Internal Dynamics Interact with Proprioceptive Feedback During Movement Execution in an RNN Model of Motor Cortex

Abstract

Neuronal population responses in motor cortex display low-dimensional rotational patterns, where neural state rotates consistently during movement execution. These patterns are thought to arise from the internal dynamics within motor cortex. However, external inputs such as proprioception also shape motor cortical activity. To investigate the roles of internal dynamics and proprioceptive feedback, we constructed an inhibitory stabilized network, with excitatory recurrence stabilized by inhibition, designed to control an arm model performing delayed reach task. The optimized network generated patterns similar to those observed in motor cortex data, demonstrating our model’s plausibility. We designed a disruption strategy to dissect the contribution of internal dynamics and proprioceptive feedback, and found that disrupting internal dynamics significantly diminished rotational dynamics, whereas disrupting proprioceptive feedback led to ceaseless movement. Thus, internal dynamics dominate, while proprioceptive feedback fine-turns cortical pattern generation. Analysis of the ablation experiment revealed that proprioceptive feedback improved the robustness against noisy initial conditions by decreasing tangling in neural activity. Finally, our network still exhibited rotational dynamics when the arm moved at a constant speed, suggesting that these rotations could be prominent features of motor cortical activity.