Nature Neuroscience

Insights from biomedical citation networks can be used to identify promising avenues for accelerating research and its downstream bench-to-bedside translation. Citation analysis generally assumes that each citation documents causal knowledge transfer that informed the conception, design, or execution of the main experiments. Citations may exist for other reasons. In this paper we identify a subset of citations that are unlikely to represent causal knowledge flow. Using a large, comprehensive feature set of open access data, we train a predictive model to identify such citations. The model relies only on the title, abstract, and reference set and not the full-text or future citations patterns, making it suitable for publications as soon as they are released, or those behind a paywall (the vast majority). We find that the model identifies, with high prediction scores, citations that were likely added during the peer review process, and conversely identifies with low prediction scores citations that are known to represent causal knowledge transfer. Using the model, we find that federally funded biomedical research publications represent 30% of the estimated causal knowledge transfer from basic studies to clinical research, even though these comprise only 10% of the literature, a three-fold overrepresentation in this important type of knowledge transfer. This finding underscores the importance of federal funding as a policy lever to improve human health. Significance statementCitation networks document knowledge flow across the literature, and insights from these networks are increasingly used to form science policy decisions. However, many citations are known to be not causally related to the inception, design, and execution of the citing study. This adds noise to the insights derived from these networks. Here, we show that it is possible to train a machine learning model to identify such citations, and that the model learns to identify known causal citations as well. We use this model to show that government funding drives a disproportionate amount of causal knowledge transfer from basic to clinical research. This result highlights a straightforward policy lever for accelerating improvements to human health: federal funding.

Mapping the relationship between neural activity and motor behavior is a central aim of sensori-motor neuroscience and neurotechnology. While most progress to this end has relied on restricting complexity, the advent of foundation models instead proposes integrating a breadth of data as an alternate avenue for broadly advancing downstream modeling. We quantify this premise for motor decoding from intracortical microelectrode data, pretraining an autoregressive Transformer on 2000 hours of neural population spiking activity paired with diverse motor covariates from over 30 monkeys and humans. The resulting model is broadly useful, benefiting decoding on 8 downstream decoding tasks and generalizing to a variety of neural distribution shifts. However, we also highlight that scaling autoregressive Transformers seems unlikely to resolve limitations stemming from sensor variability and output stereotypy in neural datasets. Code: https://github.com/joel99/ndt3

Many cortical and subcortical regions contribute to complex social behavior; nevertheless, the network level architecture whereby the brain integrates this information to encode appetitive socioemotional behavior remains unknown. Here we measure electrical activity from eight brain regions as mice engage in a social preference assay. We then use machine learning to discover an explainable brain network that encodes the extent to which mice chose to engage another mouse. This socioemotional network is organized by theta oscillations leading from prelimbic cortex and amygdala that converge on ventral tegmental area, and network activity is synchronized with brain-wide cellular firing. The network generalizes, on a mouse-by-mouse basis, to encode socioemotional behaviors in healthy animals, but fails to encode an appetitive socioemotional state in a high confidence genetic mouse model of autism. Thus, our findings reveal the architecture whereby the brain integrates spatially distributed activity across timescales to encode an appetitive socioemotional brain state in health and disease.

Social interaction is inherently asymmetric, requiring coordinated activity between non-homologous brain regions across individuals. However, the brain-wide dynamics underlying such inter-brain coordination remain poorly understood. We used multi-fiber photometry to simultaneously record from 24 brain regions in pairs of freely interacting mice, including a model of autism. Social interactions evoked widespread, dynamic activity across brains, with inter-brain synchrony, especially between non-homologous areas, exceeding intra-brain synchrony, particularly in dominant mice. Network analysis revealed three subnetworks: (1) Emotional, intra-brain enhanced in subordinates; (2) Sensory, spanning both mice; (3) Decision/consolidation, linking dominant prefrontal cortex to subordinate hippocampus. These subnetworks encoded dominance, identity, and interaction roles, and followed a clear temporal sequence around social events. In an autism model, socially evoked activity was hyperactive displaying mostly within brain synchrony but lacked inter-brain synchrony. Our results uncover dynamic inter-brain circuits as a hallmark of social behavior and reveal their disruption in autism.

Classically, psychiatric disorders have been considered to lack defining pathology, but recent work has demonstrated consistent disruption at the molecular level, characterized by transcriptomic and epigenetic alterations.1-3 In ASD, upregulation of microglial, astrocyte, and immune signaling genes, downregulation of specific synaptic genes, and attenuation of regional gene expression differences are observed.1,2,4-6 However, whether these changes are limited to the cortical association areas profiled is unknown. Here, we perform RNA-sequencing (RNA-seq) on 725 brain samples spanning 11 distinct cortical areas in 112 ASD cases and neurotypical controls. We identify substantially more genes and isoforms that differentiate ASD from controls than previously observed. These alterations are pervasive and cortex-wide, but vary in magnitude across regions, roughly showing an anterior to posterior gradient, with the strongest signal in visual cortex, followed by parietal cortex and the temporal lobe. We find a notable enrichment of ASD genetic risk variants among cortex-wide downregulated synaptic plasticity genes and upregulated protein folding gene isoforms. Finally, using snRNA-seq, we determine that regional variation in the magnitude of transcriptomic dysregulation reflects changes in cellular proportion and cell-type-specific gene expression, particularly impacting L3/4 excitatory neurons. These results highlight widespread, genetically-driven neuronal dysfunction as a major component of ASD pathology in the cerebral cortex, extending beyond association cortices to involve primary sensory regions.

Functional magnetic resonance imaging (fMRI) contains rich individual, cognitive, and pathological information, yet no universal model exists for multi-task modeling of these dimensions. Here, we introduce MAMBAxBrain, a multi-task neural framework that integrates Mamba architecture with functional connectivity analysis to jointly model the temporal dynamics and spatial coordination of neural activity. MAMBAxBrain achieves high accuracy across four distinct fMRI objectives--brain fingerprinting, cognitive task decoding, reaction time prediction, and schizophrenia classification--consistently outperforming state-of-the-art methods with robust crosssession generalization. Interpretability analyses show that each task engages distinct, biologically plausible circuitry--from higher-order association cortex for identity to subcortical-motor loops for reaction time and disrupted control-sensory connectivity for schizophrenia. These findings inform a longstanding debate: rather than operating through wholly separate or entirely shared systems, the brain preferentially recruits task-specific circuits while retaining common representational structure across functions.

Skilled movement often requires flexibly combining multiple subskills, each requiring dedicated control strategies and underlying computations. How the motor system achieves such versatility remains unclear. Using high-density Neuropixels recordings from primary motor cortex (M1) in macaques performing a challenging force-tracking task, we reveal that M1 activity is much higher-dimensional, and far more flexible, than traditionally assumed. Although our task employed only a single external degree of freedom, neural dynamics reflected transitions amongst many dimensions and multiple distinct computations. Different behavioral control strategies were associated with distinct neural locations and dimensions, sometimes used compositionally. Groups of population-level factors became active when a particular form of dynamics was needed, and remained silent otherwise. Neural activity was thus dominated by the engaged subskill, and could be very different even for matched motor output. These findings challenge prevailing views of M1, and reveal an unexpectedly flexible and high-dimensional neural system underlying skilled motor behavior.

It is frequently assumed that the phenotypic heterogeneity in autism spectrum disorder reflects underlying pathobiological variation. However, direct evidence in support of this hypothesis is lacking. Here, we leverage cross-species functional neuroimaging to examine whether variability in brain functional connectivity reflects distinct biological mechanisms. We find that fMRI connectivity alterations in 20 distinct mouse models of autism (n=549 individual mice) can be clustered into two prominent hypo- and hyperconnectivity subtypes. We show that these connectivity profiles are linked to distinct signaling pathways, with hypoconnectivity being associated with synaptic dysfunction, and hyperconnectivity reflecting transcriptional and immune-related alterations. Extending these findings to humans, we identify analogous hypo- and hyperconnectivity subtypes in a large, multicenter resting state fMRI dataset of n=940 autistic and n=1036 neurotypical individuals. Remarkably, hypo- and hyperconnectivity autism subtypes are replicable across independent cohorts (accounting for 25.1% of all autism data), exhibit distinct functional network architecture, are behaviorally dissociable, and recapitulate synaptic and immune mechanisms identified in corresponding mouse subtypes. Our cross-species investigation, thus, decodes the heterogeneity of fMRI connectivity in autism into distinct pathway-specific etiologies, offering a new empirical framework for targeted subtyping of autism.

Numerous brain disorders demonstrate structural brain abnormalities, which are thought to arise from molecular perturbations or connectome miswiring. The unique and shared contributions of these molecular and connectomic vulnerabilities to brain disorders remain unknown, and has yet to be studied in a single multi-disorder framework. Using MRI morphometry from the ENIGMA consortium, we construct maps of cortical abnormalities for thirteen neurodevelopmental, neurological, and psychiatric disorders from N = 21 000 patients and N = 26 000 controls, collected using a harmonized processing protocol. We systematically compare cortical maps to multiple micro-architectural measures, including gene expression, neurotransmitter density, metabolism, and myelination (molecular vulnerability), as well as global connectomic measures including number of connections, centrality, and connection diversity (connectomic vulnerability). We find that regional molecular vulnerability and macroscale brain network architecture interact to drive the spatial patterning of cortical abnormalities in multiple disorders. Local attributes, particularly neurotransmitter receptor profiles, constitute the best predictors of both disorder-specific cortical morphology and cross-disorder similarity. Finally, we find that cross-disorder abnormalities are consistently subtended by a small subset of network epicentres in bilateral sensory-motor, medial temporal lobe, precuneus, and superior parietal cortex. Collectively, our results highlight how local biological attributes and global connectivity jointly shape cross-disorder cortical abnormalities.

Challenges in verbal communication are a prominent feature of autism. However, gene regulatory programs in speech-related cortical regions remain poorly characterized. In parallel, it remains unclear whether the heterogeneous genetic factors underlying autism converge on shared neurobiological mechanisms. To address these gaps, we generated paired transcriptomic and epigenomic data from post-mortem human brain tissue across 100 donors. Here, we show that transcriptional differences in the speech-related Brodmann Area 22 in individuals with neurodevelopmental conditions, including autism, are strongest among those with a known genetic diagnosis. A similar but attenuated signature is observed in those without a genetic diagnosis. These transcriptional differences are most pronounced in neurons, with glutamatergic L4/5 intratelencephalic neurons affected across multiple modalities. Finally, multimodal analysis implicates altered RFX3-dependent networks as a central hub in autism, particularly among L4/5 intratelencephalic neurons in non-verbal individuals. Together, our study identifies regulatory architecture linking chromatin state, transcriptional output, and variation in verbal ability in autism.

Neural synchrony, or correlated neural activity across interacting individuals, scales with relationship quality in humans, yet how it evolves during social bond formation remains unknown. Using fiber photometry in monogamous prairie voles, we track prefrontal cortex synchrony across pair bond formation. Bonded voles show stronger synchrony with partners than strangers, mirroring human findings. A linear mixed model reveals that synchrony is jointly shaped by bond strength, inter-animal distance, and time since interaction onset, with relationship type modulating how each factor contributes. Using a machine-learning behavioral classification pipeline we developed for freely interacting voles, we demonstrate that the coupling between specific behaviors and synchrony depends on the nature of the dyadic relationship. These findings establish that neural synchrony is not a simple function of proximity or interaction time but is fundamentally shaped by relationship history--a conclusion with direct implications for understanding the synchrony in human social attachment.

1.Schizophrenia spectrum disorders (SSDs) present a profound clinical enigma, manifesting as a heterogeneous continuum ranging from the chaotic volatility of acute psychosis to the impenetrable rigidity of systematized delusions. While neurobiological research has independently implicated NMDA receptor hypofunction or dopaminergic hyperfunction as cardinal pathophysiological distinct mechanisms, a computational framework capable of bridging these distinct cellular deficits to the spectrums vast phenomenological diversity remains elusive. Here, we propose a biologically plausible neural model using a dynamic Bayesian inference with separable positive and negative prediction-error pathways. We demonstrate that NMDA hypofunction selectively blunts negative prediction errors, fostering rigid, bias-dominated beliefs, while dopaminergic hyperfunction uniformly amplifies error signals, driving volatile, observation-dominated states. Their interaction reconstructs SSDs as a continuous bias-volatility spectrum, accounting for key neurophysiological markers and offering a theoretical foundation for mechanism-based patient stratification.

The thalamus is a key forebrain structure that gates peripheral, subcortical, and cortico-cortical communication1,2. Awake thalamic bursts provide the cortex with a "wake-up" signal2-4. Paradoxically, thalamic neurons discharge tonically during cellular depolarization and activated brain states (wakefulness, REM sleep) but burst during hyperpolarization and NREM sleep5-9. It has been proposed that NREM thalamic bursts do not awaken the cortex because of their periodic and synchronized nature2-4; however, this has never been tested in vivo during natural sleep. We simultaneously recorded polysomnographic signals, local field potentials, and spiking activity from multiple thalamic neurons in the ventral anterior and centromedian nuclei of two female non-human primates during naturally occurring vigilance states. These nuclei receive GABAergic output from the basal ganglia10,11, with discharge rate and GABA outflow decreasing during NREM sleep12. We found that despite the expected thalamic depolarization, bursting increased significantly. NREM bursts were neither periodic nor highly synchronized. However, EEG activity time-locked to burst onset during NREM sleep differed markedly from that observed during wakefulness and REM sleep. These results support a modulatory, rather than a driving, relationship between the basal ganglia and thalamus. NREM thalamic bursts do not awaken the cortex, probably due to unique state-dependent thalamocortical dynamics.

Sporadic amyotrophic lateral sclerosis (sALS) lacks longitudinal molecular measurements, making it difficult to distinguish early disease-organising changes from downstream consequences. We present a training-free framework that extracts directional structure from static single-nucleus RNA-seq by applying discrete Hodge decomposition to gene co-expression dynamics across pseudotime-ordered donor states. The framework separates irreversible co-expression cascades from circular feedback structure and regresses out the component explained by the healthy co-expression network, allowing disease-specific organisation to be examined in isolation. Perturbation benchmarks show that experimentally imposed sources are recoverable from control-normalised off-diagonal covariance-response fields, whereas marginal variance and diagonal covariance controls do not recover the source. Applied to sALS primary motor cortex (24 donors, 10 cell types), the framework identifies oligodendrocytes as the most structurally upstream cell type and upper-motor-neuron-containing layers as the most structurally downstream (Oligo cell-type{varphi} = 0.900, with glial cell types preserving the healthy co-expression network topology, whereas neuronal cell types show collapse-dominant deformation). Cytoplasmic translation is the only pathway with reproducible cross-cell-type upstream enrichment. At the gene level, the ribosome-associated quality-control factor NEMF -- which appends C-terminal alanine-threonine tags ("CATylation") to nascent chains on stalled ribosomes -- shows disease-specific loss of co-expression coherence in seven of ten cell types despite essentially unchanged mRNA expression; the disease signal is decoupling from collision-response partners (GCN2, PKR), not expression-level change. Cross-cohort validation across three BA4 motor cortex cohorts (including two external cohorts; total N=107) reproduced the oligodendrocyte-upstream / upper-motor-neuron-downstream structural architecture (Oligo-preserved / ET-sink) in all three cohorts, with NEMF co-expression coherence loss replicated in two of three cohorts. These data support a brain-side, circuit-distal structural model in which oligodendrocyte-lineage stress occupies an upstream-like preserved compartment, while upper-motor-neuron-containing excitatory populations form a downstream sink. The pattern is consistent with -- but does not directly establish -- a cascade architecture in which oligodendrocyte stress structurally precedes motor neuron TDP-43 pathology, and would produce a clinical phenotype resembling dying-back (the conventional view of ALS, in which motor neuron pathology appears to begin at distal axons and spread retrogradely toward the cell body) yet originating centrally and glially. NEMF/CATylation network disruption is identified as a candidate intermediate structural node bridging oligodendrocyte stress and motor neuron TDP-43 pathology.

Adaptive learning requires distinguishing environmental volatility from observation stochasticity, two sources of uncertainty that demand opposite adjustments to the learning rate but inflate experienced variance similarly. Disentangling them is computationally difficult with no tractable closed-form solution. Particle-filter methods are the natural tool for this kind of joint inference, but their stochastic likelihoods and non-differentiable objectives force derivative-free fitting protocols and discourage the individual-difference analyses central to cognitive modeling, where small effect sizes leave little room for additional estimator noise. We introduce the Categorical Bayes Filter (CBF), a deterministic alternative that preserves the conditional structure of recent particle-filter accounts but replaces the stochastic outer layer with a categorical distribution on a quantile grid parameterized through differentiable Beta quantile functions. The procedure performs evidence maximization with an exact, deterministic marginal likelihood that is fully differentiable in the grid parameters. In a volatility-stochasticity task with N = 643 participants, fitted CBF dispersion parameters reveal a cross-over phenotyping pattern between volatility-blind and stochasticity-blind subjects that is not recoverable from particle-filter parameters fit to the same data under a state-of-the-art protocol. The deterministic structure also yields a trial-by-trial ambiguity signal that predicts response times not used in fitting. More broadly, the approach opens individual-level analyses in cognitive modeling and computational psychiatry that stochastic methods have effectively foreclosed.

Neurons interact at synapses, but they also communicate through physical contact and proximity, including diffusion, glia-mediated interactions, and ephaptic coupling. Standard connectomes map synapses, but cannot capture the full set of cell-cell contacts that can support these pathways. Here we extract contactomes from two large mouse visual cortex volumes at nanoscale resolution and quantify every cell-cell contact, the shared surface area of each contact, and the relationship between contact and synaptic connectivity. We find that contactomes are 5 - 10x denser than synaptic graphs, revealing that neurons physically contact a much larger set of potential neighbors than they synaptically connect to. We further find that most nearby potential neighbors are already in physical contact, indicating that local structural change would add few new candidate synaptic partners. Finally, we find that astrocytes form a single large syncytium-like network that spans the tissue and directly contacts nearly all neurons, and that glial processes lie within a micron or two of almost every synapse, indicating that synapses reside within a pervasive glia-shaped microenvironment. Together, these results show that physical contact forms a distinct layer of brain architecture that extends far beyond the synaptic connectome.

Dynamic interactions of neurons and glia in the ventral midbrain (VM) mediate reward and addiction behavior. We studied gene expression in 212,713 VM single nuclei from 95 human opioid overdose cases and drug-free controls. Chronic exposure to opioids left numerical proportions of VM glial and neuronal subtypes unaltered, while broadly affecting glial transcriptomes, involving 9.5 - 6.2% of expressed genes within microglia, oligodendrocytes, and astrocytes, with prominent activation of the immune response including interferon, NFkB signaling, and cell motility pathways, sharply contrasting with down-regulated expression of synaptic signaling and plasticity genes in VM non-dopaminergic neurons. VM transcriptomic reprogramming in the context of opioid exposure and overdose included 325 genes with genetic variation linked to substance use traits in the broader population, thereby pointing to heritable risk architectures in the genomic organization of the brains reward circuitry.

Working memory (WM) enables temporary retention of information essential for flexible cognition. Although persistent population activity has long been regarded as a principal mechanism of memory maintenance, continuous single-neuron firing is energetically demanding and difficult to reconcile with the heterogeneous firing properties of cortical neurons. Applying single-trial analyses to a dataset of 902 neurons recorded from 21 neurosurgical patients performing a WM task, we found that maintenance was supported by transient, burst-like episodes of coordinated activity rather than sustained firing. Cross-temporal decoding exhibited localized generalization, and decoding accuracy increased with wider temporal windows, indicating that apparent persistence can emerge from temporally interleaved activity across neurons. We further developed a feature-based, putative cell-type classifier that revealed distinct circuit contributions: pyramidal neurons expressed content in burst-aligned events during maintenance, whereas interneurons were strongly modulated by memory load and behavior. Together, these findings reconcile dynamic and persistent accounts, indicating that human WM can emerge from temporally interleaved, cell-type-specific dynamics that provide a flexible and potentially metabolically efficient substrate for maintaining information over time.

Patterns of neural activity underlie human cognition. Transitions between these patterns are orchestrated by the brains network architecture. What are the mechanisms linking network structure to cognitively relevant activation patterns? Here we implement principles of network control to investigate how the architecture of the human connectome shapes transitions between 123 experimentally defined cognitive activation maps (cognitive topographies) from the NeuroSynth meta-analytic engine. We also systematically incorporate neurotransmitter receptor density maps (18 receptors and transporters) and disease-related cortical abnormality maps (11 neurodegenerative, psychiatric and neurodevelopmental diseases; N = 17 000 patients, N = 22 000 controls). Integrating large-scale multimodal neuroimaging data from functional MRI, diffusion tractography, cortical morphometry, and positron emission tomography, we simulate how anatomically-guided transitions between cognitive states can be reshaped by pharmacological or pathological perturbation. Our results provide a comprehensive look-up table charting how brain network organisation and chemoarchitecture interact to manifest different cognitive topographies. This computational framework establishes a principled foundation for systematically identifying novel ways to promote selective transitions between desired cognitive topographies.

The ability to retrieve a single episode encountered just once is a hallmark of human intelligence and episodic memory[1]. Yet, decoding a specific memory from neuronal activity in the human brain remains a formidable challenge. Here, we develop a transformer neural network model[2, 3] trained on neuronal spikes from intracranial microelectrodes recorded during a single viewing of an audiovisual episode. Combining spikes throughout the brain via cross-channel attention[4], capable of discovering neural patterns spread across brain regions and timescales, individual participant models predict memory retrieval of specific concepts such as persons or places. Brain regions differentially contribute to memory decoding before and after sleep. Models trained using only medial temporal lobe (MTL) spikes significantly decode concepts before but not after sleep, while models trained using only frontal cortex (FC) spikes decode concepts after but not before sleep. These findings suggest a system-wide distribution of information across neural populations that transforms over wake/sleep cycles[5]. Such decoding of internally generated memories suggests a path towards brain-computer interfaces to treat episodic memory disorders through enhancement or muting of specific memories.