Separate and shared low-dimensional neural architectures for error-based and reinforcement motor learning

Abstract

AO_SCPLOWBSTRACTC_SCPLOWMotor learning is supported by multiple systems adapted to processing different forms of sensory information (e.g., reward versus error feedback), and by higher-order systems supporting strategic processes. Yet, the extent to which these systems recruit shared versus separate neural pathways is poorly understood. To elucidate these pathways, we separately studied error-based (EL) and reinforcement-based (RL) motor learning in two functional MRI experiments in the same human subjects. We find that EL and RL occupy opposite ends of neural axis broadly separating cerebellar and striatal connectivity, respectively, with somatomotor cortex, and that alignment of this axis to each task is related to performance. Further, we identify a separate neural axis that is associated with strategy use during EL, and show that the expression of this same axis during RL predicts better performance. Together, these results offer a macroscale view of the common versus distinct neural architectures supporting different learning systems.