Because VTA is especially susceptible to physiological noise, its Ivacaftor solubility dmso signal variance was greatly reduced following the removal of noise components (Figure S2A). Second, a physiological noise model was constructed using an in-house developed MATLAB toolbox (Hutton et al., 2011). Models for cardiac and respiratory

phase and their aliased harmonics were based on RETROICOR (Glover et al., 2000). The model for changes in respiratory volume was based on (Birn et al., 2006). This resulted in 17 regressors, separate ones for each slice: 10 for cardiac phase, 6 for respiratory phase, and 1 for respiratory volume. We generated these 17 regressors once with respect to every slice (n = 43 slices) to maximize their sensitivity for different slice acquisition times. To match the voxelwise input format required by FSL, each of the 17 regressors was formatted as a four-dimensional volume with identical regressors

for voxels within the same slice, but different regressors BAY 73-4506 across voxels of different slices. This resulted in 17 regressors with the following dimensions: 64 (voxels in x) × 64 (voxels in y) × 43 (slices) × 234 (volumes), importantly differing only in the “slice” and “volume” dimensions. Regressors were included in the general linear model (GLM) that led to a further reduction of the Phosphatidylinositol diacylglycerol-lyase signal variance in VTA (Figure S2B). Temporal difference models predict different patterns of dopaminergic activity in the two groups. For creating the regressors to include in the GLM, we used a hazard function, reflecting

the probability that a reward will occur at time t given that it has not yet occurred rP(t)dt(1−r)+r(1−∫0tP(t)dt),where P is a γ distribution with a mean of 6 and a standard deviation of 1.5 from which the CS-US intervals were drawn ( Figure 1). We varied the parameter r to be r = 0.5 to predict the situation when only half of the outcomes were shown (groupU), and r = 1 for when all outcomes were shown (groupS). This led to the predictions shown in Figure 3A. In groupU, where the most likely time for a reward delivery is the mean delivery time, the BOLD RPE response is predicted to be large for early and late, but smaller for midtime unexpected rewards. In groupS, it becomes more likely as time passes that each new time bin will contain a reward. The RPE signal is therefore expected to be largest for early, and smallest for late unexpected rewards. The GLM included 47 regressors in groupS and 39 regressors in groupU.