This was done separately for the SP and IP data. We then averaged the forgetting scores of the

two tests to get our index of forgetting. A 3T Siemens TIM Trio MRI scanner was used for acquisition of T2∗-weighted echoplanar images (64 × 64; 3 × 3 mm pixels; 3 mm thick, oriented to the AC-PC plane; TR: 2 s; TE: 30 ms; flip angle 78°; 133 volumes for each of the six sessions). Additionally, MPRAGE structural images were acquired (256 × 240 × 192; 1 mm3 isotropic voxels; TR: 2,250 ms; TE: 2.99 ms; flip angle 9°). Data were analyzed using SPM8 (http://www.fil.ion.ucl.ac.uk/spm/software/spm8). The volumes were realigned, corrected for different slice acquisition times, and coregistered with the structural images. These were spatially normalized and the resulting parameters served to normalize the functional SNS-032 purchase images into 3 × 3 × 3 mm3 cubic voxels by fourth degree B-spine interpolation (using the Montreal Neurological Institute reference brain). The images were then smoothed by an isotropic 8 mm FWHM Gaussian kernel. The variance in BOLD signal was decomposed in a general linear model (Friston et al., Fulvestrant mouse 1995), separately for each run. Delta functions coded the time point of reminder onsets, separately for suppress and recall events. These regressors included only those reminders whose associates had successfully been learned. Reminders for the remaining

items were coded by two additional regressors (one for each condition). A further delta function coded transient changes associated with block onset. All of those regressors were convolved with the canonical hemodynamic response function. The full model additionally comprised regressors representing the mean over science scans and residual movement artifacts. A 1/128 Hz high-pass filter was applied to the data and the model. Parameters for each regressor

were estimated from the least-mean-squares fit of the model to the data. To test our a priori predictions, we extracted contrast estimates from ROIs. These were spheres (r = 5 mm) centered on the peak coordinates discussed in the Introduction (X, Y, Z: right DLPFC: 32, 38, 26, Anderson et al., 2004; left mid-VLPFC: −50, 25, 14, Badre and Wagner, 2007; left cPFC: −52, 9, 24, Wimber et al., 2008). For the HC, we used the anatomical mask of the WFU pickatlas (Maldjian et al., 2003). To test the putative retrieval inhibition network supporting direct suppression, we modeled the effective connectivity between DLPFC and HC using DCM10. DCM explains regional effects in terms of dynamically changing patterns of connectivity during experimentally induced contextual changes (Friston et al., 2003). Importantly, this method allows inferences about the direction of causal connections, i.e., whether suppress events modulate the “top-down” connection from DLPFC to HC versus the reverse “bottom-up” connection. Therefore, we defined a standard model including both regions as nodes with bidirectional, intrinsic connections and within-region inhibitory autoconnections.