Fig. 5. Selection of functional coupling containing choice information across learning.

a Left. Hypothesis of selective retention and elimination of functional coupling across learning. Right. Schematic of functional couplings to the choice-encoding neuron (black) that are either retained (red) or eliminated (blue) across learning. Activity_retained and activity_eliminated are reconstructed activities with retained and eliminated coupling terms, respectively. b Fractions of retained and eliminated functional coupling from the intermediate to expert stage (***P < 0.001, n = 13 sessions from 7 mice, one-tailed bootstrap). Error bars indicate mean ± SEM. c Differences in decoding accuracy for individual choice-encoding neurons between activity_retained and activity_eliminated (*P < 0.05, ALM: 26; M1a: 42; M1p: 129; S1fl: 104; vS1: 85; M2: 24; RSC: 76; PPC: 105 neurons from 7 mice, one-tailed bootstrap). Error bars indicate mean ± SEM. d Choice selectivity computed by averaging pre-action population activity projected to the choice axis (1 s in duration, starting 1 s prior to the go cue, ***P < 0.001, n = 13 sessions from 7 mice, one-tailed bootstrap). Red cross indicates mean ± SEM. Inset: session-averaged reconstructed population activity prior to the go cue with retained (red) or eliminated (blue) coupling projected to the choice axis.