The functional neurobiology of negative affective traits across regions, networks, signatures, and a machine learning multiverse

Understanding the neural basis of negative affective traits like neuroticism remains a critical challenge across psychology, neuroscience, and psychiatry. Here, we investigate which level of brain organization--regions, networks, or validated whole-brain machine-learning signatures--best explains negative affective traits in a community sample of 458 adults performing the two most widely used affective fMRI tasks, viewing emotional faces and scenes. Neuroticism could not be predicted from brain activity, with Bayesian evidence against all theory-guided neural measures. However, preregistered whole-brain models successfully decoded vulnerability to stress, a lower-level facet of neuroticism, with results replicating in a hold-out sample. The neural stress vulnerability pattern demonstrated good psychometric properties and indicated that negative affective traits are best represented by distributed whole-brain patterns related to domain-general stimulation rather than localized activity. Together with results from a comprehensive multiverse analysis across 14 traits and 1,176 models-- available for exploration in an online app--the findings speak against simplistic neurobiological theories of negative affective traits, highlight a striking gap between predicting individual differences (r<.35) and within-person emotional states (r=.88), and underscore the importance of aligning psychological constructs with neural measures at the appropriate level of granularity.