Fig. 4: Increased single unit representation of the new rule with residual encoding of the irrelevant rule.

A Delta z-score of individual units in rewarded vs. non-rewarded trials, plotted for the non-match rule (X-axis) and the cue-identity rule (Y-axis), during the first (left) and last (right) sessions of problem A under the non-match rule. Marginal axes display the distributions of delta z-scores for each rule. B Proportions of neurons showing significant selectivity, for the non-match rule (blue), cue-identity rule (yellow), or both rules (gray) across different sessions of problem A under the non-match rule. C Same as (A), but for problems (A) (top) and (B) (bottom) under the cue-identity rule. Blue markers represent individual units from rats that underwent the full learning curriculum and first learned the non-match rule, while gray markers correspond to units from the control group, which was trained only on the cue-identity rule. Marginal axes show the distributions of delta z-scores for the rule-switching group (blue) and control group (gray). (*p < 0.05, **p < 0.01, Kolmogorov–Smirnov test comparing the distributions of delta z-scores in the two groups; see Supplementary Table 4). D Same as (B), but for sessions under the cue-identity rule, comparing the rule-switching group (dark bars) and control group (light bars). A two-way ANOVA revealed a significant main effect of group for the non-match rule responsive units (F(1,67) = 14.46, p < 0.001) and for the both rule responsive units (F(1,67) = 6.99, p = 0.010), indicating that the control group had significantly fewer units that were selectively responsive to this irrelevant rule compared to the rule-switching group. In contrast, there was no significant difference in the proportion of units that were selective for the relevant cue-identity rule (F(1,67) = 0.04, p = 0.841; see Supplementary Table 5).