One of the main questions in basal ganglia (BG) research addresses the role of the connections among the different nuclei. While the role of the direct pathway is well established, the interaction between the indirect and hyperdirect pathways is less clear. We addressed this questions by means of a new neuro-computational model of the basal ganglia, which through biologically plausible learning rules, develops a set of new functionalities for the different cortico-basal ganglia-thalamic pathways. The major known projections which implement the direct, hyperdirect and indirect pathway are included as well as a direct connection between the cortex and the thalamus. Learning occurrs in all the projections from the stimulus cortex to the striatum and to the subthalamic nucleus (STN), through a rule which depends on both spike timing and the level of dopamine. The effect of dopamine is reversed in striatal cells expressing different receptors (D1 or D2), enabling different functionalities for the direct and indirect pathways. The connections between the cortex and the thalamus are also plastic but learn much slower and do not depend on dopamine.
To address the connections between STN and GPe we studied the leaning in the model after a rule reversal. Due to the reduction of the dopamine level, as a consequence of the unexpected errors in performance, the striatal D2 expressing cells become more activated and learn to inhibit the previous association to allow the striatum D1 expressing cells exploring new alternatives. Once
the network discovers the new correct association the phasic dopamine level increases and the striatal D2 cells lose their capacity to prevent wrong actions. However, during this transition period, neurons in the STN, activated by the indirect pathway, now learn a direct cortex-STN association and thus transfer the information from the indirect pathway to the hyperdirect pathway.
Concluding, the model predicts that the overlapping indirect and hyperdirect pathways at the level of GPe and STN transfer the inhibition of wrong actions in the indirect pathway learned by reward dips into an inhibition in the hyperdirect pathway learned by positive reward.