Polygenic prediction of fear learning is mediated by brain connectivity

BackgroundGenetic variants may impact connectivity in the fear network such that genetically driven alterations of network properties (partially) explain individual differences in learning. Our aim was to identify genetic indices that predict physiological measures of fear learning mediated by MRI-based connectivity. MethodsWe built prediction models using exploratory mediation analysis. Predictors were polygenic scores for several psychological disorders, neuroticism, cross-disorder risk, cognitive traits, and gene expression-based scores. Candidate mediators were structural and functional connectivity estimates between the hippocampus, amygdala, dorsal anterior cingulate, ventromedial prefrontal cortex and cerebellar nuclei. Learning measures based on skin conductance responses to conditioned fear stimuli (CS+), conditioned safety cues (CS-), and differential learning (CS+ vs. CS-), for both acquisition and extinction training served as outcomes. ResultsReliable prediction of learning indices was achieved by means of conventional polygenic score construction but also by modelling cross-trait and trait-specific effects of genetic variants. A latent factor of disorder risk as well as major depressive disorder conditioned on other traits were related to the acquisition of conditioned fear. Polygenic scores for short-term memory showed an association with safety cue learning. During extinction, genetic indices for neuroticism and verbal learning were predictive of CS+ and differential learning, respectively. While mediation effects depended on connectivity modality, prediction of fear involved all regions of interest. Expression-based scores showed no associations. ConclusionsOur findings highlight the utility of leveraging pleiotropy to improve complex trait prediction and brain connectivity as a promising endophenotype to understand the pathways between genetic variation and fear expression.