Mesocorticolimbic reinforcement learning of reward representation and value provides an integrated mechanistic account for schizophrenia

Mesocorticolimbic dopamine projections are crucial for value learning, motivational control, and cognitive functions, but their precise neurocomputational roles remain elusive. Based on recent experimental and theoretical findings, we constructed a neural circuit model where dopamine neuronal populations receive differential inputs from individual rewards and encode heterogeneous reward prediction errors, which train cortical and striatal neurons to learn reward-associated state representation and value. Learning is achieved via simultaneous alignments of the cortical and striatal downstream connections to the mesocorticolimbic dopamine projections, and inhibition-dominance in the cortical recurrent network is a key for successful learning. Excessive excitation, whether pre-existing or induced by manipulations, leads to aberrant activity, which disrupts the alignments and even causes anti-alignment. This impairs both reward-specific motivational control and credit assignment, potentially explaining the negative and positive symptoms of schizophrenia, respectively. Our model thus provides a mechanistic account for schizophrenia, integrating the different causes and symptoms, with testable predictions.