Behavioral data

We quantified attentional bias effects using inverse efficiency (IE; RT/accuracy), a measure that accounts for speed-accuracy tradeoffs (Townsend and Ashby, 1983). RTs below 100 ms were excluded from further analyses; responses that were too slow (>800 ms) were coded as incorrect, which were consistent with previous study utilizing these paradigms (Anderson et al., 2011; Failing and Theeuwes, 2014). For the training task, a two-way repeated-measures ANOVA compared performance inverse efficiency (IE) for rewarded and unrewarded targets between both sessions. For the testing task, a one-way repeated measures ANOVA comparing performance (IE) for invalid PR, neutral, and valid PR trials assessed attentional bias, including facilitated capture for the valid trials and impaired disengagement for the invalid trials. A follow-up one-way repeated measures ANOVA was conducted for the PU trials to determine whether attentional bias is specific to reward or due to general experience. Similar analyses performed on RT and accuracy independently are reported in Supplementary material.

Activity within the striatum during the training task for rewarded>unrewarded correct trials was extracted for each subject individually using FSL’s feat query function. The ROI sphere was created by drawing a 4-mm sphere around the peak activation within the right caudate observed at the group level (coordinates: x = 8, y = 18, and z = 4, Supplementary Table S2). Then, a group-level GLM was constructed in which the extracted beta weight values of striatal activity from the training task (i.e., predictor variable) were associated with neural activity in the whole brain during the testing task. Finally, to examine whether striatal activity predicts attentional bias to reward in the testing task, we used a non-parametric Spearman correlation between striatal activity during the training phase and attentional bias on PR trials during the testing task (Rousselet and Pernet, 2012).