastly, the UHDRS Behavioral Total, and TFC (Total Functional Capacity) scores were correlated. Using our novel method, we obtained several tract bundles with decreased anisotropy. All bundles obtained by differential tractography were further segmented into five different white matter regions, which included cingulum, corpus callosum, corticostriatal pathway, corticospinal pathway, and the whole brain. This allowed us to study region-specific correlation.
Targeted fiber tracking analysis was performed for each scan using their corresponding differential tractography results. Quantitative data such as tract volume for each segmented region was registered as a reference for tract involvement, higher volumes indicate greater magnitude of affected tracts.
Overall, a total of 35 comparisons were performed to determine statistical correlation, which translate to 35 hypotheses, one for each longitudinal and cross-sectional analyses. For cross-sectional analysis, we correlated the volume extracted from each fiber bundle with each clinical score. Furthermore, we correlated the change in tract volume with the change in clinical scores which yielded 35 hypotheses. Each hypothesis was tested in repeat scans of pre-manifest and manifest subjects using the GEE model. We also studied these 35 correlation hypotheses for each scan time point (scans 1, 2, and 3) as three independent cross-sectional studies using the Spearman correlation model, a nonparametric method to investigate the correlation using the rank of the tract values.
The longitudinal change in tract volume and the clinical scores of the above-mentioned 35 correlation hypotheses, were also studied using the GEE model for the manifest patients. Three separate Spearman correlation analyses were conducted to study the change between scan one and scan two, scan one and scan three, and scan two and scan three. The hypothesis was tested using a one-sided tail t -test. A p-value of 0.05 was corrected using Bonferroni correction to obtain familywise significance and eliminate false positive results, yielding a p-value of 0.001 or less to be considered statistically significant. All analyses were conducted in SAS 9.3.
The statistics of this study and its interpretation were supervised by a statistician (YF. C.).
## Results
### Individual differential tractography results
shows differential tractography volume measurements of cingulum, corpus callosum, corticostriatal pathway, corticospinal pathway, and whole brain in all manifest and premanifest subjects, which were mapped automatically by differential tractography. As noted in , increased tract volumes (mm3) denote reduced tract integrity compared to normal population. The color red in helps to differentiate tracts with higher volume (dark red color) from tracts with lower volumes (light red color). The UHDRS Total Motor Score (TMS) and differential tractography results were assessed independently. Differential tractography progression was demonstrated in nine out of twelve manifest subjects (75%), and in one out of four premanifest subject (25%) with a time-dependent increased volume of affected tracts. Subjects A, B, and C were selected to demonstrate a correlation based on their UHDRS TMS, in which higher deteriorating motor function was evident ( ). Higher UHDRS TMS indicates worse performance, and all three subjects demonstrated an increased volume of affected tracts, likely correlating with decreased connectivity ( ). This progression corresponded with UHDRS TMS higher scores at each measurement, with the exception of subject C, in which an increase in the volume of degenerating tracts did not correspond with UHDRS TMS, remaining unchanged at 6-months compared to the baseline scan. To visualize inter-individual variability, please refer to .
Tract volume measurements in HD subjects.
Fiber pathways affected in three manifest subjects mapped by differential tractography. The red–greenblue colors represents the orientation of diffusion (red: left–right, green: anterior-posterior, blue: superior-inferior). Tract volumes are represented in mm The greater the volume of affected fibers, the higher the UHDRS Total Motor Score (UHDRS TMS) that subjects will display, showing a deteriorating performance in motor functions. Subject A displays a significant correlation between fiber pathways affected and UHDRS TMS (35, 41, and 58) with increasing tract volumes (13,753 mm , 13,656 mm , and 46,728 mm ) in three different scanning time points respectively (0 month, 6 months, 1 year). Subject B shows a UHDRS TMS of 45, 49, and 52 with tract volumes of 34,840 mm , 53,288 mm , and 57,560 mm at 0 month, 6 months, and 1 year, respectively. Subject C is the subject with the most change among the three, with a UHDRS TMS of 35, 35, and 64 and tract volumes of 22,168 mm , 61,384 mm , and 68,488 mm at 0 month, 6 months, and 1 year, respectively. Interestingly, subject C shows no change in the UHDRS TMS between 0 and 6 months, nevertheless, a significant increase in tract volume was observed in this time period (22,168 mm and 61,384 mm respectively), providing evidence that differential tractography can be used as a dynamic biomarker to predict pre-clinical manifestations.
### Manifest versus premanifest patients
Significant differences were observed in the manifest and premanifest group. Initial scans in symptomatic patients demonstrated a significant number of affected bundles. In contrast none or a small number of affected tracts in the premanifest group ( ). These results provide further validation of this technique in identifying affected pathways and distinguishing presymptomatic from symptomatic patients.
### Longitudinal versus cross-sectional analyses
Longitudinal data was evaluated to determine the correlation between affected tract volumes and UHDRS clinical scores. We performed two longitudinal analyses and the time frame was 6–12 months. First, we studied the correlation between UHDRS clinical scores and tract volumes in each brain region (cingulum, corpus callosum, corticostriatal pathway, corticospinal pathway, and whole brain). In addition, a second longitudinal analysis was performed to examine the correlation between change in clinical scores and the change in volumes of tracts, including cingulum, corpus callosum, corticostriatal pathway, corticospinal pathway, and whole brain.
Out of 35 correlations in our initial longitudinal analysis, twelve (34.3%) showed statistical significance between tract volume and clinical scores, which included 2 correlations (5.7%) with statistical significance ( p-value < 0.001) and ten correlations (28.6%) statistically significant ( p-value < 0.0001). In addition, all brain bundles (cingulum, corpus callosum, corticostriatal pathway, corticospinal pathway, and whole brain) significantly correlated with clinical scores as follows. UHDRS TMS was statistically significant ( p-value < 0.0001) in cingulum and corticostriatal pathway; RAM was significant with cingulum, corpus callosum, corticostriatal and corticospinal pathway ( p-value < 0.0001), and with whole brain ( p-value < 0.0001); Stroop color-word was significant with Cingulum and Corpus Callosum ( p-value < 0.0001); and UHDRS TFC was significant with cingulum and corticostriatal pathway ( p-value < 0.0001), and with corticospinal pathway ( p-value < 0.001) ( ). Supplementary represent visual components of for better visualization.
Correlation analysis between tract volume and clinical scores in longitudinal data.
In the second longitudinal analysis, seven correlations (20%) were statistically significant ( p-value < 0.0001) as follows. A statistical significance was observed in dystonia total with cingulum and corpus callosum ( p-value < 0.0001); and RAM showed significance in all brain regions ( p-value < 0.0001). Results from this analysis supports differential tractography as a practical and accurate biomarker for evaluating changes in volume of different brain regions in relation to clinical scores ( ).
show results for cross-sectional and longitudinal data in manifest patients. Correlation analysis was applied to evaluate the relationship between clinical scores and tract volumes in cross-sectional data from the first, second, and third scans, yielding a total of 105 correlations that were corrected using Bonferroni correction to consider familywise significance. In addition, shows results of correlation analysis which was applied to evaluate the relationship between the changes in all clinical scores and the changes in tract volumes in three separate groups: (1) changes observed from first to second scan, (2) changes observed from the first to the third scan, and (3) changes observed from the second to the third scan. No statistical significance was observed when tract volumes were compared to clinical scores in cross-sectional data, or when tract volumes were compared to the changes in clinical scores in longitudinal data ( , ). However, it is unlikely to achieve significance when familywise p-value is considered (Bonferroni correction) due to the small number of subjects included in the analysis, and this does not diminish the important findings obtained in the longitudinal analysis ( ). For better visualization, supplementary represent visual components of and , respectively.
Cross-sectional correlation between clinical scores and tract volume.
Cross-sectional correlation between change in clinical scores and the change in tract volumes.
Since this study does not examine individual fiber tracts, but rather white matter bundles as a group, any findings are common between subjects in majority. In addition, since subject data was normalized to a common space, the overlapped findings were examined and groupwise statistical significance was common in the population.
## Discussion
In this study we evaluated differential tractography as a clinical translational tool by conducting correlation analyses between white matter volumes measurements and clinical scores in manifest and premanifest HD patients. Overall results indicate that differential tractography appears to be a robust dynamic biomarker with high statistical significance in longitudinal data to determine changes in tract volumes of white matter tracts with the potential to supplement the UHDRS in manifest and premanifest HD. Differential tractography appears to be a highly reliable monitoring biomarker to delimit changes exhibited in cingulum, corpus callosum, corticostriatal pathway, corticospinal pathway, and whole brain when correlated with UHDRS. Moreover, an increase of volume of damaged tracts was observed before symptom onset in one particular subject (Subject C, ). This prediction power can be taken in consideration to anticipate onset at the premanifest stage to characterize disease progression, adding great value and high reliability to differential tractography as a predictive monitoring biomarker. The use of different scanners will have introduced a fixed bias in our correlation analysis, bringing the same intercept for our variable. Thus, the correlation coefficient will not have been affected, since the scanner difference is the same for all subjects, and our hypothesis would have remained the same.
It is also noteworthy the distinction that differential tractography provides when comparing manifest vs premanifest individuals, as affected tracts in manifest subjects displayed higher tract volumes as expected, in opposition to gene-positive individuals who yielded few or no tracts at all (as seen in Supplementary ), supporting the accuracy of the technique. Since differential tractography findings are associated with damage appearing in pathway trajectories, the technique provides the amount of degeneration (in volume measurements) in addition to providing a better localization of disease. In this sense, we have been able to map segments of dysconnectivity in white matter areas to find the link between lesions and grey matter to better understand functional changes due to neuronal degeneration. Therefore, our diffusion-based analysis technique exhibits a significant novelty over conventional tractography by differentiating errors in local voxels versus true findings that spread along a fiber trajectory. This in turn, provides a biological advantage for not only localizing neurodegeneration with precision, but also for tracking the evolution of disease and treatment response, as suggested by a previous study ( ). By applying deterministic tractography, we have an advantage by the novelty of the technique which is capable of resolving crossing fibers. Deterministic tractography makes use of quantitative anisotropy (QA) which relies on generalized q-sampling imaging (GQI) to estimate the orientation of individual fibers ( ), and spin distribution function (SDF) to provide the amount or density of diffusing water in any direction within a single voxel ( ), therefore posing a great value over widely used diffusivity-based estimations such as FA. The use of differential tractography paired with a robust clinical evaluation at the pre-clinical stage in gene positive asymptomatic populations, can be of utmost clinical significance in routine clinical follow-up, and although our study provides a small number of subjects, we acknowledge that future studies are granted to obtain robust measures for when assessment of new treatment and therapies are required in clinical trials.
### Implications of the clinical data
Longitudinal analysis demonstrated the highest statistical correlation with progression of clinical UHDRS scores in all brain regions (cingulum, corpus callosum, corticostriatal pathway, and corticospinal pathway), in relation to UHDRS TMS, Stroop Color-Word, UHDRS Total Functional Capacity (TFC), and especially in relation to Rapid Alternating Movements (RAM), which was statistically significant ( p-value < 0.0001) in all brain regions. Since Bonferroni correction was applied to consider familywise significance in multiple comparison analyses, results from the cross-sectional analysis did not yield significant findings. However, this does not hinder the potential of differential tractography as a tool to further understand the biological mechanism of white matter loss, and further studies with larger samples are required to determine a true significant value in cross-sectional data and larger group studies. Despite this limitation, results further confirm the role of white matter pathways involved in HD progression ( , ). Demonstrated changes on differential tractography in both premanifest and manifest HD, and particularly in the earlier stages, may be of value in future longitudinal and cross-sectional studies ( ). In premanifest HD where clinical markers of disease progression do not exist, differential tractography can be used as a non-invasive tool to dynamically monitor clinically asymptomatic disease progression. In manifest HD, the observed disease progression made by differential tractography can be used to supplement existing clinical markers of progression.
### Speculative mechanisms
Degeneration in the association, commissural and projection fibers are implicated in the course of the disease and its clinical manifestations. Degeneration of corticospinal and corticostriatal pathways white matter tracts are linked to changes in motor functions behavior, executive function, movement, and the lack of integration of motor and cognitive function resulting in progression of UHDRS TMS, RAM, Stroop Color Word, UHDRS TFC. Thus, statistical correlation of Corticospinal and corticostriatal pathways with UHDRS TMS, RAM, and UHDRS TFC supports the relationship with motor dysfunction, and studies have supported these findings in manifest HD, and several important behavioral changes such as global apathy have been recently associated with degeneration of corticostriatal pathway ( , ). Statistical correlation exhibited by the corticospinal tract in relation to UHDRS TMS, Stroop Color Word, TFC, and especially with RAM corroborates the critical relationship between corticospinal tract demyelination and motor symptoms at the premanifest and manifest stages which is associated with progression of UHDRS motor scores ( ). The highest correlation found with respect to RAM in longitudinal data studied by differential tractography, is validated by the motor involvement of the disease. Therefore, differential tractography represents a novel monitoring biomarker allowing detection of the exact anatomical location of degeneration and its subsequent correlation with loss of clinical function as measured by existing markers of progression.
### Differential tractography in relation to premanifest and manifest disease and UHDRS scores
Despite the small number of patients, significant differences were observed between the premanifest and manifest HD. Relatively few areas were affected in premanifest patients in relation to patients in the manifest group (as in Supplementary ), thereby lending further credibility to this imaging method. As expected, significant progression was observed at 6 and 12 months in manifest patients in relation to the baseline scan. The observed increase in volume of affected tracts corresponded with an increase in the UHDRS clinical scores. Despite being a reliable gold-standard to determine clinical progression in HD for many years ( ), the UHDRS assessment can be prone to variability. Differential tractography as an automated method is less prone to variability, and can supplement the use of the UHDRS in manifest HD. In premanifest patients, differential tractography can demonstrate changes in white matter preceding disease onset.
### Future directions and limitations
We demonstrate the feasibility of differential tractography as a potential biomarker to anticipate disease onset in premanifest and manifest HD. Our main limitation was the small number of subjects which prevented obtaining statistical significance in cross-sectional data, and this limitation prevented us to obtain a more homogeneous clinical and longitudinal data. Additional research with larger samples is required to obtain a clear validation. We acknowledge that we cannot estimate effect size, as the method is for individual diagnosis which places more emphasis on sensitivity and specificity. Therefore, although differential tractography has potential for group diagnosis, future studies with a greater number of subjects will be necessary to evaluate the effect size at the group level. We did not account for time between scans in the statistical model, and this will certainly be a variable which must be considered in future larger studies. In addition, premanifest HD diagnosis was made based of genetic profile and we did not acquire clinical markers such as CAP score (CAG - Age Product Scaled score) or DBS (Disease Burden Score), for which we will consider obtaining in future studies. There is a mismatch between differential tractography and the UHDRS TMS in few cases as shown by one patient (subject F, ) with decreasing volumes and progression of the UHDRS TMS. In manifest HD, differential tractography demonstrated changes or progression at an anatomical level that may not be readily discernible with UHDRS scores. At this stage we do not have a clear understanding of the nature of the mismatch and thus differential tractography will require further validation in a larger study. Lastly, although differential tractography provides encouraging results to carry future larger studies, we recognize the limitation that differences between manifest and premanifest patients cannot be generalizable due to the small number of subjects in each group. Nevertheless, the overall findings confirmed the applicability of differential tractography as a dynamic non-invasive biomarker. Differential tractography can be considered in future studies with larger cohorts with more homogenous clinical and longitudinal data to assess the efficacy of therapeutic trials particularly in premanifest HD, where future drug trials will be aimed to prevent symptomatic conversion.
## Data availability statement
Data supporting the findings of this study are not publicly available as it contains sensitive information that may compromise the privacy of the participants of this study.
## CRediT authorship contribution statement
Jessica V. Barrios-Martinez: Investigation, Writing – original draft, Writing – review & editing, Visualization. David T. Fernandes-Cabral: Formal analysis, Investigation, Writing – original draft. Kumar Abhinav: Writing – review & editing. Juan C. Fernandez-Miranda: Funding acquisition, Conceptualization, Supervision. Yue-Fang Chang: Formal analysis, Validation, Supervision. Valerie Suski: Data curation, Investigation, Supervision, Validation, Writing – review & editing. Fang-Cheng Yeh: Data curation, Funding acquisition, Methodology, Project administration, Software, Supervision, Writing – review & editing, Visualization. Robert M. Friedlander: Conceptualization, Funding acquisition, Investigation, Supervision, Writing – review & editing.
## Declaration of Competing Interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Dr. Robert M. Friedlander is on the Board of NeuBase Therapeutics and Difusion Technologies. All other authors report no conflict of interest.
|
There is evidence that vascular risk factors contribute to the pathology of Alzheimer's disease. Increased concentrations of circulating homocysteine are associated with vascular risk factors and Alzheimer's disease but the mechanisms involved are unclear. Homocysteine inhibits the hydrolysis of S-adenosylhomocysteine (SAH) which is a product inhibitor of S-adenosylmethionine (SAM) dependent methyltransferase reactions. It has been shown previously that SAH inhibits phosphatidylethanolamine N-methyltransferase (PEMT) in the liver. The activity of PEMT in the liver plays an important role in the methylation of phosphatidylethanolamine (PE) to phosphatidylcholine (PC) and the delivery of essential polyunsaturated fatty acids (PUFAs) to peripheral tissues. In the present study, the plasma concentrations of SAH, SAM and homocysteine and the erythrocyte composition of phosphatidylcholine (PC), phosphatidylethanolamine (PE) and their respective polyunsaturated fatty acid concentrations were determined in 26 patients with Alzheimer's disease and compared to those in 29 healthy control subjects. There was a significant increase in the plasma concentrations of SAH (p<0.001) and homocysteine (p<0.001) and a significant increase in the plasma concentrations of SAM (p<0.001) in the Alzheimer's patients. A significant positive correlation was found between the plasma concentrations of SAH and homocysteine (r=0.738, p<0.001). There was a negative correlation between the plasma concentrations of homocysteine and the ratio of SAM/SAH (r=-0.637, p<0.01). There was a significant decrease in the erythrocyte content of PC (p<0.001) and an increase in the erythrocyte content of PE (p<0.001) in the Alzheimer's patients. Plasma SAH concentrations were negatively related to erythrocyte PC concentrations (r=-0.286, p<0.01) and positively related to erythrocyte PE concentrations (r=0.429, p<0.001). The erythrocyte PC from Alzheimer's patients had a significant depletion of docosahexaenoic acid (DHA) (p<0.001) while there was no significant difference in the DHA content of erythrocyte PE. There was a significant negative correlation between plasma SAH and the DHA composition of erythrocyte PC (r=-0.271, p<0.001). This data may reflect the inhibition of hepatic PEMT activity by SAH in Alzheimer's disease. The decreased mobilization of DHA from the liver into plasma and peripheral tissues may increases the risk of atherosclerosis and stroke leading to chronic cerebral hypoperfusion. The evidence suggests that a metabolic link between the increased production of SAH and phospholipid metabolism may contribute to cerebrovascular and neurodegenerative changes in Alzheimer's disease. |
Alzheimer's disease (AD) is likely to disrupt the synchronization of the bioelectrical processes in the distributed cortical networks underlying cognition. We analyze the surface topography of the multivariate phase synchronization (MPS) of multichannel EEG in 17 patients (Clinical Dementia Rating (CDR) Scale: 0.5-1; Functional Assessment Staging (FAST): 3-4) compared to 17 controls by applying a combination of global and regional MPS measures to the resting EEG. In early AD, whole-head mapping reveals a specific landscape of synchronization characterized by a decrease in MPS over the fronto-temporal region and an increase over the temporo-parieto-occipital region predominantly of the left hemisphere. These features manifest themselves through the EEG delta-beta bands and discriminate patients from controls with an accuracy of up to 94%. Moreover, the abnormal MPS in both anterior and posterior clusters correlates with the Mini Mental State Examination score, binding regional EEG synchronization to cognitive decline in AD patients. The MPS technique reveals that the EEG phenotype of early AD is relevant to the clinical picture and may ultimately become its sensitive and specific biomarker. |
Mutations in the valosin-containing-protein (VCP) gene are associated with the multidisorder disease, inclusion body myopathy with Pagets and associated frontotemporal dementia. This disease is characterized pathologically by large ubiquitinated, TAR DNA Binding Protein 43 (TDP-43) positive inclusions. These inclusions are also a common feature in neurological diseases including amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTLD). Mutations in the VCP gene have been identified in ALS patients, therefore we aimed to characterize VCP variations in our own cohort of familial and sporadic ALS patients by sequencing all 17 coding exons of VCP. This study failed to detect any exonic variations in a subset of British familial and sporadic ALS patients. |
Clusterin gene (CLU), also known as apolipoprotein J (ApoJ), is a strong candidate gene for late-onset Alzheimer's disease (LOAD) according to the Alzgene database. To further characterize this association and to isolate the variants contributing to the pathogenesis of LOAD in Han Chinese, we first sequenced a small sample (n = 100) to discover variants in the promoter, exons, the 5' and 3' untranslated regions, and exon-intron boundaries of CLU. Follow-up genotyping of identified variants in a larger sample (n = 1592). Sequencing analysis identified 18 variants. Analysis in the larger population revealed that only the rs9331949 C allele was significantly associated with an increased risk of LOAD, even after adjusting for multiple testing (p = 0.026). Logistic analysis identified the rs9331949 polymorphism was still strongly associated with LOAD (additive model: p = 0.004, odds ratio = 1.274; dominant model: p = 0.039, odds ratio = 1.239; recessive model: p = 0.002, OR = 1.975) after adjusting for sex, age, and APOE ε4 status. Our findings implicate CLU as a susceptibility gene for LOAD in Han Chinese. |
We tested association of nine late-onset Alzheimer's disease (LOAD) risk variants from genome-wide association studies (GWAS) with memory and progression to mild cognitive impairment (MCI) or LOAD (MCI/LOAD) in older Caucasians, cognitively normal at baseline and longitudinally evaluated at Mayo Clinic Rochester and Jacksonville (n>2000). Each variant was tested both individually and collectively using a weighted risk score. APOE-e4 associated with worse baseline memory and increased decline with highly significant overall effect on memory. CLU-rs11136000-G associated with worse baseline memory and incident MCI/LOAD. MS4A6A-rs610932-C associated with increased incident MCI/LOAD and suggestively with lower baseline memory. ABCA7-rs3764650-C and EPHA1-rs11767557-A associated with increased rates of memory decline in subjects with a final diagnosis of MCI/LOAD. PICALM-rs3851179-G had an unexpected protective effect on incident MCI/LOAD. Only APOE-inclusive risk scores associated with worse memory and incident MCI/LOAD. The collective influence of the nine top LOAD GWAS variants on memory decline and progression to MCI/LOAD appears limited. Discovery of biologically functional variants at these loci may uncover stronger effects on memory and incident disease. |
A recent study in autopsy-confirmed Parkinson's disease (PD) patients and controls revived the debate about the role of PARK10 in this disorder. In an attempt to replicate these results and further understand the role of this locus in the risk and age at onset of PD, we decided to explore NeuroX genotyping and whole exome sequencing data from 2 large independent cohorts of clinical patients and controls from the International Parkinson's Disease Genomic Consortium. A series of single-variant and gene-based aggregation (sequence kernel association test and combined multivariate and collapsing test) statistical tests suggested that common and rare genetic variation in this locus do not influence the risk or age at onset of clinical PD. |
Brain-derived neurotrophic factor (BDNF) has been discussed to be involved in plasticity processes in the human brain, in particular during aging. Recently, aging and its (neurodegenerative) diseases have increasingly been conceptualized as disconnection syndromes. Here, connectivity changes in neural networks (the connectome) are suggested to be the most relevant and characteristic features for such processes or diseases. To further elucidate the impact of aging on neural networks, we investigated the interaction between plasticity processes, brain connectivity, and healthy aging by measuring levels of serum BDNF and resting-state fMRI data in 25 young (mean age 24.8 ± 2.7 (SD) years) and 23 old healthy participants (mean age, 68.6 ± 4.1 years). To identify neural hubs most essentially related to serum BDNF, we applied graph theory approaches, namely the new data-driven and parameter-free approach eigenvector centrality (EC) mapping. The analysis revealed a positive correlation between serum BDNF and EC in the premotor and motor cortex in older participants in contrast to young volunteers, where we did not detect any association. This positive relationship between serum BDNF and EC appears to be specific for older adults. Our results might indicate that the amount of physical activity and learning capacities, leading to higher BDNF levels, increases brain connectivity in (pre)motor areas in healthy aging in agreement with rodent animal studies. Pilot results have to be replicated in a larger sample including behavioral data to disentangle the cause for the relationship between BDNF levels and connectivity. |
Mutations in the TBK1 gene were just recently identified to cause amyotrophic lateral sclerosis (ALS), and their role in ALS in various populations remains unclear. The aim of this study was to determine the frequency and spectrum of mutations in TBK1 in a Taiwanese ALS cohort of Han Chinese origin. Mutational analyses of TBK1 were carried out by direct nucleotide sequencing in a cohort of 207 unrelated patients with ALS. Among them, the genetic diagnoses of 168 patients remained elusive after mutations in SOD1, C9ORF72, TARDBP, FUS, ATXN2, OPTN, VCP, UBQLN2, SQSTM1, PFN1, HNRNPA1, HNRNPA2B1, MATR3, CHCHD10, and TUBA4A had been excluded. We identified one nonsense mutation, p.R444X (c.1330C>T), in one patient with apparently sporadic ALS-frontotemporal dementia. In vitro functional study demonstrated the p.R444X mutation resulting in a truncated TANK-binding kinase 1 (TBK1) protein product, low protein expression, and loss of kinase function and interaction with optineurin. The frequency of TBK1 mutations in ALS patients in Taiwan is, therefore, approximately 0.5% (1/207). This study reports a novel TBK1 mutation and stresses on the importance to consider TBK1 mutation as a possible etiology of ALS. |
Frontotemporal lobar degeneration (FTLD) is a group of neurodegenerative diseases displaying high clinical, pathologic, and genetic heterogeneity. Several autosomal dominant progranulin (GRN) mutations have been reported, accounting for 5%-10% of FTLD cases worldwide. In this study, we described the clinical characteristics of 7 Italian patients, 5 with a diagnosis of frontotemporal dementia behavioral variant and 2 of corticobasal syndrome (CBS), carrying the GRN deletion g.101349_101355delCTGCTGT, resulting in the C157KfsX97 null mutation, and hypothesized the existence of a founder effect by means of haplotype sharing analysis. We performed plasma progranulin dosage, GRN gene sequencing, and haplotype sharing study, analyzing 10 short tandem repeat markers, spanning a region of 11.08 Mb flanking GRN on chromosome 17q21. We observed shared alleles among 6 patients for 8 consecutive short tandem repeat markers spanning a 7.29 Mb region. Therefore, also with this particular mutation, the elevated clinical variability described among GRN-mutated FTLD cases is confirmed. Moreover, this is the first study reporting the likely existence of a founder effect for C157KfsX97 mutation in Southern Italy. |
The aggregation of Tau protein is a hallmark of neurodegenerative diseases including Alzheimer's disease. Previously, we generated a cell model of tauopathy based on the 4-repeat domain with the FTDP-17 mutation ΔK280 (Tau<sup>4RDΔK</sup>) which is expressed in a regulatable fashion (tet-on). The deletion variant ΔK280 is highly amyloidogenic and forms fibrous aggregates in neuroblastoma N2a cells staining with the reporter dye Thioflavin S. The aggregation of Tau<sup>4RDΔK</sup> is toxic, contrary to wildtype or anti-aggregant variants of the protein. Using a novel approach for monitoring in situ Tau aggregation and toxicity by combination of microscopic analysis with FACS and biochemical analysis of cells enabled the dissection of the aggregating species which cause a time-dependent increase of toxicity. The dominant initiating step is the dimerization of Tau<sup>4RDΔK</sup> which leads to further aggregation and induces a strong increase in reactive oxygen species (ROS) and cytoplasmic Ca<sup>2+</sup> which damage the membranes and cause cell death. Tau-based treatments using Tau aggregation inhibitors reduce both soluble oligomeric and fully aggregated Tau species and decrease their toxicity. |
Elevated serum and cerebrospinal fluid concentrations of S100β, a protein predominantly found in glia, are associated with intracranial injury and neurodegeneration, although concentrations are also influenced by several other factors. The longitudinal association between serum S100β concentrations and brain health in nonpathological aging is unknown. In a large group (baseline N = 593; longitudinal N = 414) of community-dwelling older adults at ages 73 and 76 years, we examined cross-sectional and parallel longitudinal changes between serum S100β and brain MRI parameters: white matter hyperintensities, perivascular space visibility, white matter fractional anisotropy and mean diffusivity (MD), global atrophy, and gray matter volume. Using bivariate change score structural equation models, correcting for age, sex, diabetes, and hypertension, higher S100β was cross-sectionally associated with poorer general fractional anisotropy (r = -0.150, p = 0.001), which was strongest in the anterior thalamic (r = -0.155, p < 0.001) and cingulum bundles (r = -0.111, p = 0.005), and survived false discovery rate correction. Longitudinally, there were no significant associations between changes in brain imaging parameters and S100β after false discovery rate correction. These data provide some weak evidence that S100β may be an informative biomarker of brain white matter aging. |
Alzheimer's disease (AD) is an age-related neurodegenerative disorder characterized by accumulation of amyloid β-peptide (Aβ) plaques in the brain and decreased cognitive function leading to dementia. We tested if hydroxyurea (HU), a ribonucleotide reductase inhibitor known to activate adaptive cellular stress responses and ameliorate abnormalities associated with several genetic disorders, could protect rat hippocampal neurons against oxidative-, excitatory-, mitochondrial-, and Aβ-induced stress and if HU treatment could improve learning and memory in the APP/PS1 mouse model of AD. HU treatment attenuated the loss of cell viability induced by treatment of hippocampal neurons with hydrogen peroxide, glutamate, rotenone, and Aβ<sub>1-42</sub>. HU treatment attenuated reductions of mitochondrial reserve capacity, maximal respiration, and cellular adenosine triphosphate content induced by hydrogen peroxide treatment. In vivo, treatment of APP/PS1 mice with HU (45 mg/kg/d) improved spatial memory performance in the hippocampus-dependent Morris water maze task without reducing Aβ levels. HU provides neuroprotection against toxic insults including Aβ, improves mitochondrial bioenergetics, and improves spatial memory in an AD mouse model. HU may offer a new therapeutic approach to delay cognitive decline in AD. |
Niemann-Pick disease type C (NP-C) is a hereditary neurovisceral lipid storage disorder. Although traditionally considered a primary cholesterol storage disorder, a variety of glycolipids accumulate in NP-C cells, which resemble those from glycosphingolipidosis patients. Substrate reduction therapy (SRT) with miglustat, an inhibitor of glycosphingolipid biosynthesis, is a novel therapy for the glycosphingolipidoses. We report the use of SRT in a patient with NP-C. We show that depletion of glycosphingolipids by miglustat treatment reduces pathological lipid storage, improves endosomal uptake and normalises lipid trafficking in peripheral blood B lymphocytes. The demonstration that treatment with miglustat, which has no direct effect on cholesterol metabolism, corrects the abnormal lipid trafficking seen in B lymphocytes in NP-C indicates that glycosphingolipid accumulation is the primary pathogenetic event in NP-C. These observations support the use of SRT in patients with this devastating neurodegenerative disease. |
Denervation-induced myofiber atrophy can be reversed by reinnervation. Growing reinnervated myofibers upregulate numerous molecules, many of which determine the muscle fiber type. In the present study we aimed at identifying factors that might contribute specifically to myofiber growth after reinnervation. The common peroneal nerve of 15 male Wistar rats was cut and resutured without delay (9 animals) or with a delay of 4 weeks (6 animals). We studied the transcriptional repertoire of intact reinnervated tibialis anterior muscle by microarray gene analysis. We assessed SC activation by immunolabeling using anti-MyoD and -myogenin antibodies. The percentage of SC expressing MyoD reached up to 50% of M-cadherin+ cells whereas the percentage of SC expressing myogenin was normal (<10%) in all muscles examined. The values of ipsi- and contralateral muscles did not differ significantly from one another between right and left leg (p<0.05). Thirteen known genes were differentially regulated after reinnervation compared with contralateral muscles. Five of them determine the slow-twitch fiber type (four and a half LIM domains 3, cardiac beta-myosin heavy chain, calsequestrin 2, troponin C (slow), and heart myosin light chain), and three of them are neurally regulated (thrombospondin 4, transferrin receptor, cardiac ankyrin repeat protein). The results strengthen the notion that reinnervaton affects the molecular repertoire of the myofibers directly, leading to fiber type transformation and partial reversal of the denervation phenotype. By contrast, SC do not appear to be affected by reinnervation directly. They can be activated both in reinnervated and contralateral muscles, and they do not fully differentiate. This makes them unlikely to contribute to myofiber growth. |
Neuronal ceroid lipofuscinoses (NCLs) are pediatric, neurodegenerative, lysosomal storage disorders. Mutations in cathepsin D result in the most severe, congenital form of NCLs. We have previously generated a cathepsin D deficient Drosophila model, which exhibits the key features of NCLs: progressive intracellular accumulation of autofluorescent storage material and modest neurodegeneration in the brain areas related to visual functions. Here we extend the phenotypic characterization of cathepsin D deficient Drosophila and report that modest degenerative changes are also present in their retinae. Furthermore, by utilizing this phenotype, we examined the possible effect of 17 candidate modifiers, selected based on the results from other cathepsin D deficiency models. We found enhancers of this phenotype that support the involvement of endocytosis-, lipid metabolism- and oxidation-related factors in the cathepsin D deficiency induced degeneration. Our results warrant further investigation of these mechanisms in the pathogenesis of cathepsin D deficiency. |
Mutations in MECP2, encoding methyl CpG binding protein 2, cause the neurodevelopmental disorder Rett syndrome. MeCP2 is an abundant nuclear protein that binds to chromatin and modulates transcription in response to neuronal activity. Prior studies of MeCP2 function have focused on specific gene targets of MeCP2, but a more global role for MeCP2 in neuronal nuclear maturation has remained unexplored. MeCP2 levels increase during postnatal brain development, coinciding with dynamic changes in neuronal chromatin architecture, particularly detectable as changes in size, number, and location of nucleoli and perinucleolar heterochromatic chromocenters. To determine a potential role for MeCP2 in neuronal chromatin maturational changes, we measured nucleoli and chromocenters in developing wild-type and Mecp2-deficient mouse cortical sections, as well as mouse primary cortical neurons and a human neuronal cell line following induced maturation. Mecp2-deficient mouse neurons exhibited significant differences in nucleolar and chromocenter number and size, as more abundant, smaller nucleoli in brain and primary neurons compared to wild-type, consistent with delayed neuronal nuclear maturation in the absence of MeCP2. Primary neurons increased chromocenter size following depolarization in wild-type, but not Mecp2-deficient cultures. Wild-type MECP2e1 over-expression in human SH-SY5Y cells was sufficient to induce significantly larger nucleoli, but not a T158M mutation of the methyl-binding domain. These results suggest that, in addition to the established role of MeCP2 in transcriptional regulation of specific target genes, the global chromatin-binding function of MeCP2 is essential for activity-dependent global chromatin dynamics during postnatal neuronal maturation. |
Friedreich ataxia is an inherited neurodegenerative disease that leads to progressive disability. There is currently no effective treatment and patients die prematurely. The underlying genetic defect leads to reduced expression of the mitochondrial protein frataxin. Frataxin insufficiency causes mitochondrial dysfunction and ultimately cell death, particularly in peripheral sensory ganglia. There is an inverse correlation between the amount of residual frataxin and the severity of disease progression; therefore, therapeutic approaches aiming at increasing frataxin levels are expected to improve patients' conditions. We previously discovered that a significant amount of frataxin precursor is degraded by the ubiquitin/proteasome system before its functional mitochondrial maturation. We also provided evidence for the therapeutic potential of small molecules that increase frataxin levels by docking on the frataxin ubiquitination site, thus preventing frataxin ubiquitination and degradation. We called these compounds ubiquitin-competing molecules (UCM). By extending our search for effective UCM, we identified a set of new and more potent compounds that more efficiently promote frataxin accumulation. Here we show that these compounds directly interact with frataxin and prevent its ubiquitination. Interestingly, these UCM are not effective on the ubiquitin-resistant frataxin mutant, indicating their specific action on preventing frataxin ubiquitination. Most importantly, these compounds are able to promote frataxin accumulation and aconitase rescue in cells derived from patients, strongly supporting their therapeutic potential. |
Zinc transporter-3 (ZnT3) protein is responsible for loading zinc into presynaptic vesicles and consequently controls the availability of zinc at the glutamatergic synapse. ZnT3 has been shown to decline with age and in Alzheimer's disease (AD) and is crucially involved in learning and memory. In this study, we utilised whole animal behavioural analyses in the ZnT3 KO mouse line, together with electrophysiological analysis of long-term potentiation in brain slices from ZnT3 KO mice, to show that metal chaperones (clioquinol, 30 mg/kg/day for 6weeks) can prevent the age-dependent cognitive phenotype that characterises these animals. This likely occurs as a result of a homeostatic restoration of synaptic protein expression, as clioquinol significantly restored levels of various pre- and postsynaptic proteins that are critical for normal cognition, including PSD-95; AMPAR and NMDAR2b. We hypothesised that this clioquinol-mediated restoration of synaptic health resulted from a selective increase in synaptic zinc content within the hippocampus. While we demonstrated a small regional increase in hippocampal zinc content using synchrotron x-ray fluorescence microscopy, further sub-region analyses are required to determine whether this effect is seen in other regions of the hippocampal formation that are more closely linked to the synaptic plasticity effects observed in this study. These data support our recent report on the use of a different metal chaperone (PBT2) to prevent normal age-related cognitive decline and demonstrate that metal chaperones are efficacious in preventing the zinc-mediated cognitive decline that characterises ageing and disease. |
A history of mild traumatic brain injury (mTBI) is linked to a number of chronic neurological conditions, however there is still much unknown about the underlying mechanisms. To provide new insights, this study used a clinically relevant model of repeated mTBI in rats to characterize the acute and chronic neuropathological and neurobehavioral consequences of these injuries. Rats were given four sham-injuries or four mTBIs and allocated to 7-day or 3.5-months post-injury recovery groups. Behavioral analysis assessed sensorimotor function, locomotion, anxiety, and spatial memory. Neuropathological analysis included serum quantification of neurofilament light (NfL), mass spectrometry of the hippocampal proteome, and ex vivo magnetic resonance imaging (MRI). Repeated mTBI rats had evidence of acute cognitive deficits and prolonged sensorimotor impairments. Serum NfL was elevated at 7 days post injury, with levels correlating with sensorimotor deficits; however, no NfL differences were observed at 3.5 months. Several hippocampal proteins were altered by repeated mTBI, including those associated with energy metabolism, neuroinflammation, and impaired neurogenic capacity. Diffusion MRI analysis at 3.5 months found widespread reductions in white matter integrity. Taken together, these findings provide novel insights into the nature and progression of repeated mTBI neuropathology that may underlie lingering or chronic neurobehavioral deficits. |
Learning and memory processes may be influenced by fluctuations in steroid hormones, such as estrogens and progestins. In this study, we have used an animal model to investigate the effects of endogenous fluctuations in ovarian steroids in intact female rats and effects of administration of ovarian steroids to ovariectomized rats for non-spatial, working memory using the object recognition task. Performance in this task relies on cortical and hippocampal function. As such, serum, cortical, and hippocampal concentrations of estradiol (E2), progesterone (P4), and P4's metabolite, 5alpha-pregnan-3alpha-ol-20-one (3alpha,5alpha-THP), were measured by radioimmunoassay. Experiment 1: Rats in behavioral estrus, compared to those in diestrus or estrus, spent a greater percentage of time exploring a novel object concomitant with increases in serum E2, P4, and 3alpha,5alpha-THP levels. Regression analyses revealed that there was a significant positive relationship between E2 levels in the hippocampus and 3alpha,5alpha-THP levels in the hippocampus and cortex and performance in this task. Experiment 2: Administration of E2 and/or P4 immediately post-training increased the percentage of time spent exploring the novel object and produced levels of E2, P4, and 3alpha,5alpha-THP akin to that of rats in behavioral estrus. Experiment 3: Post-training administration of selective estrogen receptor modulators, including 17beta-E2, propyl pyrazole triol, and diarylpropionitrile increased the percentage of time spent exploring the novel object compared to vehicle-administration. Experiment 4: Post-training P4 or 3alpha,5alpha-THP administration, compared to vehicle, increased the percentage of time spent exploring the novel object and produced P4 and/or 3alpha,5alpha-THP levels within the physiological range typically observed for rats in behavioral estrus. Experiment 5: If post-training administration of E2 and/or P4 was delayed one hour, no enhancement in object recognition was observed. Together, these results suggest that E2 and progestins can have mnemonic effects through actions in the cortex and/or hippocampus. |
Previous evidence showed that administration of d-galactose (d-gal) increased ROS production and resulted in impairment of cholinergic system. Troxerutin, a natural bioflavonoid, has been reported to have many benefits and medicinal properties. In this study, we evaluated the protective effect of troxerutin against d-gal-induced impairment of cholinergic system, and explored the potential mechanism of its action. Our results displayed that troxerutin administration significantly improved behavioral performance of d-gal-treated mice in step-through test and morris water maze task. One of the potential mechanisms of this action was decreased AGEs, ROS and protein carbonyl levels in the basal forebrain, hippocampus and front cortex of d-gal-treated mice. Furthermore, our results also showed that troxerutin significantly inhibited cholinesterase (AchE) activity, increased the expression of nicotinic acetylcholine receptor alpha 7 (nAchRalpha7) and enhanced interactions between nAchRalpha7 and either postsynaptic density protein 95 (PSD95) or N-methyl-d-aspartate receptors subunit 1 (NMDAR1) in the basal forebrain, hippocampus and front cortex of d-gal-treated mice, which could help restore impairment of brain function. |
Recent studies on the effect of stress on modulation of fear memory in our laboratory have uncovered endogenous opioid and adrenergic based modulation systems, working in concert, that limit the strengthening or weakening of newly acquired fear memory during consolidation under conditions of mild or intense stress, respectively. The present study sought to determine if similar stress-dependent modulation, mediated by endogenous opioid and adrenergic systems, occurs during reconsolidation of newly retrieved fear memory. Rats underwent contextual fear conditioning followed 24h later by reactivation of fear memory; a retention test was administered the next day. Stress was manipulated by varying duration of recall of fear memory during reactivation. In the first experiment, vehicle or the opioid-receptor blocker naloxone was administered immediately after varied durations (30 or 120 s) of reactivation. The results indicate that (1) reactivation, in the absence of drug, has a marked effect on freezing behavior-as duration of reactivation increases from 30 to 120 s, freezing behavior and presumably fear-induced stress increases and (2) naloxone, administered immediately after 30 s (mild stress) or 120 s (intense stress) of reactivation, enhances or impairs retention, respectively, the next day. In the second experiment, naloxone and the ß-adrenergic blocker propranolol were administered either separately or in combination immediately after 120 s (intense stress) reactivation. The results indicate that separate administration of propranolol and naloxone impairs retention, while the combined administration fails to do so. Taken together the results of the two experiments are consistent with a protective mechanism, mediated by endogenous opioid and adrenergic systems working in concert, that limits enhancement and impairment of newly retrieved fear memory during reactivation in a stress-dependent manner. |
Incorporation of details from waking life events into Rapid Eye Movement (REM) sleep dreams has been found to be highest on the night after, and then 5-7 nights after events (termed, respectively, the day-residue and dream-lag effects). In experiment 1, 44 participants kept a daily log for 10 days, reporting major daily activities (MDAs), personally significant events (PSEs), and major concerns (MCs). Dream reports were collected from REM and Slow Wave Sleep (SWS) in the laboratory, or from REM sleep at home. The dream-lag effect was found for the incorporation of PSEs into REM dreams collected at home, but not for MDAs or MCs. No dream-lag effect was found for SWS dreams, or for REM dreams collected in the lab after SWS awakenings earlier in the night. In experiment 2, the 44 participants recorded reports of their spontaneously recalled home dreams over the 10 nights following the instrumental awakenings night, which thus acted as a controlled stimulus with two salience levels, high (sleep lab) and low (home awakenings). The dream-lag effect was found for the incorporation into home dreams of references to the experience of being in the sleep laboratory, but only for participants who had reported concerns beforehand about being in the sleep laboratory. The delayed incorporation of events from daily life into dreams has been proposed to reflect REM sleep-dependent memory consolidation. However, an alternative emotion processing or emotional impact of events account, distinct from memory consolidation, is supported by the finding that SWS dreams do not evidence the dream-lag effect. |
Time plays an important role in conditioning, it is not only possible to associate stimuli with events that overlap, as in delay fear conditioning, but it is also possible to associate stimuli that are discontinuous in time, as shown in trace conditioning for a discrete stimuli. The environment itself can be a powerful conditioned stimulus (CS) and be associated to unconditioned stimulus (US). Thus, the aim of the present study was to determine the parameters in which contextual fear conditioning occurs by the maintenance of a contextual representation over short and long time intervals. The results showed that a contextual representation can be maintained and associated after 5s, even in the absence of a 15s re-exposure to the training context before US delivery. The same effect was not observed with a 24h interval of discontinuity. Furthermore, optimal conditioned response with a 5s interval is produced only when the contexts (of pre-exposure and shock) match. As the pre-limbic cortex (PL) is necessary for the maintenance of a continuous representation of a stimulus, the involvement of the PL in this temporal and contextual processing was investigated. The reversible inactivation of the PL by muscimol infusion impaired the acquisition of contextual fear conditioning with a 5s interval, but not with a 24h interval, and did not impair delay fear conditioning. The data provided evidence that short and long intervals of discontinuity have different mechanisms, thus contributing to a better understanding of PL involvement in contextual fear conditioning and providing a model that considers both temporal and contextual factors in fear conditioning. |
In different vertebrate species, hippocampus plays a crucial role for spatial orientation. However, even though cognitive lateralization is widespread in the animal kingdom, the lateralization of this hippocampal function has been poorly studied. The aim of the present study was to investigate the lateralization of hippocampal activation in domestic chicks, during spatial navigation in relation to free-standing objects. Two groups of chicks were trained to find food in one of the feeders located in a large circular arena. Chicks of one group solved the task using the relational spatial information provided by free-standing objects present in the arena, while the other group used the local appearance of the baited feeder as a beacon. The immediate early gene product c-Fos was employed to map neural activation of hippocampus and medial striatum of both hemispheres. Chicks that used spatial cues for navigation showed higher activation of the right hippocampus compared to chicks that oriented by local features and compared to the left hippocampus. Such differences between the two groups were not present in the left hippocampus or in the medial striatum. Relational spatial information seems thus to be selectively processed by the right hippocampus in domestic chicks. The results are discussed in light of existing evidence of hippocampal lateralization of spatial processing in chicks, with particular attention to the contrasting evidence found in pigeons. |
Impulse control disorders (ICDs) are frequent in Parkinson's disease (PD). Aim of the present study was to investigate cognition and behaviour in PD patients with and without ICDs, in order to identify potential early clinical features which might be associated to the development of ICDs. We recruited 17 PD patients with ICDs and 17 without ICDs, matched for several clinical variables, without clinically significant cognitive deficits. Assessments included behavioural scales and a neuropsychological battery, including the Iowa Gambling Task (IGT). In patients with ICDs, the total score of the BIS and the Motor Impulsivity subscore were significantly higher than in patients without ICDs. In patients with ICDs, we observed only statistical trends towards a worse performance on neuropsychological tasks (go-no-go subtest of the Frontal Assessment Battery, oral verb naming task, copying of drawings with landmarks) sensitive to frontal lobe dysfunction (FLD) and on the IGT (loss of a greater amount of money, more risky choices). As compared to patients without ICDs, they reported a more than threefold number of errors on the interference subtest of Stroop test, which is also sensitive to FLD. Although this study did not show any significant difference between PD patients presenting ICDs as compared with patients without ICDs on neuropsychological variables, some preliminary evidence was detected suggesting a trend toward a worse performance of the PD-ICD group on few neuropsychological tasks which are at least partially sensitive to frontal lobe dysfunction, including tasks sensitive to dysfunction of ventral fronto-striatal loops. |
Vestibular-evoked myogenic potentials (VEMP), short-latency electromyographic responses elicited by acoustic stimuli, evaluate the function of vestibulocollic reflex and may give information about brainstem function. The aim of the present study is to evaluate the potential contribution of VEMP to the diagnosis of multiple sclerosis (MS). Fifty patients with MS and 30 healthy control subjects were included in this study. The frequency of VEMP p1-n1 and n2-p2 waves; mean p1, n1, n2, and p2 latency; and mean p1-n1 and n2-p2 amplitude were determined. The relation between clinical and imaging findings and VEMP parameters was evaluated. The p1-n1 and n2-p2 waves were more frequently absent in MS than in control subjects [p1-n1 wave absent: MS, 25 (25 %) ears; control, 6 (10 %) ears; P ≤ 0.02] [n2-p2 wave absent: MS, 44 (44 %) ears; control, 7 (12 %) ears; P ≤ 0.001]. The mean p1-n1 amplitude was lower in MS than in control subjects (MS, 19.1 ± 7.2 μV; control, 23.3 ± 7.4 μV; P ≤ 0.002). A total of 24/50 (48 %) MS patients had VEMP abnormalities (absent responses and/or prolonged latencies). VEMP abnormalities were more frequent in patients with than without vestibular symptoms (P ≤ 0.02) and with brainstem functional system score (FSS) ≥ 1 than FSS = 0 (P ≤ 0.02). In patients with MS, absence of p1-n1 wave was more frequent in patients with than without vestibular symptoms [absence of p1-n1 wave: vestibular symptoms, 9 (45 %) ears; no vestibular symptoms, 16 (20 %) ears; P ≤ 0.03] and patients with Expanded Disability Status Scale (EDSS) score ≥ 5.5 [absence of p1-n1 wave: EDSS ≥ 5.5, 7 (70 %) ears; EDSS <5.5, 18 (20 %) ears; P ≤ 0.001]. Abnormal VEMP may be noted in MS patients, especially those with vestibular symptoms and greater disability. The VEMP test may complement other studies for diagnosis and follow-up of patients with MS. |
We report a 47-years-old male with ischemic stroke, whose arteriographic and echocardiographic investigations did not reveal any steno-occlusive arterial disease or embolic source from the left cardiac chambers. A transesophageal echocardiogram showed a patent foramen ovale (PFO), whilst laboratory screening for coagulation abnormalities showed heterozygosity for factor V Leiden mutation. The significance of the association of PFO with factor V Leiden mutation is discussed as a possible cause of ischemic stroke through paradoxical embolism from a venous source. The high prevalence of these two conditions in the general population is emphasized and the indication for anticoagulant therapy is discussed. |
Objective.  The aim of our work was to investigate whether lateral stimulation of the spinal cord, lateral cord stimulation (LCS), results in inhibition of the spastic phenomena of upper motor lesions in an animal model. Methods.  This study was conducted using an animal model consisting of surgically brain damaged pigs subjected to unilateral cortical and subcortical brain lesions. A double laminectomy at cervical (C3-C4) and lumbar (L3-L6) was performed, and spastic thresholds of abnormal electromyographic responses, disseminated to adjacent segments, facilitated by spinal liberation, and produced by extradural electrical stimulation of the fourth lumbar root, were measured before and after cervical stimulation of the LCS. The variable studied was the minimal amount of current of LCS necessary to abolish electromyographic responses in the L7 myotome, away from the stimulated L4 nerve root. Results.  Experiments in 12 animals showed a significant increase of threshold after LCS, with a marked posteffect, signaling a less abnormal threshold. Conclusions.  This experiment demonstrated that LCS produces threshold increases to abolish abnormally propagated electromyographic evoked responses induced by the electrical stimulation of the fourth lumbar root in pigs with experimental cortical and subcortical brain lesions. |
## Introduction
Clinical response to deep brain stimulation (DBS) strongly depends on the appropriate placement of the electrode in the targeted structure. Postoperative MRI is recognized as the gold standard to verify the DBS‐electrode position in relation to the intended anatomical target. However, intraoperative computed tomography (iCT) might be a feasible alternative to MRI.
## Materials and Methods
In this prospective noninferiority study, we compared iCT with postoperative MRI (24‐72 hours after surgery) in 29 consecutive patients undergoing placement of 58 DBS electrodes. The primary outcome was defined as the difference in Euclidean distance between lead tip coordinates as determined on both imaging modalities, using the lead tip depicted on MRI as reference. Secondary outcomes were difference in radial error and depth, as well as difference in accuracy relative to target.
## Results
The mean difference between the lead tips was 0.98 ± 0.49 mm (0.97 ± 0.47 mm for the left‐sided electrodes and 1.00 ± 0.53 mm for the right‐sided electrodes). The upper confidence interval (95% CI, 0.851 to 1.112) did not exceed the noninferiority margin established. The average radial error between lead tips was 0.74 ± 0.48 mm and the average depth error was determined to be 0.53 ± 0.40 mm. The linear Deming regression indicated a good agreement between both imaging modalities regarding accuracy relative to target.
## Conclusions
Intraoperative CT is noninferior to MRI for the verification of the DBS‐electrode position. CT and MRI have their specific benefits, but both should be considered equally suitable for assessing accuracy.
## INTRODUCTION
Deep brain stimulation (DBS) is a well‐recognized and effective neurosurgical treatment for various movement disorders. The clinical effect of DBS largely depends on the appropriate placement of the electrode in the targeted structure . Therefore, a correct assessment of the electrode position with imaging techniques during or directly after the surgical procedure is crucial, since it can timely indicate a necessary repositioning of the electrode.
Magnetic resonance imaging (MRI) is considered to be the gold standard to assess the electrode position after DBS implantation , , , , , , , . MRI offers detailed visualization of relevant brain structures. However, image distortion caused by local magnetic field inhomogeneity may cause a nonconcentric artifact, usually larger than the electrode itself, which could possibly have a negative impact on the suitability of MRI for electrode position assessment , , .
Intraoperative computed tomography (iCT) offers a high spatial resolution and a good delineation of the DBS electrode, providing a precise localization of the electrode . CT is significantly cheaper and less time consuming than MRI. Furthermore, while iCT is readily available in most hospital settings, access to an intraoperative MRI is often limited.
A number of studies have been conducted regarding the most suitable imaging modality for assessing accuracy in DBS , , , . However, these studies had several methodological limitations, such as nonconsecutive inclusions , , only a retrospective character , , , uncertainty about the duration between CT and MRI scan , and the lack of sample size calculations , , , .
This study was designed to compare iCT (fused with preoperative MRI , , , ) with early postoperative MRI for electrode position verification in DBS surgery.
## MATERIALS AND METHODS
### Study Population and DBS Targets
We prospectively studied a single‐institution series of 29 consecutive patients (mean age 58 ± 13.6 years, range: 16‐76) undergoing bilateral DBS placement (58 electrodes) between November 2016 and April 2018. Thirty‐eight electrodes were implanted in the subthalamic nucleus (STN), 14 in the internal globus pallidus (GPi), 4 in the zona incerta (ZI), and 2 in the thalamic ventral intermediate nucleus (VIM).
### Imaging and Targeting
DBS targeting was based on preoperative 3 T MRI (Philips Intera, Eindhoven, the Netherlands), using the planning software iPlan 3.6 (Brainlab, Feldkirchen, Germany). Targeting was independently performed by two neurosurgeons. Preoperative stereotactic CT images (Sensation 64, Siemens, Erlangen, Germany) using the Leksell G frame (Elekta, Stockholm, Sweden) were transferred to the iPlan software and subsequently fused with the preoperative 3 T MRI to register the planned target in the stereotactic coordinate system. During the surgical procedure, immediately after bilateral lead placement, patients were brought to the Medical Imaging Unit, in which iCT images were obtained on a diagnostic CT suite (Sensation 64). A deviation from the intended target was manually calculated based on the iCT scan, using the iPlan probe view tool. If the lead was positioned <2 mm from the intended target, lead positioning was accepted. On the contrary, if a deviation of ≥2 mm off‐target was determined, lead repositioning was performed immediately. Only one iCT verification was performed per patient. Afterwards, the internal pulse generator was implanted under general anesthesia. Within 24 to 72 hours after surgery, patients underwent 1.5 T‐MRI (Aera, Siemens, Erlangen, Germany). Refer to Table for imaging protocol specification.
Imaging Protocol Specification.
The iCT and postoperative MRI datasets were fused to the preoperative stereotactic CT. The fusion of images by iPlan 3.6 runs automatically, using a mutual information algorithm for dataset correlation.
### Lead Visualization and Localization
iPlan 3.6 was used to localize the leads. To improve visualization, a new trajectory was planned along the center of the lead artifact with a diameter of 1.2 mm, corresponding to the diameter of the actual lead (1.27 mm).
Lead artifacts appear differently on iCT and MRI (Fig. ). An ellipsoid shaped artifact was seen on MRI, while a clear, well‐delineated hyperdense artifact was seen on iCT. On iCT, a specific window level setting (Houndsfield Unit (HU) level: 1100 HU, width: 50 HU) was chosen to maximize contrast between lead and surrounding tissue, improving visualization. The lead position was determined as the imaginary center of the artifact on both modalities (Fig. ). From the iCT and MRI datasets, the stereotactic coordinates of the lead tip were obtained, taking the most caudal part of the lead artifact as the lead tip position.
DBS leads of the same patient on MRI (left) and on iCT (before windowing; right). The MRI artifact is depicted as a hypodense signal whereas the CT artifact is hyperdense. [Color figure can be viewed at ]
Lead visualization and plotted lead trajectory on MRI (left) and iCT (after windowing; right) in the same patient using the probe view in iPlan. The red and green lines represent the left‐sided and the right‐sided lead‐artifact, respectively. [Color figure can be viewed at ]
To study interobserver reliability, the plotting of the electrode trajectories was repeated by a neuroradiologist for all patients.
### Direct Comparison Between Modalities: Euclidean Distance, Radial Error, and Depth
Lead tip coordinates were compared between both imaging modalities. The Euclidean distance between lead tip positions as determined on iCT and MRI was calculated for all electrodes.
Besides the Euclidean distance, radial error and depth were assessed. Radial error is defined as the 2D distance in X and Y plane. For these assessments, the iCT lead was plotted into the postoperative MRI. Via the iPlan probe view tool, radial distance and depth were determined between the centers of both lead tips.
### Indirect Comparison Between Modalities: Accuracy Relative to Target
Furthermore, the difference between both modalities and the intended target was determined. Targets were assessed according to the intended nucleus. The target is placed in the −1/+2 contact spacing of the DBS lead when the target is in the STN, VIM, or ZI, while in GPi stimulation, the intention is to insert the lead tip at target.
In the STN, VIM, ZI leads (44 leads), coordinates of the −1/+2 contact spacing of the DBS lead were calculated using vector geometry. The Euclidean distance between these coordinates and the intended target coordinates were calculated, using the aforementioned formula.
In the GPi leads (14 leads), the Euclidean distance between the lead tip coordinates and the intended target coordinates were calculated.
### Statistical Analysis
The primary hypothesis was that iCT would be noninferior to postoperative MRI for the verification of the DBS lead position. We determined a noninferiority margin of 2 mm. It has been described in literature , , that the weighted mean distance between lead tips on CT and MRI is 1.50 ± 0.50 mm. Accordingly, we estimated a difference of 0.50 mm or more to be clinically relevant for the accuracy estimation. To calculate our sample size, we used a Cohen's d (mean difference/standard deviation, d ) of 1. Based on a power (1 − β ) of 0.95 and an alpha significance level ( α ) of 0.025, we estimated that 16 electrode distances were needed for our statistical analysis.
Calculations were performed according to the formula :
In which f is the function of α and β , σ is the standard deviation, and d is the noninferiority limit.
Thus, 29 participants implanted with 58 electrodes had sufficient power determine the possible noninferiority of iCT to MRI.
Euclidean distances resulted from the direct comparison of iCT and MRI leads, were compared with the mean distance previously reported in literature using a one sample right‐tailed t ‐test. Confidence intervals (CIs) were calculated to determine whether the upper limit of the CI exceeded the noninferiority margin.
For the indirect comparison between modalities, a linear Deming regression was performed between the error from target to lead in Euclidean distance, estimated both using MRI and CT. The variances were assumed to be similar between both modalities (lambda = 1) and the level of significance alpha was determined at 0.05. The correlation between both techniques was estimated using Pearson's r .
The statistical analysis was performed using R version 3.5.1 and the statistical package mcr. Descriptive statistics are given with mean and standard deviation. Intraclass correlation (ICC) with 95% CI were obtained for two raters for interobserver reliability.
According to Dutch legislation, the local research ethical board stated that the study was not submitted to the Medical Research Involving Human Subjects Act (WMO).
## RESULTS
### Direct Comparison Between Modalities: Euclidean Distance, Radial Error, and Depth
In our cohort, one lead position was revised after initial implantation because the lead was placed 2 mm too superficial as determined on iCT. After placing the lead 2 mm more caudal, an additional iCT was not acquired. Therefore, this lead was excluded and analysis was performed in 57 leads. The average Euclidean distance between lead tips was 0.98 ± 0.49 mm; 0.97 ± 0.47 mm for the left‐sided electrode and 1.00 ± 0.53 mm for the right‐sided electrode (Table ). The calculated mean Euclidean distances were not significantly higher than the reported weighted mean ( t = −7.9335, p = 1). The upper CI (0.851‐1.112) did not exceed the noninferiority margin established.
Absolute Differences Between Lead Tip Coordinates on iCT and Postoperative MRI.
The average radial error between lead tips was 0.74 ± 0.48 mm, whereas the average depth error was determined to be 0.53 ± 0.40 mm.
### Indirect comparison between modalities: accuracy relative to target
On iCT, the average Euclidean distance between the DBS lead and the intended target was 1.71 ± 0.61 mm; 1.98 ± 0.61 mm for the left‐sided electrode and 1.40 ± 0.45 mm for the right‐sided electrode.
On MRI, the average Euclidean distance between the DBS lead and the intended target was 1.94 ± 0.74 mm; 2.19 ± 0.61 mm for the left‐sided electrode and 1.70 ± 0.79 mm for the right‐sided electrode.
The linear Deming regression indicated a good agreement between both imaging modalities (intercept 0.066, CI −0.3607 to 0.5310, slope 0.08362, CI 0.6047 to 1.0326). The CI of the intercept contains 0, which indicates no significant accuracy difference between CT and MRI, while the CI of the slope contains 1, which indicates no significant difference in the precision of both imaging techniques. Pearson's r was 0.798, showing a strong correlation between both MRI and CT (Fig. ).
Linear Deming regression plot indicating good agreement between iCT and MRI. [Color figure can be viewed at ]
### Intraclass Correlations
The lead tip two‐rater interobserver ICC showed an almost perfect agreement for both iCT and MRI measurements, 0.999 (95% CI; range: 0.699‐1.000) and 0.995 (95% CI; range: 0.940‐0.998), respectively.
## DISCUSSION
Ideally, proper visualization of the DBS lead and the nucleus borders of the target would either ensure correct lead position or shed light upon the need for revision. Unfortunately, the current imaging techniques have not reached the stage in which both DBS lead and nucleus borders can be clearly visualized. MRI offers detailed visualization of relevant brain structures, but also induces an artifact that overestimates the actual electrode. Besides, DBS systems are not always compatible with MRI and occurrence of adverse events have been described in DBS implanted patients after MRI . Also, the specific absorption rate limits implemented in MRI safety protocols as a result of these concerns can limit the quality of the images (depending on sequence), resulting in suboptimal images for lead verification. In addition, access to an intraoperative MRI suite is limited in most hospitals. All the more reason for the entry of an alternative imaging modality in DBS surgery.
Our study shows that iCT is noninferior to MRI for the verification of lead position in DBS surgery. The average Euclidean distance between lead tip position determined on iCT and MRI was 0.98 mm ± 0.49 mm, which is lower than the noninferiority margin for significant clinical relevance.
The differences found in this study are smaller than those found in other studies on lead localization in DBS. Shahlaie et al. found differences of 1.65 ± 0.19 mm between lead tips on iCT and postoperative MRI. Based on the conclusions of previous publications , , and the results obtained in this article, the authors agree that iCT could replace postoperative MRI in assessing DBS leads. Lee et al. directly fused postoperative CT and postoperative MRI and subsequently compared the lead centers at five different levels. The lead centers showed differences of 1.08 mm to 1.40 mm. Thani et al. performed intraoperative MRI with a surrogate marker (carbothane stylette, in which the lead was placed later) and fused this dataset with postoperative CT (with DBS electrodes) to calculate the discrepancy between the location of the active contact of the two images. The discrepancy found was 1.60 ± 0.20 mm. Carlson et al. reported a distance of 1.43 ± 0.66 mm comparing postoperative MRI (1‐2 weeks) and postoperative CT (12 hours).
Besides studying the difference of the lead tip position on iCT and MRI, the difference between the lead and the intended target was assessed on each modality, because this information is very important for the intraoperative decision to revise the lead or not. Larger Euclidean distances were found between target and DBS lead on MRI, compared to iCT (1.94 ± 0.74 mm vs. 1.71 ± 0.61 mm). This difference between modalities is most likely caused by the difficulty both lead tip assessors experienced identifying the lead tip on MRI. The often unclear black artifact was not well‐delineated and showed a gradual beginning at the lead tip, making it difficult to accurately mark the lead tip. Difficulty visualizing the lead tip on MRI imaging has been described before , .
### Assessing Lead Position Based on Artifact Visualization
On both CT and MRI the electrode induces an artifact that exceeds the actual electrode size. To accurately assess lead positioning on imaging, it is important to exactly know the electrode position in relation to the artifact. On MRI, DBS leads are depicted as ellipse‐shaped low signal artifacts, in which each contact point individually induces a symmetrical artifact extending approximately 1.4 mm over the proximal and distal ends of the contact and 1.16 mm over the lateral limit of the contact . This suggests that both the artifact and the relative contact have the same center , . Pollo et al. thus identified 1.4 mm to be the distance on MRI between the distal limit of the artifact and the distal limit of the first contact (contact 0). On CT, lead artifacts appear as a clear, hyperdense signal in the darkened (after windowing) intracranial space. Because of the high spatial fidelity of CT, the artifact is likely to be concentrically formed around the lead . Hemm et al. conducted a study based on lead artifacts on CT images and found distances of 1.1 mm and 1.2 mm between the beginning of the artifact and the distal limit of contact 0, in their in vivo and in vitro study, respectively . In our study, the above cited distances were taken into account when calculating the exact position of the −1/+2 contact spacing for assessing the lead position relative to target.
### Limitations
The position of the lead tip and the lead position relative to the target was compared between modalities, taken into account the imaging specific characteristics of the different lead artifacts. However, since the distance from lead tip to contact 0 varies between MRI and CT (1.4 mm vs. 1.2 mm), contact 0 might be more suitable than the lead tip to compare modalities. Nonetheless, the lead tip has been extensively used to study accuracy in the past , , , and a difference of 0.2 mm between modalities is very small.
Fusion error could have played a role in the accuracy of lead measurements. When fusion between imaging sets was being performed, fusion was always visually verified by checking ventricular and sulcal shape. In none of the cases manual fusion adjustment was necessary.
Another limitation in our study might be the difference in time between the iCT and the postoperative MRI (24–72 hours). Ideally, the postoperative MRI should be performed directly after the iCT to minimize any effects related to brain shift, but this was not possible because of logistic reasons. Therefore, brain shift could have led to a less reliable comparison between both imaging modalities.
### Future Perspectives
The results show that iCT is noninferior to postoperative MRI for lead localization in DBS surgery. Our institution may therefore have the possibility of proceeding towards an asleep DBS procedure using iCT, since iCT proves to measure up to the gold standard, MRI. Good results have been published by recent studies regarding asleep DBS surgery , , , , , , , , . Asleep DBS is proven to be safe and without differences in adverse events compared to awake DBS . The asleep procedure could possibly be even more effective and may be cheaper than operating in an awake situation. Risk of surgical complications such as hemorrhages or infections are also significantly less frequent in asleep DBS , .
In the current study, we have proven iCT to be noninferior to postoperative MRI, whereby our surgical group can advance to performing asleep surgery in our patients using iCT. The lack of MRI availability does not hinder the possibility to perform asleep DBS surgery.
## CONCLUSION
In this prospective study iCT was found to be noninferior to postoperative MRI for the verification of the lead position in DBS surgery. There were no relevant differences between the lead position on iCT and postoperative MRI. In conclusion, both modalities have their pros and cons, but either one is suitable for lead position verification in DBS surgery.
## Authorship Statements
Ms. Kremer, Dr. Oterdoom, Mr. van Hulzen, and Dr. van Dijk designed and conducted the study, including data collection and data analysis. Dr. PJ van Laar is also responsible for data collection. Ms. Kremer prepared the manuscript draft with important intellectual input from Dr. Oterdoom, Dr. Piña‐Fuentes, Dr. T van Laar, Dr. Drost, Mr. van Hulzen, and Dr. van Dijk. All authors approved the final manuscript. Statistical support in analyzing the data with input from Ms. Kremer and Dr. Piña‐Fuentes. Ms. Kremer, Dr. Piña‐Fuentes, and Mr. van Hulzen had complete access to the study data.
## COMMENT
This study compares visualization of DBS leads using two different imaging modalities: intraoperative CT (iCT) and postoperative MRI (postop MRI). Intraoperative deep brain stimulation (DBS) lead verification is clinically relevant. First, the outcomes in DBS are directly related to accurate electrode position. Second, intraoperative verification allows immediate deviation correction, whenever found. Although MRI would be the gold standard for allowing both electrode and nuclei visualization, it is not broadly available as iCT. Some of the limitations of previous studies comparing MRI and CT lead verification were properly addressed and overcome by the design of this study. I believe the topics presented and discussed here will assist in decision making during imaging‐verified DBS implantations.
Caio Matias, MD, PhD
São Paulo, Brazil
Comments not included in the Early View version of this paper.
|
Corticostriatal glutamate afferents and mesostriatal dopamine afferents commonly converge onto the same postsynaptic spines of medium projection neurons. The consequent synaptic triad provides an ideal configuration for dopamine modulation of glutamatergic transmission. In this issue of Neuron, Bamford et al. report that dopamine inhibits glutamate release in a selective manner by activating presynaptic D2 receptors. |
The class 3 Semaphorins Sema3A and Sema3F are potent axonal repellents that cause repulsion by binding Neuropilin-1 and Neuropilin-2, respectively. Plexins are implicated as signaling coreceptors for the Neuropilins, but the identity of the Plexins that transduce Sema3A and Sema3F responses in vivo is uncertain. Here, we show that Plexin-A3 and -A4 are key determinants of these responses, through analysis of a Plexin-A3/Plexin-A4 double mutant mouse. Sensory and sympathetic neurons from the double mutant are insensitive to Sema3A and Sema3F in vitro, and defects in axonal projections in vivo correspond to those seen in Neuropilin-1 and -2 mutants. Interestingly, we found a differential requirement for these two Plexins: signaling via Neuropilin-1 is mediated principally by Plexin-A4, whereas signaling via Neuropilin-2 is mediated principally by Plexin-A3. Thus, Plexin-A3 and -A4 contribute to the specificity of axonal responses to class 3 Semaphorins. |
Mammals have developed patterns of social relationships that enhance the survival of individuals and maximize the reproductive success of species. Although social stimuli and social responses are highly complex, recent studies are providing substantial insights into their neural substrates. Neural pathways employing the nonapeptides vasopressin and oxytocin play a particularly prominent role both in social recognition and the expression of appropriate social responses. New insights into social neuroscience are discussed, along with the relevance of this rapidly developing field to human relationships and disease processes. |
Filamentous tau inclusions are hallmarks of Alzheimer's disease (AD) and related tauopathies, but earlier pathologies may herald disease onset. To investigate this, we studied wild-type and P301S mutant human tau transgenic (Tg) mice. Filamentous tau lesions developed in P301S Tg mice at 6 months of age, and progressively accumulated in association with striking neuron loss as well as hippocampal and entorhinal cortical atrophy by 9-12 months of age. Remarkably, hippocampal synapse loss and impaired synaptic function were detected in 3 month old P301S Tg mice before fibrillary tau tangles emerged. Prominent microglial activation also preceded tangle formation. Importantly, immunosuppression of young P301S Tg mice with FK506 attenuated tau pathology and increased lifespan, thereby linking neuroinflammation to early progression of tauopathies. Thus, hippocampal synaptic pathology and microgliosis may be the earliest manifestations of neurodegenerative tauopathies, and abrogation of tau-induced microglial activation could retard progression of these disorders. |
In addition to establishing dendritic coverage of the receptive field, neurons need to adjust their dendritic arbors to match changes of the receptive field. Here, we show that dendrite arborization (da) sensory neurons establish dendritic coverage of the body wall early in Drosophila larval development and then grow in precise proportion to their substrate, the underlying body wall epithelium, as the larva more than triples in length. This phenomenon, referred to as scaling growth of dendrites, requires the function of the microRNA (miRNA) bantam (ban) in the epithelial cells rather than the da neurons themselves. We further show that ban in epithelial cells dampens Akt kinase activity in adjacent neurons to influence dendrite growth. This signaling between epithelial cells and neurons receiving sensory input from the body wall synchronizes their growth to ensure proper dendritic coverage of the receptive field. |
When the primary visual cortex (V1) is activated by sensory stimulation, what is the temporal correlation between the synaptic inputs to nearby neurons? This question underlies the origin of correlated activity, the mechanism of how visually evoked activity emerges and propagates in cortical circuits, and the relationship between spontaneous and evoked activity. Here, we have recorded membrane potential from pairs of V1 neurons in anesthetized cats and found that visual stimulation suppressed low-frequency membrane potential synchrony (0-10 Hz), and often increased synchrony at high frequencies (20-80 Hz). The increase in high-frequency synchrony occurred for neurons with similar orientation preferences and for neurons with different orientation preferences and occurred for a wide range of stimulus orientations. Thus, while only a subset of neurons spike in response to visual stimulation, a far larger proportion of the circuit is correlated with spiking activity through subthreshold, high-frequency synchronous activity that crosses functional domains. |
In this issue of Neuron, Malhotra and colleagues report an enrichment of de novo copy number variants in bipolar disorder and schizophrenia when compared with those of controls. The study highlights the importance of a genetic model involving rare and disruptive variants to further our understanding of complex neuropsychiatric traits. |
Learning-dependent cortical encoding has been well described in single neurons. But behaviorally relevant sensory signals drive the coordinated activity of millions of cortical neurons; whether learning produces stimulus-specific changes in population codes is unknown. Because the pattern of firing rate correlations between neurons--an emergent property of neural populations--can significantly impact encoding fidelity, we hypothesize that it is a target for learning. Using an associative learning procedure, we manipulated the behavioral relevance of natural acoustic signals and examined the evoked spiking activity in auditory cortical neurons in songbirds. We show that learning produces stimulus-specific changes in the pattern of interneuronal correlations that enhance the ability of neural populations to recognize signals relevant for behavior. This learning-dependent enhancement increases with population size. The results identify the pattern of interneuronal correlation in neural populations as a target of learning that can selectively enhance the representations of specific sensory signals. |
The ability to visualize endogenous proteins in living neurons provides a powerful means to interrogate neuronal structure and function. Here we generate recombinant antibody-like proteins, termed Fibronectin intrabodies generated with mRNA display (FingRs), that bind endogenous neuronal proteins PSD-95 and Gephyrin with high affinity and that, when fused to GFP, allow excitatory and inhibitory synapses to be visualized in living neurons. Design of the FingR incorporates a transcriptional regulation system that ties FingR expression to the level of the target and reduces background fluorescence. In dissociated neurons and brain slices, FingRs generated against PSD-95 and Gephyrin did not affect the expression patterns of their endogenous target proteins or the number or strength of synapses. Together, our data indicate that PSD-95 and Gephyrin FingRs can report the localization and amount of endogenous synaptic proteins in living neurons and thus may be used to study changes in synaptic strength in vivo. |
High-throughput operant conditioning systems for rodents provide efficient training on sophisticated behavioral tasks. Combining these systems with technologies for cellular resolution functional imaging would provide a powerful approach to study neural dynamics during behavior. Here we describe an integrated two-photon microscope and behavioral apparatus that allows cellular resolution functional imaging of cortical regions during epochs of voluntary head restraint. Rats were trained to initiate periods of restraint up to 8 s in duration, which provided the mechanical stability necessary for in vivo imaging while allowing free movement between behavioral trials. A mechanical registration system repositioned the head to within a few microns, allowing the same neuronal populations to be imaged on each trial. In proof-of-principle experiments, calcium-dependent fluorescence transients were recorded from GCaMP-labeled cortical neurons. In contrast to previous methods for head restraint, this system can be incorporated into high-throughput operant conditioning systems. |
An important strategy for efficient neural coding is to match the range of cellular responses to the distribution of relevant input signals. However, the structure and relevance of sensory signals depend on behavioral state. Here, we show that behavior modifies neural activity at the earliest stages of fly vision. We describe a class of wide-field neurons that provide feedback to the most peripheral layer of the Drosophila visual system, the lamina. Using in vivo patch-clamp electrophysiology, we found that lamina wide-field neurons respond to low-frequency luminance fluctuations. Recordings in flying flies revealed that the gain and frequency tuning of wide-field neurons change during flight, and that these effects are mimicked by the neuromodulator octopamine. Genetically silencing wide-field neurons increased behavioral responses to slow-motion stimuli. Together, these findings identify a cell type that is gated by behavior to enhance neural coding by subtracting low-frequency signals from the inputs to motion detection circuits. |
The superior colliculus, or tectum, is a key sensorimotor structure that long predates the cortex. In this issue of Neuron, Zhao et al. (2014) show that the visual cortex controls the tectum's gain precisely and retinotopically, without otherwise altering its operations. |
An emerging view posits a timescale-based cortical topography, with integration windows increasing from sensory to association areas. In this issue, Chaudhuri et al. (2015) present a cortical model wherein a hierarchy of timescales arises from local and inter-regional circuit dynamics. |
Recent evidence of unconscious working memory challenges the notion that only visible stimuli can be actively maintained over time. In the present study, we investigated the neural dynamics underlying the maintenance of variably visible stimuli using magnetoencephalography. Subjects had to detect and mentally maintain the orientation of a masked grating. We show that the stimulus is fully encoded in early brain activity independently of visibility reports. However, the presence and orientation of the target are actively maintained throughout the brief retention period, even when the stimulus is reported as unseen. Source and decoding analyses revealed that perceptual maintenance recruits a hierarchical network spanning the early visual, temporal, parietal, and frontal cortices. Importantly, the representations coded in the late processing stages of this network specifically predicted visibility reports. These unexpected results challenge several theories of consciousness and suggest that invisible information can be briefly maintained within the higher processing stages of visual perception. |
Encountering another's suffering can elicit both empathic distress and empathic care-the warm desire to affiliate. It remains unclear whether these two feelings can be accurately and differentially predicted from neural activity and to what extent their neural substrates can be distinguished. We developed fMRI markers predicting moment-by-moment intensity levels of care and distress intensity while participants (n = 66) listened to true biographies describing human suffering. Both markers' predictions correlated strongly with self-report in out-of-sample participants (r = 0.59 and r = 0.63, p < 0.00001), and both markers predicted later trial-by-trial charitable donation amounts (p < 0.05). Empathic care was preferentially associated with nucleus accumbens and medial orbitofrontal cortex activity, whereas distress was preferentially associated with premotor and somatosensory cortical activity. In tests of marker specificity with an independent behavioral sample (n = 200), the empathic care marker was associated with a mixed-valence feeling state, whereas the empathic distress marker was specific to negative emotion. |
Systems-level organization in spontaneous infra-slow (<0.1Hz) brain activity, measured using blood oxygen signals in fMRI and optical imaging, has become a major theme in the study of neural function in both humans and animal models. Yet the neurophysiological basis of infra-slow activity (ISA) remains unresolved. In particular, is ISA a distinct physiological process, or is it a low-frequency analog of faster neural activity? Here, using whole-cortex calcium/hemoglobin imaging in mice, we show that ISA in each of these modalities travels through the cortex along stereotypical spatiotemporal trajectories that are state dependent (wake versus anesthesia) and distinct from trajectories in delta (1-4 Hz) activity. Moreover, mouse laminar electrophysiology reveals that ISA travels through specific cortical layers and is organized into unique cross-laminar temporal dynamics that are different from higher frequency local field potential activity. These findings suggest that ISA is a distinct neurophysiological process that is reflected in fMRI blood oxygen signals. |
Channelopathies are disorders caused by abnormal ion channel function in differentiated excitable tissues. We discovered a unique neurodevelopmental channelopathy resulting from pathogenic variants in SCN3A, a gene encoding the voltage-gated sodium channel Na<sub>V</sub>1.3. Pathogenic Na<sub>V</sub>1.3 channels showed altered biophysical properties including increased persistent current. Remarkably, affected individuals showed disrupted folding (polymicrogyria) of the perisylvian cortex of the brain but did not typically exhibit epilepsy; they presented with prominent speech and oral motor dysfunction, implicating SCN3A in prenatal development of human cortical language areas. The development of this disorder parallels SCN3A expression, which we observed to be highest early in fetal cortical development in progenitor cells of the outer subventricular zone and cortical plate neurons and decreased postnatally, when SCN1A (Na<sub>V</sub>1.1) expression increased. Disrupted cerebral cortical folding and neuronal migration were recapitulated in ferrets expressing the mutant channel, underscoring the unexpected role of SCN3A in progenitor cells and migrating neurons. |
In this issue of Neuron, O'Sullivan et al. (2019) measured electro-cortical responses to "cocktail party" speech mixtures in neurosurgical patients and demonstrated that the selective enhancement of attended speech is achieved through the adaptive weighting of primary auditory cortex output by non-primary auditory cortex. |
Neuronal cell types are arranged in brain-wide circuits that guide behavior. In mice, the superior colliculus innervates a set of targets that direct orienting and defensive actions. We combined functional ultrasound imaging (fUSI) with optogenetics to reveal the network of brain regions functionally activated by four collicular cell types. Stimulating each neuronal group triggered different behaviors and activated distinct sets of brain nuclei. This included regions not previously thought to mediate defensive behaviors, for example, the posterior paralaminar nuclei of the thalamus (PPnT), which we show to play a role in suppressing habituation. Neuronal recordings with Neuropixels probes show that (1) patterns of spiking activity and fUSI signals correlate well in space and (2) neurons in downstream nuclei preferentially respond to innately threatening visual stimuli. This work provides insight into the functional organization of the networks governing innate behaviors and demonstrates an experimental approach to explore the whole-brain neuronal activity downstream of targeted cell types. |
Hippocampal ripples are prominent synchronization events generated by hippocampal neuronal assemblies. To date, ripples have been primarily associated with navigational memory in rodents and short-term episodic recollections in humans. Here, we uncover different profiles of ripple activity in the human hippocampus during the retrieval of recent and remote autobiographical events and semantic facts. We found that the ripple rate increased significantly before reported recall compared to control conditions. Patterns of ripple activity across multiple hippocampal sites demonstrated remarkable specificity for memory type. Intriguingly, these ripple patterns revealed a semantization dimension, in which patterns associated with autobiographical contents become similar to those of semantic memory as a function of memory age. Finally, widely distributed sites across the neocortex exhibited ripple-coupled activations during recollection, with the strongest activation found within the default mode network. Our results thus reveal a key role for hippocampal ripples in orchestrating hippocampal-cortical communication across large-scale networks involved in conscious recollection. |
A balanced and fine-tuned ratio of neuronal excitation and inhibition is a prerequisite for information processing. In this issue of Neuron, He et al. (2022) reveal a causal link between reduced input to local somatostatin-expressing, MeCP2-negative O-LM interneurons in CA1 and long-term memory impairment in a mouse model of Rett syndrome. |
Summary
Axial muscles are innervated by motor neurons of the median motor column (MMC). In contrast to the segmentally restricted motor columns that innervate limb, body wall, and neuronal targets, MMC neurons are generated along the entire length of the spinal cord. We show that the specification of MMC fate involves a dorsoventral signaling program mediated by three Wnt proteins (Wnt4, Wnt5a, and Wnt5b) expressed in and around the floor plate. These Wnts appear to establish a ventral to dorsal signaling gradient and promote MMC identity and connectivity by maintaining expression of the LIM homeodomain proteins Lhx3/4 in spinal motor neurons. Elevation of Wnt4/5 activity generates additional MMC neurons at the expense of other motor neuron columnar subtypes, whereas depletion of Wnt4/5 activity inhibits the production of MMC neurons. Thus, two dorsoventral signaling pathways, mediated by Shh and Wnt4/5, are required to establish an early binary divergence in motor neuron columnar identity.
## Introduction
Motor behaviors depend on the coordinate recruitment of different muscle groups, each activated by a specialized set of motor neurons. The activation of axial muscles controls many basic vertebrate motor programs. In aquatic vertebrates, axial muscles control the lateral undulations of the trunk and tail that underlie swimming, whereas in terrestrial vertebrates the axial musculature helps to stabilize the trunk during walking ( ). The task of innervating axial muscles has been assigned to an evolutionarily conserved set of median motor column (MMC) neurons that are distinct, anatomically and functionally, from the motor neurons that innervate limb and body wall musculature ( ).
Axial muscle innervation has its origins in the generation of subclasses of spinal motor neurons. Different classes of motor neurons are specified in modular fashion, through the actions of secreted signaling factors that assign diverse transcriptional codes to progenitor cells and postmitotic neurons ( ). The early specification of spinal motor neuron fate is initiated by a dorsoventral gradient of Sonic hedgehog (Shh) signaling activity ( ), which induces the sequential expression of a series of homeodomain (HD) transcription factors, notably the Nkx6.1/.2, Mnr2/Hb9, Lhx3/4, and Isl1/2 proteins, in ventral progenitors and postmitotic motor neurons ( ). This dorsoventral signaling program operates along the entire length of the spinal cord, ensuring that motor neurons are produced at all segmental levels ( ).
The diversification of this generic set of motor neurons depends on a second patterning system that operates along the rostrocaudal axis of the spinal cord and involves the graded signaling activities of fibroblast growth factors (FGFs) and retinoids ( ). These extrinsic signals induce the expression of a network of Hox transcription factors whose collective activities specify motor neuron columnar classes at different segmental levels of the spinal cord ( ) ( ). At brachial and lumbar levels, Hox activities direct the formation of lateral motor column (LMC) neurons, which project their axons to limb muscles ( ). At thoracic levels, preganglionic motor column (PGC) neurons innervating sympathetic neuronal targets are specified by Hox9 proteins ( ), whereas hypaxial motor column (HMC) neurons innervating body wall muscles appear to be generated in a Hox-independent manner ( ).
The general rule that motor neuron columnar classes are generated within segmentally restricted domains has one notable exception. Neurons of the median motor column (MMC) are found along the entire length of the spinal cord, a spatial profile that accommodates the need to innervate an iterated series of axial muscle groups ( ) ( ). As a consequence, nascent motor neurons at every segmental level of the spinal cord are faced with a basic decision choice about their fate: whether to generate MMC neurons or motor neurons destined to populate the other segmentally restricted motor columns.
At a transcriptional level, the assignment of MMC neuronal fate involves the postmitotic expression of two LIM homeodomain (HD) proteins, Lhx3 and Lhx4 ( ). Expression of the Lhx3/4 proteins renders motor neurons refractory to the segmental columnar patterning activities of Hox proteins ( ). Moreover, ectopic expression of Lhx3/4 proteins in spinal motor neurons is sufficient to reroute axons along a dorsal trajectory that brings them to axial muscles ( ). Thus, the Lhx3/4 proteins function as intrinsic determinants that impose the identity and connectivity of MMC neurons. Despite these advances, nothing is known about the early signaling events that ensure that a fraction of the motor neurons generated at each segmental level of the spinal cord progress to an MMC fate, rather than to segmentally restricted columnar subtypes.
The generation of MMC neurons along the entire rostrocaudal extent of the spinal cord prompted us to consider whether signals that operate along the dorsoventral axis specify MMC, as well as generic, motor neuron fate. The idea that dorsoventral signaling contributes to motor neuron diversification has received support from studies in the hindbrain showing that quantitative differences in the level of Shh signaling specifies the dorsoventral distinction between adjacent pMN and p3 progenitor domains that give rise to ventral (vMN) and dorsal (dMN) motor neuron classes ( ). Yet in the spinal cord, all motor neurons derive from the pMN domain ( ). Thus, it remains unclear whether dorsoventral signaling has any role in spinal motor neuron diversification. And if it does, is Shh or another as yet unidentified signaling factor responsible for this patterning activity?
To address these issues, we set out to examine whether inductive signals that operate along the dorsoventral axis of the spinal cord regulate the developmental decision to generate MMC neurons. Using a combination of molecular and genetic methods in chick and mouse embryos, we show that the generation of MMC neurons depends on the dorsoventral position at which motor neurons are generated—MMC neurons can be induced ventral, but not dorsal, to the normal position of motor neuron generation. The position-dependent nature of MMC generation can be traced to the activities of a triumvirate of Wnt genes expressed in and around the floor plate. Wnt4, Wnt5a, and Wnt5b act redundantly to establish a ventral to dorsal signaling gradient that specifies MMC identity and connectivity by promoting persistent expression of Lhx3/4 in postmitotic motor neurons. Elevation of Wnt4/5 activity results in the generation of additional MMC neurons, at the expense of HMC and LMC neurons, whereas depletion of Wnt4/5 activity inhibits the production of MMC neurons and generates additional HMC neurons. Together, our findings show that motor neuron generation in the spinal cord depends on two dorsoventral signaling systems—mediated by Shh and Wnt4/5 proteins—and that the concerted activity of these two systems establishes an early divergence in motor neuron columnar identity.
## Results
### Dorsoventral Position of Motor Neuron Generation Influences MMC Fate
We first examined whether the specification of MMC neuronal identity is influenced by the dorsoventral position of motor neuron generation within the spinal cord. To explore this issue, we used loss- and gain-of-function approaches in mouse and chick to elicit motor neuron differentiation at ectopic ventral or dorsal positions and assessed the columnar identity of the supernumerary, misplaced, motor neurons. MMC neurons were defined by coexpression of the HD proteins Lhx3/4, Isl1/2, and Hb9; HMC neurons by coexpression of Isl1/2 and Hb9 in the absence of Lhx3/4; and PGC neurons by expression of Isl1 in the absence of other HD proteins, as well as by their dorsal settling position ( ). At limb levels, medial LMC neurons were defined by Isl1 expression in the absence of Lhx3/4 and Hb9, and lateral LMC neurons by coexpression of Isl2 and Hb9 ( ).
To elicit motor neuron generation at an ectopic ventral position, we examined mice mutant for the homeobox gene Nkx2.2 . In Nkx2.2 mutants, progenitor cells in the p3 domain, which lies ventral to the normal domain of motor neuron differentiation, switch their fate from p3 to pMN identity, with the consequence that motor neurons rather than V3 neurons are generated ( ). The columnar subtype identity of these ectopic motor neurons has not been resolved, however. We therefore quantified the number and columnar subtype of motor neurons in wild-type and Nkx2.2 mutant mice at e13.5, after columnar identities have been consolidated ( ).
At thoracic spinal levels of e13.5 wild-type mice, the total cohort of motor neurons (mean: 66 motor neurons/ventral quadrant/15 μm section) comprised ∼35% MMC neurons, ∼40% HMC neurons, and ∼25% PGC neurons ( A, 2B, and 2I). At brachial and lumbar levels, the total motor neuron cohort (mean brachial: 125 motor neurons/ventral quadrant/15 μm section; mean lumbar: 138 motor neurons/ventral quadrant/15 μm section) comprised ∼20% MMC and ∼80% LMC neurons ( A, S1B, S1E, S1F, S1I, and S1J available online). In Nkx2.2 mutants, there was an ∼30% increase in total motor neuron number at thoracic levels and an ∼20% increase at brachial and lumbar levels ( I, I, and S1J). We detected an ∼2-fold increase in the number of MMC neurons at brachial, thoracic, and lumbar levels of Nkx2.2 mutants, whereas the number of HMC, PGC, and LMC neurons was unchanged ( C, 2D, 2I, C, S1D, and S1G–S1J). Quantitatively, the increase in total motor neuron number in Nkx2.2 mutants could be accounted for, in its entirety, by the increase in MMC number ( I, I, and S1J). These findings indicate that all of the additional motor neurons generated from an ectopic ventral position in Nkx2.2 mutants acquire an MMC identity.
To induce the differentiation of motor neurons at ectopic dorsal positions, we used in ovo electroporation in chick spinal cord to express an isoform of the Shh receptor subunit, Smoothened (Smo ), which activates the Shh transduction pathway constitutively and in a cell-autonomous manner ( ). Stage 12–14 chick thoracic neural tube was electroporated unilaterally with a Smo ::IRES::nGFP construct and the identity of GFP-labeled progenitors and postmitotic motor neurons analyzed between stages 21 and 30. Expression of Smo resulted in a dorsal expansion of the domain occupied by Olig2 pMN domain progenitors and an ∼2-fold increase in total number of Olig2 progenitor cells at stages 21 to 24 ( A, S2B, and S2E). Expression of Smo also elicited a 1.7-fold increase in the number of Isl1/2 motor neurons at stages 29 to 30 (p < 0.01 versus controls) ( H and S2J). In addition, Smo expression induced the ectopic dorsal differentiation of V2a and V3 neurons, two interneuron classes that derive from the p2 and p3 progenitor domains that flank, dorsally and ventrally, the position of motor neuron generation ( G, S2I, and S2J) ( ). The induction of V2a neurons, motor neurons, and V3 neurons in response to Smo expression presumably reflects variation in the level of activation of the Shh transduction pathway in individual progenitor cells.
We analyzed the columnar identity of motor neurons generated at ectopic dorsal positions in the thoracic spinal cord. The number of MMC neurons was unchanged after Smo expression (p > 0.05, versus control side). In contrast, the number of HMC neurons increased ∼2.5 fold, and the number of PGC neurons increased ∼2 fold ( E–2H and 2J) (p < 0.01 versus controls). Thus, few, if any, of the ectopic dorsal motor neurons induced by Smo acquire an MMC identity, despite activation of the Shh transduction pathway at levels that span the range sufficient for motor neuron induction. Together, these findings show that cell position along the dorsoventral axis of the spinal cord has a marked influence on the probability of generation of MMC neurons ( K).
### Patterned Expression of Wnt Genes in the Ventral Spinal Cord
We next considered the possible source and identity of extrinsic signals that specify MMC neuronal subtype. Since MMC differentiation is highly sensitive to dorsoventral position, we considered whether the cells of the floor plate or adjacent ventral neural tube might serve as a source of relevant inductive signals. Our data suggest that Shh signaling alone is insufficient to specify MMC fate, prompting us to examine other candidate signals. We focused on Wnt proteins because of their known expression in the ventral spinal cord ( ) and their ability to induce expression of a key MMC transcriptional determinant, Lhx3, in neuroendocrine cells ( ). We analyzed the expression of Wnts as well as soluble frizzled-related proteins (Sfrps)—a class of secreted Wnt-binding proteins that inhibit Wnt signaling—and the Frizzled (Fz) class of Wnt receptors ( ).
We examined the expression of 17 Wnt genes in the ventral spinal cord of e9.5 to e11.5 mouse embryos, the peak period of motor neuron generation. Wnt ligands have been assigned to three main classes on the basis of their signal transduction pathways. Wnt1, -3, and -8 are strong activators, and Wnt7 proteins weak activators of the β-catenin transduction pathway, whereas Wnt4/5 proteins typically fail to activate β-catenin transduction, instead engage PKC, CamKII, or intracellular Ca signaling pathways ( ).
Wnt1 , - 3 , and - 3a were expressed exclusively in the dorsal spinal cord ( M; data not shown; see ). Members of the Wnt8 family ( Wnt8a , - 8b , and - 8c ) were not expressed in chick or mouse spinal cord (data not shown; see ). Wnt7a and - 7b were expressed in a dorsal to ventral gradient within the ventral spinal cord between e9.5 and e10.5 ( N–3P). The level of expression of Wnt7b appeared greater than that of Wnt7a , but the ventral boundary of Wnt7a extended more ventrally than that of Wnt7b , approaching the floor plate ( N and 3O; ).
In contrast, Wnt4 , -5a , and - 5b were expressed in a ventral to dorsal gradient within the ventral spinal cord of mouse embryos. Wnt4 was expressed at high levels in the floor plate and p3 domain ( A and 3B) ( ), and its expression level decreased in more dorsal regions. Wnt4 was also expressed at high levels in the dorsal spinal cord ( A and 3B) ( ). At e9.5, Wnt5a was expressed at high levels by progenitor cells throughout the ventral spinal cord, but by e10.5 became largely restricted to the floor plate and pMN domain ( D and 3E). Wnt5b was expressed by the floor plate between e9.5 and e10.5 ( G and 3H). Quantitatively, analysis of the cumulative level of Wnt4 , -5a , and -5b transcripts in the ventral spinal cord of mouse embryos revealed a ventral to dorsal expression gradient at e9.5 and e10.5 ( J and 3K).
We also analyzed the expression of Wnt ligands in chick spinal cord between stages 18 to 24, the peak period of motor neuron generation. The patterns of expression of Wnt1 , -3 , -3a , -7a , and -7b were similar to those in mouse (data not shown; ). Wnt4 was not expressed by the floor plate but was detected at high levels in the p3 progenitor domain and in the dorsal spinal cord ( C) ( ). Wnt5a and - 5b were expressed at high levels in the floor plate and at lower levels within the p3 domain ( F and 3I). The cumulative distribution of chick Wnt4 , - 5a , and - 5b transcripts within the ventral spinal cord revealed a ventral to dorsal expression gradient comparable to that observed in mouse embryos ( L).
We next examined the expression of frizzled class Wnt receptors and secreted frizzled related proteins (Sfrps). Four Fz genes were expressed in the ventral spinal cord over the period of motor neuron generation. Fz2 and Fz7 were expressed by neural progenitors, in both mouse and chick ( D and S3E), with near-uniform levels of expression along the dorsoventral axis of the spinal cord ( F). Fz9 was detected in mouse, but not chick, progenitor cells (data not shown), and Fz3 was expressed broadly in both progenitor cells and postmitotic neurons (data not shown; ). Other Fz genes were not detected in ventral progenitors or motor neurons at these developmental stages (data not shown). A high level of Sfrp2 was detected in the ventral spinal cord at e9.5 to e10.5, with a ventral boundary at the interface of the p3 and pMN domains ( B; data not shown; ). The pattern of Sfrp1 expression was similar to that of Sfrp2 , although low levels of expression were also detected within the p3 domain at e9.5 ( A; data not shown). The cumulative profile of Sfrp1/2 expression along the dorsoventral axis of the ventral spinal cord was inverted when compared to that of Wnt4 , Wnt5a , and Wnt5b transcripts, exhibiting a dorsal to ventral expression gradient ( C).
### Wnt4 and Wnt5 Promote MMC Columnar Identity
The inverse dorsoventral gradients of Wnt7a / 7b and Wnt4 / 5a / 5b expression led us to test whether the specification of MMC neurons results from the evasion of dorsally derived Wnt7 ligands or from exposure to ventrally derived Wnt4/5 ligands. To resolve this issue, we explored whether any of these Wnts have MMC-inducing activity in chick spinal cord in vivo. We used in ovo electroporation to coexpress Wnt cDNAs together with a marker eGFP construct in the ventral spinal cord of stage 12–14 chick embryos and assessed the columnar identity of motor neurons by their transcriptional profile and axonal projection pattern at stages 26–29. We focused on thoracic and lumbar levels of the spinal cord for this analysis because, at these more caudal levels, transgene expression can be achieved at an earlier stage of neuronal differentiation. The activities of Wnt4 , Wnt5a , Wnt5b , and Wnt7b were compared with that of Wnt1 , a strong activator of β-catenin transduction.
Expression of Wnt4 , Wnt5a , or Wnt5b did not change the number of Olig2 motor neuron progenitors ( A–S4D and S4G; data not shown), nor was there a change in motor neuron number at brachial, thoracic, or lumbar levels of the spinal cord ( J, 4K, E, J, and S6K; data not shown). Nevertheless, Wnt4 , - 5a , or - 5b expression elicited a 1.7- to 2.0- fold increase in the number of MMC neurons at thoracic levels, assessed by coexpression of Lhx3/4 and Isl1/2 (p < 0.01; Student's t test; n = 16–20 embryos for each Wnt assayed) ( A–4F, 4J, 4K, A, S5B, and S5E). The increase in MMC neurons was accompanied by a 2-fold decrease in the number of HMC neurons (p < 0.01; Student's t test; n = 16–20 embryos), whereas the number of PGC neurons was not changed ( A–4F, 4J, 4K, A, S5B, and S5E). Expression of Wnt4 and Wnt 5a , the two Wnts analyzed at lumbar levels, increased the number of MMC neurons ∼2.7 fold, at the expense of LMC neurons which exhibited a small decrease in number (p < 0.05; Student's t test; n = 16–20 embryos) ( A–S6F, S6J, and S6K). We also examined whether there is a temporal constraint on Wnt4/5 signaling activity. We found that thoracic or lumbar electroporation of Wnt4 , - 5a , or - 5b at stage 18, rather than stage 12–14, failed to elicit a change in motor neuron columnar identities (data not shown). Thus, early, but not late, expression of each of three noncanonical Wnts in the ventral spinal cord enhances the generation of MMC neurons, at the expense of other motor neuron columnar subtypes.
We also determined if the axons of Lhx3/4 motor neurons induced by Wnt4/5 signaling pursue a trajectory that is consistent with their apparent MMC character. To assess this, we monitored the transcriptional status of retrogradely labeled motor neurons in the thoracic spinal cord of stage 29–30 chick embryos after injection of horseradish peroxidase (HRP) into axial muscles ( A). On the control side of the spinal cord, ∼60% of all Lhx3 MMC neurons accumulated HRP, and more importantly, all HRP-labeled neurons expressed Lhx3 ( B–5E, 5J, and 5K). On the side of the spinal cord transfected with Wnt5a , we detected an ∼2.1-fold increase in the total number of MMC neurons ( G and 5J). We found that ∼55% of all Lhx3 motor neurons accumulated HRP, a proportion similar to that found in controls (p = 0.58) ( F–5I and 5K). Furthermore, all HRP-labeled motor neurons expressed Lhx3 ( F–5I). Together, these findings indicate that the extra Lhx3 neurons generated in response to enhanced Wnt4/5 expression send their axons along a dorsal branch that takes them to axial muscles—the trajectory and target of MMC neurons.
Wnt1 signaling has been reported to regulate the pattern of homeodomain transcription factors in ventral progenitor cells ( ), prompting us to examine whether Wnt1 activity influences the specification of MMC neurons. We found that ectopic expression of Wnt1 in the ventral spinal cord results in an ∼2-fold increase in the number of Olig2 progenitor cells ( E–S4G) and an ∼1.5 fold increase in the number of postmitotic motor neurons (both p < 0.01 versus controls, Student's t test; n = 10 embryos) ( G–4I, 4L, G–S6I, and S6L). Similar Wnt1 inductive activities were observed at thoracic and lumbar levels of the spinal cord. At thoracic levels, Wnt1 expression did not change the number of MMC neurons (p > 0.05; Student's t test; n = 10 embryos), but led to an ∼1.7-fold increase in the number of HMC and PGC neurons (p < 0.01, Student's t test; n = 10 embryos) ( G–4I and 4L). Similarly, at lumbar levels, Wnt1 expression did not change the number of MMC neurons but led to an ∼1.5 fold increase in the number of LMC neurons ( G–S6I and S6L). Neurons in the medial and lateral divisions of the LMC exhibited a similar increase in neuronal number (p < 0.01; Student's t test; n = 10 embryos; data not shown). The selective increase in PGC, HMC, and LMC neurons upon Wnt1 expression may have its basis in the enhanced proliferation of ventral progenitor cells ( ), such that most of the additional motor neurons are generated at more dorsal positions, beyond the range of Wnt4/5 signaling ( E and S4F). Expression of Wnt7b , in contrast, did not change the number of Olig2 motor neuron progenitors or total motor neuron number ( F; data not shown). Furthermore, the fraction of motor neurons allocated to individual motor columns was not altered by thoracic or lumbar Wnt7b expression ( C, S5D, and S5F; data not shown). Thus, Wnt4/5 but not Wnt1 or Wnt7b activities enhance the generation of MMC neurons at the expense of other motor neuron columnar subtypes.
### Persistence of MMC Identity after Disruption of the Wnt Planar Polarity Pathway
We attempted to clarify the neural signaling pathway that links Wnt4/5 activity to the maintenance of Lhx3 expression. The observation that Wnt1 fails to mimic the MMC-inducing activity of Wnt4/5 argues against the involvement of the β-catenin pathway ( ). In certain cellular contexts, noncanonical Wnt ligands, including Wnt4/5, interact with Fz3, Fz5, and Fz7 receptors ( ). We therefore examined whether overexpression of Ig-modified versions of the CRD ectodomains of Fz5 and Fz7 ( ) in embryonic chick spinal cord is able to block MMC specification. We found, however, that both these reagents severely reduced the total number of motor neurons (data not shown), precluding a meaningful analysis of MMC specification. The reduction in motor neuron generation is likely to reflect the blockade of canonical Wnts that promote cell proliferation in the ventral neural tube ( ).
Many noncanonical Wnt ligands activate the vertebrate planar cell polarity pathway, a transduction system that depends on the function of Vangl2/Ltap, a vertebrate homolog of the Drosophila Van Gogh/Strabismus protein ( ). We therefore considered whether Wnt4/5 induction of MMC identity involves this signaling pathway, analyzing motor neuron differentiation in loop tail mutant mice (which carry a null mutation in the Vangl2/Ltap gene) ( ). At e13.5, the total number of motor neurons and the proportional allocation of MMC neurons were similar in the thoracic spinal cord of wild-type and loop tail embryos ( A–S7G) (p > 0.05; Student's t test; n = 3 embryos), arguing against the involvement of the Wnt planar cell polarity pathway in MMC specification.
### Switch from MMC to Segmental Columnar Subtypes in Wnt4/5 Mutant Mice
To address the requirement for Wnt4/5 signaling in the specification of MMC identity, we used mouse genetics to examine the impact of reducing Wnt activity on the assignment of motor neuron columnar identities. We assessed motor neuron columnar identity in 14 of 27 possible Wnt4 , - 5a , and - 5b allelic combinations, eliminating from one to five Wnt alleles ( ). We found that Wnt4 ; Wnt5a ; Wnt5b triple-mutant embryos died before the onset of motor neuron differentiation (data not shown). Wnt5a mutants exhibit severe defects in limb development ( ) that are likely to perturb motor neuron differentiation at limb levels of the spinal cord through other cellular mechanisms ( ). For this reason, we focused our analysis primarily on thoracic levels, assessing the impact of progressive removal of Wnt alleles on motor neuron differentiation.
Mice heterozygous for Wnt4 , Wnt5a , or Wnt5b did not show a significant difference in thoracic motor neuron number, nor in the fractional representation of motor columnar subtypes when compared to wild-type embryos ( F–6I). Similarly, we found that Wnt4 , Wnt5a , or Wnt5b single-mutant embryos exhibited no significant difference in thoracic motor neuron number, nor in the representation of motor columnar subtypes when compared with wild-type or heterozygous embryos ( F–6I).
Analysis of mice carrying three mutated Wnt alleles ( Wnt4 ; Wnt5a and Wnt4 ; Wnt5a genotypes) revealed that the total number of motor neurons was unchanged ( F). In Wnt4 ; Wnt5a mice, we detected a 46% decrease in the number of MMC neurons (24 ± 2.0 neurons in wild-types versus 12.7 ± 2.5 in Wnt4 ; Wnt5a embryos; ANOVA test; p < 0.01; n = 6 embryos) ( H). The reduction in MMC neuronal number was accompanied by an increase in the number of HMC neurons (26 ± 4.0 neurons in wild-types versus 32 ± 3.5 in Wnt4 ; Wnt5a embryos; ANOVA test; p = 0.02; n = 6 embryos), whereas the number of PGC neurons was unchanged ( G and 6I). Analysis of Wnt4 ; Wnt5a mutants revealed a smaller (22%) decrease in the number of MMC neurons (24 ± 2.0 neurons in wild-types versus 17.8 ± 2.4 in Wnt4 ; Wnt5a mutants; ANOVA test; p = 0.022; n = 5 embryos), and there was no significant change in the representation of other columnar subtypes ( G–6I).
In Wnt4 ; Wnt5a , Wnt5a ; Wnt5b , and Wnt4 ; Wnt5b double mutant embryos, there was no change in thoracic motor neuron number, but a more marked (50%–60%) decrease in the number of MMC neurons (58% decrease in Wnt4 ; Wnt5a , 58% decrease in Wnt4 ; Wnt5b , and 50% decrease in Wnt5a ; Wnt5b genotypes, compared to wild-type; ANOVA test; p < 0.01; n = 3–5 embryos per genotype) ( B and 6H). Conversely, there was a significant increase in the number HMC neurons (40 ± 4 neurons in Wnt4 ; Wnt5a , 35 ± 4 neurons in Wnt5a ; Wnt5b , and 40 ± 4 neurons in Wnt4 ; Wnt5b versus 26 ± 4 neurons in wild-type littermates; ANOVA test; p < 0.05; n = 3–5 embryos per genotype) ( B and 6I). These findings provide evidence that Wnt4 , Wnt5a , and Wnt5b each contribute to the specification of MMC identity.
Each of the three combinations of five mutated Wnt alleles exhibited a marked decrease in MMC neuronal number, compared to wild-type controls (59% in Wnt4 ; Wnt5a ; Wnt5b , 63% in Wnt4 ; Wnt5a ; Wnt5b , and 67% in Wnt4 ; Wnt5a ; Wnt5b ; ANOVA test; p < 0.01) ( C–6E and 6H). In addition, there was a compensatory increase in HMC neuronal number (62% in Wnt4 ; Wnt5a ; Wnt5b , 50% in Wnt4 ; Wnt5a ; Wnt5b , and 39% in Wnt4 ; Wnt5a ; Wnt5b ; ANOVA test; p < 0.01) ( C–6E and 6I). In contrast, the number of PGC neurons was unchanged ( G). These findings provide genetic evidence that, at thoracic levels, prospective MMC neurons switch to an alternate, HMC, fate in the absence of Wnt4/5a/5b signaling.
We also analyzed the impact of eliminating Wnt4/5 signaling on motor columnar identity at lumbar levels of the spinal cord. As discussed, this analysis was complicated by the fact that loss of Wnt4/5 signaling impairs limb development, with potential secondary consequences for motor neuron differentiation and survival. Indeed, embryos carrying four or five mutated Wnt4/5 alleles exhibited a small (∼20%) reduction in total motor neuron number at hindlimb levels, compared to wild-type controls (ANOVA test, p < 0.01 versus controls) ( A). Nevertheless, we detected a greater (∼50%) decrease in the fraction of MMC neurons at lumbar levels of Wnt4 ; Wnt5b and Wnt4 ; Wnt5a ; Wnt5b mutants (ANOVA test; p < 0.01 versus controls; n = 3 embryos) ( B). These findings support the view that Wnt4, Wnt5a, and Wnt5b signaling promotes the generation of MMC neurons at limb as well as thoracic levels of the spinal cord.
## Discussion
Vertebrates have discovered many uses for axial muscles, each of which depends on their activation by spinal motor neurons located in the MMC. But the fundamental issue of how motor neurons acquire an MMC identity that ensures axial muscle activation has not been resolved. We have found that Wnt4/5 expression by cells in and adjacent to the ventral midline of the spinal cord promotes the progression of nascent motor neurons to an MMC fate. Our findings reveal the existence of two parallel ventrodorsal signaling gradients, mediated by Hh and Wnt proteins, and show that these two signaling pathways have convergent functions in specifying the identity and connectivity of spinal motor neurons. Wnt4 signaling has also been shown to direct the rostral trajectory of the axons of spinal commissural neurons, after their passage through the floor plate ( ). Noncanonical Wnts therefore join BMPs ( ) and Shh ( ) as versatile signaling factors that specify both neuronal fate and axonal trajectory in the developing spinal cord.
### Wnt4/5 Signaling Promotes MMC Identity
Over the period that motor neurons acquire their columnar identities, cells in the ventral spinal cord express three noncanonical Wnt genes, Wnt4 , Wnt5a , and Wnt5b. The composite profile of these genes appears to establish a ventral-to-dorsal gradient of Wnt4/5 transcript expression in the ventral spinal cord. Overexpression of Wnt4/5 increases the generation of MMC neurons, at the expense of segmental columnar classes, whereas reducing Wnt4/5 expression depletes the spinal cord of MMC neurons and promotes the generation of segmental classes ( A). These findings indicate that Wnt4/5 signaling is a determinant of MMC identity.
Both Shh and Wnt4/5 signals are graded in character, but there are notable differences in the origin and operation of these gradients. Shh is expressed by floor plate cells, whereas the noncanonical Wnts are expressed over a broader ventral domain, suggesting that the strategy used to establish a Wnt4/5 signaling gradient differs from that underlying the Shh gradient. Shh's pervasive influence on ventral neuronal specification and patterning relies on the long-range spread of Shh protein within the ventral neural epithelium ( ). In contrast, the range of activity of many secreted Wnts is more limited ( ), supporting the idea that the spatial profile of Wnt4/5 activity is established primarily through the graded ventral expression of Wnt transcripts. A second difference in the logic of Shh and Wnt4/5 signaling may be the concentration dependence of their activities. Shh functions as a gradient morphogen—specifying distinct ventral cell fates at different concentration thresholds ( )—whereas the specification of MMC fate could simply require exposure to a critical threshold level of Wnt4/5 signaling. In this view, the graded expression of Wnt4/5 transcripts may merely serve as an effective strategy for ensuring that an appropriate fraction of cells within the pMN domain are exposed to a threshold level of Wnt4/5 activity.
The slope and spread of the Wnt4/5 activity gradient within the ventral spinal cord appears to be steeper and shorter than that of the Shh gradient. We infer this from the observation that all motor neurons generated in ectopic ventral positions acquire MMC character whereas few if any of the ectopic motor neurons generated dorsal to the normal pMN domain do so. The dorsal limit of Wnt4/5 signaling activity may be constrained by the high level of Sfrp expression evident within the p0, p1, and p2 progenitor domains ( ). In this view, the secretion of Sfrp proteins may block the actions of secreted Wnt4/5 proteins that manage to reach these more dorsal domains of the ventral neural tube. The inverted dorsoventral profiles of Wnt4/5 and Sfrp expression are therefore likely to contribute to the restricted range of Wnt4/5 signaling evident within the ventral spinal cord.
How does the Wnt4/5 activity gradient determine the position of generation of motor neuron columnar subtypes within the pMN domain? Were Wnt4/5 protein activity to extend throughout the pMN domain, the probability of generation of neurons of the MMC and segmental motor columns would presumably change smoothly as a function of the dorsoventral position of progenitor cells within the domain. Alternatively, the limit of Wnt4/5 signaling activity could be located within the pMN domain, such that only the most ventrally positioned pMN domain progenitors would have the opportunity to generate MMC neurons. Independent of the linear or step landscape of Wnt4/5 signaling, our findings imply that the diversification of neurons within a single ventral progenitor domain depends on a dorsoventral difference in the intensity or quality of inductive signals ( A). This position-dependent plan for motor neuron diversification differs conceptually from the mosaic, position-independent mode of Notch signaling that directs to the diversification of certain ventral interneuron subtypes ( ). Plausibly, the combination of both strategies within a single ventral progenitor domain could further enhance the diversity of neuronal subtypes.
### Wnt4/5 Signaling and the Origins of Spinal Motor Neuron Diversity
Shh and Wnt4/5 signals activate different components of the transcriptional network that controls spinal motor neuron differentiation. Shh signaling is essential for the specification of generic motor neuron character, revealed by the expression of the Nkx6.1/.2, Isl1/2, and Mnr2/Hb9 HD proteins ( ). In contrast, Wnt4/5 signaling operates only in the context of a core transcriptional profile established by Shh activity and directs the progression of generic motor neurons to an MMC fate by promoting Lhx3/4 expression. Neural progenitors destined to give rise to other motor neuron columnar subtypes also transiently express Lhx3 ( ), indicating that Wnt4/5 signaling specifies MMC fate by programming progenitor cells and/or nascent motor neurons to maintain Lhx3/4 expression after their exit from the cell cycle ( B).
Why do residual MMC neurons persist under conditions in which five of the six noncanonical Wnt alleles have been removed? The activity of the one extant Wnt4/5 allele could be sufficient to generate a significant number of MMC neurons. Alternatively, additional noncanonical Wnts could still be at work—the early expression of Wnt11 by axial mesodermal cells ( ) could transfer active protein to overlying ventral neural tissue. It is also conceivable that persistent expression of Lhx3 in motor neurons is programmed spontaneously at a low incidence, with Wnt4/5 signaling serving to increase the probability of maintained Lhx3 expression. Defining the transduction pathway through which Wnt4/5 signaling maintains expression of Lhx3 in postmitotic motor neurons may help to distinguish these possibilities. Our findings argue that Wnt4/5 signals are not mediated by canonical β-catenin or planar cell polarity pathways, but leave unresolved the relevant Wnt receptors and intracellular signals.
The discovery that Wnt4/5 signaling specifies MMC character supplies a missing link in the molecular logic of motor neuron columnar diversification and provides a more coherent view of this developmental program. Our findings, together with studies on the specification of LMC and PGC identities ( ), indicate that the program of motor neuron differentiation initiated by Shh signaling generates a set of Hb9 , Lhx3 neurons that represent a “ground-state” character of spinal motor neurons. At the time of their generation, this ground-state motor neuron cohort is exposed to two further, opponent, inductive influences: a dorsoventral Wnt4/5 signaling pathway that maintains Lhx3 expression and directs MMC character and a rostrocaudal FGF pathway that patterns Hox expression and so directs LMC and PGC character ( B). Our data suggest that selection of the Wnt4/5-Lhx3 program precludes neurons from pursuing the FGF-Hox pathway and vice versa. Motor neurons that fail to pursue either of these two options appear to progress to an HMC columnar fate ( ).
The opponent activities of the Wnt4/5-Lhx3 and FGF-Hox signaling pathways can account for the differing efficiencies of motor neuron columnar interconversion observed under conditions of altered Wnt4/5 signaling. Increasing Wnt4/5 activity results in the conversion of many prospective HMC neurons to an MMC fate. In contrast, prospective LMC and PGC neurons appear to switch fates at much lower efficiency—a consequence of the activation of Hox proteins in these cells ( ). Consistent with this view, the loss of MMC neurons observed at thoracic levels after reductions in Wnt4/5 signaling is accompanied by a preferential increase in HMC, rather than PGC, neurons. This competitive signaling scheme also provides a potential explanation for the finding that enhanced Wnt4/5 signaling does an incomplete job in converting motor neurons to an MMC fate—engagement of the FGF-Hox pathway will have begun to recruit some cells at the time of exposure to Wnt4/5 signals. However, this scheme does not explain why, in Nkx2.2 mutants, all of the extra motor neurons generated within the former p3 domain acquire MMC character. One possible reason is that the peak of Wnt4/5 expression in and around the ventral midline results in the activation of Wnt4/5 signaling in the p3 domain at much higher levels than in the pMN domain and thus more effectively recruits cells away from the FGF-Hox option. The high level of Shh signaling activity within the p3 domain could also bias cells in favor of the Wnt4/5-Lhx3 pathway.
Finally, our findings raise the possibility that regulation of Wnt4/5 signaling strength—the operation of a Wnt4/5 rheostat—constitutes a crucial step in transforming the spinal motor system from an MMC-centric plan that typifies early aquatic vertebrates (and larval teleost and amphibian forms) ( ) to the more diversified plan of columnar organization and connectivity that characterizes birds and mammals. Conditions of high-level Wnt4/5 signaling may prevail in the ventral neural tube of early aquatic vertebrates, ensuring that most or all Shh-specified motor neurons progress to an Lhx3 MMC-like character. If so, a decrease in the strength of Wnt4/5 signaling may constitute a critical, enabling step in the formation of a population of Shh-specified motor neurons that fail to acquire Lhx3 expression. This Lhx3 motor neuron ground state has recently been shown to serve as the cellular substrate for the columnar programming activities of Hox proteins and FoxP cofactors which direct the formation of LMC and PGC neurons ( ).
## Experimental Procedures
### cDNA Probes
Mouse Wnt1-7b cDNAs were provided by J. Kitajewski (Columbia University); Wnt9a-16 cDNAs were amplified by RT-PCR. Chick Wnt4 was provided by C. Tabin (Harvard University), and chick Wnt5a and Wnt5b were amplified from total mRNA. Mouse Wnt1 , - 4 , -5a , -5b , and - 7b cDNAs were amplified by PCR and cloned into the pCIG expression vector ( ). Fz2 -7 cDNAs were provided by J. Nathans (John Hopkins University); Fz9 cDNA was provided by U. Francke (Stanford University); Fz1 , Fz8 , Sfrp1 , and Sfrp2 cDNAs were amplified by RT-PCR. Partial cDNAs for chick Fz2 , -3 , -7 , and - 9 were amplified from HH stage 24 embryos. Smo cDNA ( ) was obtained from L. Zeltser (Columbia University) and cloned into the pCIG expression vector.
### In Situ Hybridization, Immunohistochemistry, and In Ovo Electroporation
In situ hybridization was performed on sections of e9.5 to e10.5 mouse or HH st.18 to 21 chick embryos ( ). Antisense mRNA probes were generated with the DIG RNA Labeling Kit (Roche Applied Science). The graphs for Wnt transcript expression were generated with Sigma Plot after measuring the levels of transcript from the ventral midline to the intermediate spinal cord (black arrow) using ImageJ (NIH). The abscissa in these plots represents the average value for each transcript from eight different thoracic sections. The ordinate indicates distance from the ventral midline (microns).
In ovo electroporation into chick neural tube was performed as described ( ). Immunohistochemistry was performed on 15 μm cryostat sections as described ( ). Retrograde labeling of MMC neurons after tracer HRP injection into axial muscles was performed as described ( ).
### Mouse Strains
Nkx2.2 , Wnt4 ; Wnt5a , and Vangl2/Ltap mice were genotyped as described ( ). Wnt5b mice were generated by insertion of the PGK-Neo cassette in exon4 using the endogenous PstI and SacI sites followed by gene targeting in ES cells. The wild-type allele of Wnt5b animals was genotyped with the following primers: Wnt5bwtF (5′GGG ACT CGA ACT CAG ATT GTC AGG3′) and Wnt5bwtR (5′ATG AGC TCG CAG CCG TCC AT3′), located on intron 3 and exon 4, respectively; this generates a 500 bp fragment. The mutant allele was screened with primers: Wnt5bwtF (5′GGG ACT CGA ACT CAG ATT GTC AGG3′) and Wnt5bmutR (5′GCA GGC ATG CTG GGG ATG CGG3′) that amplify a 250 bp band. Animals were housed in the Columbia University Animal Facility and handled according to institutional guidelines.
### Statistical Analysis
Student's t test was used to determine the significance of values of the number of motor neurons and motor columns in experiments where two genotypes or conditions were compared. Mixed effect analysis of variance (ANOVA) was used to test whether variances in the number of total motor neurons or motor columns were different between the wild-type genotype, and the allelic combinations that contained mutated Wnt4 , Wnt5a or Wnt5b alleles (a total of 17 genotypes).
|
Summary
The behavioral response to a sensory stimulus may depend on both learned and innate neuronal representations. How these circuits interact to produce appropriate behavior is unknown. In Drosophila , the lateral horn (LH) and mushroom body (MB) are thought to mediate innate and learned olfactory behavior, respectively, although LH function has not been tested directly. Here we identify two LH cell types (PD2a1 and PD2b1) that receive input from an MB output neuron required for recall of aversive olfactory memories. These neurons are required for aversive memory retrieval and modulated by training. Connectomics data demonstrate that PD2a1 and PD2b1 neurons also receive direct input from food odor-encoding neurons. Consistent with this, PD2a1 and PD2b1 are also necessary for unlearned attraction to some odors, indicating that these neurons have a dual behavioral role. This provides a circuit mechanism by which learned and innate olfactory information can interact in identified neurons to produce appropriate behavior.
## Video Abstract
Highlights
Specific Drosophila lateral horn neurons mediate innate attraction to food odors
The same neurons receive plastic odor information from the mushroom body
Recall after associative learning depends on reduced drive to lateral horn neurons
Connectomics circuit for integration of learned and innate odor representations
Sensory stimuli can engage both learned and innate behaviors. Dolan et al. identify neurons in Drosophila that directly integrate unlearned and plastic odor representations; they are required for innate approach to food odors but also learned aversive recall.
## Introduction
The action of natural selection on evolutionary timescales endows animal species with behavioral responses to stimuli of particular ethological relevance. In addition, most animals show adaptive responses based on learning during their lifetime. Learning may modify an unlearned response. However, it remains unknown how memory recall interacts with innate sensory representations to produce the most appropriate behavior. This study explores this general issue using the Drosophila olfactory system. Olfaction is a shallow sense (in terms of neural processing) with a privileged connection to memory systems in many species ( ). Genetic tractability and numeric simplicity make the Drosophila brain an ideal model to study this interaction at a neural circuit level, whereas the similarity in organization of peripheral olfactory circuits makes it possible that neurobiological principles may also be shared deeper in the brain between insects and mammals ( ).
In Drosophila , olfactory sensory neurons project to specific glomeruli in the antennal lobe ( ). Following local computations, excitatory uniglomerular projection neurons (PNs) make divergent connections to two higher processing regions, the lateral horn (LH) and the mushroom body (MB) ( ), in addition to other antennal lobe (AL) outputs ( , ). The prevailing model of olfactory processing proposes a clear functional division between these regions: the MB is required for learning, consolidation, and retrieval of olfactory memories, whereas the LH is thought to mediate innate behavior ( , ). Many studies have confirmed the necessity of the MB for associative memory, where a reward or punishment (the unconditioned stimulus [US]) is associated with one odor (the conditioned stimulus [CS+]), but not with a second odor (CS−) ( ). The role of the LH in innate behavior has been inferred from experiments that silenced the MB and observed innate olfactory responses ( , ). However, no studies to date have directly examined the behavioral functions of LH neurons in olfaction.
Mapping studies show that PNs from different glomeruli have stereotyped axonal projections in the LH ( , , ), consistent with a role in innate olfactory behaviors. Anatomical and physiological analyses have shown a role for specific Drosophila LH neurons in processing pheromone cues relevant to sex-specific behaviors such as courtship and aggression ( , , , ). Recent results have shown that some LH neurons can also show stereotyped responses to general olfactory stimuli ( , ) and are stereotypically connected to input PNs ( ). In addition, new large-scale data have confirmed response stereotypy and showed that different LH neurons have wide variations in odor tuning and may encode odor categories ( , ).
In contrast to the LH, MB neurons are extremely well characterized ( ). The dendrites of intrinsic MB neurons (Kenyon cells) are localized to a region called the calyx, where they sample incoming PN axons in an apparently random manner ( ). Kenyon cells have parallel, axonal fibers that form five different lobes, with three distinct branching patterns that define as many Kenyon cell types ( ). Anatomical analysis has subdivided the lobes into 15 compartments, each innervated by specific dopaminergic input neurons (DANs) and MB output neurons (MBONs) ( ). These compartments are anatomically and physiologically distinct ( , ), although each Kenyon cell axon synapses in all compartments of each lobe ( ).
Odors are sparsely represented in the Kenyon cell assembly, so only a subset of axon terminals will release neurotransmitters upon olfactory stimulation ( ). Electric shock, the US during aversive learning, activates a subset of DANs so that, when US and CS+ are coincident, the subset of olfaction-driven Kenyon cells also receives dopaminergic input within specific compartments. This coincident input produces compartment-specific synaptic plasticity ( , , , , ), changing the response of that compartment’s MBON to the CS+. MBONs function in valence behaviors, and a modified response to the trained odor may bias the fly’s behavior toward avoidance or attraction depending on the compartment ( , ). One of these output neurons, MBON-ɑ2sc (also known as MB-V2ɑ), projects from the MB to several brain regions, including the LH ( , ). Optogenetic stimulation of the entire V2 cluster (MBON-α2sc, MBON-α′3m, and MBON-α′3ap) drives approach behavior, but activation of MBON-ɑ2sc alone does not lead to any change in valence behavior ( ). Previous work has demonstrated that MBON-ɑ2sc is required for the retrieval of aversive olfactory memories across short, medium, and long timescales ( , ) although not necessary for the recall of appetitive memories ( ). Recordings from MBON-ɑ2sc demonstrated that it is broadly odor-responsive ( ) but depresses its response to CS+ after training ( , ). This depression to the trained odor response is thought to spread to unknown downstream neural circuits mediating aversive olfactory memory retrieval ( , , ), in addition to an increased drive of negative valence MBONs ( , , ). Given the presumed role of the LH in innate olfaction, the function of the MB to LH projection of MBON-α2sc is unclear. Is memory information transmitted to the LH, and if so, is this communication required for retrieval of the aversive memory?
In this study, we examine the behavioral function of this connection between the presumed innate and learned olfactory processing centers. We use computational anatomy and microscopy to identify two LH output neuron cell types (PD2a1 and PD2b1) postsynaptic to MBON-ɑ2sc. We use whole-brain electron microscopy connectomics ( ) to verify this synaptic connectivity and then test the function of these cell types in behavior. Contrary to the model described above, where the LH mediates only innate olfactory behavior, PD2a1 and PD2b1 are necessary for memory retrieval. We generate new split-GAL4 lines ( , ) specifically targeting these neurons to confirm their necessity for memory recall. Calcium imaging shows that PD2a1 and PD2b1 olfactory responses are depressed after training, similar to the MBON. Additional connectomics work finds direct olfactory PN input onto PD2a1 and PD2b1 dendrites, identifying these cells as responsive to food or appetitive odors. We then demonstrate that PD2a1 and PD2b1 neurons are necessary for innate olfactory attraction for several odors. This work provides a model for the interaction of innate and learned sensory information.
## Results
### Identifying LH Neurons Postsynaptic to MBON-ɑ2sc
To understand the role of information flow from the MB and LH, we first sought to identify postsynaptic neurons in the LH that receive input from MBON-ɑ2sc. We developed a computational pipeline to find MBON-ɑ2sc postsynaptic candidates. We used in silico overlap of GAL4 expression patterns to identify candidate postsynaptic partners of MBON-ɑ2sc. Using image registration ( ), we created a mask of the MBON-ɑ2sc axonal terminals expressing a presynaptically localized marker ( ). We then calculated pixel overlap of the mask with registered images of published GAL4 lines ( , ). We ranked lines by a relative “overlap score” for each brain that compared the GFP signal within the MB peduncle to exclude lines with MB Kenyon cell expression, which could complicate behavioral analysis. Scores for approximately 3,500 GAL4 lines ( A) were mostly close to zero or negative (having little or no LH overlap but strong peduncle expression). We focused on the top ∼100 lines (97th percentile). After excluding lines labeling MBON-ɑ2sc, the top hits identified 5 cell types putatively postsynaptic to MBON-ɑ2sc in the dorsal LH. Many lines were excluded because of broad expression, so there are likely other LH neurons that we could not analyze.
PD2a1 and PD2b1 Are Postsynaptic to MBON-ɑ2sc and Necessary for Memory Retrieval
(A) Distribution of LH overlap scores for MBON-ɑ2sc axon mask versus 3,500 GAL4 lines. Scores > 97 percentile are labeled in red, y axis clipped <−2,000.
(B) Sparsest GAL4 line labeling cell type PD2a1 and PD2b1, R37G11-GAL4 (image from ). Scale bar, 30 μm.
(C) z-projection of double labeling. MBON axons are labeled in magenta, and PD2a1 and PD2b1 are labeled with membrane-bound GFP (in green). This LexA line contains both MBON-ɑ2sc (dorsal) and MBON-α′3ap (ventral). Scale bar, 5 μm. The image is representative of n = 4.
(D–D”) Flies with R37G11-GAL4 driving Shi and genotypic controls were trained and tested with the illustrated protocols (restrictive temperature indicated in red). Silencing PD2a1 and PD2b1 neurons impaired immediate memory after single-cycle training (D; n = 12–13, F = 3.79, p = 0.033), 3-hr memory after single-cycle training (D’; n = 9, F = 12.07, p = 0.0002), and long-term memory after spaced training (D”; n = 9, F = 6.28, p = 0.0064).
(E) Flies expressing Shi driven by the 37G11-GAL4 driver showed normal olfactory avoidance to octanol (Oct) and methylcyclohexanol (Mch) compared with their controls at the restrictive temperature (Oct, n = 14, F = 2.41, p = 0.10; Mch, n = 14, F = 0.23, p = 0.79). Data are presented as mean ± SEM.
(F) Confocal z-projection of PD2a1 and PD2b1 driving both membrane-bound GFP (green) and Synaptotagmin-HA (gray). PD2a1 and PD2b1 has been manually segmented. The orange rectangle represents the inset. Inset: a single slice of PD2a1 and PD2b1 dendrites showing punctate Synaptotagmin-HA, indicating dendritic presynapses. The image is representative of n = 5.
(G–G”). ChAT immunohistochemistry demonstrating that PD2a1 and PD2b1 neurons are cholinergic. The images show a representative slice (n = 4 stacks). Scale bars, 5 μm.
See also .
We next generated a LexA line to orthogonally control MBON-ɑ2sc ( A). Double-labeling of MBON presynapses and various LH cell types furthered the number of candidates. Two cell types had potential synaptic sites identified by double labeling and high-resolution confocal microscopy: LH output neuron cell types posterior dorsal 2a1 and b1 (PD2a1 and PD2b1) ( B and 1C; see below for single-neuron data) and anterior ventral 6a1 (AV6a1) ( A and S2C). These names are based on a hierarchical nomenclature for over 150 LH cell types ( ). We also repeated this analysis for MBON axonal processes in the superior intermediate protocerebrum (SIP), identifying only one candidate postsynaptic cell type, SIP-1 ( B and S2D).
### PD2a1 and PD2b1 Are Necessary for Memory Retrieval
We identified the sparsest GAL4 lines for the three selected cell types identified and screened for memory retrieval defects when the neurons were silenced in an aversive olfaction-associative conditioning paradigm. LH cell types expressed the temperature-sensitive silencer shibire ( ), which inhibits neuronal signaling at high temperatures (33°C, the restrictive temperature). By raising the temperature during a memory test 3 hr after aversive olfactory conditioning, we could silence these neurons to probe their role in memory recall ( ).
Silencing the AV6a1 and SIP cell type GAL4 lines had no detectable effect on memory ( G and S2H). However, silencing PD2a1 and PD2b1 neurons with R37G11-GAL4 impaired 3-hr memory retrieval relative to genotype ( D’) and temperature ( B) controls. We extended these analyses of PD2a1 and PD2b1 to include immediate and long-term memory, which also require MBON-ɑ2sc ( , ). Silencing PD2a1 and PD2b1 neurons attenuated memory retrieval for both memory phases ( D and 1D”) versus controls ( A and S3C). Surprisingly, PD2a1 and PD2b1 inhibition had no effect on naive olfactory avoidance to the two training odors at the concentrations used in our memory assay ( E), so the observed phenotype was not due to defective innate olfactory processing, the proposed function of LH neurons. These results indicate that PD2a1 and PD2b1 activity is necessary during memory recall.
We confirmed that PD2a1 and PD2b1 are primarily an LH output cell type by expressing hemagglutinin (HA)-fused synaptotagmin (Syt::HA) to label presynapses ( ; F). We also observed some presynapses in the presumptive LH dendrites ( F). We next determined their neurotransmitter profiles. PD2a1 and PD2b1 was ChAT-immunoreactive ( G–G”) but gamma-aminobutyric acid (GABA)- and Drosophila vesicular glutamate transporter (dVGlut)-negative ( B and S1C; ). These neurons, therefore, appear to be excitatory cholinergic LH outputs, a conclusion we confirmed using a genetic approach to label cholinergic neurons ( ; D).
### Generation and Characterization of Cell-Type-Specific Split-GAL4 Lines
Although R37G11-GAL4 is relatively specific, it contained some other cell types that could confound our behavioral results. To confirm that PD2a1 and PD2b1 neurons are responsible for the memory retrieval deficit, we generated split-GAL4 lines ( , ) specific to PD2a1 and PD2b1 in the central brain ( A and 2B). We focused on two split-GAL4 lines, LH989 and LH991, that used the same R37G11 enhancer as the original GAL4 line, reasoning that they were most likely the same neurons. Both of these split-GAL4 lines also labeled neurons in the ventral nerve cord (VNC); however, these VNC cell types were different between lines ( A and 2B). We compared the number of PD2a1 and PD2b1 neurons labeled by each line; R37G11-GAL4 labeled 6.9 ± 0.6 cells, whereas LH989 and LH991 contained 5.25 ± 0.5 and 5.67 ± 0.8 neurons, respectively.
Specific Control with the Split-GAL4 System Confirms PD2a1 and PD2b1’s Role in Memory Retrieval, but Not Innate Behavior
(A and B) Confocal z-projections of split-GAL4 lines targeting PD2a1 and PD2b1 neurons, LH989 (A) and LH991 (B). mVenus membrane stain, green; neuropil, magenta. Flies expressing Shi by the split-GAL4 lines LH989 or LH991 were trained and tested according to the illustrated protocols along with genotypic controls (restrictive temperature in red).
(C and D) Silencing PD2a1 and PD2b1 neurons using LH989 (C; n = 14–15, F = 4.13, p = 0.02) or LH991 (D; n = 18, F = 7.27, p = 0.0017) impaired immediate memory after single-cycle training.
(E and F) Silencing PD2a1 and PD2b1 neurons during the retrieval phase 3 hr after single-cycle training using LH989 (E; n = 14, F = 6.73, p = 0.0031) or LH991 (F; n = 11–13, F = 8.23, p = 0.0013) caused a memory defect.
(G and H) Silencing PD2a1 and PD2b1 neurons during the retrieval phase 24 hr after spaced training using LH989 (G; n = 7–9, F = 9.79, p = 0.0010) or LH991 (H; n = 19–23, F = 10.83, p < 0.0001) abolished performance.
(I and J) Silencing PD2a1 and PD2b1 neurons using LH989 (I; Oct, n = 8–12, F = 0.63, p = 0.54; Mch, n = 10, F = 0.44, p = 0.65) or LH991 (J; Oct, n = 7–8, F = 0.25, p = 0.78; Mch, n = 7, F = 0.068, p = 0.93) had no effect on naive avoidance of Oct or Mch.
p < 0.05, p < 0.01, p < 0.001. Data are presented as mean ± SEM. See also and .
To confirm that PD2a1 and PD2b1 are involved in the retrieval of several memory phases, immediately after single-cycle training, on the middle-term timescale (∼3 hr), and 24 hr after spaced training, we repeated our behavioral experiments with these sparse split-GAL4 lines. When flies were tested at the restrictive temperature to silence PD2a1 and PD2b1, memory performance was impaired under all three conditions compared with genotype controls ( C–2H). This ranged from mild attenuation immediately after training ( C and 2D) to full impairment for long term memory (LTM) retrieval ( G and 2H), similar to phenotypes silencing MBON-ɑ2sc ( , ). This defect was due to neuronal silencing because identical flies at the permissive temperature had no memory recall deficits ( ). Finally, we verified that silencing PD2a1 and PD2b1 neurons with split-GAL4 lines had no effect on innate olfactory avoidance for the two training odors ( I and 2J), confirming that this is a specific defect in memory recall. Output from cell type PD2a1 and PD2b1 are therefore necessary for retrieval of aversive olfactory memory, with the same characteristics as MBON-ɑ2sc.
To understand the anatomy of PD2a1 and PD2b1 cells, we labeled single neurons in R37G11-GAL4 and the two split-GAL4 lines with MultiColor FlpOut (MCFO) ( ; A–S5C), isolating 22 single neurons from the PD2a1 and PD2b1 cell type. 3 of 22 labeled neurons also projected to the MB calyx (this projection is also visible in R37G11-GAL4, LH989, and LH991), whereas all other neurons appeared indistinguishable ( B–S5D). Therefore, these lines label two distinct cell types, PD2a1 (without calyx projections) and PD2b1 (with calyx projections). The calyx is the site of PN input to the MB, upstream of the site of associative olfactory memory, arguing against a role for this connection in our memory retrieval phenotype. Because we could not separately manipulate these two cell-types with our driver lines, we refer to them as PD2a1 and PD2b1. PD2a1 and PD2b1 neurons are morphologically similar to a large group of cells named “type I” ( ).
### MBON-ɑ2sc Drives Activity in PD2a1 and PD2b1
Double labeling experiments suggested that MBON-ɑ2sc is presynaptic to PD2a1 and PD2b1, but light microscopy does not have the resolution to confirm synaptic connectivity. We used GFP reconstitution across synaptic partners (GRASP) ( ) as a measure of the proximity of PD2a1 and PD2b1 dendrites and MBON axons. The experimental genotype displayed clear GFP reconstitution in the dorsal LH ( A), indicating that processes are close enough to form synapses; no signal was detected in control brains ( B and 3C).
MBON-ɑ2sc Is Functionally Connected to PD2a1 and PD2b1
(A–C) GRASP signal in the dorsal LH (green circles, dashed lines indicate midlines) for the experimental genotype (A) and two controls (B and C). Genotypes and controls are represented in the schematics above each figure. Images are representative of n = 3.
(D) GCaMP6f was expressed in PD2a1 and PD2b1 neurons with the R37G11-GAL4 driver (scale bar, 10 μm). Fluorescence was recorded in vivo from the axonal compartment of PD2a1 and PD2b1 neurons while the temperature was shifted from 20°C to 31°C (dashed line on F, except for the blue trace).
(E) The calcium increase of PD2a1 and PD2b1 neurons because of thermal activation of V2 MBONs (red trace) was stronger than that because of temperature shift only in the genotypic controls (green and purple traces).
(F) Quantification of calcium increase from the traces (n = 10 flies per condition, except 71D08-LexA/+ [n = 8], F = 9.09, p = 0.0001).
p < 0.01. Data are presented as mean ± SEM. See also .
Because MBON-ɑ2sc is cholinergic ( , ), we would expect that stimulating this neuron would drive activity in PD2a1 and PD2b1 if these neurons are connected. We expressed the heat-activated ion channel dTRPA1 ( ) in MBON-ɑ2sc ( D–3F) while recording calcium transients in PD2a1 and PD2b1. We used R37G11-GAL4 to express GCaMP6f ( ) and our R71D08-LexA line to drive dTRPA1 ( D). We imaged PD2a1 and PD2b1 axons in vivo to determine whether driving MBON-ɑ2sc could induce calcium transients in PD2a1 and PD2b1. In a control experiment, we observed a small temperature-dependent increase in calcium in the absence of the LexAop2-dTRPA1 transgene, indicating that temperature alone weakly stimulates these neurons ( E and 3F). We also observed a small calcium increase in flies carrying only LexAop-dTRPA1 ( E and 3F). However, increasing temperature in flies expressing dTRPA1 in MBON-ɑ2sc yielded a much larger calcium increase in calcium, indicating a functional connection ( E and 3F). We confirmed that dTRPA1 was expressed in MBON-ɑ2sc by expressing a LexAop2-TdTomato reporter in the same landing site as the LexAop2-dTRPA1 transgene ( E). These thermogenetic activation data, together with the double labeling and GRASP results, suggest that MBON-ɑ2sc connects to the PD2a1 and PD2b1 LH cell type necessary for memory retrieval.
### Synaptic Resolution Analysis of MBON-ɑ2sc and PD2a1 and PD2b1 Connectivity
A GRASP signal indicates that PD2a1 and PD2b1 dendrites and MBON-ɑ2sc axons are in close proximity but does not demonstrate the existence of synapses. We therefore leveraged a new whole female brain serial section electron microscopy (EM) volume ( , ) to study connectivity with synaptic resolution. We first identified the single MBON-ɑ2sc with a soma and dendrite in the right hemisphere of this volume by tracing downstream of Kenyon cells in the MB ɑ2 compartment. We then used NBLAST combined with light EM bridging registrations to match its backbone structure with light-level image data ( , ; A and 4A’). We repeated this procedure to identify the contralateral (left) MBON-ɑ2sc because their axons project bilaterally to both LHs.
Electron Microscopy Reconstruction of PD2a1 and PD2b1
(A) Reconstruction of the right-side MBON-ɑ2sc in a whole brain EM volume. The cell body is represented as a sphere, and the primary neurite (yellow-green), primary dendrite (green), dendrite (blue), and axon (orange) compartments are separately colored. Neuropils: LH in green, MB in purple. Inset: position of presynapses (red spheres) and postsynapses (cyan spheres) on the right-side MBON-ɑ2sc. Neuropils: SLP in yellow, SIP in orange, SMP in red.
(A’) Comparison of different metrics for the reconstructions of the contralateral and ipsilateral MBON-ɑ2sc within the LH (green in A). Inset: example of a polyadic synapse with a single T-bar (red dot) and multiple postsynapses (blue dots), referred to as “output connections” in the bar chart. Scale bar, 500 nm.
(B) Dorsal view of co-registered PD2a1 and PD2b1 MCFO data (top two panels, respectively) and EM reconstructions (bottom two panels, respectively). Cells are individually colored. Ipsilateral MBON-ɑ2sc is shown in black.
(B’) Dorsal view of single PD2a1 and PD2b1 neurons reconstructed in the EM volume. Yellow-green spheres represent somata, whereas ipsilateral and contralateral MBON-ɑ2sc synaptic connections are represented in dark and light purple, respectively.
(C) Schematic of synaptic connectivity from the two MBON-ɑ2sc neurons onto each PD2a1 and PD2b1 cell. The PD2a1 and PD2b1 cells are clustered according to the NBLAST score of their axons and dendrites, identifying two main groups, PD2a1 and PD2b1. Numbers beside each arrow indicate the number of outgoing connections made onto PD2a1 and PD2b1 neurons dendro-dendritically (blue) and axo-axonically (orange). Contra, contralateral; ipsi, ipsilateral; LH, lateral horn; CA, mushroom body calyx; SIP, superior intermediate protocerebrum; SLP, superior lateral protocerebrum; SMP, superior medial protocerebrum.
See also and .
We reconstructed the right LH axonal arbors to completion for both MBON-ɑ2sc neurons, marking pre- and postsynapses, and annotating the connections each presynapse makes in the right LH ( A’). We identified 183 and 190 presynapses for the left and right MBON-ɑ2sc, respectively, in the right LH ( A’). Each individual presynapse was polyadic, connecting to 7.8 ± 4.6 (mean ± SD) postsynaptic targets. We sampled 25% of these connections ( A”, inset) and identified 70 large target arbors (>300 μm of neuronal cable; data not shown), each likely belonging to different neurons. We found that two of these target neurons had the distinctive morphology of the PD2a1 and PD2b1 cells. Based on these two candidate cells, we located the PD2 primary neurite tract (purple dots in F) and coarsely reconstructed all neurons in this tract ( F) to identify a total of five PD2a1 (PD2a1#1–5) and two PD2b1 (PD2b#1–2) cells ( B and 4B’; ). Comparison of MCFO and EM data confirmed the identity of PD2a1 and PD2b1 neurons ( B’ and G). This was corroborated by NBLAST cluster analysis, indicating no clear separation between EM, FlyCircuit ( ), and MCFO data ( H). PD2a1 dendritic arbors contained some presynapses in the LH but at lower density than their axons. For both PD2b1 neurons, the LH and calycal projections were exclusively post-synaptic ( B’ and D). We confirmed the existence of these two types of neurons by clustering NBLAST scores derived from dendritic and axonal compartments, which yielded two distinct groups for PD2a1 and PD2b1 ( C). PD2a1 neurons could be further subdivided into two groups, one of which (PD2a1#1 and PD2a1#2) received greater MBON input per neuron ( C). Consistent with observations in the larva ( ), the vast majority of postsynapses were found on microtubule-free lower-order branches ( C). Summary data for pre- and postsynaptic sites, in addition to cable length for MBON-ɑ2sc and PD2a1 and PD2b1, is presented in . PD2a1 and PD2b1 presynapses contained only clear-core vesicles, suggesting that they do not release catecholamine or peptide neurotransmitters (data not shown).
All PD2a1 and PD2b1 cells received input from the ipsilateral MBON-ɑ2sc axon, and most received input from both MBONs ( C). In sum, these observations confirm that PD2a1 and PD2b1 neurons are a direct synaptic partner of MBON-ɑ2sc in the LH.
### PD2a1 and PD2b1 Neurons Have Decreased Responses to the CS+
After training, MBON-ɑ2sc depresses its response to the CS ( , ). We next examined whether PD2a1 and PD2b1 neurons downstream of MBON-ɑ2sc also modulate their response to the CS+ odor. We expressed the GCaMP3 calcium indicator ( ) in PD2a1 and PD2b1 ( A). In naive flies, PD2a1 and PD2b1 neurons responded to 3-octanol (Oct) and 4-methylcyclohexanol (Mch), the two odorants alternately used as CS+ in our behavioral experiments ( D, 1E, and ).
PD2a1 and PD2b1 Decrease Response to the CS+ after Training
(A) GCaMP3 was expressed in PD2a1 and PD2b1 with R37G11-GAL4. Olfactory responses to Oct and Mch were recorded in vivo from the axonal compartment of PD2a1 and b1 neurons.
(B and C) In naive flies, the calcium increase in PD2a1 and PD2b1 neurons in response to Oct was larger than Mch (average traces from n = 6 flies; t test, p = 0.015; B, average time trace; C, bar chart of response integral).
(D–D”) Odor responses were recorded 3 hr after single-cycle training using Oct as CS+ (n = 19 flies) or after the corresponding unpaired control protocol (n = 20 flies) ( A).The integral of the odor responses (D’; t test, p = 0.023) and the calculation of the difference between Oct and Mch responses (D”; t test, p = 0.024) revealed a decreased response to the CS+ after the associative protocol.
(E–E”) Odor responses were recorded 3 hr after single-cycle training using Mch as CS+ (n = 22 flies) or after the corresponding unpaired control protocol (n = 21 flies) ( B). The integral of the odor responses (E’; p = 0.047) and the calculation of the difference between Mch and Oct responses (E”; t test, p = 0.041) revealed a decreased response to the CS+ after the associative protocol.
(F–F”) Odor responses were recorded 24 hr after spaced training using Oct as CS+ (n = 9 flies) or after the corresponding unpaired control protocol (n = 11 flies) ( C). The integral of the odor responses (F’; t test, p = 0.036) and the calculation of the difference between Oct and Mch responses (F”; t test, p = 0.035) revealed a decreased response to the CS+ after the associative protocol.
(G–G”) Odor responses were recorded 24 hr after spaced training using Mch as CS+ (n = 9 flies) or after the corresponding unpaired control protocol (n = 9 flies) ( C). The integral of the odor responses (G’; t test, p = 0.047) and the calculation of the difference between Mch and Oct responses (G”; t test, p = 0.010) revealed a decreased response to the CS+ after the associative protocol.
p < 0.05. Data are presented as mean ± SEM. Gray bars indicate periods of olfactory stimulation. See also .
We next looked for training-induced changes in odor responses, comparing PD2a1 and PD2b1 responses following either associative training or a control, unpaired protocol that matched the odor sequence of the associative training, but temporally separated electric shock and odor delivery (see for the protocol). We performed these experiments either 3 hr after single-cycle training ( A and S7B) or 24 hr after spaced training ( C), using either Oct or Mch as the CS+. We found that pairing CS+ and electric shock during single-cycle training resulted in a decreased CS+ response in PD2a1 and PD2b1 axons 3 hr later, either compared with unpaired controls ( D’ and 5E’) or the CS− response in the same fly ( D” and 5E”). Similar results were observed 24 hr after spaced training ( F and 5G). These data suggest that PD2a1 and PD2b1 neurons receive memory-relevant information (the decreased CS+ response), resulting from depression at Kenyon cell to MBON-ɑ2sc synapses.
### PD2a1 and PD2b1 Also Receive Input from Uniglomerular PNs Encoding Attractive Odors
PD2a1 and PD2b1 dendrites in the LH are poised to receive input from PNs as well as MBON-ɑ2sc. Antennal lobe PNs have been identified in the EM volume ( ), enabling us to identify the specific input from each AL glomerulus to PD2a1 and PD2b1 dendrites in the LH and calyx ( A’). We annotated LH presynapses for each uniglomerular excitatory mALT PN (n = 112 PNs, 51 glomeruli; R.J.V.R., P.S., A.S.B., D.B., G.S.X.E.J., and S. Lauritzen, unpublished data). Most PD2a1 and PD2b1 neurons received synaptic input from several glomeruli, chiefly DM1, DP1m, DM4, VA2, DP1l, and VM3 ( A), although some differences were observed across cells.
PD2a1 and PD2b1 Receive Input from Appetitive PNs and Are Broadly Tuned
(A) Summary heatmap of antennal lobe glomeruli with uniglomerular, excitatory PN connectivity to individual PD2a1 and PD2b1 neurons as determined by EM reconstruction. The connectivity heatmap is separated by neuropil location: PD2a1 and PD2b1 LH dendrites, PD2b1 MB calyx dendrites and total across all PD2a1 and PD2b1 dendrites. Cell numbers represent the number of synapses, and heatmap coloring represents the synapse count normalized by the total number of postsynapses in that neuropil. Uniglomerular PNs with no connectivity are not shown. Uniglomerular PNs from connected glomeruli or MBON-ɑ2sc are ordered by connection strength. PN names are colored by their behavioral significance based on published studies.
(A’) The number of synaptic inputs for all PD2a1 and PD2b1 dendrites traced in this study. Input is either undefined (gray), uniglomerular PN (oranges), or MBON-ɑ2sc (purple).
(B) Reconstruction of all presynaptic partners to PD2a1#1 in the EM volume. Shown is the PD2a1#1 EM-reconstructed skeleton with dendritic postsynapses highlighted in blue.
(B’) Right: stacked bar chart showing the percentages of postsynapses contributed by different types of input neurons (different colors). Left: histogram showing the number of upstream postsynpatic partners against their synaptic weight (number of synapses onto PD2a1#1). The gray box highlights that 50% of PD2a1#1’s postsynapses are spent on neurons that only input PD2a1#1 by less than 10 synaptic connections. MBON-ɑ2sc is indicated by purple arrowheads.
(C) Electrophysiological recording raster plot from a representative PD2a1 neuron. The responses of each cell to the different odors are stacked, black squares represent action potentials, and there are 4 presentations of each odor. The red block represents the odor stimulation period.
(C’) Tuning curve of PD2a1 and PD2b1 neurons. Responses are shown in hertz. Data are mean ± SEM; n = 7 cells, consisting of one PD2b1, one PD2a1 or PD2b1, and five PD2a1 neurons. Odors in the text are shown in cyan.
(D) Schematic for imaging experiments with MBON-ɑ2sc silencing. Flies express Shi in MBON-ɑ2sc and GCaMP3 in PD2a1 and PD2b1 neurons for calcium imaging. At the permissive temperature (left), there is no effect on MBON-ɑ2sc neurotransmission, and PD2a1 and PD2b1 neurons receive input from both MBON-ɑ2sc and directly from the antennal lobe. MBON-ɑ2sc is silenced at the restrictive temperature (right), although the PD2a1 and PD2b1 neurons still receive input from the antennal lobe.
(E) Response of PD2a1 and PD2b1 axons to Oct with or without MBON-ɑ2sc silencing. Left: time traces of normalized GCaMP3 fluorescence ( ) are shown at permissive (blue) and restrictive (red) temperature in response to Oct stimulation (light blue bar). Right: the integral of the absolute odor responses for each fly at the permissive (blue) and restrictive (red) temperatures are plotted, which revealed decreased response to Oct after MBON-ɑ2sc silencing (n = 6, paired t test = 0.044).
(F) Response of PD2a1 and PD2b1 axons to Mch with or without MBON-ɑ2sc silencing. The layout of the data is the same as in (E). This revealed a decreased response to Mch after MBON-ɑ2sc silencing (n = 6, paired t test = 0.0015).
(G) Response of PD2a1 and PD2b1 axons to vinegar with or without MBON-ɑ2sc silencing. The layout of the data is the same as in (E). There was no change in response to vinegar after MBON-ɑ2sc silencing (n = 6, paired t test = 0.67).
(H) Response of PD2a1 and PD2b1 axons to ethyl acetate with or without MBON-ɑ2sc silencing. The layout of the data is the same as in (E). This revealed a decreased response to ethyl acetate after MBON-ɑ2sc silencing (n = 6, paired t test = 0.039).
(I) Response of PD2a1 and PD2b1 axons to isoamyl acetate with or without MBON-ɑ2sc silencing. This revealed a decreased response to isoamyl acetate after MBON-ɑ2sc silencing (n = 6, paired t test = 0.012).
p < 0.05, p < 0.01. See also .
To better understand this connectivity matrix, we annotated the behavioral function of input PNs according to published studies. The dorsal LH, where PD2a1 and PD2b1 dendrites are located, has been associated with coding of food odors ( ). Consistent with this, the top synaptically connected glomeruli (DM1, DP1m, DM4, and VA2) are responsive to appetitive and food odors ( , , ; A), indicating that PD2a1 and PD2b1 receives direct PN input mostly from appetitive olfactory channels. Furthermore, input to both DM1 and VA2 glomeruli is required for approach behavior to vinegar in hungry flies ( ).
PD2b1 cells have a dendritic branch in the MB calyx. We found that these dendrites’ largest inputs are from the same top four glomeruli (DM1, DP1m, DM4, and VA2) that target PD2a1 and PD2b1 dendrites in the LH ( A). This is even true for PD2b1#1, a cell that receives negligible DP1m and DM4 input in the LH but many synapses from these PNs in the calyx ( A’).
Uniglomerular PNs provide 36% of the total inputs to PD2a1 and PD2b1 dendrites in the LH, whereas MBON-ɑ2sc contributes 2.5% on average ( A’). This varies across individual neurons, with some PD2a1 and PD2b1 neurons receiving up to 15% of their known excitatory input from MBON-ɑ2sc ( A’; see below). To compare the significance of direct MBON to LH output neuron (LHON) connectivity with other dendritic input, we traced every neuron upstream of PD2a1#1’s 732 LH postsynapses to identification. All but 4 synapses could be matched to one of 165 partner neurons, which we divided into 6 major groups ( B).
We found that PNs and MBON-ɑ2sc provided 26.5% and 4.6% of the dendritic input, respectively. The great majority of the remaining input originated from within the LH (either local neurons, 33%, or reciprocal synapses from LH output neurons, 18.4%). There was also a small group of inhibitory PN connections (1.9%). The remaining 15.6% of input was from previously undescribed neuronal classes originating from the rest of the protocerebrum; we do not know whether these are inhibitory or excitatory. From these results, we can conclude that MBON-ɑ2sc provides between 9.8% and 14.7% of the direct excitation to this PD2a1 neuron and is the fourth largest input. Therefore, together, uniglomerular PNs and MBON-ɑ2sc provide the large majority of the driving cholinergic input to PD2a1 and PD2b1.
### PD2a1 and PD2b1 Integrate Input from MBON-ɑ2sc and PNs during Olfactory Stimulation
Our anatomical data indicate that PD2a1 and PD2b1 integrate olfactory information from the very broadly tuned MBON-ɑ2sc and PNs encoding food odors. To directly measure the olfactory tuning of PD2a1 and PD2b1 neurons, we performed whole-cell electrophysiology, which is more sensitive than calcium imaging. We targeted GFP-labeled PD2a1 and PD2b1 neurons for in vivo recording, followed by stimulation with a large battery of different odorants ( C; ). As expected, we found that PD2a1 and PD2b1 neurons were broadly tuned, responding to almost all odors at the test concentrations ( C’). Response variability was not noticeably greater than other LH neurons ( ). Apple cider vinegar drove the highest response, consistent with strong DM1 and/or VA2 inputs identified by EM. Although most other strong responses were to appetitive odors, benzaldehyde, which is innately aversive, drove the second highest response. We do note that benzaldehyde is also sensed through a non-olfactory pathway ( ) that could act via the LH or MB, complicating interpretation. The conditioning odors Mch and Oct, which are naively aversive ( , ), elicited intermediate responses.
One explanation for this broad PD2a1 and PD2b1 odor tuning is that PD2a1 and PD2b1 integrates direct PN input that is relatively tuned to food odors together with broad, odor non-specific input from MBON-ɑ2sc. We know that artificial MBON-ɑ2sc stimulation can drive PD2a1 and PD2b1 calcium responses ( D–3F); is this connection strong enough to have an effect on more naturalistic activity?
We designed an experiment to test the effect of MBON-ɑ2sc on odor-evoked activity and to provide functional evidence that PD2a1 and PD2b1 indeed integrates both direct AL input from PNs as well as indirect input from the MB. We reversibly silenced MBON-ɑ2sc neurotransmission with shibire while imaging PD2a1 and PD2b1 calcium odor responses in vivo ( D). Silencing MBON-ɑ2sc strongly attenuated PD2a1 and PD2b1 responses to both Mch and Oct ( E and 6F) compared with genotype controls ( A and S8B), indicating that MBON depression can significantly reduce PD2a1 and PD2b1 responses to our training odors.
Because both Oct and Mch are innately aversive, we tested the effect of MBON-ɑ2sc signaling on responses to apple cider vinegar (ACV). Silencing MBON-ɑ2sc had no effect on PD2a1 and PD2b1 responses to apple cider vinegar ( G; see C for the genotypic control). This is likely because apple cider vinegar very strongly activates the major PNs upstream of PD2a1 and PD2b1 neurons, reducing the effect of MBON-ɑ2sc on PD2a1 and PD2b1 coding. We also tested two attractive monomolecular odorants, ethyl acetate and isoamyl acetate. We again found that silencing MBON-ɑ2sc attenuated odor responses in PD2a1 and PD2b1 ( H and 6I; see D for the genotypic control). These results confirm that PD2a1 and PD2b1 integrate input from PNs and MBON-ɑ2sc. They also show that direct (PN) and indirect (MBON) pathways have different relative strengths for different odors.
### PD2a1 and PD2b1 Neurons Are Required for Olfactory Approach Behavior
Our functional and behavioral data demonstrate that PD2a1 and PD2b1 are modulated by and necessary for aversive olfactory memory retrieval. However, our EM reconstruction and electrophysiological characterization revealed that these neurons respond strongly to apple cider vinegar, an appetitive odor. This suggests that PD2a1 and PD2b1 neurons may mediate innate olfactory attraction. To test whether these neurons are necessary for approach behavior, we silenced PD2a1 and PD2b1 neurons in naive, starved animals, for which apple cider vinegar is an appetitive stimulus ( ). Silencing PD2a1 and PD2b1 neurons completely abolished vinegar attraction compared with the genotype controls ( A). At the permissive temperature, no difference was observed in the behavior of experimental and control genotypes ( E). To determine whether PD2a1 and PD2b1 was necessary for approach to other odors, we used ethyl acetate and isoamyl acetate, both of which are monomolecular, attractive odors. We found that PD2a1 and PD2b1 neurotransmission was required for attraction to ethyl acetate ( B; see F for permissive temperature controls) but dispensable for approach to isoamyl acetate ( C). This odor specificity is likely a combination of two factors. First, the PNs providing direct input to PD2a1 and PD2b1 appear more responsive to ethyl acetate than isoamyl acetate ( ). Second, there are likely additional LH neurons that promote attraction, including neurons that receive PN inputs that are selective for isoamyl acetate over ethyl acetate.
PD2a1 and PD2b1 Mediate Innate Olfactory Attraction, Leading to a Model of Aversive Memory Retrieval
(A) Flies expressing Shi driven by either LH989 or LH991 showed impaired attraction to apple cider vinegar relative to their genotype controls at the restrictive temperature (n = 9, F = 12.10, p < 0.0001).
(B) Flies expressing Shi driven by either LH989 or LH991 showed impaired attraction to ethyl acetate relative to their genotype controls at the restrictive temperature (n = 13-16, F = 6.34, p = 0.0002).
(C) Flies expressing Shi driven by either LH989 or LH991 showed impaired attraction to isoamyl acetate relative to their genotype controls at the restrictive temperature (n = 8-9, F = 0.53, p = 0.72).
(D and E) Model for how PD2a1 and PD2b1 functions in naive and trained animals.
(D) In naive animals, PD2a1 and PD2b1 receives input from both the MB (black sphere, via broadly tuned MBON-ɑ2sc) and directly from the AL food-related PNs (yellow sphere within AL). PD2a1 and PD2b1 activity is necessary for approach behavior to some olfactory stimuli.
(E) After conditioning, the response of MBON-ɑ2sc to the CS+ is reduced via synaptic depression at the MB-to-MBON synapse. This results in a decreased response to the CS+ in PD2a1 and PD2b1. Because PD2a1 and PD2b1 are cholinergic and excitatory, this reduces the input onto downstream approach circuits, resulting in decreased attraction to the CS+ during memory recall.
p < 0.05, p < 0.01. See also .
### A Model for Memory Retrieval by MBON-ɑ2sc Modulation of PD2a1 and PD2b1
Our results indicate that PD2a1 and PD2b1 neurons play a dual role in olfaction; they are necessary for both aversive memory retrieval and innate olfactory attraction. We show (anatomically) that PD2a1 and PD2b1 receives direct appetitive odor information from the AL and provide anatomical and functional evidence for a pathway from the MB to the LH that is depressed after learning. Together, these data led us to propose a circuit model for memory retrieval in our assay ( D and 7E), based on integration of innate and learned sensory representations by PD2a1 and PD2b1 neurons.
In naive animals, PD2a1 and PD2b1 integrates innate and learned olfactory representations and interfaces with approach circuitry ( D). After aversive olfactory conditioning, MBON-ɑ2sc depresses its response to the trained odor, which results in a reduced excitatory drive to PD2a1 and PD2b1 during CS+ sensation relative to naive animals ( ). Because PD2a1 and PD2b1 are cholinergic ( G), this depression results in decreased stimulation of downstream partners of PD2a1 and PD2b1 that mediate approach. This depression reduces the attractive bias to the CS+, leading to net avoidance of the trained odor ( E). Our experiments used a T maze memory paradigm, where flies choose between two arms containing odors that are initially of similar valence; after training, a relatively small decrement in the appetitive drive in the CS+ arm should be sufficient to bias flies to choose the CS− arm.
### PD2a1 and PD2b1 Interdigitates with DAN Dendrites and MBON Axons in MB Convergence Zones
To obtain some initial clues regarding how PD2a1 and PD2b1 neurons mediate olfactory attraction, we identified potential downstream targets of this LH cell type. Light and EM characterization of PD2a1 and PD2b1 axons suggested that they transmit information from the LH to the crepine (CRE), superior medial protocerebrum (SMP), and SIP ( ). The CRE, SMP, and SIP have been identified as convergence zones for the dendrites of DANs and MBON axons ( , , ). This raises the possibility that PD2a1 and PD2b1 may interact with input and output neurons of the MB assembly that drive valence behavior ( ).
We searched for potential contact sites by computational alignment of light microscopy data, generating a percentage overlap score of PD2 with DAN dendrites ( A) or MBON axons ( C). We investigated all neurons with more than 15% overlap in this coarse analysis, using double labeling with R37G11-LexA, expressing in PD2a1 and PD2b1 neurons ( A and 8C, black lines).
PD2a1 and PD2b1 Axons Interdigitate and Interact with DAN Dendrites and MBON Axons
(A) Histogram of light microscopy overlap between a mask of PD2a1 and PD2b1 axons and masks of the dendrites of DANs (along the x axis).
(B and B’) Confocal imaging of double labeling between PD2a1 and PD2b1 axons (labeled with GFP, green) and DAN dendrites (labeled with RFP, magenta). PAM-β’1 dendrites (B) and PAM-β’2m dendrites (B’).
(C) Histogram of light microscopy overlap between a mask of PD2a1 and PD2b1 axons and masks of the dendrites of most MBONs (along the x axis).
(D–D”) Confocal imaging of double labeling between PD2a1 and PD2b1 axons (labeled with GFP, green) and MBONs (labeled with RFP, magenta). Shown are MBON-β’2mp axons (D), MBON-y2ɑ’1 axons (D’), and MBON -ɑ′2 axons (D”).
(E) Visualization of MBON-ɑ’2 axons interdigitating with PD2a1 and PD2b1 axons (black) in the EM volume. Other PD2a1 and PD2b1 neurons are shown in gray. Inset: positions of axo-axonic connections from MBON-ɑ’2 onto PD2a1 and PD2b1 neurons, shown as cyan spheres.
(E’) Summary of ipsilateral MBON-ɑ’2′s axo-axonic connectivity onto PD2a1 and PD2b1 cells. Double labeling images are examples from n = 3 brains. For double labeling, the scale bar represents 5 μm.
We examined three DANs using double labeling. Both paired anterior medial (PAM)-β′1 and PAM-β′2 m dendrites interdigitated and exhibited potential synaptic contacts with PD2a1 and PD2b1 axons ( B and B’). PAM-β′2p had dendrites proximal to PD2a1 and PD2b1 axons but did not interdigitate (data not shown). PD2a1 and PD2b1-to-DAN connectivity may allow coordination of compartment activity across the MB ( ). PAM-β′2 m together with PAM-β′2p can drive approach behavior when stimulated ( ). Double labeling of MBON axons and PD2a1 and PD2b1 axons revealed close co-projection for MBON-β’2mp, MBON-γ2ɑ′1, and MBON-ɑ′2 ( D–8D”), indicating common postsynaptic partners or, possibly, axo-axonic synapses. MBON-β’2mp receives input from the MB compartment innervated by PAM-β’2 m and plays a role in appetitive and aversive memory retrieval ( ). MBON-γ2ɑ′1 drives approach when stimulated ( ). Silencing MBON-ɑ′2 throughout training and testing abolishes appetitive memories ( ).
Spatial convergence of PD2a1 and PD2b1 and MBON axons could imply the existence of common downstream targets and/or axo-axonic synaptic interactions. To test this and validate our light-level double labeling, we returned to EM. We reconstructed MBON-ɑ′2, the MBON that gave the highest PD2a1 and PD2b1 axon overlap score for MBONs ( C). We discovered that MBON-ɑ′2 makes axo-axonic connections onto PD2a1 and PD2b1 neurons ( E–E’), indicating that PD2a1 and PD2b1 output may be modulated by MBON-ɑ′2. The close proximity between axonal arbors required to make multiple axo-axonic synapses means that PD2a1 and PD2b1 and MBON-ɑ′2 are well placed to share downstream targets. These data show that PD2a1 and PD2b1 axons interact with or converge with MB-associated neurons that drive approach behavior and memory retrieval.
## Discussion
In this study, we set out to identify how innate and learned representations interact using the tractable Drosophila brain. Previous work had identified an olfactory learned-to-innate axonal projection of neurons necessary for memory retrieval ( ). Although MBON-ɑ2sc also projects to several downstream brain regions, we hypothesized the existence of LH neurons that integrate innate and learned olfactory codes.
Using light and EM, we identified PD2a1 and PD2b1, an LH cell type that integrates both hardwired input and plastic memory information from the MB. By combining this analysis with double labeling, GRASP, thermogenetic mapping, and, eventually, neuronal reconstruction from EM, we confirm that PD2a1 and PD2b1 are directly postsynaptic to MBON-ɑ2sc. Delineation of upstream PN connectivity also revealed that PD2a1 and PD2b1 dendrites in the dorsal LH mostly receive input from PNs encoding food or appetitive odors ( , ); this includes uniglomerular PNs from the DM1 and VA2 glomeruli, which are necessary for attraction to vinegar ( ). This connectivity matched the tuning of PD2a1 and PD2b1 cells, which was broad but included strong responses to apple cider vinegar, an appetitive odor. This suggests that PD2a1 and PD2b1 integrate innate and learned information and then pass this calculation to downstream circuits. We confirmed this by demonstrating that MBON-ɑ2sc contributes significantly to the olfactory response of PD2a1 and PD2b1 for most odors.
Mirroring these anatomical and functional results, we found that PD2a1 and PD2b1 neurons are necessary for both aversive memory recall and innate olfactory attraction. Using specific split-GAL4 control of PD2a1 and PD2b1 neurons in the brain, we found that PD2a1 and PD2b1 signaling is necessary for memory retrieval across all phases but dispensable for innate olfactory aversion to the training odors (which are innately aversive). However, when animals were presented with food-related odors, which robustly generates olfactory attraction, silencing the PD2a1 and PD2b1 neurons abolished approach to a subset of odors. For the first time, to our knowledge, we have directly interrogated the role of LH neurons in olfactory behavior in adult Drosophila , discovering an LH cell type that is both necessary for innate attraction and, contrary to the assumption that the LH solely mediates innate behavior, also required for memory retrieval. Although information from the LH and MB must converge at some point in the fly brain to produce behavior, it is surprising that this integration happens within the LH rather than downstream of both the LH and MB. Indeed, MBON-ɑ2sc mostly projects to other brain regions where MB and LH neurons converge ( , ). This early convergence may minimize redundant circuitry (see below). We stress, however, that this does not preclude a role for other LH cell types in innate avoidance.
We developed a model for how this MB-to-LH circuit mediates aversive olfactory memory retrieval in the T maze assay ( D and 7E). As previous work has demonstrated, aversive olfactory conditioning induces synaptic depression at the MB-to-MBON synapse, which is thought to mediate memory retrieval ( , , ). However, the downstream circuits mediating the memory retrieval were unknown. We confirmed that PD2a1 and PD2b1 also depresses its response to the CS+, indicating that LH neurons can be modulated by MB activity. PD2a1 and PD2b1 are necessary for attraction, so the reduced drive in response to the CS+ results in less drive onto the approach circuits downstream (we have shown that PD2a1 and PD2b1 neurons are cholinergic) ( E). In accordance with the prevailing view of how memory retrieval modulates the MB-to-MBON circuit, this model suggests that aversive olfactory memory retrieval is a result of modulating hardwired attraction circuits in response to the CS+ rather than the recruitment of a dedicated aversion module. However, we note that, in the T maze, the memory test is between two odors of similar innate valence. It is possible that other memory paradigms may recruit distinct aversion circuits; this may be a reason why a second MBON pathway for aversive memory recall exists in the Drosophila brain ( ).
The identity of neurons downstream of PD2a1 and PD2b1 and their relationship to motor behavior is currently unknown. However, we demonstrate that PD2a1 and PD2b1 axons converge with MBONs implicated in memory and olfactory attraction. Downstream neurons may therefore read out both MB and PD2a1 and PD2b1 codes to guide the animal’s choice. Future connectomics and functional approaches should identify these downstream neurons and their relationship to learned and unlearned sensory representations of different valence.
What are the implications of this circuit arrangement for learned and innate behavior? First, early integration of learned and innate pathways likely economizes neuronal hardware. Second, direct integration of learned and innate stimulus representations provides a simple mechanism to resolve the potentially conflicting behavioral drives that might exist after learning. Furthermore, this integration happens at a stage when neuronal activity is clearly sensory in character; this may be simpler than carrying out parallel sensory motor transformations downstream of both the MB and LH. One interesting hypothesis raised by the specific circuitry that we uncovered is that the balance between direct PN and indirect MBON-ɑ2sc pathways onto PD2a1 and PD2b1 may constrain stimuli that can undergo aversive conditioning. Under our experimental conditions, apple cider vinegar odor responses were not altered by manipulating MBON-ɑ2sc activity (whereas representations of some monomolecular appetitive odors could be modified). This may reflect selection on an evolutionary timescale of PN to LH connectivity to ensure that approach behavior produced by odors very highly predictive of food (and associated social interactions) is hard to reverse. Finally, it will be exciting to see whether a similar learned-to-innate circuit connectivity is involved in appetitive memory recall of other sensory modalities, such as taste and vision ( , ).
The olfactory systems of flies and mammals share the same basic blueprint ( , ). In mice, the piriform cortex is required for learning and memory ( ) and responds sparsely to odors ( ) and samples from the whole olfactory bulb ( , ), similar to the MB. In contrast, the olfactory amygdala is necessary and sufficient to instruct innate olfactory behavior ( ) and receives stereotyped input from the olfactory bulb ( , ), drawing a comparison to the LH. Intriguingly, there are uncharacterized connections between the piriform cortex and olfactory amygdala ( ). A similar model of the piriform cortex modulating hardwired representations has been hypothesized in the mouse ( ). We speculate that these connections may play a role in memory retrieval in the mammalian brain by enabling integration of learned and innate olfactory representations within the amygdala.
## STAR★Methods
### Key Resources Table
### Contact for Reagent and Resource Sharing
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Gregory Jefferis ( ).
### Experimental Model and Subject Details
Standard techniques were used for fly stock maintenance and construction. For imaging and immunohistochemistry flies were raised at 25°C on standard Drosophila food. For MultiColor FlpOut (MCFO) experiments ( ), the MCFO stock (see below) was crossed to either R37G11-GAL4, LH989 or LH991. Flies were collected after eclosion, transferred to a new food vial and incubated in a 37°C water bath for 20-25 minutes.
Transgenic lines used for behavior were outcrossed for five generations to a w1118 strain in a wild-type Canton-Special (CS) background. For behavioral experiments flies were raised at 18°C and 60% humidity under a 12-hr:12-hr light-dark cycle.
For a list of all genotypes used in each figure of the paper, see .
### Method Details
In all cases, sample size was based on previous studies ( , , ). Experimenter blinding was not performed for experiments. No data was excluded from further analysis.
#### Molecular Biology
The pBP-R71D08 gateway entry construct was a kind gift from Heather Dionne. The insert was transferred to the pBPLexA::p65Uw destination vector (Addgene) via a Gateway LR recombination (Invitrogen).
The enhancers used to create split-GAL4 hemidrivers were created based on annotations for PD2a1 and PD2b1 in a GAL4 expression pattern database ( ). The enhancer hemidriver lines were created using Gateway cloning. All transgenic fly lines were generated by either Bestgene or Genetic Services.
#### Immunohistochemistry
Throughout this study we used two different immunohistochemistry (IHC) protocols. F, A, 2B, and A used Protocol 2 while all other IHC data was processed using Protocol 1. For neurons filled during electrophysiology, see protocol for electrophysiological recording below. See for antibodies used.
##### Protocol 1
IHCs were performed as described ( ). Fixation was in 4% paraformaldehyde for 20 minutes. Blocking was performed with normal goat serum overnight at 4°C. Primary and secondary antibody stains were incubated at 4% for 48 hours each. After incubation with both primary and secondary antibodies, the brains were washed with 0.5% Triton X-100 at room temperature. All specimens were mounted in Vectashield (H-1000) (Vector Laboratories, Burlingame, CA, USA).
##### Protocol 2
These IHCs were performed as described ( ). Dissected brains were fixed in 2% paraformaldehyde for 55 minutes at room temperature. Fix was removed and washed with 5% Triton X-100 at room temperature. Primary antibodies were incubated for 48 hours and secondary antibodies were incubated for 72 hours. A full step-by-step protocol can be found at . Following the IHC protocol the brains were fixed again in 4% paraformaldehyde for four hours at room temperature. The brains were mounted on poly-L-lysine-coated coverslips and dehydrated through a series of ethanol baths (30%, 50%, 75%, 95%, and 3 × 100%) each for 10 min. Following dehydration they were submerged in 100% Xylene three times for 5 minutes each. Samples were embedded in DPX (DPX; Electron Microscopy Sciences, Hatfield, PA).
#### IHC Image Acquisition
All images for IHC were acquired using a Zeiss 710 Confocal Microscope ( , ). We used three modes of imaging: 20x, 40x and 63x.
##### For 20x imaging,
whole mount brain and VNCs were imaged using a Plan-Apochromat 20x/0.8 M27 objective (voxel size = 0.56 × 0.56 × 1.0 μm; 1024 × 1024 pixels per image plane).
20x imaging was used for A and 2B.
##### For 40x imaging,
whole mount brains were imaged using an EC Plan-Neofluar 403/1.30 oil objective with 768 × 768 pixel resolution at each 1 μm, 0.6-0.7 zoom factor.
40x imaging was used for A–3C, A, S1B, S1D, E, and .
##### For 63x imaging,
whole mount brains were imaged using a Plan-Apochromat 63x/1.40 oil immersion objective (voxel size = 0.19 × 0.19 × 0.38 μm; 1024 × 1024 pixels). For certain images, tiles of regions of interest were stitched together into the final image. 63x imaging was used for C, 1F, 1G”, B, 8B’, 8D–8D”, C, C, S2D, and A.
#### Generation of split-GAL4 lines
Each split-GAL4 line consists of two hemidrivers, the p65ADZp in attP40 and the ZpGAL4DBD in attP2 ( ). The lines were screened by combining these two hemidrivers with a copy of 20xUAS-IVS-csChrimson::mVenus (attP18). The brains of females from each line were dissected and screened with an epifluorescence microscope. Split-GAL4 combinations with favorable expression patterns (sparse expression of PD2a1 and PD2b1) were double balanced to make a stable stock.
#### Behavior: Olfactory Assays
See for details of all olfactory stimuli used in behavior. For all behavior experiments, 0–2 day-old flies were transferred to fresh food vials the day before conditioning. Conditioning and tests of memory performance and of olfactory acuity were performed as described previously ( ). Groups of 40-50 flies were trained with either one cycle of aversive training (single-cycle training), or five cycles spaced by 15 min inter-trial intervals (spaced training). During one cycle of training, flies were first exposed to an odorant (the CS+) for 1 min while 12 pulses of 1.5 s-long 60V electric shocks were delivered every 5 s; flies were then exposed 45 s later to a second odorant without shocks (the CS–) for 1 min. The odorants 3-octanol (Oct) and 4-methylcyclohexanol (Mch), diluted in paraffin oil at 0.360mM and 0.325mM respectively, were alternately used as CS. The test of memory performance was performed in a T-maze. Flies were placed at the convergence point of two airflows interlaced with Oct or Mch from either arm of the T-maze. After 1 min in the dark, flies were collected from both arms of the T-maze for subsequent counting, yielding a score calculated as (N – N )/ (N + N ). A single value of the performance index is the average of two scores obtained from two groups of genetically identical flies conditioned in two reciprocal experiments, using either odorant as CS+, and tested consecutively in the T-maze. Flies were maintained on food at all times, with the exception of during conditioning and memory test. Memory test occurred 10 ± 5 minutes after conditioning, 3h ± 30 minutes after conditioning, and 24 ± 1.5 h after conditioning to assay immediate memory, 3-h memory and long-term memory, respectively. For long-term memory, flies were stored at 18°C after training which maximizes memory scores ( ). For experiments involving neuronal blockade with Shi , the time courses of the temperature shifts are provided alongside each graph of memory performance, and periods of neurotransmission blockade are highlighted in red. Flies were transferred to the restrictive temperature (33°C) 30 min before the targeted time, to allow for acclimatization to the new temperature.
To measure innate odor avoidance toward Oct or Mch, naive flies were placed at the convergence point of two airflows, one interlaced with Oct or Mch and the other from a bottle with paraffin oil only. The odor-interlaced side was alternated for successive groups that were tested. Odor concentrations used in this assay were the same as for memory assays. At these concentrations, both odorants are innately repulsive. The avoidance index was calculated the same way as the performance index in memory assays.
To measure innate odor approach, we used the avoidance assay with the same flow rate to deliver attractive odors. For apple cider vinegar experiments, the olfactory stimulus choice was between apple cider vinegar or water alone. Flies were starved on mineral water for 29h prior to experiments. The odor concentrations used were:
Ethyl acetate: 10 in paraffin oil
Isoamyl acetate: 10 in paraffin oil
Apple cider vinegar: 6.1x10 in Evian mineral water
Starvation time and odor concentrations were determined beforehand using wild-type flies (data not shown) to obtain robust attractive behavior. The attraction index was calculated as the performance index multiplied by −1.
Performance, aversion and attraction indices are displayed as means ± SEM. A single value of the performance index is the average of two scores obtained from two groups of genetically identical flies conditioned in two reciprocal experiments, using either odorant as CS+, and tested consecutively in the T-maze. The indicated ‘n’ is the number of independent values of the performance index or avoidance index for each genotype. Memory graphs were subjected to statistical analysis using 1-way ANOVA followed by Newman-Keuls pairwise comparisons between the experimental group and its controls. ANOVA is robust against slight deviations from normal distributions or the inequality of variances if the sample sizes are similar between groups which was the case in our experiments. Therefore, we did not systematically test our data for normality or verify variance homogeneity prior to statistical tests, but we rather adopted a uniform analysis strategy for all our data ANOVA results are given as the value of the Fisher distribution F(x,y) obtained from the data, where x is the number of degrees of freedom between groups and y is the total number of degrees of freedom of the distribution. Statistical tests were performed using the GraphPad Prism 5 software. In the figures, asterisks illustrate the significance level of the t test, or of the least significant pairwise comparison following an ANOVA, with the following nomenclature: p < 0.05; p < 0.01; p < 0.001; NS: not significant, p > 0.05). See for a detailed list of all odors used for behavioral and calcium imaging experiments.
#### Calcium Imaging: Functional Connectivity
The genetically encoded GCaMP6f calcium reporter ( ) ( UAS-GCaMP6f in attp18 ) was driven by R37G11-GAL4 (attP2) . The thermosensitive cation channel dTrpA1 ( ) ( LexAop2-dTrpA1 VK00005 ) was expressed in the V2 neurons by the 71D08-LexA driver (attP40). Female flies of the indicated genotypes were prepared for in vivo imaging as described above, and mounted on a custom-made chamber with controlled temperature through a Peltier cell and an analog electronic PID circuit. The baseline setpoint for the temperature was 20°C. Imaging was performed on the same setup as for olfactory responses, images were acquired at a rate of one image every 640 ms. During an acquisition with thermal activation, the setpoint of the temperature control circuit was shifted to 31°C for 30 s after a baseline recording of 10 s, and then back to 20°C. The measured risetime of the temperature from 20°C to 29°C was ∼8 s, and temperature reached 31°C within ∼11 s. The temperature decrease was slower, taking ∼15 s from 31°C to 22°C and ∼25 s in total to decrease down to 20°C. For each fly, three such acquisitions were recorded, and the resulting time traces from visible hemispheres and from all these recordings were pooled and averaged. In R71D08LexA > LexAop2-TrpA1 flies, acquisitions with activation were alternated with acquisitions without activation as a permissive temperature control. The magnitude of activation was calculated as the mean of the time trace over a 20 s-time windows starting 5 s after the change in temperature setpoint.
#### Calcium Imaging: Olfactory Responses
To monitor the olfactory responses in PD2a1 and PD2b1 neurons, the genetically encoded GCaMP3 calcium reporter ( ) was driven by R37G11 GAL4 driver. We used a transgenic line carrying the UAS-IVS-GCaMP3-p10 construct inserted on the third chromosome in VK00005 ( , ). For in-vivo imaging, one female fly was prepared for each n ( , , ). A cuticle window was removed in the back of the fly head. The fly was then placed under the objective lens (25x, 0.95 NA) of a confocal microscope under a constant airflow of 1.5 L·min-1. Images were acquired at a rate of one image every 128 ms. The emitted light was collected from transverse sections of the brain showing presynaptic terminals of PD2a1 and PD2b1 neurons. In general, both hemispheres could be recorded simultaneously. Olfactory stimuli were triggered by switching a valve to direct 30% of the total flow for 2 s through bottles containing odorants diluted in paraffin oil. Final dilution in the airflow was 1:2000. We recorded two series of responses to octanol and methylcyclohexanol, in alternating order, each separated by a 2 min interval, but only the first response to each odorant was kept for analysis. Data analysis was performed with MATLAB software. For each recording, a ΔF/F0 time trace trace was calculated from an ROI surrounding the PD2a1 and PD2b1 projections. The baseline F0 value was calculated from the 2 s period preceding the switch of the valve. The response integral was calculated as the integral of the time trace during 10 consecutive time points following the onset of odor response (approx. 2 s). The comparison of the response to a given odor between two groups, and of the response difference (Oct–Mch or Mch–Oct) between two groups, was performed using unpaired t test. The sample size was chosen according to the experiment, with olfactory response experiments having n = 6, similar to other naive imaging experiments ( , ) For training and imaging experiments we chose a higher n, n = 19-22 or n = 9-11 for MTM and LTM respectively. As MTM is only partially abolished with PD2a1 and PD2b1 silencing we chose a higher n compared to LTM, which is entirely dependent on PD2a1 and PD2b1 (see and ).
#### Calcium Imaging: Olfactory Responses with MBON-ɑ2sc silencing
Flies were prepared for imaging as described above and imaging was performed within the same imaging cell as described above. Flies expressed GCaMP3 (attP40) through 37G11-GAL4 and LexAop2-Shi (VK00005) through 71D08-LexA (genotype controls had no LexA driver).
The concentrations used for imaging were:
Ethyl acetate: 10μL in 100mL paraffin oil;
Isoamyl acetate: 50μL in 100mL paraffin oil;
Oct: 30 μL in 100mL oil;
Mch: 100μL in 100mL oil;
Apple Cider Vinegar: 5mL in 100mL mineral water.
See for more information on these odors.
To avoid interactions between odorants, each fly received only one odor, 3 trials at low (23°C), 3 trials at high (33°) temperature. Each trial consisted of 2 s of odor stimulation. For each odor, half of the flies started at high temperature and the other half at low temperature. Trials were separated by 3 minutes, and after temperature shift, 8-10 minutes were left to get used to the new temperature. All trials at a given temperature were averaged, to give a single trace (DeltaF/F) and a single value of response integral per fly per temperature. For each fly both traces were normalized to the maximum value of the low temperature trace. The calcium traces displayed are the normalized time traces across all flies at each temperature. This normalization procedure better highlighted the effect of temperature shift independently of the absolute magnitude of the response. A paired t test was used to compare the response integral between the permissive and restrictive temperature.
#### Electrophysiology and olfactory stimulation
Recordings were carried out from PD2a1 and PD2b1 neurons in LH989 and LH991 split-GAL4 animals crossed to a UAS-CD8::GFP reporter. We recorded from n = 7 cells in total, 5 PD2a1 neurons, 1 PD2b1 neuron and one cell which which had an inconclusive dye-fill. Some odour concentrations that we eventually presented were not tested for all cells, hence n = 2-7 in total.
Female flies were sorted for correct genotype on day of eclosion using CO anesthesia. One or two days later, the fly was cold-anesthetized and placed in a custom recording chamber for dissection as described previously ( ). The setup used for these experiments had a total of 64 channels. A full list of odors, solvents and dilutions used is provided in below. The length of the valve opening stimulus was 2 s. The recording electrodes were 5 to 8 MΩ.
Odor stimuli were diluted in either mineral oil or water and were delivered via a custom odor delivery system ( ) (see ). Unless otherwise indicated, liquid odors were diluted 1:500 (2 μl in 1ml) in either mineral oil or water. Solid odors were dissolved at 2mg in 1ml of solvent. During stimulus presentation, a portion of the airstream was switched from a solvent control to a selected odorant. The odorized air stream was then mixed with a clean carrier air stream at a 1:8 ratio to give a notional final dilution of 2.5 × 10 for most odors.
For labeling filled and recorded neurons, we used Alexa Fluor 568 (A11031, 1/1000) for the detection of mouse anti-nc82 and streptavidin Alexa fluor 647 (Thermo Fisher S-21374 1/4000) for detection of filled neurons. See for list of odors used for electrophysiology.
#### Sparse EM Reconstruction and Neuron Identification
Neurons were reconstructed by ‘tracing’ in a full female adult Drosophila melanogaster brain volume (x,y,z resolution 4 nm x 4nm x 40 nm) that had been acquired by serial section transmission EM ( ), wherein the authors provide detailed sample preparation, EM acquisition and volume reconstruction protocols. Tracing aimed to generate a neuronal skeleton that represents the branching of neurons and the locations of their synapses, rather than a volumetric reconstruction. Manual neuronal tracing through EM serial sections was performed in CATMAID ( ) ( ), a Web-based environment for working on large image datasets that has been optimized for tracing and online analysis of neuronal skeletons ( ). Neuronal skeleton reconstruction was performed consistent with . Presynapses and postsynapses were annotated for all neurons traced in this study. Polyadic synapses were marked consistent with the criterion of other CATMAID-based Drosophila connectomic studies (e.g., ). Briefly, synapses must have a clear presynaptic density, multiple vesicles in the vicinity of the density and a cleft between the pre- and postsynaptic membranes. Postsynapses were marked if they had a (though often unclear or faint) postsynaptic density or otherwise distinctive morphology in apposition to the synaptic cleft. Additionally, for PD2a1/1 neurons, the point at which microtubules ceased to be apparent in a branch was also annotated. Microtubules appear as thin dark filaments that flow contiguously from the cell body and terminate before the lowest order branches. Ambiguities and uncertainties in each neuron were flagged as it was traced, all neurons were subsequently and iteratively proofread and edited by an expert tracer until completion at least in the region of interest (see below). Gap junctions could not reliably be identified in this dataset.
MBONs-ɑ2sc and MBON-ɑ′2 were found by tracing downstream of extant reconstructed Kenyon cells ( ) within the appropriate MB compartment ( ). Identity was verified with visual comparison to confocal stacks collected in . Identification of PD2a1 and PD2b1 cell types began with tracing downstream of right-side MBON-ɑ2sc. 23.95% of 1837 total outgoing connections from the right-side MBON-ɑ2sc axon in the LH were traced into 70 substantial neuronal arbors (> 300 μm of cable; data not shown). Visual inspection identified candidates for two PD2a1 neurons, which were traced to identification. This provided the location of the PD2 primary neurite tract (see ; nomenclature from ). In insect brains, the majority neuronal cell bodies are positioned outside of the neuropil proper, in the cortex, and invaginate the neuropil via a primary neurite before branching. The primary neurite tract that a neuronal cell type takes is consistent between members of the type and between brains (S.F. and G.S.X.E.J., unpublished data). No similar tract that might have also contained our neurons of interest could be found after thorough visual scanning through the EM data, nor was there any indication from NBLAST clustering of LH neurons in the FlyCirciuit database ( ) or MCFO data that neurons similar to PD2a1 and PD2b1 could take multiple primary neurite tracts (data not shown). 185 neuronal profiles fasciculated within the PD2 tract, all of which were traced until their morphology made them an apparent PD2a1 and PD2b1 candidate or evidently not. Neurites for all candidates (34) were traced to or near ‘completion’ (see below). PD2a1 neurons must have 1) dendrite largely confined the the dorsal LH, 2) primary neurite tract in the PD2 bundle, 3) an axon that circumvents around the MB vertical lobe. Additionally PD2b1 neurons must have a process in the calyx. Two neurons met criterion 2 and 3, but were borderline on 1 and failed to receive similar projection neuron input to the 7 convincing members of the group, and were dropped from analysis. Identity was further verified by NBLAST ( ) of reconstructed skeletons against MCFO data from this study and the FlyCircuit database ( ). Scores for our 7 putative PD2a1 and PD2b1 neurons were higher than for other candidate neurons in the PD2 tract and other MBON-ɑ2sc targets (data not shown).
All 7 PD2a1 and PD2b1 neurons and the ipsilateral MBONs-ɑ2sc and MBON-ɑ′2 were fully traced ‘to completion’ ex nihilo , with synapse annotation. The contralateral MBONs-ɑ2sc was traced to identification, but completed within the LH. ‘Completion’ does not necessarily mean that absolutely all cable has been reconstructed and postsynapses and presynapses annotated, as a small minority of processes and connections may not have been resolved due to ambiguities in the image data. Many uniglomerular, excitatory projection neurons of the medial antennal lobe tract had been identified in the present EM volume, and traced outside the MB calyx only to identification, not completion ( ). These PNs have since been reconstructed to completion in the LH (P.S. and A.S.B., unpublished data). For this study, we proofread, edited and annotated synapses for PN arbor in the right-side LH for all 20 uniglomerular PN types in the vicinity of PD2a1 and PD2b1 dendrite and those determined to have significant overlap at a light level (data not shown). At first, one representative PN was chosen for each glomerulus that produced more than one uniglomerular, excitatory PN. If this PN was found to synapse onto PD2a1 and PD2b1 neurons, its sister cells were also completed within the LH, as the morphology of sister PNs in the LH are extremely similar ( ).
To identify neurons innervating PD2a1#1, we traced upstream of all of its dendritic postsynapses (i.e., postsynapses within the LH). Each upstream skeleton was traced to identification, i.e., the inclusion of a soma tract and main arbours, so as to ascertain whether it was a type of PN (axonic arbor in the LH, dendritic arbour in known second-order sensory neuropils), a LHON (axonic arbor leaving the LH), and LHLN (no significant arbor outside the LH), centrifugal neuron (axonic arbor within the LH, dendrites elsewhere in the superior protocerebrum) or MBON.
### Quantification and Statistical Analysis
For all double labeling and imaging experiments, each n represents either a single slice or a volume from a single brain. For behavioral experiments, each n represents a group of 40-50 flies analyzed together in an olfactory assay. For functional connectivity and calcium imaging experiments, each n represents the response of a single recorded fly. For electrophysiology data, each n represents a recorded neuron from an individual fly (one neuron was recorded per fly).
#### Image Processing and analysis
To accurately label the presynapses of the LH-projecting MBONs, the 71D08-LexA driver was crossed to LexAop2-Brp(d3)::mCherry resulting in axon-specific labeling. For the region of the MBON under investigation (MBON-a2sc axons in the dorsal LH) a mask was created. Eight 71D08 > Brp(d3)::mCherry brains were immunostained and registered onto a common template brain (JFRC2, ) using the nc82 counterstain. Image registration was carried out as described ( ) using the CMTK registration suite ( ). The boundary of the overlaid neurites for each region of interest from each brain was segmented manually as a mask in Fiji ( ), using the Segmentation Editor function. The overlap was calculated against a large database of GAL4 expression patterns ( , ) also registered against JFRC2 ( ) using the cmtk.statistics function in the open source nat (NeuroAnatomy Toolbox) package ( ) for R ( ). To control for the background signal of the brain we created a mask of the peduncle and performed the same overlap calculation for each GAL4 line. This peduncle overlap score was used normalize the MBON axon masks to produce the final overlap score for each GAL4 line. This allowed us to select lines with high signal-to-noise within the MBON masks and excluded expression patterns with Kenyon Cell expression which would confound behavioral analysis. The stacks GAL4 line expression patterns in the top 0.97 quartile were further analyzed manually to identify LH neurons.
For counting the number of cells in each line, images of R37G11-GAL4, LH989 and LH991 crossed to 20xUAS-csChrimson::mVenus (attp18) were used and cells manually counted.
MCFO brains were imaged in 63x mode (see above) and the stitched final image registered to the JFRC2013 template brain. Single neurons were manually annotated and segmented in 3D using Fluorender. For comparison with the data reconstructed from EM we automatically skeletonized MCFO image data using the filament editor tool provided by the image analysis software Amira 6.2.0, followed by manual editing. Morphological analysis was performed using NBLAST (see below). For analysis, neurons were segregated into soma, primary neurite (the neurite that leads to the cell body), dendrite, primary dendrite (the neurite connecting the dendritic and axonal arbors) and axon by visual inspection using insight from our EM data. We isolated 5 cells from R37G11-GAL4, 13 cells from LH989 and 5 cells from LH991. All lines contained neurons which projected to the MB calyx.
To examine overlap between PD2a1 and PD2b1 axons and MB neurons or PD2a1 and PD2b1 dendrites and PN axons, high resolution images (63x) of PD2a1 and PD2b1 split-GAL4 lines driving both a membrane and presynapse markers were segmented using Fluorender ( ). We compared this to published segmentations of the DANs and MBONs ( ). All data were registered to the JFRC2013 template brain ( ). For all cell-types in addition to the entire membrane stain, the axons and dendrites were segmented separately for at least n = 2 well-registered brains. For each category of segmentation (dendrite-only, axon-only) we created a mask from their different samples by overlaying all the examples of each line. This was followed by contrast enhancement, Gaussian blurring and auto-thresholding to create a mask. All image processing was performed using Fiji. Overlap comparisons for pairs of masks were compared in R using the cmtk.statistics function in the “nat” package.
Double labeling images performed with R37G11-LexA and different MBON and DAN Split-GAL4 lines were processed with a median filter using the despeckle command in Fiji. This was necessary to remove background due to the weak expression levels of the R37G11-LexA line.
#### Neuronal Skeleton Data Analysis
Neuronal skeleton data from CATMAID were analyzed in R. Open source R packages for NBLAST ( ), and R tools for accessing the CATMAID API are available on github by following links from . The catmaid and elmr R packages provide a bridge between a CATMAID server and the R statistical environment and bridging registration tools respectively. They include several add-on packages from the NeuroAnatomy Toolbox (nat see ) suite enabling statistical analysis and geometric transformation of neuronal morphology. Further analysis relied on unreleased custom R code developed by A.S.B and G.S.X.E.J.
The elmr package provides tools for transforming data from the present EM whole female Drosophila melanogaster brain volume into different light level template brains for inspection of co-registered data. Neuronal skeleton reconstructions were brought from the EM brain space into the virtual flybrain template ( ; dubbed JFRC2, the brain is divided into neuropils via the methods employed in ). FlyCircuit PD2a1 and PD2b1 neurons, identified through NBLAST clustering, were brought into the JFRC2 brain space using the Computational Morphometry Toolkit ( ).
MBONs-ɑ2sc, MBON-ɑ′2 and PD2a1 and PD2b1 neurons were segregated into axon and dendrites using a tcentrifugal synapse flow centrality algorithm ( )counting polyadic presynapses once). We verified that neurons were suitably polarized by calculating their axon-dendrite segregation index ( ), which is a quantification for the degree of segregation of postsynapses and presynapses (0, totally unsegregated, 1, completely polarized). The mean ± SD segregation index for PD2a1 and PD2b1 neurons was 0.27 ± 0.09 indicating that these neurons are polarized but receive heavy axo-axonic modulation as well as outputting significantly in the LH. MBON were highly polarized, for example right-side MBONs-ɑ2sc had a segregation index of 0.72. Again we counted polyadic presynapses once, rather than using the number of outgoing connections these make, which would have been more expensive in terms of tracing time.
For morphological analysis of PD2a1 and PD2b1 neurons NBLAST ( ) was performed on either the dendritic and/or the axonal arbors of neuronal skeletons. Primary neurite tracts and the primary dendrites connecting dendritic and axonal arbors were removed because their fasciculation, especially in the single EM brain space, made NBLAST less sensitive to dendritic and axonal differences. Clustering was performed using functions for hierarchical clustering in base R on euclidean distance matrices of NBLAST scores, employing Ward’s clustering criterion.
#### Data Presentation
All images of neuronal skeletons are shown in the JFRC2 brain space used by Virtual Fly Brain. Graphs were generated using the open source R package ggplot2 and related packages or GraphPad Prism 5 software.
### Data and Software Availability
#### Data Availability
SWC files for the skeletonized multi-color flip-out data, and EM reconstructions for PD2a1 and PD2b1 neurons, and right-side MBON-ɑ2sc and MBON-ɑ′2 are available as a supplement to this paper ( ). Other data supporting the findings in this study are available upon request. A spreadsheet of glomeruli and published behavioral significance/functions is available upon request.
#### Code Availability
All R packages described above are available by following links from . Packages include full documentation and sample code. Custom scripts used to to generate figures can be made available upon request.
|
Odorant receptors (ORs) form one of the largest gene families in the genome. However, the vast majority are orphan receptors as the ligands that activate them remain unknown. Deorphaning approaches have generally focused on finding ligands for particular receptors expressed in homologous or heterologous cells; these attempts have met with only partial success. Here, we outline a conceptually different strategy in which we search for odorant receptors activated by a known odorant. Intrinsic signal imaging of the main olfactory bulb is first used to locate activated glomeruli in vivo, followed by retrograde tracing to label the sensory neurons in the olfactory epithelium projecting to the activated glomerulus. Subsequently, single cell RT-PCR is used to reveal the identity of the odorant receptors expressed in retrogradely labeled neurons. To demonstrate the applicability of this method, we searched for candidate ORs responding to the aldehyde odorant butanal. This method may be a useful tool to decipher specific ligand--OR interactions in the mouse olfactory bulb. |
Our study examined ethanol self-administration and accumbal dopamine concentration during kappa-opioid receptor (KOPr) blockade. Long-Evans rats were trained to respond for 20 min of access to 10% ethanol (with sucrose) over 7 days. Rats were injected s.c. with the long-acting KOPr antagonist, nor-binaltorphimine (NOR-BNI; 0 or 20 mg/kg) 15-20 h prior to testing. Microdialysis revealed a transient elevation in dopamine concentration within 5 min of ethanol access in controls. NOR-BNI-treated rats did not exhibit this response, but showed a latent increase in dopamine concentration at the end of the access period. The rise in dopamine levels correlated positively with dialysate ethanol concentration but not in controls. NOR-BNI did not alter dopamine levels in rats self-administering 10% sucrose. The transient dopamine response during ethanol acquisition in controls is consistent with previous results that were attributed to ethanol stimulus cues. The altered dopamine response to NOR-BNI during ethanol drinking suggests that KOPr blockade temporarily uncovered a pharmacological stimulation of dopamine release by ethanol. Despite these neurochemical changes, NOR-BNI did not alter operant responding or ethanol intake, suggesting that the KOPr is not involved in ethanol-reinforced behavior under the limited conditions we studied. |
Cannabinoids, the active components of Cannabis sativa L. and their derivatives, inhibit tumor growth in laboratory animals by inducing apoptosis of tumor cells and inhibiting tumor angiogenesis. It has also been reported that cannabinoids inhibit tumor cell invasiveness, but the molecular targets of this cannabinoid action remain elusive. Here we evaluated the effects of cannabinoids on the expression of tissue inhibitors of metalloproteinases (TIMPs), which play critical roles in the acquisition of migrating and invasive capacities by tumor cells. Local administration of Delta(9)-tetrahydrocannabinol (THC), the major active ingredient of cannabis, down-regulated TIMP-1 expression in mice bearing subcutaneous gliomas, as determined by Western blot and immunofluorescence analyses. This cannabinoid-induced inhibition of TIMP-1 expression in gliomas (i) was mimicked by JWH-133, a selective CB(2) cannabinoid receptor agonist that is devoid of psychoactive side effects, (ii) was abrogated by fumonisin B1, a selective inhibitor of ceramide synthesis de novo, and (iii) was also evident in two patients with recurrent glioblastoma multiforme (grade IV astrocytoma). THC also depressed TIMP-1 expression in cultures of various human glioma cell lines as well as in primary tumor cells obtained from a glioblastoma multiforme patient. This action was prevented by pharmacological blockade of ceramide biosynthesis and by knocking-down the expression of the stress protein p8. As TIMP-1 up-regulation is associated with high malignancy and negative prognosis of numerous cancers, TIMP-1 down-regulation may be a hallmark of cannabinoid-induced inhibition of glioma progression. |
As chemical entities, lipoamino acids have been known for some time. However, more recently their occurrence and importance in mammalian species has been discovered. They appear to have close relationships with the endocannabinoids not only structurally but also in terms of biological actions. The latter include analgesia, anti-inflammatory effects, inhibition of cell proliferation and calcium ion mobilization. To date about 40 naturally occurring members of this family have been identified and, additionally, several synthetic analogs have been prepared and studied. To facilitate their identity, a nomenclature system has been suggested based on the name elmiric acid (EMA). The prototypic example, N-arachidonoyl glycine, does not bind to CB1, however it does inhibit the glycine transporter GLYT2a and also appears to be a ligand for the orphan G-protein-coupled receptor GPR18. It may also have a role in regulating tissue levels of anandamide by virtue of its inhibitory effect on FAAH the enzyme that mediates inactivation of anandamide. Its concentration in rat brain is several-fold higher than anandamide supporting its possible role as a physiological mediator. Future studies should be aimed at elucidating the actions of all of the members of this interesting family of molecules. |
Alzheimer's disease (AD) is the most common cause of dementia, clinically characterized by loss of memory and progressive deficits in different cognitive domains. An emerging disease-modifying approach to face the multifactorial nature of AD may be represented by the development of Multi-Target Directed Ligands (MTDLs), i.e., single compounds which may simultaneously modulate different targets involved in the neurodegenerative AD cascade. The structure of tacrine, an acetylcholinesterase (AChE) inhibitor (AChEI), has been widely used as scaffold to provide new MTDLs. In particular, its homodimer bis(7)tacrine represents an interesting lead compound to design novel MTDLs. Thus, in the search of new rationally designed MTDLs against AD, we replaced the heptamethylene linker of bis(7)tacrine with the structure of cystamine, leading to cystamine-tacrine dimer. In this study we demonstrated that the cystamine-tacrine dimer is endowed with a lower toxicity in comparison to bis(7)tacrine, it is able to inhibit AChE, butyrylcholinesterase (BChE), self- and AChE-induced beta-amyloid aggregation in the same range of the reference compound and exerts a neuroprotective action on SH-SY5Y cell line against H(2)O(2)-induced oxidative injury. The investigation of the mechanism of neuroprotection showed that the cystamine-tacrine dimer acts by activating kinase 1 and 2 (ERK1/2) and Akt/protein kinase B (PKB) pathways. This article is part of a Special Issue entitled 'Post-Traumatic Stress Disorder'. |
L-3,4-Dihydroxyphenylalanine (l-DOPA), the gold standard therapy for Parkinson disease (PD), is associated with motor fluctuations and dyskinesias. This study sought to prevent the development of l-DOPA-induced dyskinesias (LID) with the metabotropic glutamate receptor type 5 (mGlu5 receptor) antagonist 2-methyl-6-(phenylethynyl)pyridine (MPEP) in the de novo treatment of monkeys lesioned with 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) as a PD model. MPTP-lesioned monkeys were treated once daily for one month with either l-DOPA or l-DOPA + MPEP (10 mg/kg). MPEP (administered 15 min before l-DOPA) plasma concentrations were elevated during all the l-DOPA motor activation and did not accumulate during a month. The antiparkinsonian effect was maintained throughout the treatment period in MPTP-lesioned monkeys treated with l-DOPA + MPEP, while the duration of this effect decreased over time in MPTP-lesioned monkeys treated with l-DOPA alone, suggesting wearing-off. Over the month-long treatment, the mean dyskinesia score increased in l-DOPA-treated monkeys; interestingly, this increase was reduced by overall 72% in the l-DOPA + MPEP group. Mean dyskinesia scores of monkeys correlated inversely with plasma MPEP concentrations. Normal control and saline-treated MPTP-lesioned monkeys were also included for biochemical analyses. All MPTP-lesioned monkeys were extensively and similarly denervated. [(3)H]ABP688 specific binding to mGlu5 receptors increased in the putamen of l-DOPA-treated monkeys compared to control, saline or l-DOPA + MPEP-treated monkeys. Mean dyskinesia scores of MPTP-lesioned monkeys correlated positively with [(3)H]ABP688 specific binding in the putamen. This study showed a beneficial chronic antidyskinetic effect of MPEP in de novol-DOPA-treated MPTP-lesioned monkeys, supporting the therapeutic use of mGlu5 receptor antagonists in PD to prevent LID. This article is part of a Special Issue entitled 'Metabotropic Glutamate Receptors'. |
Agonists and positive allosteric modulators (PAMs) of α7 nicotinic acetylcholine receptors (nAChRs) are currently being considered as novel therapeutic approaches for managing cognitive deficits in schizophrenia and Alzheimer's disease. Though α7 agonists were recently found to possess antinociceptive and anti-inflammatory properties in rodent models of chronic neuropathic pain and inflammation, the effects of α7 nAChRs PAMs on chronic pain and inflammation remain largely unknown. The present study investigated whether PAMs, by increasing endogenous cholinergic tone, potentiate α7 nAChRs function to attenuate inflammatory and chronic neuropathic pain in mice. We tested two types of PAMS, type I (NS1738) and type II (PNU-120596) in carrageenan-induced inflammatory pain and chronic constriction injury (CCI) neuropathic pain models. We found that both NS1738 and PNU-120596 significantly reduced thermal hyperalgesia, while only PNU-120596 significantly reduced edema caused by a hind paw infusion of carrageenan. Importantly, PNU-120596 reversed established thermal hyperalgesia and edema induced by carrageenan. In the CCI model, PNU-120596 had long-lasting (up to 6 h), dose-dependent anti-hyperalgesic and anti-allodynic effects after a single injection, while NS1738 was inactive. Systemic administration of the α7 nAChR antagonist MLA reversed PNU-120596's effects, suggesting the involvement of central and peripheral α7 nAChRs. Furthermore, PNU-120596 enhanced an ineffective dose of selective agonist PHA-543613 to produce anti-allodynic effects in the CCI model. Our results indicate that the type II α7 nAChRs PAM PNU-120596, but not the type I α7 nAChRs PAM NS1738, shows significant anti-edematous and anti-allodynic effects in inflammatory and CCI pain models in mice. |
Postictal refractoriness may be taken as an expression of lasting activity of inhibitory systems arresting seizures. We tested drugs interfering with GABAergic inhibitory system in pairs of cortical epileptic afterdischarges induced with 1-min interval in rats. Under control conditions the second stimulation failed to elicit an afterdischarge. This postictal refractoriness was not affected by antagonists of GABAA receptors acting at three binding sites (bicuculline, picrotoxin, benzodiazepine inverse agonist Ro 19-4603) as well as by a less specific antagonist pentetrazol. In contrast, antagonist of GABAB receptors CGP35348 partially blocked the refractoriness. Cooperation of different inhibitory systems is probably necessary to abolish postictal refractoriness in neocortex. This article is part of the Special Issue entitled 'GABAergic Signaling in Health and Disease'. |
The cAMP signaling pathway has emerged as an important modulator of the pharmacological effects of ethanol. In this respect, the cAMP-dependent protein kinase has been shown to play an important role in the modulation of several ethanol-induced behavioral actions. Cellular levels of cAMP are maintained by the activity of adenylyl cyclases and phosphodiesterases. In the present work we have focused on ascertaining the role of PDE4 in mediating the neurobehavioral effects of ethanol. For this purpose, we have used the selective PDE4 inhibitor Ro 20-1724. This compound has been proven to enhance cellular cAMP response by PDE4 blockade and can be administered systemically. Swiss mice were injected intraperitoneally (i.p.) with Ro 20-1724 (0-5 mg/kg; i.p.) at different time intervals before ethanol (0-4 g/kg; i.p.) administration. Immediately after the ethanol injection, locomotor activity, loss of righting reflex, PKA footprint and enzymatic activity were assessed. Pretreatment with Ro 20-1724 increased ethanol-induced locomotor stimulation in a dose-dependent manner. Doses that increased locomotor stimulation did not modify basal locomotion or the suppression of motor activity produced by high doses of this alcohol. Ro 20-1724 did not alter the locomotor activation produced by amphetamine or cocaine. The time of loss of righting reflex evoked by ethanol was increased after pretreatment with Ro 20-1724. This effect was selective for the narcotic effects of ethanol since Ro 20-1724 did not affect pentobarbital-induced narcotic effects. Moreover, Ro 20-1724 administration increased the PKA footprint and enzymatic activity response elicited by ethanol. These data provide further evidence of the key role of the cAMP signaling pathway in the central effects of ethanol. |
Anaplastic lymphoma kinase (ALK) is a receptor tyrosine kinase that is expressed in the brain and implicated in alcohol abuse in humans and behavioral responses to ethanol in mice. Previous studies have shown an association of human ALK with acute responses to alcohol and alcohol dependence. In addition, Alk knockout (Alk -/-) mice consume more ethanol in a binge-drinking test and show increased sensitivity to ethanol sedation. However, the function of ALK in excessive drinking following the establishment of ethanol dependence has not been examined. In this study, we tested Alk -/- mice for dependence-induced drinking using the chronic intermittent ethanol-two bottle choice drinking (CIE-2BC) protocol. We found that Alk -/- mice initially consume more ethanol prior to CIE exposure, but do not escalate ethanol consumption after exposure, suggesting that ALK may promote the escalation of drinking after ethanol dependence. To determine the mechanism(s) responsible for this behavioral phenotype we used an electrophysiological approach to examine GABA neurotransmission in the central nucleus of the amygdala (CeA), a brain region that regulates alcohol consumption and shows increased GABA signaling after chronic ethanol exposure. GABA transmission in ethanol-naïve Alk -/- mice was enhanced at baseline and potentiated in response to acute ethanol application when compared to wild-type (Alk +/+) mice. Moreover, basal GABA transmission was not elevated by CIE exposure in Alk -/- mice as it was in Alk +/+ mice. These data suggest that ALK plays a role in dependence-induced drinking and the regulation of presynaptic GABA release in the CeA. |
The overexpression of Kir3.2, a subunit of the G protein-gated inwardly rectifying K(+) channel, is implicated in some of the neurological phenotypes of Down syndrome (DS). Chemical compounds that block Kir3.2 are expected to improve the symptoms of DS. The purpose of this study is to develop a cell-based screening system to identify Kir3.2 blockers and then investigate the mode of action of the blocker. Chemical screening was carried out using a K(+) transporter-deficient yeast strain that expressed a constitutively active Kir3.2 mutant. The mode of action of an effective blocker was electrophysiologically analyzed using Kir channels expressed in Xenopus oocytes. Proflavine was identified to inhibit the growth of Kir3.2-transformant cells and Kir3.2 activity in a concentration-dependent manner. The current inhibition was strong when membrane potentials (Vm) was above equilibrium potential of K(+) (EK). When Vm was below EK, the blockage apparently depended on the difference between Vm and [K(+)]. Furthermore, the inhibition became stronger by lowering extracellular [K(+)]. These results indicated that the yeast strain serves as a screening system to isolate Kir3.2 blockers and proflavine is a prototype of a pore blocker of Kir3.2. |
Voltage-gated potassium channels play a key role in human physiology and pathology. Reflecting their importance, numerous channelopathies have been characterised that arise from mutations in these channels or from autoimmune attack on the channels. Voltage-gated potassium channels are also the target of a broad range of peptide toxins from venomous organisms, including sea anemones, scorpions, spiders, snakes and cone snails; many of these peptides bind to the channels with high potency and selectivity. In this review we describe the various classes of peptide toxins that block these channels and illustrate the broad range of three-dimensional structures that support channel blockade. The therapeutic opportunities afforded by these peptides are also highlighted. This article is part of the Special Issue entitled 'Venom-derived Peptides as Pharmacological Tools.' |
The role of nitric oxide (NO) in nociceptive transmission at the spinal cord level remains uncertain. Increased activity of spinal N-methyl-d-aspartate (NMDA) receptors contributes to development of chronic pain induced by peripheral nerve injury. In this study, we determined how endogenous NO affects NMDA receptor activity of spinal cord dorsal horn neurons in control and spinal nerve-ligated rats. Bath application of the NO precursor l-arginine or the NO donor S-nitroso-N-acetylpenicillamine (SNAP) significantly inhibited NMDA receptor currents of spinal dorsal horn neurons in both sham control and nerve-injured rats. Inhibition of neuronal nitric oxide synthase (nNOS) or blocking the S-nitrosylation reaction with N-ethylmaleimide abolished the inhibitory effects of l-arginine on NMDA receptor currents recorded from spinal dorsal horn neurons in sham control and nerve-injured rats. However, bath application of the cGMP analog 8-bromo-cGMP had no significant effects on spinal NMDA receptor currents. Inhibition of soluble guanylyl cyclase also did not alter the inhibitory effect of l-arginine on spinal NMDA receptor activity. Furthermore, knockdown of nNOS with siRNA abolished the inhibitory effects of l-arginine, but not SNAP, on spinal NMDA receptor activity in both groups of rats. Additionally, intrathecal injection of l-arginine significantly attenuated mechanical or thermal hyperalgesia induced by nerve injury, and the l-arginine effect was diminished in rats treated with a nNOS inhibitor or nNOS-specific siRNA. These findings suggest that endogenous NO inhibits spinal NMDA receptor activity through S-nitrosylation. NO derived from nNOS attenuates spinal nociceptive transmission and neuropathic pain induced by nerve injury. |
Exposure to prenatal insults has been associated with an increased risk for neuropsychiatric disorders, including depression, but the mechanisms are still poorly understood. Persistent alterations of the HPA axis feedback mechanism as well as adult impaired neurogenesis are believed to play a relevant role in the etiology of depression. In addition, growing evidence points at epigenetic reprogramming as a key factor. We have previously shown that prenatal exposure to the synthetic glucocorticoid dexamethasone (DEX) impairs neurogenesis and leads to late onset of depression-like behavior that does not respond to the SSRI antidepressant fluoxetine (FLX). The aims of this study were to assess the effect of DEX prenatal exposure on the morphology of hippocampal granule neurons and on the expression of genes related to plasticity; and to test whether the SNRI antidepressant desipramine (DMI), unlike FLX, could counteract the effect of prenatal-DEX. C57Bl/6 mice were exposed to DEX (0.05 mg/kg/day) in utero and received intra-hippocampal injection of GFP expressing retroviral vector for labeling of newborn granule cells at eleven months. By twelve months, DEX mice showed depression-like behavior associated with decreased neurogenesis and morphological alterations of the newborn granule cells in the dentate gyrus (DG). Furthermore DEX mice displayed altered expression of genes controlling neurogenesis and neuronal morphology, such as Cdkn1c, p16, TrkB, DISC1 and Reelin. Chronic treatment with DMI led to a significant decrease in immobility time in the forced swim test. In addition, DMI restored neurogenesis, neuronal morphology in the DG, as well as the expression of all related genes. Our results suggest that (1) prenatal DEX induces early and persistent reprogramming effects resulting in altered neurogenesis and neuronal morphology; and (2) DMI treatment reverses DEX-induced depression by restoring the expression of genes relevant to neuronal plasticity. |
Excessive alcohol intake induces an inflammatory response in the brain, via TNFα, TLR4 and NF-κB signaling pathways. It has been proposed that neuroinflammation would play a very important role in the development of alcohol addiction. In addition to stimulating the synthesis of inflammatory mediators such as IL-6, IL-1β and TNFα, NF-κB is capable of reducing the anti-inflammatory activity of PPARα and PPARγ. Reciprocally, PPARα, PPARγ and melanocortin 4 receptor (MC4R) can decrease the proinflammatory activity of NF-κB, establishing an interplay of inactivations between such nuclear factors and receptors. In this review, we hypothesize that one of the mechanisms by which alcohol produces neuroinflammation is through NF-κB-mediated decrease in PPARα and PPARγ anti-inflammatory activities; in addition, ethanol negatively affects MC4R activity, decreasing the ability of this receptor to activate PPARγ. PPARα, PPARγ and MC4R can be pharmacologically activated by synthetic ligands (fibrates, thiazolidinediones and synthetic peptides, respectively); in this context, we propose that the administration of such ligands would decrease neuroinflammation produced by alcohol intake. The advantage of this approach is that fibrates and thiazolidinediones are FDA-approved drugs that have been used for years in other clinical conditions, and now may offer a new perspective for the treatment of alcoholism. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.