Neural Circuit Mechanisms Underlying Dopamine Reward Prediction Errors

Abstract

Dopamine neurons are thought to facilitate learning by signaling reward prediction errors (RPEs), the discrepancy between actual and expected reward. However, how RPEs are calculated remains unknown. In Chapter 1, I tested the hypothesis that RPE signals in dopamine neurons are inherited entirely from the lateral habenula, by examining how lesions of the habenular complex affect the response of optogenetically-identified dopamine neurons in mice. I found that despite large lesions of habenula, dopamine neurons maintained features of RPE coding pertaining to phasic dopamine responses. Interesting, a specific aspect of RPE signaling— the inhibitory responses caused by reward omission—was greatly diminished in habenula lesion animals. These results suggested that the RPE signals in dopamine neurons were not simply relayed from habenula and that multiple mechanisms underlie RPE signaling.

In Chapter 2, I systematically examined how RPE is computed at a neural circuit level, by combining rabies virus-based monosynaptic retrograde tracing with optogenetic cell identification during electrophysiological recording. We characterized the firing patterns of 205 neurons presynaptic to dopamine neurons (“input neurons”) from 7 major VTA input areas in behaving mice. We found that relatively few input neurons signaled purely ‘actual’ reward or ‘expected’ reward. Instead, many input neurons across brain areas signaled combinations of these types of information. We also found that some input neurons signaled already-computed RPEs. These results demonstrate that the information required for dopamine neurons to compute RPE is not localized to specific brain areas; rather, the computation is distributed across multiple nodes in a brain-wide network. Together, these results provide new insights on the neural circuits involved in the computation of RPE signals in dopamine neurons.