A TAN-dopamine interaction mechanism based computational model of basal ganglia in action selection.

The basal ganglia (BG) plays a key role in action selection. Physiological experiments have suggested that the reciprocal interaction between tonically active neurons (TANs) and dopamine (DA) is closely related to reward-based behaviors. However, the functional role of TAN-DA interaction in action selection remains unclear. In this study, a cortico-BG model including TAN-DA interaction mechanism is developed to explore the action selection mechanism of BG. The results show that in the default case, direct, indirect, and hyperdirect pathways are responsible for promoting, suppressing, and stopping the formation of stimulus-action associations, respectively. In the case of reinforcement learning, a single rewarded action is selected according to the combination of the TAN-DA dependent reinforcement mechanism and Hebbian mechanism with a gradual transfer from the former to the latter. Besides, a longer exploratory phase occurs when switching the reward to a new action because additional trials are required to overcome the habituation previously induced by the Hebbian mechanism. In the Parkinsonian state, the reinforcement mechanism is disrupted, and the resting tremor occurs due to dopamine deficiency. Although the model's performance significantly improves due to the levodopa treatment, it is still inferior to the healthy state. This phenomenon is consistent with the experimental results and is explained theoretically via the TAN pause duration and phasic DA release. Furthermore, the model's performances in multi-action selection further verify the rationality of the TAN-DA-dependent reinforcement mechanism. Our work provides a more complete framework for studying the action selection mechanism of basal ganglia.