Brain-wide, specialized and state-dependent cortical encoding of reward, value and action switching during reversal learning

In reversal learning tasks, large-scale circuits in multiple brain areas are involved in encoding multiple decision variables, such as trial outcomes, action values and action switching. It is unknown how spatiotemporal patterns of cortical encoding differentially encode these variables and whether the encoding depends on the level of engagement in the task. Here, we used a combination of large-scale imaging techniques and computational methods to investigate the representation of reward-related decision variables at the widefield and single-neuron levels. Widefield imaging and unsupervised analysis of cortical activity segment cortical activation into cortex-wide neural modes, each associated with a different spatial footprint, temporal dynamics and encoding of reward and error. Three-photon imaging of single-neurons across superficial and deep layers in the ACC, RSC, motor and visual cortex revealed brain-wide but specialized processing of reward-related variables. We discovered widespread representation of outcome, value, and switching in all four regions, with layer-specific enrichment of outcome encoding in the retrosplenial cortex (RSC) and of action values encoding in the anterior cingulate cortex (ACC). Outcome representation was enhanced in high-performance behavioral states but only weakly represented in low-performance states. Together, these results revealed specialized and state-dependent distributions of outcome, value, and switch representations across large-scale cortical circuits.