bioRxiv Subject Collection: Neuroscience's Journal

Patch and matrix striatonigral neurons differentially regulate locomotion
Striatonigral neurons, known to promote locomotion, reside in both the patch and matrix compartments of the dorsal striatum. However, their compartment-specific contributions to locomotion remain largely unexplored. Using molecular identifier Kremen1 and Calb1, we showed in mouse models that patch and matrix striatonigral neurons exert opposite influences on locomotion. Matrix striatonigral neurons reduced their activity before the cessation of self-paced locomotion, while patch striatonigral neuronal activity increased, suggesting an inhibitory function. Indeed, optogenetic activation of patch striatonigral neurons suppressed ongoing locomotion with reduced striatal dopamine release, contrasting with the locomotion-promoting effect of matrix striatonigral neurons, which showed an initial increase in dopamine release. Furthermore, genetic deletion of the GABA-B receptor in Aldehyde dehydrogenase 1A1-positive (ALDH1A1+) nigrostriatal dopaminergic neurons completely abolished the locomotion-suppressing effect of patch striatonigral neurons. Our findings unravel a compartment-specific mechanism governing locomotion in the dorsal striatum, where patch striatonigral neurons suppress locomotion by inhibiting ALDH1A1+ nigrostriatal dopaminergic neurons.

Salience signaling and stimulus scaling of ventral tegmental area glutamate neuron subtypes
Ventral tegmental area (VTA) glutamatergic neurons participate in reward, aversion, drug-seeking, and stress. Subsets of VTA VGluT2+ neurons are capable of co-transmitting glutamate and GABA (VGluT2+VGaT+ neurons), transmitting glutamate without GABA (VGluT2+VGaT- neurons), or co-transmitting glutamate and dopamine (VGluT2+TH+ neurons), but whether these molecularly distinct subpopulations show behavior-related differences is not wholly understood. We identified that neuronal activity of each VGluT2+ subpopulation is sensitive to reward value but signaled this in different ways. The phasic maximum activity of VGluT2+VGaT+ neurons increased with sucrose concentration, whereas VGluT2+VGaT- neurons increased maximum and sustained activity with sucrose concentration, and VGluT2+TH+ neurons increased sustained but not maximum activity with sucrose concentration. Additionally, VGluT2+ subpopulations signaled consummatory preferences in different ways. VGluT2+VGaT- neurons and VGluT2+TH+ neurons showed a signaling preference for a behaviorally-preferred fat reward over sucrose, but in temporally-distinct ways. In contrast, VGluT2+VGaT+ neurons uniquely signaled a less behaviorally-preferred sucrose reward compared with fat. Further experiments suggested that VGluT2+VGaT+ consummatory reward-related activity was related to sweetness, partially modulated by hunger state, and not dependent on caloric content or behavioral preference. All VGluT2+ subtypes increased neuronal activity following aversive stimuli but VGluT2+VGaT+ neurons uniquely scaled their magnitude and sustained activity with footshock intensity. Optogenetic activation of VGluT2+VGaT+ neurons during low intensity footshock enhanced fear-related behavior without inducing place preference or aversion. We interpret these data such that VTA glutamatergic subpopulations signal different elements of rewarding and aversive experiences and highlight the unique role of VTA VGluT2+VGaT+ neurons in enhancing the salience of behavioral experiences.

Synaptic alterations in pyramidal cells following genetic manipulation of neuronal excitability in monkey prefrontal cortex
In schizophrenia, layer 3 pyramidal neurons (L3PNs) in the dorsolateral prefrontal cortex (DLPFC) are thought to receive fewer excitatory synaptic inputs and to have lower expression levels of activity-dependent genes and of genes involved in mitochondrial energy production. In concert, these findings from previous studies suggest that DLPFC L3PNs are hypoactive in schizophrenia, disrupting the patterns of activity that are crucial for working memory, which is impaired in the illness. However, whether lower PN activity produces alterations in inhibitory and/or excitatory synaptic strength has not been tested in the primate DLPFC. Here, we decreased PN excitability in rhesus monkey DLPFC in vivo using adeno-associated viral vectors (AAVs) to produce Cre recombinase-mediated overexpression of Kir2.1 channels, a genetic silencing tool that efficiently decreases neuronal excitability. In acute slices prepared from DLPFC 7-12 weeks post-AAV microinjections, Kir2.1-overexpressing PNs had a significantly reduced excitability largely attributable to highly specific effects of the AAV-encoded Kir2.1 channels. Moreover, recordings of synaptic currents showed that Kir2.1-overexpressing DLPFC PNs had reduced strength of excitatory synapses whereas inhibitory synaptic inputs were not affected. The decrease in excitatory synaptic strength was not associated with changes in dendritic spine number, suggesting that excitatory synapse quantity was unaltered in Kir2.1-overexpressing DLPFC PNs. These findings suggest that, in schizophrenia, the excitatory synapses on hypoactive L3PNs are weaker and thus might represent a substrate for novel therapeutic interventions.

Brain Age Prediction: Deep Models Need a Hand to Generalize
In the pursuit of studying brain aging, numerous models to predict brain age from T1-weighted MRI have been developed. Recently, many of these models take advantage of deep learning architectures, and although they have shown some success in terms of reported Mean Absolute Error in predicted age, doubts linger regarding their generalization ability due to the large number of trainable parameters in deep models and the limited training data available in medical imaging datasets. Unfortunately, the evaluation details of the proposed models' generalization ability are often overlooked in the literature. In this study, we assess a deep model, SFCN-reg, based on the VGG-16 architecture, and examine whether we can enhance its generalizability and robustness by (1) employing a more than minimal preprocessing pipeline, (2) incorporating more extensive data augmentation during training, and (3) applying regularization to the model. We found that the age prediction mean absolute error in independent out-of-distribution publicly available cohorts is reduced from 5.25 years to 2.96 years (44% improvement) in the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset and from 4.35 years to 3.40 years (22% improvement) in the Australian Imaging, Biomarker and Lifestyle (AIBL) dataset, closing the generalization gap to less than 0.9 years when better preprocessing is used. Moreover, data augmentation and model regularization reduce the scan-rescan error from 0.86 years to 0.61 years (29% improvement), increasing the model robustness. The final model is also much less sensitive against registration errors, showing almost no bias in comparison with other models when facing incorrectly registered images.

Interactively Integrating Reach and Grasp Information in Macaque Premotor Cortex
Successful reach-to-grasp movements necessitate the integration of both object location and grip type information. However, how these two types of information are encoded in a single brain region and to what extend they interact with each other, remain largely unknown. We designed a novel experimental paradigm that sequentially prompted reach and grasp cues to monkeys and recorded neural activity in the dorsal premotor cortex (PMd) to investigate how the encoding structures change and interact during arm reaching and hand grasping movements. This paradigm required monkeys to retain the first prompted cue when the second one arrived, and integrate both to accomplish a final goal movement. PMd neurons represented both reach and grasp to similar extend, yet the encodings were not independent. Upon the arrival of second cue, PMd continued to encode the first cue, albeit with a significantly altered structure, as evidenced by more than half of the neurons displaying incongruent modulation. At a population level, the encoding structure formed a distinct subspace that differed from, but was not entirely orthogonal to, the original one. Employing canonical correlation analysis, we identified a subspace that consistently preserved the encoding of the initial cue, potentially serving as a mechanism for downstream brain regions to extract coherent information. Furthermore, this shared subspace comprised a diverse population of neurons, including both congruent and incongruent units. these findings support the argument that reach and grasp information are interactively integrated within PMd, with a shared subspace likely underpinning a consistent encoding framework.

Direct modulation of CRH nerve terminal function by noradrenaline and corticosterone.
Nerve terminals are the final point of regulation before neurosecretion. As such, neuromodulators acting on nerve terminals can exert significant influence on neural signalling. Hypothalamic corticotropin-releasing hormone (CRH) neurons send axonal projections to the median eminence where CRH is secreted to stimulate the hypothalamic-pituitary-adrenal (HPA) axis. Noradrenaline and corticosterone are two of the most important neuromodulators of HPA axis function; noradrenaline excites CRH neurons and corticosterone inhibits CRH neurons by negative feedback. Here, we used GCaMP6f Ca2+ imaging and measurement of nerve terminal CRH secretion using sniffer cells to determine whether these neuromodulators act directly on CRH nerve terminals. Contrary to expectations, noradrenaline inhibited action potential-dependent Ca2+ elevations in CRH nerve terminals and suppressed evoked CRH secretion. This inhibitory effect was blocked by 2-adrenoreceptor antagonism. Corticosterone also suppressed evoked CRH peptide secretion from nerve terminals, independent of action potential-dependent Ca2+ levels. This inhibition was prevented by the glucocorticoid receptor antagonist, RU486, and indicates that CRH nerve terminals may be a site of fast glucocorticoid negative feedback. Together these findings establish median eminence nerve terminals as a key site for regulation of the HPA axis.

Reevaluating the Neural Noise Hypothesis in Dyslexia: Insights from EEG and 7T MRS Biomarkers
The neural noise hypothesis of dyslexia posits an imbalance between excitatory and inhibitory (E/I) brain activity as an underlying mechanism of reading difficulties. This study provides the first direct test of this hypothesis using both indirect EEG power spectrum measures in 120 Polish adolescents and young adults (60 with dyslexia, 60 controls) and direct glutamate (Glu) and gamma-aminobutyric acid (GABA) concentrations from magnetic resonance spectroscopy (MRS) at 7T MRI scanner in half of the sample. Our results, supported by Bayesian statistics, show no evidence of E/I balance differences between groups, challenging the hypothesis that cortical hyperexcitability underlies dyslexia. These findings suggest alternative mechanisms must be explored and highlight the need for further research into the E/I balance and its role in neurodevelopmental disorders.

A 5-day course of rTMS before pain onset ameliorates future pain and increases sensorimotor peak alpha frequency
Repetitive transcranial magnetic stimulation (rTMS) has shown promise as an intervention for pain. An unexplored research question is whether the delivery of rTMS prior to pain onset might protect against a future episode of prolonged pain. The present study aimed to determine i) whether 5 consecutive days of rTMS delivered prior to experimentally-induced prolonged jaw pain could reduce future pain intensity and ii) whether any effects of rTMS on pain were mediated by changes in corticomotor excitability (CME) and/or sensorimotor peak alpha frequency (PAF). On each day from Day 0-4, forty healthy individuals received a single session of active (n = 21) or sham (n = 19) rTMS over the left primary motor cortex. PAF and CME were assessed on Day 0 (before rTMS) and Day 4 (after rTMS). Prolonged pain was induced via intramuscular injection of nerve growth factor (NGF) in the right masseter muscle after the final rTMS session. From Days 5-25, participants completed twice-daily electronic dairies including pain on chewing and yawning (primary outcomes), as well as pain during other activities (e.g. talking), functional limitation in jaw function and muscle soreness (secondary outcomes). Compared to sham, individuals who received active rTMS subsequently experienced lower pain on chewing and yawning. Although active rTMS increased PAF, the effects of rTMS on pain were not mediated by changes in PAF or CME. This study is the first to show that rTMS delivered prior to pain onset can protect against future pain and associated functional impairment. Thus, rTMS may hold promise as a prophylactic intervention for persistent pain.

Cortical tracking of naturalistic music and speech across frequency bands and brain regions: functional mapping and temporal dynamics
Music and speech encode hierarchically organized structural complexity at the service of human expressiveness and communication. Carefully controlled experiments have suggested that populations of neurons in cortical auditory regions track temporal modulations within rhythmic acoustic signals, physiologically supporting perception of both music and speech. However, whether cortical tracking of music and speech extends to less controlled (i.e., naturalistic) signals remains contentious. Here, we investigated whether cortical tracking can be observed under more natural perceptual scenarios, and how stimulus type, frequency band or anatomical localization modulates this effect. We analyzed intracranial recordings from 30 subjects while they passively watched a movie where visual scenes were accompanied by either music or speech stimuli. Cross-correlation between brain and acoustic signals, along with density-based clustering analyses and linear mixed effect modeling, revealed both anatomically overlapping and functionally distinct mapping of the tracking effect as a function of stimulus type and frequency band. We observed widespread tracking of music and speech signals in the Slow Frequency Band (1-8Hz), with near zero temporal lags and high mixed-selectivity. In contrast, High Frequency Band (70-120Hz) tracking was higher during speech perception, was more densely concentrated in classical language processing areas, and showed a clear frontal-to-temporal gradient in lag values that was not observed during perception of musical stimuli. Our results highlight the recruitment of domain-general and domain-specific mechanisms during perception of naturalistic music and speech signals, as well as a complex interaction between cortical region and frequency band that shapes temporal dynamics during processing of hierarchically organized temporal structures in speech.

The effect of a dominant kinase-dead Csf1r mutation associated with adult-onset leukoencephalopathy on brain development and neuropathology.
Amino acid substitutions in the kinase domain of the human CSF1R protein are associated with autosomal dominant adult-onset leukoencephalopathy with axonal spheroids and pigmented glia (ALSP). To model the human disease, we created a disease-associated mutation (Glu631Lys; E631K) in the mouse Csf1r locus. Previous analysis demonstrated that heterozygous mutation (Csf1rE631K/+) had a dominant inhibitory effect on CSF1R signaling in vitro and in vivo but did not recapitulate the pathology of the human disease. We speculated that leukoencephalopathy in humans requires an environmental trigger and/or epistatic interaction with common neurodegenerative disease-associated alleles. Here we examine the impact of heterozygous Csf1r mutation on microglial phenotype, normal postnatal brain development, age-related changes in gene expression and on two distinct pathologies in which microgliosis is a prominent feature, prion disease and experimental autoimmune encephalitis (EAE). The heterozygous Csf1rE631K/+ mutation reduced microglial abundance and the expression of microglial-associated transcripts relative to wild-type controls at 12 weeks and 43 weeks of age but had no selective effect on homeostatic markers such as P2ry12. An epistatic interaction was demonstrated between Csf1rE631K/+ and Cxc3r1EGFP/+ genotypes leading to dysregulated microglial and neuronal gene expression in both hippocampus and striatum. Heterozygous Csf1rE631K mutation reduced the microgliosis associated with both diseases. There was no significant impact on disease severity or progression in prion disease. In EAE, induced expression of inflammation-associated transcripts in the hippocampus and striatum was suppressed in parallel with microglia-specific transcripts, but spinal cord demyelination was exacerbated. The results support a dominant-negative model of CSF1R-associated leukoencephalopathy and likely contributions of an environmental trigger and/or genetic background to neuropathology.

Differential contribution of direct and indirect pathways from dorsolateral and dorsomedial striatum to motor symptoms in Huntington's disease mice
The alterations in the basal ganglia circuitry associated with motor symptoms in Huntington's Disease (HD) have been extensively investigated. Yet, the specific contribution of the direct and indirect striatal output pathways from the dorsolateral (DLS) and dorsomedial striatum (DMS) to the motor dysfunction is still not fully understood. Here, using the symptomatic R6/1 male mouse model of HD, strong functional connectivity alterations between DMS and DLS regions with the rest of brain were observed by fMRI, particularly pronounced in the DLS. Then, we systematically evaluated how the selective optogenetic stimulation of the direct and indirect pathways from DLS and DMS influences locomotion, exploratory behavior, and motor learning. In wild type (WT) mice, optogenetic stimulation of the direct pathway from DLS and the indirect pathway from DMS elicited subtle locomotor enhancements, while exploratory behavior remained unaltered. Additionally, stimulation of the indirect pathway from DLS improved the performance in the accelerated rotarod task. In contrast, in HD mice, optogenetic stimulation of the distinct striatal pathways did not modulate these behaviors. Overall, this study points to deficits in the integration of neuronal activity in HD mice, while it contributes to deeper understanding of the complexity of motor control by the diverse striatal subcircuits.

Transcriptomics and proteomics of projection neurons in a circuit linking hippocampus with dorsolateral prefrontal cortex in human brain
RNA-sequencing studies of brain tissue homogenates have shed light on the molecular processes underlying schizophrenia (SCZ) but lack biological granularity at the cell type level. Laser capture microdissection (LCM) can isolate selective cell populations with intact cell bodies to allow complementary gene expression analyses of mRNA and protein. We used LCM to collect excitatory neuron-enriched samples from CA1 and subiculum (SUB) of the hippocampus and layer III of the dorsolateral prefrontal cortex (DLPFC), from which we generated gene, transcript, and peptide level data. In a machine learning framework, LCM-derived expression achieved superior regional identity predictions as compared to bulk tissue, with further improvements when using isoform-level transcript and protein quantifications. LCM-derived co-expression also had increased co-expression strength of neuronal gene sets compared to tissue homogenates. SCZ risk co-expression pathways were identified and replicated across transcript and protein networks and were consistently enriched for glutamate receptor complex and post-synaptic functions. Finally, through inter-regional co-expression analyses, we show that CA1 to SUB transcriptomic connectivity may be altered in SCZ.

Decoding movie content from neuronal population activity in the human medial temporal lobe
Neurons of the medial temporal lobe (MTL) form the basis of semantic representation in the human brain. While known to contain category-selective cells, it is unclear how the MTL processes naturalistic, dynamic stimuli. We studied 2286 neurons recorded from the hippocampus, parahippocampal cortex, amygdala, and entorhinal cortex of 29 intracranially-implanted patients during a full-length movie. While few neurons responded preferentially to semantic features, we could reliably predict the presence of characters, location, and visual transitions from the neuronal populations using a recurrent neural network. We show that decoding performance differs across regions based on the feature category, and that the performance is driven by feature-selective single neurons when decoding visual transitions such as camera cuts. These findings suggest that semantic representation in the MTL varies based on semantic category, with decoding information embedded in specific subsets of neurons for event-related features or distributed across the entire population for character and location-related features.

DCX knockout ferret reveals a neurogenic mechanism in cortical development
Doublecortin (DCX) is one of the major causal proteins leading to lissencephaly and subcortical band heterotopia in human patients. However, our understanding of this disease, as well as the function of DCX during neurogenesis, remains limited due to the absence of suitable animal models that accurately represent human phenotypes. Here, we conducted a comprehensive examination of the neocortex at different stages in DCX knockout ferrets. We corroborated the neurogenic functions of DCX in progenitors. Loss of function of DCX led to the over-proliferation of neural progenitors and the truncation of basal processes of radial glial cells, which contributed to the thickening of cortices and the stalling of neurons underneath the cortical plate during neurogenic stages, respectively. We also present the first-ever cell atlas of the lissencephaly disease model, which embraces an almost reversed neuronal lamination distribution in the neocortex compared to the normal controls. Furthermore, we discovered alterations in molecular signatures tied to epilepsy, a condition frequently observed in lissencephaly patients. We also provided compelling evidence that the distribution of GABAergic inhibitory neurons in the cortex is intricately linked to glutamatergic excitatory neurons in a subtype-specific manner. In conclusion, our research offers new insights to expand our understanding of DCX's functions and enrich our comprehension of lissencephaly's intricacies.

Mitochondrial damage triggers concerted degradation of negative regulators of neuronal autophagy
Mutations in genes that regulate mitophagy, a key mitochondrial quality control pathway, are causative for neurological disorders including Parkinsons. Here, we identify a novel stress response pathway activated by mitochondrial damage that regulates mitophagy in neurons. We find that increasing levels of mitochondrial stress triggers a graded, concerted response that induces proteasomal degradation of negative regulators of autophagy. These include Myotubularin-related phosphatase 5 (MTMR5), MTMR2 and Rubicon. This Mitophagic Stress Response (MitoSR) pathway is neuron-specific and acts in parallel to the classical Pink1/Parkin mediated mitophagy pathway. While MTMR5/MTMR2 inhibits autophagosome biogenesis, we find that Rubicon inhibits lysosomal function and thus blocks autophagosome maturation. Targeted depletion of these negative regulators is sufficient to enhance mitophagy, promoting autophagosome biogenesis and facilitating the fusion of mitophagosomes with lysosomes. Our work suggests that therapeutic activation of the MitoSR pathway to induce degradation of negative regulators of autophagy may enhance mitochondrial quality control in stressed neurons.

Scale matters: Large language models with billions (rather than millions) of parameters better match neural representations of natural language
Recent research has used large language models (LLMs) to study the neural basis of naturalistic language processing in the human brain. LLMs have rapidly grown in complexity, leading to improved language processing capabilities. However, neuroscience researchers haven't kept up with the quick progress in LLM development. Here, we utilized several families of transformer-based LLMs to investigate the relationship between model size and their ability to capture linguistic information in the human brain. Crucially, a subset of LLMs were trained on a fixed training set, enabling us to dissociate model size from architecture and training set size. We used electrocorticography (ECoG) to measure neural activity in epilepsy patients while they listened to a 30-minute naturalistic audio story. We fit electrode-wise encoding models using contextual embeddings extracted from each hidden layer of the LLMs to predict word-level neural signals. In line with prior work, we found that larger LLMs better capture the structure of natural language and better predict neural activity. We also found a log-linear relationship where the encoding performance peaks in relatively earlier layers as model size increases. We also observed variations in the best-performing layer across different brain regions, corresponding to an organized language processing hierarchy.