Identification and independent validation of white matter subtypes in major depressive disorder and their role in predicting treatment response.

Major depressive disorder (MDD) is a complex psychiatric disorder characterized by diverse clinical profiles and variable treatment responses. Parsing the neuroanatomical heterogeneity among patients with MDD holds great promise for improving the predictive accuracy for treatment outcome and developing more effective treatments. In this study, we utilized a semi-supervised machine learning method called heterogeneity through discriminative analysis (HYDRA) to delineate patterns of neuroanatomical heterogeneity characterized by fractional anisotropy (FA) from 48 white matter bundles within a discovery dataset (130 patients with MDD and 128 healthy controls [HC]). Among these patients, 92 MDD patients underwent an 8-week antidepressant treatment. We further investigated the generalizability of these identified subtypes in an independent dataset comprising 84 MDD patients. Two stable subtypes of patients with MDD were identified (p < 0.05, ARI = 0.878). Subtype 1 was defined by widespread increases in FA. Subtype 2 was characterized by more severe depressive symptoms, an earlier age of onset, decreased FA, and decreased resting-state functional connectivity. Furthermore, differentiating these subtypes improved the accuracy of prediction of remission following antidepressant treatment. The independent dataset reproduced the subtype-related white matter and resting-state functional connectivity patterns, supporting the biological reproducibility of the identified subtype framework. Our study identifies two neuroanatomically distinct MDD subtypes and suggests that subtype-specific analyses may improve treatment-response prediction in a subset of patients (p = 0.012). These findings support the presence of neurobiological heterogeneity in MDD and provide a preliminary framework for future studies of pathophysiology and treatment response.