Research papers

In this review, we draw insight from the fields of neurobiology and computational neuroscience to address a fundamental question: Why does the correlation between structural and functional connectivity vary across the human cortex? We begin by summarizing empirical studies that reveal the heterogeneous expression of structure-function coupling across brain regions and among individuals. We then identify potential biological factors that mediate this variability, focusing on the roles of evolution, myeloarchitecture, cytoarchitecture, and neuromodulation in sculpting the dynamic and diverse structure-function landscape of the human cortex. We next turn to computational modeling to deepen our understanding of the relationship between a system's structural architecture and functional expression. We investigate biologically inspired computational models that map structure to function in human brain networks, paying special attention to studies that simulate external perturbations and structural lesions, and discuss the insights these approaches offer into the causal mechanisms governing the heterogeneous interplay between structural and functional connectivity. We close with a discussion of future directions, emphasizing efforts to bridge neurobiology and computational modeling to design biologically accurate, individualized models of the human brain. In particular, we highlight the potential of multi-layered networks informed by individual-specific microstructural and neuromodulatory gradients and governed by non-linear dynamics as a particularly fruitful direction. Such personalized models accounting for the synergistic effects of biological gradients could be experimentally validated to assess their predictive efficacy, ultimately bringing us one step closer to non-invasive, connectome-based clinical treatments tailored to the individual.

Flight maneuvers in the fruit fly Drosophila have long served as a model for studying principles underlying visual information processing. Advances in genetic targeting of individual types of neurons for manipulation and recording, as well as the publication of the complete connectome, have greatly expanded our knowledge of how behavior is controlled by the fly's nervous system. In this review, I summarize recent findings on how visual information relevant to flight is transformed into a behavioral output, ranging from fast stabilizing reflex-like responses to longer-lasting goal-directed behaviors. I argue that flexibility in the processing of visual information and a hierarchical recruitment of different behavioral modules enable the control of this complex behavior with a comparatively small number of neurons.

Recent interest in the neurobiology of the C. elegans male has driven exciting progress in understanding sexual dimorphism in the nervous system of this important model system. From these studies, a framework emerges in which the effects of biological sex fall into three overlapping categories. First, sex modulates developmental programs of cell lineage, neurogenesis, and cell fate specification to generate neuronal classes and structures that appear only in one sex. Second, sex modifies synaptic connectivity to alter the topology of neural circuits. Third, sex acts as a dimension of internal state, developmentally and dynamically modulating the physiology of sex-shared neurons and circuits. Together, these effects program sex-specific behaviors, tune sex-shared behaviors, and implement sexually dimorphic behavioral plasticity. Importantly, C. elegans offers the opportunity to understand these effects with uniquely high mechanistic resolution. Although mechanisms of sex determination differ markedly by species, these principles are likely conserved across evolution.

Understanding neuronal connectivity at single-cell resolution remains a fundamental challenge in neuroscience, with current methods particularly limited in mapping long-distance circuits and preserving cell type information. Here we present Connectome-seq, a high-throughput method that combines engineered synaptic proteins, RNA barcoding and parallel single-nucleus and single-synaptosome sequencing to map neuronal connectivity at single-synapse resolution. This adeno-associated virus-based approach enables simultaneous capture of both synaptic connections and molecular identities of connected neurons. We validated this approach in the mouse pontocerebellar circuit, identifying both established and potentially uncharacterized synaptic connections. Through integrated analysis of connectivity and gene expression, we identified molecular markers enriched in connected neurons, suggesting potential molecular determinants of circuit-specific connectivity. By enabling systematic mapping of neuronal connectivity across brain regions with single-cell precision and gene expression information, Connectome-seq provides a scalable platform for comprehensive circuit analysis across different experimental conditions and biological states.

Abstract Connectomic reconstruction from large image volumes produces segmentation and synaptic-assignment errors that must be resolved to support downstream analyses. As datasets have grown larger and teams more distributed, proofreading has become a critical operational bottleneck. Workflows for proofreading and error correction have not scaled commensurately with connectomic data production and may not accommodate heterogeneous proofreader expertise and machine-generated candidate edits. New tools are therefore needed to organize, prioritize, and coordinate proofreading at volume scale. Here we present NeuVue, a task-management and prioritization framework that operationalizes proofreading through atomic, auditable tasks for individual and team review, multistage routing across proofreader cohorts, performance and volume-state tracking, and integration with community annotation, visualization, and analysis services. We report the use of NeuVue across two volumetric datasets, supporting scalable proofreading by over forty proofreaders and producing over fifty thousand edits. NeuVue provides a reproducible human-in-the-loop framework for generating, validating, and maintaining large connectomic datasets.

Volume electron microscopy (EM) now enables nanometric-scale 3D reconstructions of neural tissue, opening the door to quantitative, morphology-driven neuroscience beyond connectivity alone. While previous studies relied on handcrafted descriptors and classical machine learning for morphology analysis, recent progress in deep learning for 3D shape understanding offers new opportunities to learn robust, task-specific representations directly from geometric data. In this paper we present NeMoCo , a geometry learning framework that targets the key practical bottleneck in connectomics and ultrastructural analysis: the scarcity and cost of dense expert annotations for the long tail of neurite and organelle phenotypes. NeMoCo formulates representation learning for EM-derived neurite meshes in a self-supervised Momentum Contrast (MoCo) style. We use DiffusionNet (Sharp et al., 2022) as a mesh encoder with intrinsic spectral descriptors (HKS) and train with a momentum-updated teacher encoder and a large memory bank of negatives. To learn invariances that are essential in practice, we generate paired geometric views via controlled affine transformations and resolution changes (including mesh decimation), encouraging embeddings to be stable under nuisance variability while remaining discriminative. We provide an extensive study of augmentation strength and temperature, and evaluate learned representations through frozen retrieval and non-parametric classification (frozen kNN), as well as downstream supervised fine-tuning under limited labels. NeMoCo demonstrates that MoCo-style self-supervision yields robust neurite morphology embeddings on EM meshes, improving label-efficiency and offering a scalable foundation for retrieval, clustering, and phenotype discovery in ultrastructural neuroscience. All the data and the code used for NeMoCo are available at https://github.com/Uzshah/NeMoCo .

Viral projection tracing strategies help build regional connectomes of mammalian brains. G-deleted rabies virus (ΔG-RV) establishes monosynaptic input connectivity but cannot distinguish networks at cell resolution. Here, we implement a barcoded ΔG-RV for network tracing by quantitative RNA-sequencing. At optimized library complexity and uniformity, barcode detection reliably distinguished individual monosynaptic input networks of multiple infected neurons in parallel. To scale this approach to hundreds of cells and full-brain inputs, we develop regions-of-interest network sequencing (ROInet-seq)-an accessible, scalable, and low-cost spatial assay. ROInet-seq combines routine fluorescent imaging of fixed tissue sections with targeted barcode sequencing, enabling brain-wide mapping of single-neuron networks. In the somatosensory cortex, the assay reveals dominant regional contributions that inputs from a handful of thalamic neurons diverged to multiple neurons and frequent convergent ipsilateral and contralateral cortical inputs. Integration with commercial spatial transcriptomics assays improves resolution, together establishing a scalable framework to resolve single-cell network architectures.

Abstract The continued expansion in size and resolution of volumetric datasets generated by electron microscopy (3DEM) is making high-performance computing (HPC) an essential component. HPC provides the scalable, parallelized infrastructure required to process and analyze these increasingly large datasets. The Texas Advanced Computing Center’s (TACC) Core Experience Portal (CEP) is a science gateway specialized for leveraging HPC. The CEP can be adapted into customized versions based on a research community’s needs. Here we have constructed 3dem.org as a gateway for the community exploring volumetric data generated by electron microscopy (3DEM). This gateway provides the 3DEM community with user-friendly browser-based access to raw and processed datasets including both private and shared data, image processing tools for alignment and segmentation, and simulation and analysis environments. All this is linked to the underlying HPC environment at TACC with the ability to connect to other data storage and compute systems. 3dem.org bridges advanced electron microscopy with HPC, providing the research community with scalable, accessible infrastructure for discovery.

ABSTRACT Volumetric electron microscopy (EM) has become a critical approach to generating high-resolution reconstructions of brain tissue. As the size of EM volumes increase, use of automated image segmentation within the reconstruction pipeline has become essential, although it generates errors that need correction. The proofreading and correcting of these errors has since become the dominant cost driver in the pipeline, but precisely estimating the sufficient number of proofreading edits to enable meaningful scientific analyses of the reconstructed neuronal networks remains a challenge. We present a fast, computationally inexpensive way to estimate the progress of a connectomic proofreading effort without requiring a priori knowledge of ground truth. We show that simple global graph invariants converge predictably to asymptotic limits with increasing numbers of proofreading edits, informing a quantitative “pencils down” criterion for proofreading completeness. We illustrate our method on two datasets in different stages of proofreading progress, a zebrafish spinal cord and the hemibrain Drosophila melanogaster dataset. Our method reduces the uncertainty associated with the planning and prioritization of proofreading activities and enables data owners to accurately predict and budget the amount of proofreading necessary for their scientific questions.

Large-volume serial electron microscopy (vEM) has transformed neuroscience by enabling reconstructions of neural circuits at synaptic resolution. Approaches to collecting libraries of ultrathin sections required for producing vEM datasets have been devised (e.g., automated tape-collecting ultramicrotome [ATUM], MagC, and GAUSS-EM), which has advanced the scalability of vEM in distinct ways, yet widespread adoption remains constrained by cost, hardware requirements, or infrastructure demands. Here, we introduce direct retrieval by ionizer-facilitated transfer for electron microscopy (DRIFT-EM), a low-cost and easily implemented platform designed to broaden access to direct wafer-based collection of ultrathin sections. Off-the-shelf static ionizers clear newly cut sections from the knife edge, a thermoplastic boundary confines sections, and an open-source software pipeline automates region of interest identification to guide imaging. DRIFT-EM achieves section packing densities comparable to established magnetic methods, with minimal setup cost and maintenance. Its modular style, 3D-printed components, and open design facilitate integration with other vEM workflows, helping lower barriers to creating connectomes.

Cellular function depends on the precise deployment and distribution of nanoscale structures, but these features remain difficult to measure and compare between cells and datasets. This challenge is pronounced for astrocytes, whose intricate nanostructures interface with neurons, glia, and vasculature, to control brain development, synaptic development/plasticity, homeostasis, and responses to injury/disease. Here, we developed deep learning approaches with curvature consistency for automated astrocyte segmentation across volume electron microscopy datasets, reducing reconstruction time from manual or semi-automatic methods by 12-fold and enabling brain region and cross-species interrogation of astrocytic nanoarchitecture. This allowed us to uncover organizing principles and motifs alongside ultrastructural divergence between species. While both species exhibit a wide but shallow topological network, marmoset astrocytes display increased process thickness and branching. We further identified extrasynaptic neuronal engulfment and a robust astrocytic endosomal system across species and brain regions. Together, our findings demonstrate previously inaccessible structural principles of astrocytes, offering a framework for understanding structure-function relationships in the central nervous system.

Vision foundation models have substantially advanced computer vision, enabling state-of-the-art performance in zero- and few-shot settings. They have been successfully applied to biomedical imaging tasks ranging from organ segmentation in computed tomography to cell segmentation in light microscopy. Electron microscopy (EM) is a central modality for analyzing cellular ultrastructure due to its nanometer-scale resolution. However, the application of foundation models in EM has so far been limited to specific organelles, such as mitochondria, largely due to the diversity of segmentation tasks and the scarcity of comprehensively annotated data. As a result, EM segmentation still predominantly relies on supervised learning, requiring extensive manual annotation and limiting ultrastructural analysis. To address this gap, we propose $μ$Match, a framework for semi-supervised learning and domain adaptation that leverages foundation models. We implement state-of-the-art student-teacher-based methods and evaluate multiple foundation models (SAM, SAM2, $μ$SAM, DINOv2/v3) on challenging EM tasks, including mitochondrion, nucleus, and neurite segmentation. Our results demonstrate consistent improvements over strong baselines and highlight a path toward substantially reducing the annotation effort in EM.

Abstract Understanding the neuromuscular basis for behavior requires measuring both kinematic and physiological data at high resolution in unconstrained conditions: a technically challenging goal. Here we present an integrated experimental-computational pipeline for measuring and quantifying body part kinematics and muscle activity in freely behaving Drosophila melanogaster . We first present Spotlight , a closed-loop videography system that performs real-time tracking to record untethered flies at high resolution (6 µm/pixel) and high frame rate (330 Hz) while also enabling optical recordings of limb muscle activity via a fluorescent calcium reporter. To analyze these massive datasets without manual image annotation, we introduce PoseForge , a synthetic-data-driven framework that exploits morphologically accurate biomechanical simulations to generate synthetic data, and contrastive self-supervised learning to infer 3D keypoints and dense body-part segmentation from a single camera view. Using resulting 3D kinematic data, we can replay recorded behaviors in a biomechanical digital twin, NeuroMechFly, to infer forces generated and experienced by the fly’s limbs. Finally, we illustrate the capability of our system to optically record muscle activity. We show how the legs’ long-tendon muscles activate upon mechanical vibration, possibly to activate gripping and to maintain a stable posture. Taken together, this workflow enables scalable, high-resolution measurement and modeling of unconstrained, natural behavior.

Three-dimensional (3D) alignment is a key step in volume electron microscopy (vEM), aimed at addressing misalignment during data acquisition, thereby recovering the correct biological structures. However, automated 3D alignment has long been challenged by the dilemma between eliminating nonlinear distortions and capturing natural morphological variations inherent to biological specimens. Here, we present vEMRec, a paradigm-shifting, fully automated algorithm for vEM 3D alignment. vEMRec redefines the 3D alignment problem by decoupling it into high-frequency and low-frequency subproblems. In this framework, precision rigid alignment is applied to correct rigid distortions, while a Gaussian filter-driven elastic registration algorithm addresses nonlinear distortions, all the while faithfully preserving biologically plausible deformations. Extensive experiments demonstrate that vEMRec achieves a paradigm shift in 3D alignment. Serving as a critical preprocessing step, vEMRec enhances performance in downstream isotropic reconstruction and 3D segmentation tasks by improving axial continuity in anisotropic data while preserving the structural integrity of ultrastructural details. Moreover, vEMRec accomplishes this through efficient computation, enabling TB-scale specimen analysis with biologically relevant throughput. vEMRec successfully optimized six representative large-scale real-world datasets, demonstrating its applicability, accuracy, and robustness for large-scale data processing.

Neural cell types are often clustered into inhibitory, excitatory, or neuromodulatory populations. However, the signaling mechanism between two neurons ultimately depends on their neurotransmitter-receptor pairings. Here, we demonstrate an example of a glutamatergic population of neurons that appears to either excite or inhibit their downstream partners depending on the type of glutamate receptor expressed in the downstream cell type. The upstream population encodes a sinusoidal head direction signal. The downstream partners that are excited by glutamate encode the same signal while the downstream partners that are inhibited by glutamate encode a 180° phase-shifted copy. The upstream population therefore appears to pass on two different phases of the same signal by making either excitatory or inhibitory connections to different downstream partners.

Climbing fibre (CF) inputs to Purkinje cells (PCs) instruct plasticity and learning in the cerebellum1–3. Paradoxically, CFs also excite molecular layer interneurons (MLIs)4,5, a cell type that inhibits PCs and can restrict plasticity and learning6,7. However, two types of MLI with opposing influences have recently been identified: MLI1s inhibit PCs, reduce dendritic calcium signals and suppress plasticity of granule cell to PC synapses2,6–9, whereas MLI2s inhibit MLI1s and disinhibit PCs8. To determine how CFs can activate MLIs without also suppressing the PC calcium signals necessary for plasticity and learning, we investigated the specificity of CF inputs onto MLIs. Serial electron microscopy reconstructions indicate that CFs contact both MLI subtypes without making conventional synapses, but more CFs contact each MLI2 through more sites with larger contact areas. Slice experiments indicate that CFs preferentially excite MLI2s through glutamate spillover4,5. In agreement with these anatomical and slice experiments, in vivo Neuropixels recordings show that spontaneous CF activity excites MLI2s, inhibits MLI1s and disinhibits PCs. By contrast, learning-related sensory stimulation produces more complex responses, driving convergent CF and granule cell inputs that could either activate or suppress MLI1s. This balance was robustly shifted towards MLI1 suppression when CFs were synchronously active, in turn elevating the PC dendritic calcium signals necessary for long-term depression. These data provide mechanistic insight into why CF synchrony can be highly effective at inducing cerebellar learning2,3 by revealing a critical disinhibitory circuit that allows CFs to act through MLIs to enhance PC dendritic calcium signals necessary for plasticity. Synchronous activation of climbing fibres engages disinhibitory circuitry to promote large dendritic calcium signals in Purkinje cells that are necessary to promote cerebellar learning.

Neural networks maintain stable activity levels by compensating for perturbations through homeostatic plasticity. However, homeostatic mechanisms operating at different levels may conflict with each other. For example, in inhibitory feedback circuits, if inhibitory neurons receive excess excitation, compensation at a 'local' level (e.g. reducing inhibitory neurons' activity) could conflict with 'network-level' compensation (e.g. suppressing the excitatory neurons responsible for overexciting the inhibitory neurons). We studied this problem in the Drosophila mushroom body, where excitatory Kenyon cells (KCs) receive feedback inhibition from the anterior paired lateral (APL) neuron. Dual-colour calcium imaging revealed that prolonged (24 h) artificial activation of KCs causes APL to become less sensitive to KC activity. Meanwhile, KCs compensate for their excess activity by reducing excitation, yet this change is opposed by reduced inhibition from APL. This conflict meant that KCs did not consistently show the expected homeostatic reduction in odour responses. Our findings show that neurons sometimes adapt their activity locally in a way that counteracts broader adaptations in the network. KEY POINTS: Neural networks maintain stable activity levels through homeostatic plasticity - but what physiological variables are stabilised? In inhibitory feedback circuits, local and network-level compensation might conflict. For example, if excitatory neurons are overactive, they might compensate by becoming less excitable. But if inhibitory neurons compensate for the excess excitation by also becoming less excitable, this would decrease inhibition onto the excitatory neurons and increase their activity. We tested this idea in the fruit fly brain, where excitatory Kenyon cells (KCs) get negative feedback from an inhibitory neuron called anterior paired lateral (APL). After overactivation of KCs, APL becomes less sensitive to KCs. The resulting loss of inhibition onto KCs counteracts KCs' attempts to reduce their activity. These results show that adaptation at local and network levels can conflict with each other.

The functional architecture of the primary visual cortex (V1)-manifesting as orientation selectivity maps (OS maps) in higher mammals and modular tuning clusters in rodents-provides a window into the rules governing cortical circuit organization. Although OS maps emerge from intrinsic activity and the structured sampling of retinothalamic inputs, their maturation and lifelong adaptability depend on cellular mechanisms that extend far beyond excitatory wiring. Recent advances indicate that inhibitory interneurons and glial cells play critical modulatory roles in map assembly, stabilization, and experience-dependent plasticity. Inhibitory interneurons regulate excitatory-inhibitory (E/I) balance, stabilize population responses, and gate developmental and adult plasticity through coordinated local and long-range interactions. In parallel, astrocytes modulate circuit excitability and experience-dependent refinement by regulating synaptic signaling, metabolic and ionic homeostasis, and plasticity-related receptor function, whereas microglia influence map formation and maintenance indirectly through activity-dependent synaptic remodeling and network homeostasis. This minireview synthesizes emerging insights into how inhibitory networks and glial-neuron crosstalk jointly orchestrate the formation and experience-dependent remodeling of orientation maps, offering a cellular framework for understanding the construction and flexibility of cortical sensory representations.

Abstract Neural circuit function depends not only on synaptic connectivity but also on the molecular composition and subcellular organization of neurotransmitter receptors. Here, we examine the expression, localization, and functional relevance of nicotinic acetylcholine receptors (nAChRs), the primary mediators of fast excitatory transmission in the Drosophila central nervous system (CNS). Functional nAChRs are pentamers assembled from 10 subunits (α1–α7 and β1–β3), yet how this diversity is deployed within defined circuits remains poorly understood. Using T2A-Gal4 reporters and endogenous protein tagging, we identify eight nAChR subunits (α1–α3, α5–α7, β1, and β2) expressed in larval motor neurons (MNs). MN-specific knockdown of individual subunits produces impairments in crawling, peristaltic timing, and protopodium dynamics, demonstrating that multiple nAChR subtypes contribute to motor output. Colocalization analyses reveal a wide range of spatial relationships, identifying subunit pairs with high, intermediate, and low overlap within MN dendritic and postsynaptic domains. Across subunit combinations, spatial organization correlates with pair-specific functional interactions: spatially segregated pairs tend to produce stronger locomotor defects when knocked down together, suggesting largely nonredundant contributions to cholinergic excitation of MNs, whereas highly colocalized pairs often show limited additional impairment. Notably, some colocalized pairs also exhibit additive effects, indicating that spatial proximity alone does not fully predict functional interaction. Dual knockdown of selected subunit pairs also reduces muscle contraction amplitude, linking receptor organization to motor output at the effector level. Together, these results indicate that MNs deploy multiple nAChR populations whose spatial arrangement shapes how cholinergic inputs contribute to locomotor output.

The process by which neocortical neurons and circuits amplify their response to an unexpected change in stimulus, typically referred to as deviance detection (DD), has traditionally been thought to be the product of specialized cell types and/or routing from distinct brain areas. Here, we explore a different theory, whereby DD emerges intrinsically from local network-level interactions driven by a deviant increase or decrease in exogenous input to a neocortical column. We propose that deviance-driven neural dynamics are generated by ensembles of excitatory and inhibitory neurons that have a fundamental inhibitory connectivity motif: competitive inhibition between reciprocally connected neural representations under modulation from feed-forward selective (dis)inhibition. Implementing this motif in two computational models with different levels of biophysical abstraction, we were able to simulate a variety of phenomena pertaining to the experimentally observed shifts in neural tuning during DD across neurons, time, and stimulus history. We further tested hypotheses related to our theory and examined the robustness of emergent phenomena consistent with prior experimental observations. Our results show that ensemble priming via competitive inhibition under modulation from selective (dis)inhibition can serve as a local mechanism for encoding short-term stimulus memory, enabling deviance-driven shifts in stimulus representation. This work establishes a theoretical paradigm that resolves previously confounding aspects of predictive sensory processing in Neocortex, and we provide a number of corollary predictions that can be tested in future in vivo studies.