Abstract
How does the emergence of a neural representation of affective salience differ between individuals? Previous research has shown that with successful aversive conditioning visual representations of two conditioned stimuli can converge [1]. Here we used representational similarity analysis (RSA) of fMRI data to study individual differences in the changes that accrue to the representations of visual stimuli over the course of emotional learning. Participants were scanned during a Pavlovian conditioning task using both aversive and pleasurable reinforcers in a slow event-related design. In separate blocks, participants learned to associate two individual faces with either a painful pinch or pleasurable brush stroke (CS+), while a third face remained unreinforced (CS-). Faces were rated for likability before and after conditioning and participants were grouped into conditioners and non-conditioners based on changes in likability ratings. Ventromedial prefrontal cortex (VMPFC) and ventral visual cortex regions of interest (ROIs) were anatomically defined. For each ROI, multivariate representational similarity matrices were calculated using the patterns of activation for each stimulus presentation. Finally, inferential statistics were performed to evaluate similarity of adjacent trials within each stimulus category and of corresponding trials between different stimulus categories. Results showed that in the VMPFC, similarity of representations for CS+ stimuli steadily increased during learning for conditioners only. Furthermore, the representations of the two CS+ stimuli became more similar for conditioners than non-conditioners. This suggests such a convergence in stimulus representation is driven by affective processes. This pattern was not observed in the ventral visual ROI, suggesting a dissociation between how affect information accumulates over time in PFC compared to ventral visual regions. The nature of such affect-driven information accumulation was further probed through exploratory RSA on ROIs defined by specific resting state networks.
Meeting abstract presented at VSS 2017