Abstract

Phasic activity of dopaminergic neurons in the ventral tegmental area or substantia nigra compacta has been suggested to encode reward-prediction error signal for reinforcement learning. Recent studies have shown that the lateral habenula neurons exhibit a similar response, but for nonrewarding or punishment signals. Hence, the transient signaling role of LHb neurons is opposite that of DA neurons and also that of several other brain nuclei such as the border region of the globus pallidus internal segment and the rostral medial tegmentum. Previous theoretical models have investigated the neural circuit mechanism underlying reward-based phasic activity of DA neurons, but the feasibility of a larger neural circuit model to account for the observed reward-based phasic activity in other brain nuclei such as the LHb has yet to be shown. Here, we propose a large-scale neural circuit model and show that parallel excitatory and inhibitory pathways underlie the learned neural responses across multiple brain regions. Specifically, the model can account for the phasic neural activity observed in the GPb, LHb, RMTg, and VTA/SNc. Based on sensitivity analysis, the model is found to be robust against changes in the overall neural connectivity strength. The model also predicts that striosomes play a key role in the phasic activity of VTA/SNc and LHb neurons by encoding previous and expected rewards. Taken together, our model identifies the important role of parallel neural circuit pathways in accounting for phasic activity across multiple brain areas during reward and punishment processing.