Texture similarity network uncover schizophrenia subtypes with unique molecular signatures

Abstract

Abstract Schizophrenia manifests complex heterogeneity across multiple dimensions, posing major challenges for precise diagnosis and treatment. Developing objective biomarkers to stratify patients into stable subtypes is essential for heterogeneity resolving and precision medicine. This study aimed to identify neuroimaging subtypes of schizophrenia through a novel individualized radiomic-based texture similarity network (TSN) approach and investigate the biological signatures of these subtypes. A total of 437 individuals with schizophrenia (male: 263, female: 174) and 431 healthy controls (male: 246, female: 185) were recruited from four sites using five scanners. K-means clustering identified two schizophrenia TSN-subtypes, which were validated across different samples, datasets, and disease stages. The subtypes exhibited distinct TSN disruption patterns, primarily involving the prefrontal-sensorimotor, prefrontal-limbic, subcortical-sensorimotor, subcortical-limbic, and within-subcortical networks. Additionally, these disruption patterns of the two subtypes were differentially associated with the expression of schizophrenia risk genes, which are enriched in protein binding, nervous system development, and neuron-specific structures. Finally, the density distributions of several neurotransmitter receptors, including M1, 5HT6, and mGluR5, showed broadly diverse associations with TSN disruption between subtypes. In summary, the TSN-defined subtypes are highly reproducible and reflect distinct neurobiological mechanisms at both the network and the molecular levels, highlighting divergent biological origins between subtypes.