Vibhu Sahni ¹, Nicolo Macellari ², Eleni Sinopoulou ³, Monica Perez ⁴

¹ Weill Cornell Medicine, ² University of Pittsburgh, ³ University of California, San Diego, ⁴ Shirley Ryan Ability Lab, Northwestern University

Discussant: Elvira Pirondini

Descending cortical projections to the brainstem and spinal cord form the essential substrate through which the cortex shapes skilled movement. A central unanswered question is how reorganization within these circuits contributes to recovery after injury, and whether insights into their developmental and evolutionary origins can guide strategies to promote repair. However, emerging work across species now reveals how the organization of this circuitry diverges across mammals, raising new questions about how these differences shape motor control and recovery potential.  Integrating these cross-species insights might allow us to infer the roles these pathways play in both the intact and injured CNS promoting the development of novel intervention and technologies. This symposium brings together new molecular, anatomical, and functional insights from mice, primates, and humans to refine our understanding of these descending projections in motor control and their role in recovery after cerebral damage.

Dr. Vibhu Sahni will present findings from the mouse that dissect how descending corticobrainstem and corticospinal pathways emerge during development, integrating molecular delineation, axonal anatomy, and circuit-level function. These developmental principles provide a framework for understanding both the consequences of early disruptions on movement control as well as the opportunities they provide for recovery after adult injury. Dr. Eleni Sinopoulou will then present findings comparing rodent and primate descending systems, highlighting conserved features as well as key species-specific divergences in connectivity that potentially shape motor capabilities and recovery potential. Finally, Dr. Nicolo Macellari and Dr. Monica Perez will turn the focus to evidence from primates and humans demonstrating how plasticity within corticofugal circuits supports functional recovery after cortical and spinal lesions, and how plasticity mechanisms within these networks can be leveraged to design more precise and effective therapeutic interventions to restore motor function.

Together, the talks will illustrate how a cross-species, developmental-to-clinical perspective on descending control systems can uncover fundamental principles of motor circuit organization and resilience, offering a framework to understand normal function and inform strategies for restoring function after injury.

Movement, like perception and cognition, is fundamentally shaped by the pursuit of reward. Across species, reward is known to be a powerful modulator of action, guiding action selection, movement invigoration and reinforcing successful behavior. Most work on reward has leveraged decision-making paradigms, in which agents have to learn to select among a discrete number of actions through reinforcement. Yet growing evidence indicates that reward is also instrumental in tasks with richer motor demands, such as motor learning, where individuals refine movement kinematics through practice. Despite the clear potential of incorporating reward into motor rehabilitation, the precise behavioral and neural mechanisms by which reinforcement shapes motor learning remain underexplored.

In this talk, I will first present evidence that specific properties of reinforcement feedback such as its extrinsic value, timing, and the quality of concurrent sensory feedback profoundly influence how humans control, learn, and retain motor skills. I will also discuss the feasibility, efficiency, and constraints of delivering personalized reinforcement in real time during continuous motor control in healthy adults and patients with chronic stroke.

In the second part, I will present work investigating the causal role of the striatum in motor and reinforcement learning, using transcranial Temporal Interference Stimulation (tTIS)—a non-invasive method for deep-brain neuromodulation in humans—in combination with fMRI. In particular, I will discuss data supporting the causal involvement of specific striatal rhythms in reinforcement and sensory-based motor learning, and highlight their possible implications for neuropsychiatric disorders that impact the motor and reward systems. Overall, these results illustrate the breadth of mechanisms by which reward shapes movement and delineate the promise—as well as the potential limitations—of reward-based approaches to motor rehabilitation.

Kazuhiko Seki ¹, Martyn Goulding ², Aya Takeoka ³

¹ National Center of Neurology and Psychiatry, ² Salk Institute for Biological Studies, ³ RIKEN Center for Brain Science

Discussant: Simon Giszter

For more than a century, the field of sensorimotor control has aimed to clarify the organization of central nervous system circuits. The field is now entering a pivotal phase, as multiple research streams converge on the same conclusion: to move forward, we need large-scale recordings from spinal circuits during natural behavior.

This shift is driven by advances in molecular-genetic circuit dissection. Since Sherrington, spinal physiology has identified fundamental circuit modules using electrophysiological approaches. Modern genetic tools have now refined these modules with remarkable precision, enabling causal tests by selectively removing specific components. Transcriptomic profiling further reveals extensive cellular heterogeneity. Yet real-world behavior is clearly not governed by isolated modules but by their coordinated interaction. To understand what each module contributes—and how they work together—we need direct recordings during natural movement.

A similar momentum comes from population-level research in the cerebral cortex. Since Evarts, systems neuroscience has shown that individual cortical neurons regulate force, direction, and motor patterns. Recent advances in large-scale recording and computational analysis now link population dynamics to behavior. Yet cortical commands are rarely executed directly; most are routed through spinal circuits. Without understanding this spinal “middleware,” long treated as a black box, we cannot fully explain motor control. Large-scale spinal recordings in behaving animals will therefore be essential.

In this session, we will discuss—together with the audience—how such recordings are beginning to revise long-standing textbook views of the neural control of movement, drawing on the speakers’ recent experimental findings and their implications.

The session will open with Kazuhiko Seki outlining the motivation and historical context of this Perspective session, followed by his latest experimental results showing how spinal interneuron dynamics are differentially recruited during rat and monkey reaching, depending on various movement parameters. Martyn Goulding will then discuss how sensory information that is used to control movement is represented in the spinal cord and brainstem, followed by Aya Takeoka’s talk demonstrating how the multiunit activity recorded from the spinal cord opens new avenues for the circuit mechanisms underlying movement adaptation and learning. The session will conclude with a discussion led by Simon Giszter, exploring—with audience participation—how the spinal cord serves as a “middleware” for the sensorimotor control and motor learning, and what new insights are emerging from this line of research.

O1.1 – An output-null neural manifold for planning-on-the-fly during free manual control

Nicolas Meirhaeghe ¹, Julio Rodino ², Shrabasti Jana ³, Lucio Condro ³, Frédéric Barthélemy ³, Junji Ito ², David Dahmen ², Alexa Riehle ¹, Sonja Grün ⁴, Thomas Brochier ¹

¹ CNRS, ² Institute for Advanced Simulation (IAS-6), Jülich Research Centre, ³ Aix-Marseille University, ⁴ Institute for Advanced Simulation (IAS-6) Jülich Research Centre

Presenting Author: Nicolas Meirhaeghe

Studies of the neural basis of motor planning have largely focused on scenarios in which a single movement is specified in advance and prepared “offline” prior to execution. However, much of natural behavior cannot be reduced to a sequence of discrete, preplanned steps. Instead, everyday actions unfold as a continuous stream of movement that is controlled “online,” updated in real time as errors are detected and goals evolve. This raises two key unresolved questions for the field of motor control: How are unconstrained, naturalistic actions planned at the neural level, and do offline and online planning rely on shared neural principles?To address these questions, we analyzed motor cortical population activity from two macaque monkeys trained to perform both offline- and online-prepared reaches to visual targets. In both tasks, animals initiated behavior with a center-out reach to a randomly chosen peripheral target. They were then either (1) free to reach the remaining simultaneously presented targets in any order (unconstrained task), or (2) required to follow a sequence imposed by targets that appeared one at a time (constrained task).In the constrained task, we first replicate and extend classical findings: preparatory activity associated with offline planning of center-out reaches is (1) low-dimensional, (2) expressed in output-null dimensions forming a preparatory subspace, and (3) organized as direction-specific preparatory states lying on a circular manifold. We then show that this manifold is preserved across different initial hand positions and generalizes to any reach direction planned offline within sequences of the constrained task. This invariance suggests that the manifold provides a substrate for context-independent encoding of reach direction—analogous to the ring-like structure observed in the mammalian head-direction system.Crucially, we demonstrate that this same manifold is re-engaged during the unconstrained task, when movements are planned online and continuously updated. The strongest evidence for such reuse comes from our ability to predict upcoming reach direction hundreds of milliseconds in advance during free movement, using only activity projected onto the manifold defined from the center-out preparatory epoch. We further confirm that this predictive signal reflects genuine planning rather than motor execution: although hand trajectory can also be decoded from the execution subspace, these signals exhibit systematically shorter temporal lags than predictions derived from the orthogonal preparatory subspace.Together, these results provide a unified view of how motor cortex supports both discrete and continuous action planning. They reveal shared neural mechanisms underpinning offline and online planning and offer new opportunities for leveraging invariant preparatory structure to design more robust and efficient brain–machine interface decoders suited for naturalistic movement.

O1.2 – Neural population geometry underlying sequential reach in the macaque motor cortex

Di Zhu ¹, Tianwei Wang ², Yu-Qi You ¹, Yao Chen ¹, He Cui ³, Ru-Yuan Zhang ¹

¹ Shanghai Jiao Tong University, ² Lin Gang Laboratory, ³ Chinese Institute for Brain Research, Beijing(CIBR)

Presenting Author: Di Zhu

Natural motor behavior relies on the smooth concatenation of discrete actions. When generating multiple actions in sequence, the motor cortex must encode not only the physical features of each action (e.g., reach direction) but also integrate them into a coherent behavioral sequence. Previous studies have proposed three distinct hypotheses regarding how sequencing influences neural representations of individual actions. The competition hypothesis suggests that representations of adjacent actions interfere with one another due to increased cognitive load or limited neural resources. In contrast, evidence for coarticulation in some motor tasks indicates that blending actions into a sequence actually enhances the representation of each component. A third account, the independence hypothesis, posits that individual actions are generated separately and thus sequencing should not alter individual’s neural encoding. However, these competing hypotheses have not been systematically tested at the neural population level.Here we analyzed population activity from the primary motor cortex of two rhesus macaques (monkey C: 118 units; monkey G: 44 units, Utah array) performing a sequential reaching task. The task was a variant of the center-out delayed-reach paradigm and included both single-reach (SR) and double-reach (DR) trials. In SR trials, one target was presented, followed by a random delay, after which the monkey executed one reach. In DR trials, two targets were presented and remembered, and the monkeys executed two reaches sequentially without an inter-movement pause. Comparing neural activity between SR and DR trials allowed us to examine how planning and executing a subsequent action influences the neural representation of the first action.We report three main findings. First, decoding accuracy for the first reach direction in the DR condition was significantly higher than that in the SR condition throughout most of the preparatory and execution periods. This enhancement indicates that planning an additional action strengthens the representation of the first action, consistent with the coarticulation hypothesis. Second, neural population geometry revealed that improved representations arose from increased separation between manifolds corresponding to different reach directions, driven by systematic tuning changes at the single-neuron level. Third, population subspace analyses showed that this manifold separation originated primarily within the preparatory subspace rather than the execution subspace. In addition, analyses of single-neuron responses uncovered several nonlinear tuning modulations that further contributed to the strengthened population representation.These results provide direct neural evidence supporting the coarticulation hypothesis in the coding of sequential actions. Our findings also highlight the importance of examining neural population geometry to understand how action sequencing can enhance the encoding of individual movements in the motor cortex by optimizing neural resources.

O1.3 – Disrupting PnC Activity Impairs Skilled Hand Reaching and Grasping in Non-Human Primates

Yiping Sun ¹, Reona Yamaguchi ¹, Tadashi Isa ¹

¹ Kyoto University

Presenting Author: Yiping Sun

Dexterous finger control is important for daily hand functions in nonhuman primates and humans. Maintaining balanced muscle strength and tone is necessary for precise finger coordination, but its control mechanisms are less clear compared to the mechanisms of individual finger movement, primarily attributed to the corticospinal tract. Recent evidence suggests that reticulospinal neurons (RSNs) play a critical role in this process.In this study, we recorded single-unit activity in two healthy Macaca fuscata to localize startle-reaction neurons, presumably RSNs, in the caudal pontine reticular formation (PnC). The identified location was consistent with previous studies. Based on this anatomical information, we manipulated the RSNs by microinjections of muscimol, a GABAA receptor agonist, into the PnC to investigate their contribution to hand dexterity. To quantify the behavioral effects of PnC inactivation, we developed a markerless 3D kinematic reconstruction pipeline based on DeepLabCut™, enabling frame-by-frame extraction of multi-joint kinematics of the index finger and thumb—including joint angles, fingertip trajectories, and apertures—during natural reach-and-grasp tasks.Transient inactivation of the PnC produced clear impairments in hand dexterity without overt changes in overall posture. Across grasp types, monkeys showed significantly prolonged grasping time, distinct aperture modulation patterns across individuals, and compromised thumb–index coordination. Although the two monkeys adopted different natural grasping strategies, both exhibited task-dependent abnormalities in aperture control during the pre-shaping phase. To determine whether these differences reflected a shared deficit, we applied dynamic time warping (DTW) to all time-varying joint angles and apertures. Increased DTW distances reflected lower similarity between movement time-series. Following muscimol injection, both monkeys exhibited significantly greater DTW distances, indicating that PnC inactivation disrupted finger coordination during grasping.These findings suggest that PnC neurons contribute to fine finger control by regulating the coordinated activation of finger muscles. Future physiological studies will further elucidate the role of RSNs in sensorimotor integration and brainstem contributions to dexterous motor control.

O1.4 – Deep Brain Stimulation of the Motor Thalamus Alleviates Post-Stroke Upper-Limb Motor Deficits

Arianna Damiani ¹, Nicolo Macellari ¹, Mel Xu ¹, Miles Uribe ¹, Lucy Liang ¹, Catherine Jezerc ¹, Jordyn Ting ¹, Sirisha Nouduri ¹, Erinn Grigsby ¹, Lilly Tang ¹, Marco Capogrosso ¹, Jorge Gonzalez-Martinez ¹, Elvira Pirondini ¹

¹ University of Pittsburgh

Presenting Author: Arianna Damiani

Stroke is among the leading causes of permanent disability in the United States, affecting approximately 795,000 people each year. Infarcts mostly damage subcortical areas, thus interrupting long-range connections between cortical and spinal centers that regulate movements, i.e., the corticospinal tract (CST). Consequently, humans with stroke invading the CST suffer from the most severe and permanent upper-limb motor deficits, including loss of strength and dexterity. Despite the large population size of affected individuals, intense physical therapy remains the only significant intervention with limited impact on moderate to severe paresis. Therefore, novel and effective therapeutic approaches are necessary. We previously demonstrated that Deep Brain Stimulation targeting the motor thalamus (mThDBS), a subcortical region with direct excitatory projections to the motor cortex, facilitates residual CST axons and consequently increases motor output in anesthetized monkeys. Yet, whether motor thalamus stimulation leads to functional improvements for post-stroke hemiparesis remains unclear. Here, we hypothesized that mThDBS increases the excitability of the CST also when voluntarily activated to perform a movement and thus alleviates post-stroke upper-limb motor deficits. To test this hypothesis, we trained n=2 monkeys to perform a battery of tasks to measure hand and arm dexterity and strength. We then performed a controlled thermocoagulation of the CST inducing chronic stroke with moderate upper-limb paresis. We also implanted a DBS electrode in the motor thalamus and assessed behavioral performances and kinematics with and without DBS, both in subacute and chronic stages. Following the CST thermocoagulation, the monkeys developed hallmark post-stroke motor symptoms, including mild loss of arm dexterity, and moderate loss of finger and hand control. mThDBS immediately facilitated 3D arm and finger control, as evidenced by higher task success rate and improvement in several kinematic variables, such as velocity, joints range of motion and endpoint precision. Finally, we replicated these results in n=1 human patient with post-stroke hemiparesis, who received a temporary DBS implant in the motor thalamus and was stimulated for 4 weeks, 2h per day during motor tasks. Our intervention led to immediate improvements in strength across multiple joints, increased dexterity and range of motion. Notably, electrophysiological testing confirmed enhanced corticospinal excitability after only 4 weeks of DBS, as evidenced by the emergence of transcranial magnetic stimulation-induced motor evoked potentials in the forearm muscles. Importantly, we also observed clinically relevant improvement in clinical scales (+8 points in Fugl-Meyer). These findings demonstrate for the first time the potential of motor thalamus DBS to alleviate motor symptoms developed after stroke and pave the way towards novel therapies for upper-limb hemiparesis.

O1.5 – Behavioral and neural determinants of speech motor memory

Nishant Rao ¹, Rosalie Gendron ², Timothy Manning ², David Ostry ²

¹ Yale University, ² McGill University

Presenting Author: Nishant Rao

In speech motor behavior, the involvement of auditory and somatosensory streams offers a unique opportunity to probe the relationship between sensory inputs and motor output. In this study, we sought to determine the behavioral and neural underpinnings of speech motor memory using speech formant frequency perturbation as a model task to induce motor learning. Akin to a visuomotor reach adaptation task, the speech formant perturbation task records formant frequencies (constituents of vowels in any language) via a microphone, alters the first formant frequency, and plays back the modified speech in real time to the participants via headphones. When subjected to several such trials, participants learn to compensate for the perturbed formant frequency by shifting the first formant in their speech output in an opposite direction. In a first experiment (n=14), we show that this speech motor learning paradigm enables participants to produce new speech movements leading to speaking a new vowel, which is sufficiently different from the well-learned vowels. In a second experiment (n=58), participants first performed the speech motor learning task and returned either 8- or 24-hours later for retention assessment. We find that the newly acquired motor memory is substantially retained for at least 24 hours post learning. The retention is similar to that when probed 8 hours following learning, and with abrupt or gradual formant shifts. Importantly, the memory was retrieved only when speech feedback was available, but not in presence of noise feedback, highlighting the context dependence of the speech motor memory upon presence of speech feedback. In a third experiment (n=60), the neural underpinnings of speech motor memory were investigated by separately disrupting either the primary motor, somatosensory, or auditory cortex using continuous theta burst transcranial magnetic brain stimulation (cTBS). Participants first underwent the speech motor learning task, immediately after which they received cTBS, and returned 24 hours later for retention assessment. We found that disruption of either auditory or somatosensory cortex impaired the retention of speech motor memory. In contrast, retention following disruption of motor cortex was not different than a no-TMS control, indicating the causal involvement of sensory cortex in retaining speech motor memory. Taken together, the study documents the behavioral and neural underpinnings of speech motor memory and underscores the relationship between sensory inputs and motor output when both auditory and somatic sensory streams are involved in guiding motor behavior.

Sponsored by

All trainees welcome to join us for a casual, networking social following the conclusion of the day.  Network in a casual environment, get to know new people, and enjoy this trainee focused event.

Adrian Haith ¹, Nidhi Seethapathi ², Eric Yttri ³, Alexander Mathis ⁴

¹Unaffiliated, ² Massachusetts Institute of Technology, ³ Carnegie Mellon University, ⁴ EPFL

Discussant: Joshua Cashaback

Humans can learn to perform a seemingly limitless array of complex motor skills. But how are we able to learn these skills? This panel will explore reinforcement learning as a potential theoretical framework for understanding learning of new motor skills. Existing computational theories of learning are largely founded on principles of supervised learning, appealing to notions of internal models and error-driven learning. Such theories account very well for short-term adaptation of existing behaviors, but don’t seem to account for how we learn new motor skills over timescales of days, weeks or months. A lot of work has examined reinforcement-based learning in operant-learning paradigms and in simple, adaptation-like tasks, and it has often been proposed that reinforcement learning may play an important role in more challenging forms of motor learning, but few concrete theories have been proposed. Meanwhile, the robotics community has leveraged reinforcement-learning-based approaches to power remarkable advances in recent years. This panel will present emerging work examining how these same approaches can be applied to understand biological learning. The panel will describe key ideas and approaches used in modern RL, and show how they can be applied to model biological learning. Adrian Haith will introduce policy-gradient reinforcement learning, a simple and widely-used model-free method, and show how this can account for dynamics of learning through practice over thousands of trials across a range of motor learning tasks. Nidhi Seethapathi will extend these ideas to consider reinforcement learning in continuous control settings, focusing on the interplay between exploration and stability during locomotor learning. Eric Yttri will provide causal evidence from animal studies demonstrating how cell types within the basal ganglia exert continuous control through policy-based RL across a range of motor tasks. Alexander Mathis will argue that, despite advances in RL, insights from psychology are currently necessary to learn skills with high-dimensional musculoskeletal models, and will furthermore illustrate what can be concluded about biological motor control from these models. We will conclude with a panel discussion in which we will consider the feasibility and scope of reinforcement learning as a theoretical framework for understanding motor skill learning in humans and animals, potential underlying learning mechanisms, and challenges associated with learning complex repertoires of skills through reinforcement learning.

O2.1 – Network Features Underlying Motor Learning and Skill Generalization in Marmosets

Samantha Johnson ¹, Jeffrey Walker ², Nicholas Hatsopoulos ¹

¹ University of Chicago, ² Yale University

Presenting Author: Samantha Johnson

Motor skill learning occurs across multiple timescales, yet most studies focus on short-term adaptations within single-session learning paradigms. Consequently, the network-level mechanisms underlying long-term skill acquisition remain poorly characterized. Emerging rodent work suggests that connectivity dynamics may pass through distinct phases during extended learning periods but has not been extensively examined with primates. Moreover, it is unclear how these network dynamics support generalization, the process by which previously acquired skills are adapted to new but related task contexts. The neural basis of generalization, especially across multiple interconnected sensorimotor areas, remains largely unknown.In this study, we trained marmosets on two sequential motor tasks requiring similar behavioral profiles. In the first task, animals performed a reach–grasp–carry movement toward a static food reward, reaching asymptotic performance over two weeks of daily exposure. We then introduced a more complex version of the task in which the target moved dynamically, requiring animals to generalize previously acquired movement strategies to a new context. Faster learning in the dynamic as compared to the static version of the task provided behavioral evidence of generalization. Throughout both tasks, we wirelessly recorded neural activity from chronically implanted Utah arrays spanning dorsal premotor, primary motor, and somatosensory cortices in unrestrained animals. Using spike timing of simultaneously recorded single units, we constructed functional networks to capture changes in sensorimotor connectivity over learning and generalization.We found that functional networks exhibited structured, biphasic changes across weeks of learning. Trial-to-trial network variability started high, consistent with exploration, decreased during the early stages of initial skill acquisition, and then increased as the skill consolidated. When animals transitioned to the dynamic task, we saw a return to variability, mirroring the early learning phase of the first task. These findings suggest that motor cortex and broader sensorimotor networks become heavily structured during the early phase of long-term learning, but may reduce their engagement after a skill becomes stable. Importantly, generalization appears to rely not on modifying an existing stable connectivity pattern but on reinstating a variability-rich network state. This indicates that generalization recruits similar neural mechanisms as initial skill acquisition rather than simple adaptation of an established motor program.Together, this work provides rare long-timescale, multi-area recordings in a primate model and reveals network-level signatures of exploration, consolidation, and generalization. These results highlight how sensorimotor connectivity reorganizes across sequentially acquired skills and motivate further comparison of functional network structure during expert phases of related behaviors.

O2.2 – Distinct premotor and motor cortical population dynamics support adaptive control in brain-machine interfaces in nonhuman primates

Alessia Sepe ¹, Ophelie Saussus ¹, Sofie De Schrijver ¹, Irene Caprara ², Renaud Detry ¹, Pinhao Song ¹, Thomas Decramer ³, Peter Janssen ¹

¹ Katholieke Universiteit Leuven, ²Massachusetts General Hospital, ³ Universitair Ziekenhuis Leuven

Presenting Author: Alessia Sepe

Motor brain-machine interfaces (BMIs) translate cortical activity into device control, offering a platform not only for restoring movement but also for investigating the neural principles underlying motor behavior. Although natural action requires rapid adjustments to changing environmental conditions, most BMI paradigms rely on simple, unperturbed movements, leaving the neural basis of adaptive control unsolved and limiting the development of more flexible motor BMIs. Here we recorded neural activity from primary motor cortex (M1), ventral premotor cortex (PMv), and dorsal premotor cortex (PMd) using three 96-channel Utah arrays in three macaques performing BMI-control in a 3D virtual environment in Unity. Monkeys guided a sphere toward one of five distant targets using the neural activity of all three areas, with trials occasionally including two environmental perturbations, obstacle appearance and target displacement. In obstacle appearance trials, an obstacle appeared unpredictably in half of the trials, midway between the starting point and the target location, and the monkeys had to move the sphere around this obstacle to reach the target. In target displacement trials, the target changed location in half of the trials when the sphere was at one third of the distance to the target. We quantified firing rates and across-unit variance within the subpopulation of units directly used by the BMI decoder (based on Preferential Subspace Identification) to determine how each area contributes to online corrective control. M1 activity was largely unaffected by the presence or type of perturbation. By contrast, both PMv and PMd exhibited marked reductions in across-unit variance early after the perturbation, indicating increased population-level coherence during adaptive control. Notably, PMv responses decreased for both perturbation types, whereas PMd showed enhanced responses selectively in target displacement trials.These findings demonstrate that adaptive visuomotor control in BMIs relies predominantly on premotor circuits, with PMd playing a specialized role in updating movement plans when goal locations shift. M1 contributes minimally to early perturbation processing, consistent with a division of labor in which premotor areas compute corrective commands that M1 implements downstream. This framework provides a foundation for designing BMIs tailored to flexible, real-world motor behavior.

O2.3 – Linking Neural Control and Muscle Mechanics: Feedforward and Feedback Contributions in Healthy and Neuropathic Balance Control

Hansol Ryu ¹, Surabhi Simha ², Gregory Sawicki ¹, Lena Ting ²

¹ Georgia Institute of Technology, ² Georgia Institute of Technology & Emory University

Presenting Author: Hansol Ryu

Effective control of perturbed movement relies on the interplay between feedforward and feedback control. Computational models are needed to dissociate their respective contributions to perturbed movement control because they are difficult to separate experimentally. Standard Hill-type muscle models, though useful for steady-state force estimation, may lead to inaccurate simulated neural control strategies as they miss critical transient behaviors such as short-range stiffness (SRS). As shown previously [1], Hill-type models without SRS predict unrealistically large initial movements before corrective activity engages, limiting interpretations of neural control in health and impairments.To develop a mechanistic simulation of perturbed movement, we incorporated a 3-state cross-bridge model [2] into a closed-loop inverted pendulum system. Our muscle model under baseline activation reproduces rapid, history-dependent force rise observed in single fibers during stretch [3], but its implications for perturbed movement control have not been explored. We coupled agonist-antagonist muscle pair to an inverted pendulum and simulated responses to sudden translational perturbations. Both muscles were controlled with feedforward tonic activation, modulated by fiber length, velocity, and acceleration feedback with delays.Cross-bridge mechanics markedly reduces initial displacement compared to Hill-type models due to SRS. Simulations reproduce the rapid, SRS-driven rise in joint torque at perturbation onset, followed by a distinct initial burst and plateau in muscle activation from sensorimotor feedback, as observed in cats and humans [4,5]. The model also captures sensory neuropathy signatures: removing velocity and acceleration feedback causes loss of balance, as in acute neuropathy, while restoring partial feedback yields large excursions, diminished initial bursts and increased co-contraction, consistent with chronic neuropathy in cats [5].Our novel biophysical muscle model in a closed-loop simulation dissociates feedforward and feedback contributions to perturbed balance control, incorporating SRS tuned by feedforward co-contraction. Because SRS naturally emerges from cross-bridge dynamics under baseline activation rather than being phenomenologically imposed, the model can be applied to different tasks such as reaching and walking. This framework allows for better understanding of muscle–neural interactions in perturbed movement control and may help generate hypotheses about how movement is controlled in health and altered in conditions like neuropathy, aging, fatigue, or disease. [1] De Groote F, Allen JL, Ting LH. (2017) J. Biomech. [2] Campbell KS. (2014) J. Gen. Physiol. [3] Horslen, BC., et al. (2023) J. Exp. Biol. [4] Jakubowski, KL, et al. (2025) J. Neurophysiol. [5] Lockhart, DB., Ting, L. H. (2007) Nat. Neurosci.

O2.4 – Cerebellar signals to the motor cortex link preparation to execution of voluntary movements

Nirvik Sinha ¹, Ora Ben Harosh ¹, Henn Kramer ¹, Ran Harel ², Julius Dewald ³, Jonathan Kadmon ¹, Yifat Prut ¹

¹ Hebrew University of Jerusalem, ² Tel Aviv University, ³ Northwestern University

Presenting Author: Nirvik Sinha

Performing rapid, coordinated, and accurate voluntary movements relies on predictions that guide movements before sensory feedback can be used. The cerebellum is essential for calculating these predictive signals based on efferent copies of motor commands and a forward internal model. Although this view emphasizes the role of the cerebellum in the initial phase of movements, recent studies have shown cerebello-thalamo-cortical interactions as early as the movement preparation period. This suggests that cerebellar predictions may already shape emerging motor commands during planning. To test this hypothesis, we trained two monkeys to perform center-out delayed reaching movements on a vertical touchscreen. During task performance, we reversibly disrupted cerebellar output using high-frequency (130 Hz) stimulation of the superior cerebellar peduncle while recording muscle and neural activity. Neurons were sampled using high-density linear probes inserted into the task-related regions of the primary motor (M1) and premotor (PM) cortices. Blocking the cerebellar outflow altered motor behavior, leading to delayed onset (p<0.001), diminished peak speed (p=0.013) and increased variability in hand position (p<0.001). At the neural level, cerebellar block produced changes in the structure of preparatory activity, including a rotation of the preparatory subspace away from the control condition (45% alignment, p< 0.001) and an increase in its dimensionality (p<0.001). These results indicate an impaired pattern of inter-neuron coordination, with effects emerging earlier in PM than in M1. Next, we examined changes in the predictive power of the preparatory state under these conditions. To this end, we used reduced rank regression on the single-trial data to identify the low dimensional linear mapping linking the neural activity during preparation to that of execution. Cerebellar block increased the dimensionality of this mapping (quantified as the minimum rank needed to reach the R² asymptote of the full rank regression, p=0.021) even though its capacity to predict the execution-related activity was reduced (i.e., lower R2, p = 0.004), indicating a less efficient transformation of the preparatory state into movement-related activity. Consistent with this impaired mapping at the neural level, cerebellar block markedly reduced the ability of preparatory activity to predict movement kinematics (peak speed and hand position; p < 0.001), whereas the prediction of these parameters from execution-related activity was unchanged (p = 0.286; p = 0.979). This indicates that cerebellar signals primarily support the transformation of preparatory neural states into accurate movement plans rather than their execution. Taken together, our findings show that cerebellar predictive signals relayed through motor thalamus play a dual role in controlling voluntary movements. First, these signals are necessary for organizing the structure of preparatory dynamics in the motor cortex. Second, these signals play a critical role in sustaining the mapping that links preparation to execution and carries kinematic information essential for accurate and timely multi-joint reaching.

O2.5 – Cereballar circuits anticipate dopamine rewards

Benjamin Filio ¹, Mark Wagner ²

¹ National Institute of Neurological Disorders and Stroke, ² National Institutes of Health

Presenting Author: Benjamin Filio

Theories of motor control, especially skill acquisition and reward-driven motor learning, place heavy emphasis on the role of cortical and striatal circuits, while the cerebellum is most famous for motor adaptation and refining movements. Extensive evidence demonstrates that the cerebellum is functionally interconnected with cortical and striatal circuits, highlighting a need to update brain-wide computational frameworks. The cerebellum is also causally involved in functions traditionally ascribed to forebrain and midbrain, one of these functions being reward-based motor learning. Cerebellar circuits can predict rewards such as food and water, but it is unclear whether the cerebellum also encodes reward outside of conditions where (1) the reward satisfies homeostatic craving and (2) when a physical action is required to consume the reward. Here, we developed a reward-driven operant task where mice pushed a lever to receive dopamine (DA) self-stimulation, which was either optogenetic activation of DA neurons in the ventral tegmental area or electrical activation of the medial forebrain bundle. This satisfies the requirements to answer our questions: first, the mice did not need to physically consume a reward, disentangling neural representations of reward from consummatory movements. Second, the reward was temporally delayed, allowing us to separate reward predictive signals from neural representations of action. Third, the stimulation paradigm was surgically, genetically, and optically compatible with two-photon cerebellar imaging. Mice performed robustly for both types of DA reward. In expert mice, 22% of GrC continually ramped up activity for a 1-s delay between push and expected DA reward. Further, 65% of CFs spiked with short latency when DA stimulation occurred. When we subsequently trained mice on a 2-s delay, we observed 31% of individual GrCs “stretched” their activity profiles to match the expected time of DA release. When we trained mice sequentially with water reward either preceding or following DA reward, both rewards elicited similar population GrC and CF representations. Individually, 32% of GrCs had significantly correlated activity between water and DA, while 33% of CFs activated similarly for water and DA. Causally, mice learned operant reinforcement to push for CF optogenetic activation, with GrCs anticipating the predicted self-stimulation. Finally, inhibiting GrC activity in the delay period between push and DA reduced operant task performance, push quality, and motivation compared to control mice. We detail that previously characterized GrC-CF computations for homeostatic reward also extends to direct mesolimbic dopamine stimulation, displaying the cerebellum uses a general computation to link motor programs to expected rewards. This broadens the involvement of cerebellum in brain-wide computational networks involved in sensorimotor behavior and motor control.

Arkarup Banerjee ¹, Britton Sauerbrei ², Daniela Vallentin ³, Gregg Castellucci ⁴

¹ Cold Spring Harbor Laboratory, ² Case Western Reserve University School of Medicine, ³ Max Planck Institute for Biological Intelligence, Germany, ⁴ University of Rochester

Discussant: Andrew Pruszynski

The control of voluntary movements (e.g., reaching) is modeled as a top-down process in which cortical commands are filtered through spinal networks to determine motor output. Behaviors like speech, singing, and locomotion, however, require bidirectional coordination between higher brain areas and subcortical circuits which can generate motor patterns independently. In this session, we explore such coordination in a range of vertebrate species and behaviors.

Arkarup Banerjee, using the singing mouse, will discuss how motor cortex hierarchically interacts with brainstem pattern-generator circuits to control vocal production. Using circuit perturbations, electrophysiological data, and computational modeling, he will show that motor cortex activity adjusts song duration, while downstream circuits determine note-level features. These findings outline a systems-level framework illustrating how higher-level cortical controllers shape lower-level pattern generators to flexibly adapt behavior—a challenge shared by natural and artificial agents.

Daniela Vallentin, using the nightingale, will reveal a hierarchical mechanism by which temporal and spectral features are integrated during vocal matching. Wild nightingales show real-time pitch matching, and manipulating syllable duration shifts their own whistle durations toward the presented structure. When exposed to unnatural pitch–duration pairings, they flexibly trade off spectral and temporal imitation. A computational model formalizes this hierarchy, showing that syllable duration provides the temporal scaffold for pitch adjustments and shapes real-time matching behavior.

Gregg Castellucci will introduce a conceptual model of multi-effector action that places behaviors along a continuum from multitasking (simultaneous, competing actions) to unitary whole-body movements aimed at a single goal. He will discuss how sequential multi-effector behaviors across species—including speech and gesture in humans and wingspread displays in cowbirds—fit within this framework, and how different levels of the motor hierarchy and biomechanical factors contribute to their coordination. Finally, he will show how the kinematics of multi-effector actions can be used to infer their underlying planning and control strategies and illuminate their neural basis.

Britton Sauerbrei will discuss coordination between neural dynamics in motor cortex and the spinal central pattern generator (CPG) in mice performing an obstacle traversal task. Cortical dynamics consist of a large preparatory transient as the animal nears the barrier, oscillations driven by an efference copy from the CPG, and small, muscle-like signals resembling corticospinal commands. A simple model is then proposed which transforms inputs conveying obstacle proximity and locomotor phase into motor commands. These results reveal a regime in which higher brain areas must sculpt an ongoing, spinally-generated program to flexibly control behavior.

Join us to learn more about the society, the financial position, incoming board members and more!

Join us to hear from key members of the NCM community as they discuss their professional journey in industry and academia.  Following the presentation, stay for networking and focused group discussions, facilitated by leaders in the community and society.  This event is welcome to all attendees who want to hear, learn, and support others in the community.

Zekun Sun ¹, Katja Kornysheva ², Lucas Tian ³, Yuto Makino ⁴

¹ Yale University, ² University of Birmingham, ³ Rockefeller University, ⁴ National Institute of Information and Communications Technology

Discussant: Samuel McDougle

A violinist can move between instruments with only minor adjustment, a tennis player can switch rackets with little effort, and a child can draw the same shape using crayons, markers, or finger paint. Our ability to generalize motor skills so efficiently suggests that the mind can encode actions in an abstract format — one that preserves their essential structure while freeing them from the constraints of specific muscle commands and contexts. This panel will present recent theoretical advances and empirical findings that significantly modify and extend beyond traditional accounts of motor abstraction.

First, Zekun Sun will present behavioral evidence showing that motor abstractions support novel skill learning at the very earliest stages of acquisition. Using a novel handwriting paradigm, Sun will show that the motor system recruits genuinely motoric, high-level representations that generalize across substantial kinematic variation.

Building on this theme of generalization, Yuto Makino will then introduce a new theoretical framework for how the motor system retrieves and applies learned skills across contexts. Makino shows that motor learning is best explained by representations organized around the relative phase of a movement, rather than absolute state or timing parameters.

Transitioning from behavioral to neural mechanisms, Katja Kornysheva will examine how skilled actions are constructed in real time. Kornysheva argues that behavioral fusion in memory-guided sequences cannot be taken as a proxy for neural fusion of sequential elements, but that skilled actions are dynamically assembled from coordinated interactions between motor, premotor, parietal, and hippocampal systems. This perspective highlights a flexible, distributed control architecture that can rapidly reorganize skilled actions as task demands evolve.

Finally, Lucas Tian will address how the brain represents and manipulates the symbolic building blocks of action. Using a compositional drawing task paired with large-scale neural recordings in macaques, Tian identifies neural populations that encode discrete, recombinable action symbols in premotor cortex, and neurons in preSMA that represent variables in grammatical rules. The structured population geometry of these signals explains macaques’ capacity to generalize learned grammars.

Together, these four perspectives converge on a new understanding of motor abstraction as a transferable, flexible, dynamic, and generative phenomenon. Rather than relying on rigid templates or invariant schemata, the motor system appears to build actions from flexible symbolic units, modular neural components, and phase-based representations that support rapid learning, adaptive recombination, and robust generalization. This panel highlights how behavioral experiments, neural analyses, and computational modeling are reshaping foundational theories of how the brain constructs and generalizes skilled actions.

O3.1 – Understanding strategic sensorimotor adaptation as a process of hypothesis testing

Anjuli Niyogi ¹, Elizabeth Cisneros ², Wei Ding ³, Richard Ivry ⁴, Jonathan Tsay ¹

¹ Carnegie Mellon University, ² University of California, Berkeley, ³ Tsinghua University, ⁴ University of California

Presenting Author: Anjuli Niyogi

Picture a surgeon, mid-procedure, suddenly forced to switch from a trusted instrument to an unfamiliar one. In that moment, success—or catastrophe—hinges on the ability to rapidly adapt to the new tool’s weight and dynamics. This capacity for motor adaptation is fundamental to human competence, allowing us to adjust to changes in our bodies (e.g., injury) and environments (e.g., novel tools). Indeed, motor adaptation emerges from multiple learning processes—some implicit and automatic, others explicit and strategic—that jointly enable flexible, goal-directed behavior. For decades, motor learning research has largely revolved around the mechanisms supporting implicit adaptation. The dominant computational account, the error reduction model, casts adaptation as a gradual, trial-by-trial minimization of motor error. This framework has captured many core phenomena of implicit motor adaptation and has guided extensive work identifying the cerebellum as its core neural substrate. However, the processes underlying strategic motor adaptation remain largely unknown. To close this gap, we developed a psychophysical paradigm that isolates strategic adaptation and measures the full two-dimensional structure of participants’ movements (via an unconstrained x–y workspace). This approach uncovers a striking pattern: although group-averaged behavior looks gradual, individuals produce large, diverse, and highly structured errors before a sudden insight that brings performance into a stable solution space. These idiosyncratic learning functions contradict the smooth, incremental learning predicted by Error Reduction, pointing to a qualitatively different process. We propose a new computational account in which strategic adaptation arises from learners generating, evaluating, and revising causal explanations for their motor errors (e.g., rotation, translation, reflection), and devising strategies to counteract the inferred perturbation. Our model accounts for a broad range of strategic phenomena across diverse perturbation types (rotations, reflections, gain changes). The model outperforms leading alternatives in the motor-learning literature (error reduction, reinforcement-learning models, win–stay/lose–shift heuristics, “moment-of-insight’’) in capturing individual learning functions.Moreover, the model reveals how uncertainty in feedback—symbolic feedback offering abstract, numeric guidance vs. sensory feedback offering precise, location-based information—attenuates strategy discovery. Additionally, it can also explain how environmental structure (e.g., target arrangement) biases learners toward specific hypotheses, producing dramatically different motor-error distributions. It also provides a mechanistic account of why strategic adaptation deteriorates in both healthy and pathological aging. Finally, we discuss how, in more complex sensorimotor tasks (e.g., riding a bicycle), the relevant hypotheses may extend beyond geometric primitives to other conceptual primitives. Altogether, these findings define a new, unifying theory of how humans discover successful sensorimotor strategies, bridging action closer to cognition.

O3.2 – Cortex-dependent motor control can be learned from subcortical demonstration

Jason Keller ¹, Joshua Dudman ²

¹ HHMI Janelia Research Campus, ² Howard Hughes Medical Institute

Presenting Author: Jason Keller

Motor control in mammals is traditionally viewed as hierarchical, with the frontal cortex at the top. This framework suggests that cortex learns to control subcortical circuits through trial-and-error, a process that is both inefficient and inconsistent with the rapid contextual learning observed with innate behaviors. To address this, we studied cortical and subcortical interactions as head-fixed mice learned contextual control of a hindlimb extension motor primitive. Naïve mice performed reactive extensions to a cold air stimulus within seconds and, using predictive cues, learned to avoid the stimulus altogether in tens of trials. Using 3D kinematic tracking, we show that while extension movements were remarkably consistent across avoidance learning, predictive avoid extensions were subtly slower than reactive ones, reminiscent of cortically-controlled motor preparation. We thus examined the cortical dependence of hindlimb extension using intermittent local optoinhibition of the frontal cortex and found that it prevented avoid but not react extensions. However, mice with such inhibition on every trial easily adapted to this perturbation, suggesting distributed cortical control of the contextual, predictive extensions. To test this, we also inhibited the majority of the dorsal cortex using a novel “acute optogenetic decortication” and found that this perturbation could indeed inhibit all avoid extensions when applied on every trial, but still left reactive extensions intact with only minor posture changes. Surprisingly, though, when we released mice from this decortication on every trial, we observed an immediate appearance of avoid trials, indicating that mice can “covertly” learn to use predictive cues to elicit avoid extensions, even in the absence of any prior cortically-mediated avoidance experience. Simultaneous Neuropixels recordings in several brain areas confirmed the extent of optogenetic inhibition and were consistent with distributed control of hindlimb extension movements and widespread motor corollary activity. Furthermore, they revealed that avoid extensions were marked by persistent cue activity in the frontal cortex that was linearly separable from movement-related activity. These findings support a distributed cortical-subcortical, heterarchical control logic in which the frontal cortex can learn contextual, predictive motor control from subcortical demonstration of innate motor primitives.

O3.3 – Coordinated spinal locomotor network dynamics emerge from cell-type-specific connectivity patterns

Frank Wandler ¹, Benjamin Lemberger ¹, James Murray ¹, David Mclean ²

¹ University of Oregon, ² University of Edinburgh

Presenting Author: Frank Wandler

Vertebrate locomotion is accomplished by rhythmic muscle activity that is generated and precisely coordinated by a dedicated neural circuit in the spinal cord, called the locomotor central pattern generator (CPG). Despite decades of research, how this rhythmic activity emerges from the locomotor CPG remains incompletely understood. Existing models tend to rely on local (often single-cell) oscillators to generate rhythmic activity, which is then patterned by network-level mechanisms. However, such models have not fully accounted for recent experimental results in zebrafish and other organisms that point to the importance of cell-type-specific intersegmental connectivity patterns and recruitment of speed-selective subpopulations of interneurons. To address these limitations, we develop a hierarchy of increasingly detailed models of the locomotor CPG in larval zebrafish that iteratively incorporate these observations. Surprisingly, in these models, rhythmogenesis occurs at the network level, without the need for local oscillators. In particular, we find that coordinated locomotion emerges in an inhibition-dominated network in which connectivity is determined by intersegmental phase relationships among interneurons and variable-speed control is implemented by recruitment of speed-selective subpopulations. Further, while structured excitatory connections are not necessary for rhythmogenesis, they are useful for increasing peak locomotion frequency, albeit at the cost of smooth interpolation between frequencies, suggesting a basic computational trade-off between speed and control. We verify these network-level mechanisms for rhythmogenesis by reproducing our results in a spiking model. Finally, by training the weights in the model using gradient descent, we find that hierarchical and modular connectivity motifs emerge as a solution to reproduce experimentally observed dynamics from intracellular recordings from various cell types. Our results establish a novel mechanistic understanding of rhythm generation and control in CPGs based on biologically-motivated connectivity motifs.

O3.4 – Cortico-subcortical dynamics underlying adaptive locomotion

Martin Esparza ¹, Ioana Lazar ¹, Catia Fortunato ¹, Mostafa Safaie ¹, Juan Gallego ²

¹ Imperial College London, ² Champalimaud Foundation

Presenting Author: Martin Esparza

Locomotion is a fundamental behaviour common to all animal species. As such, its neural underpinnings have been extensively characterized, with particular emphasis on spinal and brainstem circuits during stereotyped locomotion. Yet, the neural processes mediating adaptive locomotion, which likely involve higher-order cortical and subcortical areas, remain elusive.To address this gap, we designed a task based on delivering rapid, unpredictable mechanical perturbations to head-fixed mice running on a spherical treadmill using 12 evenly distributed actuators. This allowed us to elicit a broad range of behavioural corrections, determined by the actuator’s location and the animals’ ongoing gait cycle. We recorded from sensorimotor cortices, downstream basal ganglia projections, and relay centres in the motor thalamus, while also collecting whole-body 3D kinematics.To investigate the role of each region during adaptive locomotion, we projected their firing rates onto the dominant patterns of covariation to obtain region-specific latent dynamics. Linear decoders trained on these latent dynamics were able to classify the perturbation direction from all regions as well as from the kinematics. Yet, the peak decoding time was variable across brain regions, with the primary motor and somatosensory cortices leading downstream regions. Additionally, perturbations selectively increased inter-region interaction strength when compared to unperturbed running, with interaction peaks happening at different time shifts, suggesting “information flow” from primary motor to somatosensory cortex and then basal ganglia and thalamus.To disentangle how sensory responses and motor corrections contributed to these results, we trained linear regressors to predict latent dynamics from kinematics (sensory encoding) and vice-versa (motor decoding). Primary motor cortex presented both the best encoding and decoding of limb kinematics in absolute terms. Although both types of predictions improved across all regions following a perturbation, the magnitude of this increase was region-specific. Particularly, motor cortex showed the largest improvement in predictive accuracy when encoding kinematics, while the downstream dorsolateral striatum and thalamus showed the largest increase in decoding kinematics.Finally, as animals engaged in perturbed trials, decoding accuracy of kinematics from latent dynamics increased compared to a ~10 min pre-perturbation period. This marked improvement was accompanied by a region-specific change in the leading covariation patterns across neurons, which spanned directions nearly orthogonal in between these two periods. These observations suggest a shift in “control mode”, where adaptive locomotion recruits higher-order regions when facing unpredictable perturbations. Ongoing pharmacological manipulations will causally define these regions’ contributions, shedding light on how sensorimotor cortical and subcortical interactions enable adaptive locomotion.

Get out and experience Kobe!  Further information regarding excursion options can be found on the Excursions page.

Mayu Takahashi ¹, Ziad Hafed ², Jeffrey Schall ³, Martin Bohlen ⁴

¹ Tohoku University, ² Centre for Integrative Neuroscience, ³ York University, ⁴ Duke University

Discussant: Neeraj Gandhi

Rapid eye movements (saccades) are among the most elementary and time-critical motor actions performed by the nervous system. Such eye movements are under a perpetual influence of endogenous cognitive demands as well as exogenous sensory drives, and this is reflected in decades of research largely investigating these two classes of oculomotor system influences in isolation. However, even simple retrospection reveals that exogenous sensory inputs, arriving asynchronously with respect to internal brain state, must compete with endogenous processes to drive behavior. This symposium will highlight novel perspectives on the enduring tension between peripheral sensory drives and top-down control of eye movements. Using recent advances in anatomical, physiological, and theoretical approaches, we address a fundamental question in systems neuroscience: how distinct pathways—direct retinal projections to midbrain and brainstem circuits, and cortical sensory and decision networks—interact, and potentially even compete, to govern rapid sensorimotor decisions. Starting from the final gate for allowing saccade generation to proceed, the brainstem omnipause neurons, we will show how feature-tuned sensory modulations of these neurons’ activity jumpstart a highly reflexive and precise coordination between exogenous sensory events and endogenous motor plans; remarkably, these sensory-driven ocular reflexes depend entirely on the geniculostriate pathway (Hafed). This dominance of the geniculostriate pathway starkly contrasts with novel and compelling tracing and high-density electrophysiology evidence revealing diverse direct projections from the retina to the superior colliculus in primates (Bohlen), raising essential questions about when sensory pathways that bypass the geniculostriate system are functionally engaged. Upstream of the omnipause neurons, we will address the question of how the decision to initiate, suppress, or interrupt an eye movement requires top-down impacts from parietal and frontal cortical areas on the brainstem, and ultimately the omnipause neurons (Schall). Theoretically, similar coordination principles to those involved in sensory-driven modulations of omnipause neurons emerge, but with clearly distinct functions and time scales. These distinctions are further elucidated by integrated physiological and anatomical evidence defining the connectivity among the frontal eye field, superior colliculus, and omnipause neurons (Takahashi), revealing a shared brainstem switching mechanism through which fixation and saccade initiation are implemented. By aligning bottom-up, intermediate, and top-down perspectives on rapid eye-movement control, this session will redefine and sharpen classic questions in oculomotor neuroscience and highlight where the next major conceptual breakthroughs are likely to arise.

O4.1 – Dual descending pathways drive the initiation of skilled movements

Yutaka Yoshida ¹, Akimasa Ishida ¹, Esther Lai ¹, Teruko Danjo ², Fumiyasu Imai ², Jay Bikoff ³, Samuel Sober ⁴, Amanda Jacob ⁴, Matthew Williams ⁴, Kyle Thomas ⁴

^{1 2} Okinawa Institute of Science and Technology, ² Weill Cornell, ³ St. Jude, ⁴ Emory

Presenting Author: Yutaka Yoshida

The initiation of skilled movements is regulated by descending neurons in the sensorimotor cortex and brainstem projecting to the spinal cord. However, the mechanisms by which these pathways convey initiation signals remain unclear. This uncertainty arises from two key previous observations. First, inhibition of each descending pathway does not abolish the initiation of skilled movements such as reach–grasp behaviors. Second, brief stimulation of descending neurons fails to evoke the sequence of these movements. In this study, we demonstrate that combined inhibition of corticospinal neurons (CSNs) in the rostral forelimb area (rCSNs) and reticulospinal neurons (ReSNs) in the reticular formation effectively blocks the initiation of skilled reaching behaviors. Conversely, simultaneous brief activation of both rCSNs and ReSNs is sufficient to not only initiate reaching, but also to evoke sequential, naturalistic reach–grasp movements. These findings suggest that rCSNs and ReSNs converge on a “reach–grasp generator” located in the cervical spinal cord. Supporting this, we find that both rCSNs and ReSNs are activated prior to movement initiation, and that these two descending circuits synergistically operate, with no apparent crosstalk in the central nervous system. Together, these findings reveal a cooperative mechanism by which two parallel descending pathways drive the initiation and execution of skilled motor sequences in mammals.

O4.2 – The influence of using finger-extending exoskeletons on proprioceptive and tactile localization

Dominika Radziun ¹, Siebe Geurts ¹, Valeria Peviani ¹, Luke Miller ¹

¹ Donders Institute for Brain, Cognition and Behaviour, Radboud University

Presenting Author: Dominika Radziun

Humans show a remarkable ability to incorporate tools and artificial extensions into their sensorimotor control loops. Yet the perceptual and computational consequences of such integration remain poorly understood. We investigated how finger-extending exoskeletons reshape proprioceptive and tactile localization, providing a controlled test case for probing the flexibility of somatosensory body maps during technological augmentation. In the first experiment, twenty participants completed a high-density proprioceptive mapping task across four phases: before wearing the exoskeleton, after donning it, after active use, and after device removal. We identified three dynamic phases of plasticity. Wearing the device initially contracted the perceived length of the biological finger. After active use, both biological and exoskeletal fingers exhibited significant representational stretching, an effect absent in a non-augmenting control condition. Finally, a post-removal aftereffect persisted on the biological finger. In the second experiment, seventeen adults performed tactile localization on their index and middle fingers before and after 45 minutes of training with 10-cm finger-extending exoskeletons. Using a probabilistic model of tactile localization, we isolated changes in sensory processing, decision variables, and body representation. Exoskeleton use selectively altered the spatial remapping between mechanoreceptor and physical space, producing a ~24% stretch in tactile maps accompanied by a proportional increase in localization uncertainty. All other sensory and decisional parameters remained stable. These results show that wearable augmentations drive structured updates to somatosensory representations rather than simple spatial shifts, revealing principled computational rules governing representational expansion. Together, these findings demonstrate that tactile and proprioceptive maps rapidly adapt to artificial extensions, with changes governed by structured probabilistic computations and shaped by both the structural and functional properties of the device. This work highlights the flexibility of the sensorimotor system in accommodating technological augmentations and provides a computational framework for understanding how artificial extensions become integrated into the body representation.

O4.3 – Spatially independent and temporally robust somatosensory hand representations in expert pianists

Masato Hirano ¹, Sachiko Shiotani ², Shinichi Furuya ¹

¹ Sony Computer Science Laboratories, Inc., ² NeuroPiano Institute

Presenting Author: Masato Hirano

In the sensorimotor system, population neural activity is constrained by the physical structure of the musculoskeletal system and its kinematics. A seminal study has shown that the geometry of hand representation, the covariation structure across fingers, mirrors the covariation patterns of finger use in daily life (Ejaz et al. 2015), suggesting that long-term usage or training can reorganize hand representations. Classical studies in musicians suggested that long-term instrumental practice increases the spatial independence of cortical motor and somatosensory hand representations (Elbert et al., 1995). However, recent fMRI work has challenged this view, claiming no clear expansion and even reduced representational distances between fingers in musicians (Ogawa et al., 2019), and preserved somatosensory maps after amputation (Schone et al., 2025). The controversy raises the question of whether long-term training of dexterous finger movements reorganizes somatosensory hand representations.An additional open question is whether structured somatosensory mappings are preserved or distorted when multiple fingers are moved in rapid succession, as in musical performance. In such situations, somatosensory signals from different fingers arrive close in time, requiring the nervous system to separate temporally overlapping inputs while maintaining a segregated mapping of which finger moved and how. This challenge is particularly pronounced in expert pianists who, through years of practice, execute rapid sequences of multi-finger movements. Pianists therefore provide a unique model to elucidate how extensive training overcomes such a challenge through adaptation of somatosensory hand representations.Here we tested these ideas by comparing expert pianists with matched nonmusicians. EEG was recorded while a custom-made hand exoskeleton robot delivered precisely controlled passive flexion/extension movements to individual fingers and to multiple fingers in rapid succession. By systematically varying movement tempo, we characterized both the spatial organization and temporal robustness of somatosensory hand representations by using multivariate pattern analyses of evoked responses. When the robot delivered isolated passive movements to individual fingers, nonmusicians exhibited a characteristic representational geometry, with shared activity patterns between the middle, ring, and little fingers. In contrast, expert pianists showed a more differentiated spatial geometry with increased separation across these digits. When multiple fingers were moved in rapid succession, nonmusicians showed a distortion of this structure, with finger- and motion-specific representational patterns largely collapsing. However, pianists maintained clear finger- and motion-specific geometry during rapid multi-finger movements. These findings suggest that long-term piano training induces somatosensory hand representations that are both spatially more independent and temporally robust, supporting accurate processing of proprioceptive inputs during high-speed sequential actions.

O4.4 – Setting the gains: Neural and behavioral evidence for pre-movement feedback controller configuration

Dominique Delisle-Godin ¹, Pierre-Michel Bernier ¹

¹ Université de Sherbrooke

Presenting Author: Dominique Delisle-Godin

Humans routinely perform skilled actions that involve automatic, context-appropriate adjustments to movement, such as reaching for a cup of coffee while reading this abstract and seamlessly correcting the hand’s trajectory if bumped by a co-worker. These rapid feedback corrections are thought to rely on the engagement of controllers whose feedback gains are tuned to the specific demands of the task. For instance, previous work using mechanical perturbations has shown that feedback gains scale with accuracy requirements, yielding stronger long-latency (LL) muscle responses when targets are narrower. Despite the contention that these controllers are set up before movement, the neural evidence for such preparatory configuration remains limited because prior studies have predominantly focussed on movement execution. Here, we tested whether neural activity preceding movement reflects the preparation of a feedback controller with increased feedback gains. A secondary objective was to test whether the availability of online visual feedback influences preparatory neural activity. Twenty-eight healthy adults performed a delayed, goal-directed reaching task toward either a narrow (2.5 cm) or wide (80 cm) target. On a subset of trials, mechanical perturbations (±12 N) were applied to probe feedback responses. To assess the role of vision, blocks of trials were performed both with and without online visual feedback of the cursor. Surface electromyography (EMG) and electroencephalography (EEG) were recorded simultaneously to characterize muscle and neural activity. We hypothesized that preparing a controller with increased feedback gains (narrow target) would be accompanied by stronger beta-band (13–30 Hz) event-related desynchronization (ERD) during the delay period, reflecting greater investment of neural resources in motor preparation. As previously reported, LL responses were significantly larger for the narrow compared to the wide target for both +12 N and -12 N perturbations (p < 0.001), confirming the modulation of feedback gains by accuracy demands. Critically, and consistent with our predictions, EEG analyses revealed a frontoparietal cluster showing significantly stronger beta ERD in the later portion of the delay period for the narrow target condition (p < 0.01). Interestingly, the same qualitative pattern of EMG and EEG results held true irrespective of the vision condition, indicating that similar controllers were engaged with or without visual feedback. By linking pre-movement neural activity to feedback responses during movement, this work provides evidence that feedback controllers are configured prior to movement and identifies beta ERD as a marker of preparing a high-gain controller.

O4.5 – Hierarchical and Context-Dependent Population Codes for Intended and Observed Actions in Human Parietal and Motor Cortex

Celia Bougou ¹, Jorge Gamez ², Emily Rosario ³, Charles Liu ⁴, Kelsie Pejsa ¹, Ausaf Bari ⁵, Richard Andersen²

¹ California Institute of Technology, ² Division of Biology and Biological Engineering, California Institute of Technology, ³ Casa Colina Hospital and Centers for Healthcare, ⁴ Keck School of Medicine of USC, ⁵ University of California, Los Angeles

Presenting Author: Celia Bougou              

How the human brain represents actions across intention and observation remains unknown, largely because direct intracortical comparisons across cortical levels have been exceptionally rare. We addressed this gap by recording neural activity from Utah arrays implanted in motor cortex (MC) and posterior parietal cortex (PPC) of two individuals with tetraplegia participating in an ongoing brain–machine interface study. Across five implants, we obtained more than 4,500 single units, providing an unprecedented opportunity to examine how action features are encoded across cognitive formats and cortical hierarchies. Participants either internally generated or passively viewed the same set of actions in a fully crossed design manipulating hand, action type, and movement direction. We identified a clear functional dissociation between areas. MC robustly encoded internally generated actions yet exhibited only sparse and unreliable tuning during observation, offering the first direct human evidence that single-unit mirroring in MC is minimal. However, population analyses revealed a latent overlap between intention and observation: observed actions occupied a weak but geometrically aligned subspace of the intention manifold, reconciling the sparse responses at the single-unit level with longstanding imaging reports of MC activity during action observation. PPC showed a contrasting profile. Single-unit selectivity and population activity both demonstrated that action identity was represented invariantly across intention and observation. Representational similarity analysis, cross-format decoding, and low-dimensional trajectory alignment all converged on a shared, generalizable code for action features in PPC, suggesting that it supports an abstract representation of actions. To examine how these representations behave when observed and intended actions occur simultaneously, one participant performed a dissociation paradigm in which the viewed action and the instructed action were presented simultaneously and could be either congruent or incongruent. When the participant simply intended an action while observing another, PPC predominantly encoded the internal motor plan, with the observed action largely suppressed. However, when the task required the participant to intend an action, observe a potentially conflicting action, and subsequently report the observed one, PPC represented both the intended and the observed action. MC, in contrast, encoded only the instructed action in all conditions. These findings provide the first single-unit evidence in humans that the posterior parietal cortex flexibly regulates visual–motor coupling based on behavioral relevance and together support a hierarchical framework in which motor and parietal cortices occupy distinct and complementary roles in action encoding.

Naama Kadmon Harpaz ¹, Alice Mosberger ², Vivek Athalye ³, Karunesh Ganguly ⁴, Michael Long⁵

 Harvard University, ² NYU Langone Health, ³ Allen Institute, ⁴ University of California, San Francisco, ⁵ New York University Langone Medical Center

Discussant: Alice Mosberger

Animals can expand their behavioral repertoire by flexibly combining and sequencing existing motor elements into novel actions. Such behavioral versatility is essential for adapting to a changing environment and new conditions. However, only with practice, an initially slow, discontinuous, and variable sequence of movements is transformed into a fast, smooth, and stereotyped skilled behavior. These behavioral changes are thought to arise from plasticity within motor circuits and the reorganization of neural dynamics and cross-area interactions. This raises a central question: How and when do neural circuits reorganize during learning to support a new action composed of a sequence of motor elements, and how does this learning alter the representation of those elements themselves?

This panel will bring together work addressing these questions across diverse model organisms, including non-human primates, mice, rats, and parrots. We will examine how interactions among premotor and primary motor cortical regions and subcortical structures such as the striatum support the acquisition and execution of learned actions. The panel will also highlight recent methodological advances that allow manipulation and tracking of the activity of single neurons during learning. Together, we will explore circuit-level mechanisms that may be shared across species and that enable animals to expand their behavioral repertoire.

First, Athalye will discuss how movements are encoded in the striatum, a key region for learning and controlling skilled actions. He will show that the minimal movements required for an isometric push–pull task in mice – opposing muscle contractions – are represented in striatal neurons even prior to learning, providing a substrate upon which learning can build to support more complex actions.

Ganguly will then focus on mechanisms of circuit reorganization that stabilize a newly learned action composed of multiple motor elements. He will demonstrate how cross-area reactivation during NREM sleep drives cortico-striatal plasticity in mice and non-human primates practicing a reach-to-grasp task and discuss a specific NREM rhythm spanning cortex and striatum that causally contributes to the emergence of skilled performance.

Next, Kadmon Harpaz will discuss how animals learn and execute multiple actions composed of partially overlapping motor sequences. She will present work tracking behavioral changes in rats learning several lever-press sequences and the accompanying reorganization of the cortico-striatal circuit.

Finally, Long will explore how motor regions interact to generate multiple learned vocalization sequences in the budgerigar, a parrot species with a versatile vocal repertoire. He will describe how neurons in a premotor region generate temporally structured activity that is relayed to a motor output region, forming a hierarchical model that supports the acquisition and execution of versatile vocal behaviors.

Abstract to come

Following the Distinguished Career Award presentation and talk, join us for a drink to celebrate NCM 2026!