Matthew Tyler Kaufman

Grasping objects is a complex behavior requiring high-dimensional hand control and dynamic interaction with the environment, yet its neural mechanisms remain poorly understood. Dynamical systems approaches have provided key insights into how neural populations generate control signals for arm movements such as reaching and cycling but have proven insufficient to account for neural population activity during grasping. Here, we re-analyzed multi-area neural recordings from rhesus monkeys performing a reach-to-grasp task to many objects and found two key population-level features. First, the component of the population trajectory that was common to all grasps formed a small angle with the subspace containing grasp condition-specific (object) tuning. Second, the neural activity was well fitted by a recent, flexible dynamical model developed for reaching, called location-dependent rotations (LDR). Neural activity in M1 and F5 during grasping exhibited conserved population-level rotational frequencies, but the state-space planes in which these rotations occurred varied systematically with grasp condition. The rotational center (the location) and the orientation of the rotations related both to each other and to the kinematics of movement. While these rotations were reoriented in high dimensional space, reflecting the high-dimensional nature of hand control, the extent of their tilt into additional dimensions were small, in contrast to reaching. This structure may reflect the nature of grasping as modulations of an overall open-close motif. Together, these results indicate that the LDR dynamics framework applies to grasping as well as reaching and provides an entry point to understanding how grasp commands are generated.

Cortical control of movement is a distributed computation spanning multiple densely interconnected regions. Although we have rich anatomical atlases and a coarse understanding of how function maps to areas and subregions, we lack a detailed account of how behaviorally relevant activity is organized across the cortical sheet. Here, we trained head-fixed mice to perform a 15-target reach-to-grasp task while we performed cellular-resolution, two-photon calcium imaging across five regions of sensorimotor cortex (>39,000 layer 2/3 neurons). We characterized each neuron’s trial-averaged peri-event activity with interpretable metrics and mapped these response properties across areas, revealing large-scale spatial structure. Neuronal response profiles often shifted abruptly at anatomical borders: motor areas showed sharper tuning and more linear relationships with target location, whereas somatosensory areas displayed more heterogeneous response patterns. Neural response properties also differed according to somatotopic representation. Nonlinear dimensionality reduction of the neural feature matrix revealed that areas varied in their average response profiles, but that areas did not have well-separated feature distributions; instead, each area contained subpopulations. Neurons in each subpopulation had characteristic response profiles and were distributed across multiple cortical areas. The spatial distributions of the subpopulations overlapped, with neurons from different subpopulations salt-and-pepper intermingled in the overlap zones. Together, these results describe novel activity structure across sensorimotor cortex and identify several distinct but spatially overlapping subpopulations with characteristic activity patterns during reach-to-grasp behavior.

Cortical control of movement is a distributed computation spanning multiple densely-interconnected regions. Although we have rich anatomical atlases and a coarse understanding of how function maps to areas and subregions, we lack a detailed account of how behaviorally-relevant activity is organized across the cortical sheet. Here, we trained head-fixed mice to perform a 15-target reach-to-grasp task while we performed cellular-resolution, two-photon calcium imaging across five regions of sensorimotor cortex (>39,000 layer 2/3 neurons). We characterized each neuron’s trial-averaged peri-event activity with interpretable metrics and mapped these response properties across areas, revealing large-scale spatial structure. Neuronal response profiles often shifted abruptly at anatomical borders: motor areas showed sharper tuning and more linear relationships with target location, whereas somatosensory areas displayed more heterogeneous response patterns. Neural response properties also differed according to somatotopic representation. Nonlinear dimensionality reduction of the neural feature matrix revealed that areas varied in their average response profiles, but that areas did not have well-separated feature distributions; instead, each area contained subpopulations. Neurons in each subpopulation had characteristic response profiles and were distributed across multiple cortical areas. The spatial distributions of the subpopulations overlapped, with neurons from different subpopulations salt-and-pepper intermingled in the overlap zones. Together, these results describe novel activity structure across sensorimotor cortex and identify several distinct but spatially-overlapping subpopulations with characteristic activity patterns during reach-to-grasp behavior.

Coordinated movement along the body axis is critical to locomotion. In segmented, limbless animals, anterior (head) and posterior (tail) segments play different roles in locomotion, leading to a need for flexible coordination across body regions. Larval Drosophila melanogaster present a tractable experimental model for limbless, segmented crawling given the extensive genetic tools available and the optical clarity of the body. Prior work has suggested that, during crawling, all larval body segments contract similarly, despite the fact that each crawl cycle comprises two overlapping phases: a piston involving the most posterior segments and a peristaltic wave involving all body segments. To test whether coordination varies regionally during locomotion, we expressed GCaMP in body wall muscles of larvae of either sex and recorded segmental contraction kinematics and muscle recruitment during many cycles of locomotion in linear channels. Facilitated by machine vision techniques, we discovered new features of larval crawling at multiple scales. First, the propagation of both contraction and recruitment waves slowed approaching mid-body segments, then sped up toward the head. Second, the timing relationship between contraction and recruitment waves could be highly variable in anterior segments. Third, contraction durations showed particularly strong intersegmental correlations among posterior segments. These data suggest posterior segments coordinate to power the piston phase, while anterior segments tolerate greater flexibility to enable reorienting behaviors. Our results depict an unanticipated degree of axial heterogeneity in the coordination of limbless crawling, opening new avenues to study the origins of whole-body coordination and the consequences of segmental diversity for locomotion.

Coordinated forelimb actions, such as reaching and grasping, rely on motor commands that span a spectrum from abstract target selection to detailed instantaneous muscle control. The sensorimotor cortex is central to controlling these complex movements, yet how the detailed command signals are distributed across its numerous subregions remains unclear. In particular, in mice, it is unknown if the primary motor (M1) and somatosensory (S1) cortices represent low-level joint angle details in addition to high-level signals like movement direction. Here, we combine high-quality markerless tracking and two-photon imaging during a reach-to-grasp task to quantify movement-related activity in the mouse forelimb M1 (M1-fl) and forelimb S1 (S1-fl). Linear decoding models reveal a strong representation of proximal and distal joint angles in both areas, and both areas support joint angle decoding with comparable fidelity. Despite shared low-level encoding, the time course of high-level target-specific information varied across areas. M1-fl exhibited early onset and sustained encoding of target-specific signals, while S1-fl was more transiently modulated around lift onset. These results reveal both shared and unique contributions of M1-fl and S1-fl to reaching and grasping, implicating a more distributed cortical circuit for mouse forelimb control than has been previously considered.

Coordinated forelimb actions, such as reaching and grasping, rely on motor commands that span a spectrum from abstract target specification to detailed instantaneous muscle control. The sensorimotor cortex is central to controlling these complex movements, yet how the detailed command signals are distributed across its numerous subregions remains unclear. In particular, in mice it is unknown if the primary motor (M1) and somatosensory (S1) cortices represent low-level joint angle details in addition to high-level signals like movement direction. Here, we combine high quality markerless tracking and two-photon imaging during a reach-to-grasp task to quantify movement-related activity in the mouse forelimb M1 (M1-fl) and forelimb S1 (S1-fl). Linear decoding models reveal a strong representation of proximal and distal joint angles in both areas, and both areas support joint angle decoding with comparable fidelity. Despite shared low-level encoding, the time course of high-level target-specific information varied across areas. M1-fl exhibited early onset and sustained encoding of target-specific signals while S1-fl was more transiently modulated around lift onset. These results reveal both shared and unique contributions of M1-fl and S1-fl to reaching and grasping, implicating a more distributed cortical circuit for mouse forelimb control than has been previously considered.

Posterior Parietal Cortex (PPC) exhibits tuning to many variables, including strong representations of visual information, movement, and behavioral biases. Whether PPC communicates all these variables to other areas is less clear. We examined PPC activity in mice performing a novel, closed-loop, 2D visuomotor joystick task that required animals to act exclusively on a task-relevant axis of visual motion. To determine what components of PPC's representation were sent to M1, we performed two-photon calcium imaging of layer 2/3 neurons in contralateral PPC of expert mice with PPC–M1 projection neurons identified via retrograde tracing. Consistent with previous results, PPC neurons exhibited random mixed selectivity and were typically most strongly modulated by joystick movement. Most of the visually responsive neurons were more strongly modulated by task–relevant than task–irrelevant visual motion. Encoding in labeled PPC–M1 neurons was similar to encoding in unlabeled neurons, with one major exception: unlike the task-relevant visual enrichment in unlabeled PPC neurons, task-relevant and task–irrelevant visual motion were encoded at similarly weak levels in PPC–M1 neurons. This argues that although PPC encodes a mix of visual, movement and other information, the PPC–M1 pathway is dominated by movement information and does not propagate PPC's learned enrichment of task-relevant visual signals.

Cortical control of movement is a distributed computation spanning multiple densely-interconnected regions. Although we have rich anatomical atlases and a coarse understanding of how function maps to areas and subregions, we lack a detailed account of how behaviorally-relevant activity is organized across the cortical sheet. Here, we trained head-fixed mice to perform a 15-target reach-to-grasp task while we performed cellular-resolution, two-photon calcium imaging across five regions of sensorimotor cortex (>39,000 layer 2/3 neurons). We characterized each neuron's trial-averaged peri-event activity with interpretable metrics and mapped these response properties across areas, revealing large-scale spatial structure. Neuronal response profiles often shifted abruptly at anatomical borders: motor areas showed sharper tuning and more linear relationships with target location, whereas somatosensory areas displayed more heterogeneous response patterns. Neural response properties also differed according to somatotopic representation. Nonlinear dimensionality reduction of the neural feature matrix revealed that areas varied in their average response profiles, but that areas did not have well-separated feature distributions; instead, each area contained subpopulations. Neurons in each subpopulation had characteristic response profiles and were distributed across multiple cortical areas. The spatial distributions of the subpopulations overlapped, with neurons from different subpopulations salt-and-pepper intermingled in the overlap zones. Together, these results describe novel activity structure across sensorimotor cortex and identify several distinct but spatially-overlapping subpopulations with characteristic activity patterns during reach-to-grasp behavior.

Coordinated movement along the body axis is critical to locomotion. In segmented, limbless animals, anterior (head) and posterior (tail) segments play different roles in locomotion, leading to a need for flexible coordination across body regions. Larval Drosophila melanogaster present a tractable experimental model for limbless, segmented crawling given the extensive genetic tools available and the optical clarity of the body. Prior work has suggested that, during crawling, all larval body segments contract similarly, despite the fact that each crawl cycle comprises two overlapping phases: a piston involving the most posterior segments, and a peristaltic wave involving all body segments. To test whether coordination varies regionally during locomotion, we expressed GCaMP in body wall muscles of larvae of either sex, and recorded segmental contraction kinematics and muscle recruitment during many cycles of locomotion in linear channels. Facilitated by machine vision techniques, we discovered new features of larval crawling at multiple scales. First, the propagation of both contraction and recruitment waves slowed approaching mid-body segments, then sped up towards the head. Second, the timing relationship between contraction and recruitment waves could be highly variable in anterior segments. Third, contraction durations showed particularly strong intersegmental correlations among posterior segments. These data suggest posterior segments coordinate to power the piston phase while anterior segments tolerate greater flexibility to enable reorienting behaviors. Our results depict an unanticipated degree of axial heterogeneity in the coordination of limbless crawling, opening new avenues to study the origins of whole-body coordination and the consequences of segmental diversity for locomotion.