Figure 5.

Amygdala and Noncontingent Learning

(A) In experiment 1, stronger amygdala responses to noncontingent rewards correlate with better contingent relative to noncontingent learning.

(B) Extraction of parameter estimates from the peak coordinate of the above contrast in (A) shows that amygdala responses to noncontingent rewards in experiment 1 correlate negatively with all three noncontingent learning mechanisms, albeit the correlation with SoE_Rew was more pronounced than that with either SoE_Ch or PROX (see main text). The correlations are partial correlations, that is, after regressing out the effects of the respective other learning mechanisms from both parameters of interest.

(C) Following a choice, the amygdala signal was suppressed (left). Amygdala activity after a choice (in anticipation of a contingent reward) correlated positively with SoE_Ch and SoE_Rew (right), meaning that a lack of amygdala suppression was associated with misassignment of the following reward via one of these noncontingent mechanisms.

(D) On a trial-by-trial level, credit for a reward following choice of A was likely to be misassigned to one of the noncontingent bins when amygdala activity was high in the period between choice and reward.

(E) In experiment 2, amygdala was exclusively reward sensitive in the FORWARD condition, the only condition where learning was only possible from spreading credit for a reward to the average choice history. The graph shows the evolution of a simple “reward-no reward” contrast over intratrial time as in Figure 4, taken from the peak coordinate from experiment 1 shown in (A).

Solid lines in (C) and (E) show the mean; shaded areas and error bars in (D) represent the SEM of the contrast estimates across subjects. The black vertical lines in (E) represent the time of choice and outcome delivery, respectively.