Content area
Variability, a ubiquitous feature of neural activity, is thought to play an integral role in behavior. However, studying neural variability is difficult because measuring all the possible neural activity patterns that produce a single behavior is challenging. By implementing a brain-computer interface (BCI), we circumvented this challenge by establishing a direct, causal relationship between select neurons (BCI neurons) and behavior. We trained monkeys (Macaca mulatta) in a BCI task in which they continuously altered (modulated) neural spiking activity to control a computer cursor. We then challenged our monkeys to adapt to novel BCI protocols (i.e., different task perturbations) and determined how components of neural variability constrained (or supported) subsequent behavioral adaptation. In the first project, we found that how the BCI neuron population covaries remains highly similar before and after adaptation. Additionally, neural populations readily exploit this property to regain proficient cursor control after perturbation – even when this strategy appears behaviorally sub-optimal. Finally, we found evidence that neural variability is disruptive to stable behavior. However, specific components of this variability can be leveraged to adapt behavior when behavioral contexts change and can be used as a metric to predict the amount of behavioral adaptation that is possible within a day. In the second project, we found that individual BCI neurons exhibited a variety of changes in response to task perturbations. Still, monkeys were able to adapt their neural activity enough to regain sufficient cursor control. We found that neurons with high levels of co-variability within the BCI population and neurons that contributed the most to behavioral output changed their activity the least. The mismatch between these measured changes and the changes required to counteract the perturbation fully reliably predicts the amount of behavioral recovery. Overall, we developed a neural variability-based framework that explains and predicts neural limitations of self-modulation.