System-wide dissociation of reward and aversive dopaminergic signals
Dopamine is central to reinforcement learning, classically linked to reward prediction error signaling. However, recent findings suggest a more complex picture, with dopamine neurons responding to aversive or non-reward-related events and displaying diverse projection patterns. To investigate how dopamine release varies across the brain, we used multi-site fiber photometry to record the fluorescent dopamine sensor GRAB-DA in mice performing tasks involving both rewarding and aversive outcomes. We found that while reward-related dopamine signals were broadly distributed, responses to aversive events were region-specific, enabling classification of projection targets from these dopamine dynamics. By using dimensionality reduction to examine the main axis of covariance of dopamine release across the brain, we found that reward prediction error signals are encoded by parallel manifolds, whereas aversive stimulus signals are encoded by orthogonal manifolds. Thus, our findings support a distributed, vector-valued model of dopaminergic signaling, in which anatomically and functionally distinct pathways contribute to encoding of reinforcement-related variables across the brain.