Asymmetric representation of aversive prediction errors in Pavlovian threat conditioning

Abstract

Learning to predict threat is important for survival. Such learning may be driven by differences between expected and encountered outcomes, termed prediction errors (PEs). While PEs are crucial for reward learning, the role of putative PE signals in aversive learning is less clear. Here, we used functional magnetic resonance imaging in humans to investigate neural PE signals. Four cues, each with a different probability of being followed by an aversive outcome, were presented multiple times. We found that neural activity only at omission - but not at occurrence - of predicted threat related to PEs in the medial prefrontal cortex. More expected omission was associated with higher neural activity. In no brain region did neural activity fulfill necessary computational criteria for full signed PE representation. Our result suggests that, different from reward learning, aversive learning may not be primarily driven by PE signals in one single brain region.