Representations of active grasp maintenance emerge during reach-grasp-carry learning in mouse motor cortex

Abstract

Object manipulation is a dexterous behavior requiring ongoing sensorimotor coordination to establish and maintain a grasp during consecutive movements. Although grasp initiation is known to modulate activity in primary motor cortex (M1), it remains unclear if sustained grasp maintenance is represented in M1, whether it corresponds to an intrinsic M1 state or one that emerges through learning, and how this state contributes to skilled behaviors involving object manipulation. We addressed these questions using two-photon calcium imaging in mice learning a self-initiated, self-paced, reach-grasp-carry task. This task allowed mice to control trial structure and timing, revealing two kinematically similar but functionally distinct carry behaviors: active grasps, in which mice maintained control of a sucrose pellet and received sensory feedback from the held object, and empty grasps, in which mice completed the same carry movement after failing to grasp the pellet. As behavior was refined, empty grasps became less frequent and the M1 representation of active grasp maintenance became more reliable and more distinct from empty grasp. This distinction was evident in single-neuron activity, population activity, and functional connectivity between neurons. These differences were sufficiently robust to support accurate decoding of active and empty grasp state from neural activity and improved encoding of single neuron activity when models incorporated condition-specific functional network coupling. Together, these findings show that learning refines M1 activity into a distinct and reliable representation of active grasp maintenance, revealing sensorimotor integration during object contact as a key neural substrate of learned dexterous behavior.