Reward-driven adaptation of movements requires strong recurrent basal ganglia-cortical loops

The basal ganglia (BG), a set of subcortical nuclei involved in motor control, sensorimotor integration, and procedural learning, modulate movement through tonic inhibition of thalamo-cortical networks. While essential for sensorimotor integration and learning, the BG are not necessary for executing well-learned movements. During skill learning, they guide behavioral corrections via dopamine-dependent cortico-striatal plasticity, but these corrections become BG-independent as modifications occur in the motor cortex. Existing models of BG function often overlook the feedback dynamics of cortico-BG-thalamo-cortical loops, and do not address the relative role of cortex and the BG in the generation and adaptation of movement. In this work, we develop a theoretical model of this multiregional network, integrating anatomical, physiological, and behavioral evidence to explore how its dynamics shape movement execution and reward-based adaptation. We show that the BG-thalamo-cortical network influences motor output through three key factors: (i) the rich dynamics of its closed-loop architecture, (ii) attractor dynamics from recurrent cortical connections, and (iii) classical reinforcement learning via dopamine-dependent cortico-striatal plasticity. Our study highlights that efficient visuomotor adaptation requires strong feedback from the BG to the cortex. Finally, we propose a mechanism for initial movement learning through motor babbling. This model suggests the BG-cortical network shapes motor output through its intricate closed-loop dynamics and cortico-striatal dopamine-dependent plasticity.