Abstract
<p>Social anxiety is reliably characterized by biases toward avoidance and aversive learning. Frontal-midline theta (FM-theta) oscillations have been implicated as a key neural mechanism underlying these learning biases. Here we combined a dynamic network neuroscience approach with computational modeling to investigate how FM-theta oscillatory activity coordinates large-scale functional network interactions on a trial-to-trial basis during feedback-guided learning. Additionally, we examined how these neural dynamics are altered in social anxiety across different social contexts. A sample of 146 human participants (129 female, mean age = 20.4 years) with varying levels of social anxiety completed a probabilistic selection task while performing alone and under scrutiny. Using non-negative matrix factorization, we identified two functional neural subgraphs supported by FM-theta. Fronto-parietal connectivity increased following aversive outcomes and scaled with positive prediction errors, consistent with error-monitoring and belief-updating. Fronto-occipital connectivity scaled with prediction errors when performing alone but not under scrutiny, indicative of context-sensitive disruption of belief-updating. Participants with elevated social anxiety showed increased fronto-occipital connectivity, as well as stronger subgraph coupling, indexing cross-network integration. This latter effect facilitated learning of simple stimulus-reward associations, but hampered learning of difficult stimulus-reward associations. Together, these findings reveal that social anxiety amplifies fronto-occipital FM-theta connectivity and cross-network integration, facilitating learning under low demand while hampering learning under high demand.</p>