Subtyping psychotic disorders using a data-driven approach reveals divergent cortical and cellular signatures

Structural brain abnormalities in psychosis are well-replicated but heterogenous posing a barrier to uncovering the pathophysiology, etiology, and treatment of psychosis. To parse neurostructural heterogeneity and assess for the presence of anatomically-derived subtypes, we applied a data-driven method, similarity network fusion (SNF), to structural neuroimaging data in a broad cohort of individuals with psychosis (schizophrenia spectrum disorders (SSD) n = 280; bipolar disorder with psychotic features (BD) n = 101). SNF identified two transdiagnostic subtypes in psychosis (subtype 1: n = 158 SSD, n = 75 BD; subtype 2: n = 122 SSD, n = 26 BD) that exhibited divergent patterns of abnormal cortical surface area and subcortical volumes. Compared to controls (n = 243), subtype 1 showed moderate enlargement of surface area in frontal and parietal areas and larger dorsal striatal volumes, whereas subtype 2 demonstrated markedly smaller surface areas in frontal and temporal areas and subcortical volumes, including hippocampus, amygdala, thalamus and ventral striatum. When comparing subtypes on clinical characteristics, subtype 2 demonstrated more severe negative symptoms, greater neuropsychological impairment, and lower estimated premorbid intellectual functioning. Integrating cell-type data imputed from gene expression in the Allen Human Brain Atlas revealed an association between interregional glial cell and parvalbumin-expressing interneuron abundance and surface area expansion in subtype 1, yet cortical thinning in subtype 2. The largest deviations from controls were present in subtype 2 surface area that spatially coupled with layer 5 glutamatergic neuron abundance, critical for corticostriatal connectivity. These outcomes indicate early diverging neurodevelopmental trajectories of psychosis subtypes, as evidenced by opposing patterns in surface area, cellular fingerprints, and subcortical volumes.