Reward Prediction Error Signaling during Reinforcement Learning in Social Anxiety Disorder is altered by Social Observation

ABSTRACT BACKGROUND Dysfunctional expectations of impending social or performance outcomes are core features of Social Anxiety Disorder (SAD) but often lack formal definition in clinical research. Reinforcement learning (RL) models offer a framework to define changes in outcome expectations in a formal way by computing the prediction error (PE). This study quantifies the updating of expectations by PEs in SAD and investigates alterations in RL regions associated with PE signaling. METHOD 48 adult participants (24 diagnosed with SAD and 24 age-, gender-, and education- matched healthy controls (HC)) underwent event-related functional magnetic resonance imaging while learning from probabilistic feedback. Crucially, both groups completed two parallel versions of the task: one in which they learned under scrutiny (social observation) and one in which they learned without being overtly evaluated (non-social control condition). RESULTS Coupling to prediction errors in SAD was elevated in dorsomedial prefrontal cortex (DMPFC) when learning under observation. These findings provide the first evidence that PE signaling during social performance situations in SAD is associated with hypersensitive response signatures in DMPFC, a brain region associated with using others’ value standards as a proxy for one’s own value standards. Dynamic Causal Modelling further revealed that RPE-modulated connectivity from ventral pallidum to DMPFC during observation was reduced in SAD. CONCLUSIONS The present results corroborate a crucial role of DMPFC in SAD, which corresponds to dysfunctional expectations about others’ alleged performance standards that play a prominent role in current models of the disorder.