Thalamocortical credit routing from the basal ganglia and cerebellum: anatomical constraints and circuit principles

Animals possess the ability to learn and choose actions that yield the greatest reward in intricate settings. A key challenge in this learning process is credit assignment: the outcome of a behavior must be traced back to the neural cells involved in generating the pertinent actions. One potential pathway for this assignment is through thalamocortical routes, which transport learning signals from the basal ganglia and cerebellum, where learning modifications are influenced by reward prediction errors.

Nevertheless, the precise manner in which these pathways transmit such signals to specific cortical circuits remains uncertain. To investigate this, we integrated anatomical circuit mapping in mice with closed-loop circuit simulations. Anterograde transsynaptic tracing indicated that thalamic neurons receiving input from the substantia nigra pars reticulata or the lateral cerebellar nucleus were largely distinct, with a slight overlap in the ventromedial thalamus.

Nonetheless, their cortical axons demonstrated a significant concentration in layer 1a, particularly in the secondary motor cortex. In a brain-wide circuit model with interconnected loops, learning was amplified when both pathways delivered both additive and gain-modulating feedback that corresponded with the cortical action readout.

Calibration of local activity-based processes confirmed this alignment, while plasticity at both the dendritic spike and reward-prediction error levels further fine-tuned behavioral output. The specificity of feedback from distinct modules was crucial for a brain-machine interface task that necessitated the simultaneous reinforcement of multiple target neurons.

Overall, these discoveries elucidate a credit-routing framework in which laminar convergence, plastic alignment, and modular feedback connect parallel subcortical learning systems to cortical representations of actions.