Scientists Trace Social Anxiety to a Hidden Brain Wiring Pattern

Social anxiety is one of the most common psychiatric conditions worldwide, yet its biological roots have remained frustratingly elusive. Decades of brain imaging studies have pointed to the amygdala, the insula, and prefrontal circuits, but the findings have been scattered and often contradictory. Now, a team at the University of Münster has taken a different route: instead of asking which single brain region misfires in social anxiety, they mapped the brain’s entire wiring diagram, the structural connectome, and searched it for a network that predicts how socially anxious a person is. Their results, published in Translational Psychiatry, offer the first connectome-wide look at social anxiety as a continuous trait rather than a binary diagnosis.

The research hinges on a conceptual shift that has been gaining momentum in psychiatry. Rather than comparing healthy controls against patients with social anxiety disorder, the researchers treated social anxiety as a spectrum that cuts across mental health conditions. Trait social anxiety, a stable disposition to feel fear or discomfort in social and performance situations, appears at every level of severity and frequently co-occurs with conditions such as major depressive disorder. By studying it dimensionally, the team hoped to identify neural signatures shared across the full range, from subclinical shyness to diagnosable disorder, rather than signatures unique to one diagnostic category.

To do this, the investigators drew on the Transdiagnostic Investigation of Emotional Processing in Depression and Anxiety cohort at the University of Münster. One hundred sixty participants between the ages of 18 and 65 underwent structured clinical interviews, completed the Liebowitz Social Anxiety Scale, and received magnetic resonance imaging. Sixty-four of them met diagnostic criteria for social anxiety disorder, and nearly half of those also carried a diagnosis of acute or partially remitted depression. The remaining ninety-six participants had no current social anxiety disorder but spanned a wide range of social anxiety scores, including forty-nine people with no lifetime psychiatric history at all.

The imaging protocol combined two techniques. High-resolution T1-weighted scans parcellated each participant’s cortex into 114 regions using the Cammoun subdivision of the Desikan-Killiany atlas, while diffusion tensor imaging traced the white matter fiber tracts linking those regions. After motion and eddy-current correction, the researchers used the RESTORE algorithm for robust tensor estimation and deterministic tractography to reconstruct the fibers, retaining only connections supported by three or more streamlines. The result was a structural connectivity matrix for every participant, a numerical portrait of how each brain region is physically wired to every other.

Onto these matrices the team applied connectome-based predictive modeling, a data-driven machine learning framework that links individual differences in brain connectivity to behavior. In a refinement of the standard approach, they used a repeated cross-validation scheme with ten inner and one hundred outer folds, selecting edges whose connectivity strength correlated significantly with social anxiety scores and then applying the Network-Based Statistic to keep only the largest connected component of those edges. Hyperparameters were tuned in an inner loop, with the model choosing between linear regression and support vector regression, and age and sex were controlled as covariates. Prediction accuracy was evaluated out-of-sample and tested against a null distribution generated by shuffling the outcome labels thousands of times.

The search paid off, though modestly. The analysis identified a network of just forty-six edges, a mere 0.18 percent of the entire connectome, that significantly predicted Liebowitz Social Anxiety Scale scores with a correlation of 0.141. Most of these edges were positively associated with social anxiety, and the network’s strongest connection ran between the left lateral occipital cortex and the lingual gyrus, deep in the visual processing system. Negative connections linked the fusiform, inferior temporal, and occipital regions, hinting at decoupling among higher-order visual areas. When the analysis was repeated on the clinical subsample of 111 participants, excluding the healthy controls, the network grew to fifty-nine edges and prediction improved to a correlation of 0.172, and the same pattern held when a different cortical parcellation was used.

That visual-temporal core is the study’s most intriguing clue. The occipital, lingual, and fusiform regions are central to perceiving faces, bodies, and social cues, and previous work has consistently documented hypervigilance to social signals in socially anxious individuals. Altered structural coupling among these areas could reflect differences in how socially relevant visual information is processed and integrated, a key component of social anxiety. The authors caution, however, that the network’s limited stability means the findings should be read as exploratory network-level associations rather than definitive markers, and that larger samples with independent validation are needed before any clinical use.

Follow-up analyses probed the network’s internal architecture. Using a preregistered hub definition that required a node to rank in the top 15 percent of both degree and betweenness centrality, no hubs emerged, likely because the network was too small and its edges too evenly distributed. An exploratory, more lenient definition based on degree alone pointed to hubs in the left and right precuneus and the right superior parietal cortex, regions tied to visuospatial attention and perspective taking. The researchers speculate that these areas could support the heightened monitoring of others’ positions, movements, and postures characteristic of social anxiety, but they emphasize that these are candidate regions at best, not established neural markers.

The team also compared whole-brain graph metrics between the social anxiety disorder group and everyone else. The groups differed in only one measure: the number of modules, which was higher in the disorder group, suggesting a more segregated network architecture. But when social anxiety was treated as a continuous score, a richer picture appeared. Higher trait social anxiety was associated with more connections, longer path lengths, greater small-worldness, and reduced efficiency and streamline counts, a pattern the authors interpret as increased connectivity alongside less efficient information transfer across the brain. Such a shift in the balance between network integration and segregation could affect how emotional regulation, attention, and social cognition communicate, though the authors note these interpretations remain tentative because graph metrics depend heavily on how the network is constructed.

The study has clear limits. The sample included only healthy controls, social anxiety disorder, and depression, with no dedicated subclinical group, and participants showed generally high global functioning, so the findings may not extend to more severely impaired populations. Comorbid depressive symptoms may have shaped the identified connectivity pattern, raising questions about specificity. The effect sizes, while statistically significant and comparable to others in psychiatric neuroimaging, are small, and the authors explicitly state the network should not be considered an established biomarker or clinically applicable at this stage. Still, as the first connectome-wide investigation of trait social anxiety in a transdiagnostic sample, the work marks a genuine step forward. If future studies in larger, more diverse cohorts confirm the visual-temporal signature, structural brain networks could eventually help identify people at risk for social anxiety early and point clinicians toward the mechanisms most worth targeting.

Subject of Research: Structural connectome signatures of trait social anxiety in a transdiagnostic sample

Article Title: Identifying a structural brain network for social anxiety: connectome-based predictive modeling and network analyses in a transdiagnostic sample

Article References: Meinert, H., Gruber, M., Brügge, E. D. E., Brüwer, L., Sindermann, L., Winter, A., Slump, T. M., Breuer, F., Schrammen, E., Meinert, S., Goltermann, J., Grotegerd, D., Repple, J., Redlich, R., Bonnekoh, L. M., Bauer, J., Dannlowski, U., Böhnlein, J., & Leehr, E. J. (2026). Identifying a structural brain network for social anxiety: connectome-based predictive modeling and network analyses in a transdiagnostic sample. Translational Psychiatry, 16(1), Article 495. https://doi.org/10.1038/s41398-026-04466-9

Image Credits: AI Generated

DOI: 10.1038/s41398-026-04466-9

Keywords: social anxiety, structural connectome, connectome-based predictive modeling, diffusion tensor imaging, transdiagnostic psychiatry, graph theory, social anxiety disorder, white matter connectivity, biomarkers, Translational Psychiatry, network neuroscience, Identifying