They were instructed to learn how to associate three CS shapes with three possible outcomes (40p, 0p, 40/0p), and two colors with either more or less

predictable reward timing. All subjects had learned the associations successfully after the training as shown in a brief questionnaire. However, one subject was excluded because he reported nonexistent changes in color-timing associations after scanning. The scanning session consisted of four experimental blocks learn more of 48 normal and 8 test trials each. The order of trials was randomized and different in each block. Subjects were paid according to the number of successful timing estimates given in test trials. More precisely, the sum of all rewards collected during the experiment (amounting to £30 if no trials were missed) was multiplied by the percentage of test trials in which the time they indicated was within 1 s of the true reward time. On average, subjects earned £15 on the task (min £5, max £26), and were paid an extra £10 for their participation. We carried out t tests and Kolmogorov-Smirnov tests on the timing estimates subjects gave in instrumental test trials. Comparisons were done both between and across groups. We acquired T2∗-weighted EPI images on a 3 T TRIO scanner (Siemens) using a 12-channel head coil. Each of the four blocks consisted of 237 volumes with 43 slices, a 70 ms echo time (TE), resulting in a repetition time (TR) of 3.01 s; the voxel size was 3 ×

3 × 3 mm, flip angle −30°. We used a sequence optimized for orbito-frontal and midbrain regions to minimize signal dropout. We also acquired a high resolution structural find more scan (1 × 1 × 1 mm; 176 partitions, FoV = 256 × 240, TE = 2.48 ms, TR = 7.92 ms, FA = 16°, TI = 910 ms, 50% TI ratio) and a field map (TE1 = 10 ms and TE2 = 12.46 ms, 3 × 3 × 2 mm resolution, 1 mm gap). During scanning peripheral measurements of subject pulse, breathing, Rolziracetam and skin conductance responses were made together with scanner slice synchronization pulses. FMRI analysis was implemented using FMRIB Software

Library (FSL) (Smith et al., 2004). Data were preprocessed using the default options in FSL: Images were motion corrected (Jenkinson et al., 2002) and unwarped using the acquired field maps. Brain matter was segmented from nonbrain (Smith, 2002) before applying Gaussian spatial smoothing with a 5 mm FWHM kernel. Images were high-pass filtered and registered to the high-resolution structural image (7 degrees of freedom) and then the standard MNI152 template using affine registration (12 degrees of freedom) (Jenkinson and Smith, 2001). Further to using a sequence that minimized signal drop-out in midbrain regions, we performed two steps to increase the sensitivity to BOLD responses in the midbrain. These steps were taken because the anatomical location of the VTA makes BOLD signals in the region exquisitely sensitive to both physiological noise and subject motion.