2026-10-28

Understanding how distributed neural circuits generate complex behaviors requires long-duration recordings from populations of individual neurons across multiple brain regions. High-density Neuropixels probes have revolutionized rodent systems neuroscience. Extending these capabilities to non-human primates (NHPs) for chronic, multi-regional recordings requires compact, implantable microdrive hardware and reliable automated spike-sorting software. NHPs provide access to complex and flexible behaviors that are difficult to study in other model systems. Extending the Neuropixels revolution to NHPs therefore depends on integrating hardware and software for long-duration, multi-regional recordings. Here, we present a scalable hardware-software platform co-designed for high-density neural data acquisition and reliable single-unit identification in NHPs. The Multi-regional Implantable NeuroDrive (MIND) is a lightweight, implantable microdrive that provides stable recordings from multiple independently positioned Neuropixels probes and is designed to support chronic, ambulatory, head-free recordings. We paired MIND with UnitAssay, an automated post-sorting curation framework that identifies and curates characteristic spike-sorting errors with minimal computational cost. Together, MIND and UnitAssay provide an integrated platform for implanted, multi-regional recordings in NHPs performing complex and naturalistic behaviors. We tested MIND and UnitAssay across parietal, prefrontal, and motor cortices in rhesus macaques performing reaching and eye-movement tasks. MIND enabled stable multi-probe recordings across cortical regions and durations. UnitAssay enabled automated curation of large-scale spike-sorting output and improved reliable single-unit identification relative to Kilosort 4 labels and standard SpikeInterface quality control. Together, MIND and UnitAssay provide a scalable solution for high-density, multi-regional, long-duration electrophysiology in NHPs.

Focal neuronal loss after stroke, traumatic brain injury or spinal cord injury can disrupt circuits and result in persistent motor deficits. Which cells and connections must be restored to recover a specific behavior is difficult to establish experimentally. Here we use a connectome-constrained spiking model of the fruit-fly nervous system and decode motor-neuron output to simulate visually guided steering. After loss of 4,000 neurons, path straightness ranged from 0.04 to 0.95 depending on lesion location. At a fixed sensorimotor site, a 2,000-neuron lesion reduced simulated displacement to 6.5% of intact. Across 100 stochastic simulation realizations of the connectome, replacing 100 cells with their original connections increased displacement to 62.5% of intact when motor-control pathways were prioritized, compared with 10.1% after random selection of 100 cells and 48.5% after random selection of 1,000 cells. When connections were instead inferred from cell type, the same 100 intended replacements reached 60.0% of intact with matched identities and 6.7% with non-target identities, indicating that cell identity and type-dependent integration are required for recovery. Our model generates testable predictions for genetic ablation experiments in fruit flies and identifies candidate cellular and circuit requirements for motor repair that can inform cell replacement therapies and future models of mammalian locomotion.

Neurobiological circuits may be energy efficient in part because there is a division of labor: different subsystems compute or represent different things. In light of this, it is interesting that a major difference between subsystems is the time scale on which they typically vary. Here, we argue that temporal smoothness constraints---the idea that it may be energetically costly for systems to vary much more quickly than their typical time scale---imply a certain division of labor with respect to temporal stimulus features. In particular, slow subsystems ought to represent slowly varying stimulus features, and fast subsystems ought to represent quickly varying features. We formulate a novel efficient coding model that we use to investigate this claim, and exactly solve for the optimal division of labor. In addition to finding that slower subsystems ought to represent more slowly varying temporal features, we find surprising structure of mathematical interest: optimal codes have subsystems which represent orthogonal temporal features, and which correspond to the eigenfunctions of a Sturm-Liouville problem.

Functional near-infrared spectroscopy (fNIRS) offers a promising approach for monitoring the hemodynamic correlates of neural activity in naturalistic settings. However, existing modeling methods for fNIRS are largely confined to small-scale, single-session data with low sampling rates, driven by high heterogeneity in acquisition devices and optode montages. In addition, the complex spatiotemporal structure of fNIRS signals further restricts model scalability and transferability across settings. As a result, it remains unclear whether a single fNIRS encoder can learn representations that are usable across subjects, tasks, devices, and montages. Here, we show that a montage-aware masked autoencoder, pretrained on 12 public fNIRS datasets spanning 268 subjects and 567 recording hours, learns generalizable neurovascular representations that transfer across three external datasets. Our fNIRS foundation model, MtReNIRS, jointly encodes the slow hemodynamic nature of fNIRS, chromophore identity and cortical location by mapping heterogeneous montages onto a shared anatomical reference. MtReNIRS achieves cross-subject performance on par or better than fully supervised baselines and excels in few-shot within-subject calibration, where it outperforms prior state-of-the-art fNIRS models, while remaining robust to sparse montages. Our results indicate that montage-aware self-supervision yields physiologically grounded, transferable fNIRS representations, providing a scalable foundation for label-efficient brain-computer interfaces.

The Buice-Cowan field theory [PRE 75, 051919 (2007)] describes fluctuations and correlations of neural activity through a path integral whose mean-field limit is the Wilson-Cowan equation and whose critical behavior belongs to the directed percolation class. In its continuum limit the neurons are the points of a Euclidean space and activity spreads by ordinary diffusion, which encodes neither the hierarchical organization of cortical networks nor their small-world property. We formulate a version of the theory in which the neurons are organized in an infinite, self-similar, tree-like structure, described by the p-adic numbers. The path integral carries over unchanged, and finite networks are exact restrictions of the continuum theory. The main change is in the linear response: in the Euclidean theory every short-range kernel reduces to the Laplacian, while in the p-adic theory each connectivity kernel gives its own nonlocal p-adic diffusion operator. For couplings that decay by a fixed factor per level of the hierarchy this is the Taibleson-Vladimirov operator, and the critical theory is an ultrametric Reggeon field theory whose upper critical dimension is twice the decay exponent of the coupling. Hence the decay of the coupling up the hierarchy, and not the dimension of space, decides whether mean-field exponents apply. We obtain a Ginzburg-type criterion, loop-corrected hierarchical Wilson-Cowan equations, log-periodic corrections to scaling and power-law correlations. Exact simulations on a tree support the predictions: the avalanche-size exponent takes the mean-field value 3/2 when the coupling decays slowly and decreases toward the directed-percolation value when it decays fast. The hierarchical network is also small-world when the decay exponent is smaller than the dimension, a property that the Euclidean theory obtains only by assuming long-range coupling.

How do local changes at individual synapses give rise to structured recurrent connectivity? Experiments have revealed diverse forms of spike-timing-dependent plasticity (STDP) across the brain, making it difficult to predict which circuit structures different rules will generate. We developed an analytical framework mapping excitatory Hebbian, anti-Hebbian, and symmetric STDP rules to directed three-neuron motifs. Individual homogeneous rules generate restricted subsets of motifs, and no single rule produces all motifs. Combining different rules generates motifs inaccessible to individual rules and yields the full motif repertoire, and these predictions generalize to larger recurrent networks. The mapping is nevertheless highly non-unique, with distinct rules and parameters producing similar motif distributions, including distributions compatible with experimental connectivity. Thus, circuit structure constrains candidate plasticity rules but does not uniquely identify them. Finally, networks with similar motif distributions can nonetheless differ in their firing rates and pairwise correlations, suggesting that activity may provide an additional constraint on candidate plasticity rules.

Brains do not simply add up every signal they receive: they select which routes dominate. Yet connectomes are usually analyzed with tools that sum over all paths and place neural states in Euclidean spaces, blurring the feedback circuits that may sustain internal states. Here we treat neural signaling as routing instead of diffusion and describe it with max-plus algebra, in which signals accumulate along a path and the strongest route prevails. We propose a model in which the functional connectome of Caenorhabditis elegans emerges from the time-varying combination of electrical, synaptic, monoaminergic, neuropeptidergic and hormonal layers, and analyze five networks within this framework. Each collapses onto a handful of dominant feedback loops, which recover well-characterized circuits and point to new candidates for experimental study. Unlike linear spectra, these structures keep the biological meaning of the data, and their order-based description suggests that brain states live in lattices rather than Euclidean spaces.