Supplementary Figure 10: PFC and HPC neuronal trial outcome signals are distinct from subcortical reward prediction error signals

(a) Roughly equal numbers of neurons show stronger activity for correct and incorrect outcomes. Plots show the percent of neurons in PFC (red; n = 319) and HPC (blue; n = 199) with a significant preference for correct (Cor) or incorrect (Inc) outcomes (P < 0.05, 2-sided permutation test). Values are plotted separately for early, mid, and late learning stages (bottom to top), and time epochs capturing transient (left; 100–500 ms) and sustained (right; 600–1350 ms) response components. These results are in contrast to previous results from the ventral tegmental area and lateral habenula, which show strong biases toward positive and negative reward prediction errors (roughly, uncertain correct and incorrect outcomes), respectively. (b) No robust transfer of outcome signals to earlier trial events. Population mean percent variance explained by trial outcome (correct vs. incorrect) is plotted as a function of time during (left) and after (right) the trial, separately for PFC and HPC and learning stages (see legend). Tick marks at bottom indicate time points with significant explained variance during the late learning stage (P < 0.05, uncorrected, 1-sample bootstrap test). In the ventral tegmental area and lateral habenula, as reward becomes more predictable during learning, activation shifts from the post-response outcome feedback epoch to earlier trial events predictive of reward. In contrast, post-response trial outcome information in PFC and HPC is present throughout learning, with little shift to earlier time points. These properties, and the learning-related shift from bias toward encoding incorrect to correct outcomes in HPC (main text Fig. 3c), distinguish the outcome signals we report from the static reward prediction error signals found in areas such as the ventral tegmental area and lateral habenula.