Latent Brain-Symptom Dimensions Characterize Heterogeneity in Clinical High Risk for Psychosis

BACKGROUND: Clinical high risk (CHR) for psychosis is characterized by substantial heterogeneity in symptoms, functioning, and outcomes. Although distributed resting-state functional connectivity (RSFC) abnormalities have been reported, it remains unclear whether this heterogeneity can be organized into consistent multivariate brain-symptom dimensions and clinically meaningful subgroups. METHODS: We analyzed RSFC and 34 clinical variables from 701 CHR participants in the Accelerating Medicines Partnership Schizophrenia Program (AMP SCZ). Partial least squares analysis identified latent brain-symptom dimensions, which were further examined using transcriptomic annotation, independent psychosocial and cognitive measures, hierarchical clustering, and Month 2 follow-up data. RESULTS: Two significant brain-symptom dimensions emerged. The first reflected general psychopathology and was associated with somatomotor-salience and higher-order connectivity, with enrichment for genes related to neuronal development. The second, capturing symptom differentiation, spanned negative/internalizing features with psychobehavioral overactivation and involved cerebellar-thalamic-somatomotor circuitry, with enrichment for genes related to synaptic signaling. These dimensions showed dissociable associations with broader psychosocial and cognitive measures, and their loading patterns were substantially reproducible at Month 2. Clustering based on these dimensions identified two biotypes with distinct clinical and connectivity profiles that remained broadly similar at Month 2 despite divergent short-term RSFC changes. CONCLUSIONS: These findings support complementary dimensional and categorical representations of CHR heterogeneity that link clinical variation with distributed brain connectivity and molecular signatures, while demonstrating short-term reproducibility of the identified dimensions and divergent RSFC changes across biotypes. This multilevel framework may help characterize biologically meaningful variation in psychosis risk.