M1 dynamics share similar inputs for initiating and correcting movement

Motor cortex is integral to generating voluntary movement commands. However, as a dynamical system, it is unclear how motor cortical movement commands are informed by either new or sensory-driven corrective instructions. Here, we examine population activity in the primary motor cortex of macaques during a continuous, sequential arm movement task in which the movement instruction is updated several times over the course of a trial. We use Latent Factor Analysis via Dynamical Systems (LFADS) to decompose population activity into a portion explainable via dynamics, and a stream of inferred inputs required to instruct that dynamical system. The time series of inferred inputs had several surprising properties. First, input timing was more strongly locked to target appearance than to movement onset, suggesting that variable reaction times may be a function of how inputs interact with ongoing dynamics rather than variability in instruction timing. Second, inferred inputs were tuned nearly identically for both initial and corrective movements, suggesting a commonality in the structure of inputs across visually-instructed and corrective movements that was previously obscured by the complexity of the dynamical system that is M1.