Determining the Sources of Motor-Cortex Response Complexity

PROJECT ABSTRACT Motor cortex is the brain area most responsible for voluntary movement. In humans, damage to motor cortex or its outputs – for example due to stroke or disease – causes profound and often permanent movement deficits. Treating dysfunction, or finding ways to bypass it, requires a principled understanding of normal function. Fundamental to that goal is the need to characterize the relationship between motor-cortex neural activity and movement. Attempts to do so must contend with a striking feature of motor cortex neural responses: their complexity and heterogeneity. Motor cortex responses are floridly multiphasic and remarkably diverse. Each recorded neuron has a unique response pattern, different from the last hundred neurons one observed. Attempts to understand motor cortex computation must therefore embrace, and hopefully explain, response complexity. Our field was initially resistant to this path – response complexity seemed like a bug rather than a feature. Yet as the nature and extent of response complexity became inescapable, hypotheses have increasingly sought to explain complexity as a natural consequence of the ‘job’ performed by motor cortex. Three possible sources of response complexity have been considered: (1) representation of many movement parameters, (2) biomechanical complexity, and (3) complexity of neural dynamics in which motor cortex participates. While all three sources presumably contribute, current hypotheses make very different assumptions regarding which source predominates. Distinguishing amongst these hypotheses – and informing new ones – has been difficult for a straightforward reason: most laboratory tasks engage all three sources of complexity. To make headway, we combine multiple innovations. These include very large-scale neural recordings (using many Neuropixels probes simultaneously) and recordings of motor-unit (MU) populations from the muscles. Critically, we leverage a novel task that minimizes physical complexity while maintaining – or even increasing – the complexity of internal computations needed for accurate performance. We use recently developed analysis tools that are ideally suited to addressing the central questions, and will compare results with network models that embody competing hypotheses. The resulting datasets are expected to be unusually rich, and thus capable of both testing current hypotheses and informing new ones.