Manifold Interactions between the Action-Mode Network and Sensorimotor Cortex during Human Motor Learning

Abstract

Effective motor learning requires a dynamic interplay between specialized sensorimotor circuits and higher-order control networks. How these systems coordinate their activity across the distinct phases of learning—from initial adaptation to consolidated performance and subsequent relearning—remains poorly understood. Here, we investigated the evolving functional coupling between the action-mode network (AMN), a domain-general system for goal-directed action, and the somatomotor network (SMN) throughout a multiday visuomotor adaptation task in human participants (19 females, 13 males). Using manifold learning techniques to characterize the low-dimensional geometry of changes in AMN–SMN functional connectivity, we observed a series of robust, task-dependent network reconfigurations. We found that initial, error-driven learning was marked by significant manifold contraction, reflecting heightened functional integration between the AMN and SMN with the broader higher-order association cortices. However, as learning performance plateaued, the AMN disengaged, leading to manifold expansion and greater functional segregation, while the SMN remained in a persistently integrated state, forming a latent signature of the newly acquired motor memory. Notably, this entire temporal pattern of network dynamics was reinstated during relearning on the next day. Furthermore, we found that the degree of functional integration during the initial learning phases on both days was associated with individual differences in learning and relearning performance. Together, these findings establish a hierarchical framework where the AMN dynamically couples and decouples with the SMN to meet the changing demands of skill acquisition, consolidation, and memory-guided retrieval, providing new insights into the large-scale sensorimotor network mechanisms that guide motor learning.