Neural trajectories improve motor precision

Populations of neurons in motor cortex signal voluntary movement. Most classic neural encoding models and current brain-computer interface decoders assume individual neurons sum together along a neural dimension to represent movement features such as velocity or force. However, large population neural analyses continue to identify trajectories of neural activity evolving with time that traverse multiple dimensions. Explanations for these neural trajectories typically focus on how cortical circuits learn, organize, and implement movements. However, descriptions of how these neural trajectories might improve performance, and specifically motor precision, are lacking. In this study, we proposed and tested a computational model that highlights the role of neural trajectories, through the selective co-activation and selective timing of firing rates across the neural populations, for improving motor precision. Our model uses experimental results from a center-out reaching task as inspiration to create several physiologically realistic models for the neural encoding of movement. Using a recurrent neural network to simulate how a downstream population of neurons might receive such information, like the spinal cord and motor units, we show that movements are more accurate when neural information specific to the phase and/or amplitude of movement are incorporated across time instead of an instantaneous, linear tuning model. Our finding suggests that precise motor control arises from spatiotemporal recruitment of neural populations that create distinct neural trajectories. We anticipate our results will significantly impact not only how neural encoding of movement in motor cortex is described but also future understating for how brain networks communicate information for planning and executing movements. Our model also provides potential inspiration for how to incorporate selective activation across a neural population to improve future brain-computer interfaces.