Motor cortex activity across movement speeds is predicted by network-level strategies for generating muscle activity

Abstract

Learned movements can be skillfully performed at different paces. What neural strategies produce this flexibility? Can they be predicted and understood by network modeling? We trained monkeys to perform a cycling task at different speeds, and trained artificial recurrent networks to generate the empirical muscle-activity patterns. Network solutions reflected the principle that smooth well-behaved dynamics require low trajectory tangling. Network solutions had a consistent form, which yielded quantitative and qualitative predictions. To evaluate predictions, we analyzed motor cortex activity recorded during the same task. Responses supported the hypothesis that the dominant neural signals reflect not muscle activity, but network-level strategies for generating muscle activity. Single-neuron responses were better accounted for by network activity than by muscle activity. Similarly, neural population trajectories shared their organization not with muscle trajectories, but with network solutions. Thus, cortical activity could be understood based on the need to generate muscle activity via dynamics that allow smooth, robust control over movement speed.

Simple decoding of behavior from a complicated neural manifold

“…[61,62]). For example, noise-robust solutions require low trajectory tangling [63], resulting in neural trajectories that look very different from the outputs they create [64].…”

“…the middle of circular trajectories) remain unoccupied [63,64]. Precisely because this geometry is difficult to describe with a model, MINT approximates the manifold using the empirical trajectories as a 'scaffolding', with interpolation producing a mesh of states that are considered during decoding.…”

“…This might even be done in an unsupervised way: a trajectory library could be built without behavioral labels, then appropriately linked to a 'reference' behavioral library via knowledge of task-specific neural geometry. As a simple example, cycling across different speeds produces a tube-shaped manifold with different velocities at each end [64]. An empirically observed tube might be given behavioral labels based on this knowledge.…”

Pathophysiology of Dyt1 dystonia is mediated by spinal cord dysfunction

“…Neural circuits that control movement are distributed across the neuraxis and are comprised of multiple interconnected loops involving the cerebral cortex 1 , basal ganglia 2,3 , thalamus 4 , cerebellum 5 , brain stem 6 , and spinal cord 7 . While each of these loops has its own function, it is the collaboration of the ensemble that ultimately produces functional movement and hence behaviour.…”

Direct neural perturbations reveal a dynamical mechanism for robust computation

“…9,10,[28][29][30][31] Low-dimensional subspace-structure can also emerge in task-optimized artificial systems when appropriately regularized. [32][33][34] Lastly, motor cortex may be governed by path-following dynamics (H3), in which the neural state is constrained to move along an externally configured path. In the context of motor control, previous work has proposed that the motor cortex might serve to activate specific motor programs implemented by recurrent circuitry in the spinal cord.…”