Associative learning increases trial-by-trial similarity of BOLD-MRI patterns

Associative learning is a dynamic process that allows us to incorporate new knowledge within existing semantic networks. Even after years, a seemingly stable association can be altered by a single significant experience. Here, we investigate whether the acquisition of new associations affects the neural representation of stimuli and how the brain categorizes stimuli according to preexisting and emerging associations. Functional MRI data were collected during a differential fear conditioning procedure and at test (4-5 weeks later). Two pictures of faces and two pictures of houses served as stimuli. One of each pair coterminated with a shock in half of the trials (partial reinforcement). Applying Multivoxel Pattern Analysis (MVPA) in a trial-by-trial manner, we quantified changes in the similarity of neural representations of stimuli over the course of conditioning. Our findings show an increase in similarity of neural patterns throughout the cortex on consecutive trials of the reinforced stimuli. Furthermore, neural pattern similarity reveals a shift from original categories (faces/houses) toward new categories (reinforced/unreinforced) over the course of conditioning. This effect was differentially represented in the cortex, with visual areas primarily reflecting similarity of low-level stimulus properties (original categories) and frontal areas reflecting similarity of stimulus significance (new categories). Effects were not dependent on overall response amplitude and were still present during follow-up. We conclude that trial-by-trial MVPA is a useful tool for examining how the human brain encodes relevant associations and forms new associative networks.