UC Berkeley

Research on brain-machine interface (BMI) systems has flourished in recent years, with motor BMIs showing great promise as a therapeutic option for patients suffering from limb loss or immobility. In addition to this direct clinical application, BMI tasks also serve as a powerful research tool, in that they enable the researcher to directly define which cells are relevant for behavioral output and the ways in which activity in these cells affects the external world. Neuroprosthetic tasks also serve as a completely novel motor-like learning paradigm for subjects, as they invoke the motor system but do not involve natural body movements or muscle activity. Intriguingly, a large body of work has nevertheless suggested striking similarities between natural motor learning and neuroprosthetic learning, implying that the two forms of learning may share common neural mechanisms.

We developed a novel rodent paradigm to study neuroprosthetic learning in which rodents controlled a one-dimensional auditory cursor by modulating activity in primary motor cortex (M1) in order to hit one of two targets. We first use this paradigm to explore corticostriatal representations of neuroprosthetic skills, the ways in which these representations change over the course of learning, and the necessity of plasticity in the corticostriatal network for abstract learning. We then investigate fine-scale temporal coordination between M1 and the dorsal striatum over the course of neuroprosthetic learning, demonstrating the development of coherent interactions between M1 spikes and the striatal local field potential with high temporal precision. Importantly, we found these interactions to be specifically present in output-relevant neurons, despite close proximity to other neuronal populations. Finally, we modified this behavioral paradigm for use in conjunction with two-photon calcium imaging in head-fixed mice to examine the fine-scale spatial characteristics of network adaptations during neuroprosthetic learning. We demonstrate the development of coordinated network activity and the sparsening of task-relevant modulations over the course of learning. This novel imaging-based BMI paradigm also allows for a number of new techniques to now be applied to research on neuroprosthetic learning. Together, these data suggest striking similarities between BMI and natural motor learning, demonstrating an important computational role for corticostriatal plasticity, neuronal coherence, and sparsening of cortical representations over the course of neuroprosthetic learning.