Abstract
In a departure from current models of autonomous cortical activity, we examined the effect of continuous input on the dynamic activity of motor cortical neurons throughout a reaching task. We found a series of distinct state transitions in the firing rates of these recorded neural populations. We then asked whether these changes in state were due to internal processing in the motor cortex, as previous models would suggest, or to continuous external inputs. In other words, were signals coming from outside the motor cortex the main drivers of neural activity during reaching behavior? To answer this question, we used hybrid neural networks (HNNs)-consisting of a mixture of artificial connectivity and empirical firing rates-to construct realistic model systems. The HNNs faithfully produced the firing rates of the individual neurons and the state transitions of the populations we recorded, with extrinsic input consisting of episodically modulated neurons. Instead of the reported primacy of intrinsic action, we found that input from extrinsic sources was responsible for these results. Episodic external drive produced consistent changes in the statistics of pre-threshold input integration to cause the state transitions. By using HNNs with empirically constrained connectivity, we have shown that continuous input is a plausible agent for broad system functionality.