Figure 6. History coding of retrosplenial cortex reflects ongoing behavioral strategy.

(A-B) Decoding accuracy for current and history information based on population activity of 138 neurons from each of the 4 cortical areas in early (A) or expert (B) sessions. Only longitudinally tracked neurons were included in this analysis. Ready period activity (between −1.9 and −0.1 s from go cue) was used for history decoding ((t-10) to (t-1)), but post-choice period activity (between 0 and 1 s from choice) was used for current trial information decoding (t). Bar graphs show mean ± s.e.m. Two-way ANOVA with Tukey’s post-hoc test for (t-10)~(t-2), One-way ANOVA with Tukey’s post-hoc test for (t-1) and (t). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. All error bars are s.e.m.

(C) The relationship between the behavioral sensitivity to value difference based on the RL model and population encoding of history events (t-1 to t-10 trials) in mean ready period activity (between −1.9 and −0.1 s from go cue) in the expert sessions from the 4 cortical areas. Only the population encoding of rewarded and unrewarded choice history by RSC significantly correlates with the ongoing behavioral strategy (Spearman correlation).

(D) The relationship between the behavioral sensitivity to value difference based on the RL model and population encoding of the current trial events in mean post-choice period activity (between 0 and +1 s from choice) in the expert sessions from the 4 cortical areas. None of the comparisons shows significant correlations (Spearman correlation).

See also Figure S6.