Final Report Walter Benjamin Project Dr. Gina-Isabelle Henze

Acute psychosocial stress is a key transdiagnostic risk factor for mental disorders. Although psychosocial stress can be experimentally induced in the MRI environment (e.g., ScanSTRESS, MIST), the reported brain response patterns have been heterogeneous across studies, complicating the identification of robust neurobiological stress markers and limiting translation to (pre-)clinical contexts. The overarching aim of this Walter Benjamin project was therefore to identify normative neural stress response patterns and to establish a methodological framework for quantifying how strongly individuals deviate from these patterns. The project comprised three work packages (WPs). In WP1, we conducted a systematic synthesis of the literature and performed a preregistered, harmonized mega-analysis of ScanSTRESS datasets (N=459) using a unified preprocessing and modelling pipeline. This work refined the “triple network” framework of stress processing by demonstrating that psychosocial stress reliably engages all three major brain networks — salience (SN), default mode (DMN), and central executive (CEN) — in a differentiated manner. SN and DMN responses were primarily linked to endocrine (cortisol), cardiovascular (heart rate), and affective (negative affect) stress responses, whereas CEN and DMN responses were associated with task performance. In addition, we identified an age-related effect (increasing DMN activation with age) and replicated an exposure-time effect (decreasing activation during repeated stress exposure), suggesting candidate markers of stress regulation and resilience. In WP2, the project expanded strategically toward large-scale structural brain–endocrine associations, motivated by the observation that robust multi-site evidence on brain structure and acute cortisol reactivity was scarce. In a large structural MRI study (N=291), we showed that smaller cingulate cortex surface area — particularly within anterior cingulate subregions — is associated with stronger cortisol stress responses, supported by complementary confirmatory, whole-brain, and machine learning approaches. A second preregistered study (N=281) used trajectory modelling to demonstrate that limbic structure and function predict distinct cortisol and affect response profiles, highlighting meaningful interindividual heterogeneity beyond average stress responses. Pre- versus post-stress resting-state functional connectivity analyses are being preregistered and will be conducted next, building directly on empirically derived network hypotheses from WP1. In WP3, we implemented voxel-wise normative modelling of task-evoked stress responses across nine sites (N=696). The models achieved excellent out-of-sample performance and demonstrated that robust group-level stress signatures can coexist with substantial person-level variability: up to ~10% of individuals showed extreme deviations in specific cortical areas. Applying deviation maps to a subclinical burnout sample indicated that while group-level differences were subtle, person-level deviation patterns carried discriminative information. Overall, the project delivered large-scale, harmonized evidence on brain stress processing and established a person-centered normative modelling framework. These advances provide an empirical and methodological foundation for follow-up work aiming to interpret stressrelated alterations in clinical populations as deviations from normative stress regulation.