BACKGROUND Individuals at clinical high risk for psychosis (CHR) present with heterogeneous combinations of positive, negative, and affective symptoms, cognitive difficulties, and functional impairment. However, reproducible multivariate links between this multidomain phenotype and brain morphology remain limited. This study explored the multivariate brain-behavior associations in CHR using a data-driven approach. METHODS Baseline data from 720 CHR participants from the Accelerating Medicines Partnership® Schizophrenia program (AMP® SCZ) were analyzed using sparse canonical correlation analysis to identify associations between 95 regional brain volumes and 28 behavioral variables indexing symptoms, cognition, and functioning. Generalizability was tested using leave-one-network-out cross-validation (PRoNET and PRESCIENT), and stability was assessed using bootstrapping and leave-one-site-out analyses. Associations with demographic, clinical and site-related variables were examined. RESULTS Two brain-behavior dimensions generalized across networks. A positive-affective axis (r=0.21-0.23), linking higher depression, anxiety/stress, positive symptoms and anhedonia with predominantly frontal-cingulate-insular volume reductions and relatively larger corpus callosum and putamen volumes. This axis was associated with comorbid depression, cannabis use and family history of depression or affective psychosis. A negative-cognitive axis (r=0.17-0.18) linked greater asociality, alogia, emotional withdrawal, blunted affect, and motor retardation, poorer cognition and functioning with reduced cerebellar and parietal volumes, and relatively larger lateral ventricles and corpus callosum. This axis was associated with lower IQ and poorer premorbid adjustment. CONCLUSIONS Two partially dissociable brain-behavior dimensions were identified: a positive-affective axis capturing the co-expression of affective and attenuated positive symptoms, and a negative-cognitive axis suggestive of neurodevelopmental vulnerability.
