White Matter Stratification in Depression Predicts Multidimensional Antidepressant Responses

Background Major depressive disorder (MDD) is clinically heterogeneous, posing a persistent challenge for personalized treatment. While neuroimaging offers a promising path, existing symptom-based stratification schemes have proven inadequate in predicting antidepressant response. Crucially, studies focusing on white matter (WM) heterogeneity — a potential source of neurobiological subtypes— have failed to address this critical gap. Here, we bridge this divide by investigating WM-based MDD subtypes and their predictive value for treatment outcomes. Methods We used non-negative matrix factorization biclustering of diffusion MRI data from 311 MDD patients (discovery: n=209; validation: n=102) to identified neuroanatomical subgroups with distinct WM microstructural signatures. Subgroups were characterized via neuroanatomical profiling, clinical phenotyping (symptom domains/treatment responses), and WM-symptom associations. Baseline WM features predicted 4-week treatment outcomes (overall/dimension-specific symptom reduction) across five antidepressant therapies using support vector regression. Results Three robust MDD subgroups emerged: (1) frontoparietal-corticospinal alterations linked to anxiety/hopelessness; (2) cerebellar-visual circuit disruptions tied to cognitive-psychomotor deficits; (3) fornix-centered abnormalities associated with attenuated symptom severity. Subgroup-specific WM networks predicted treatment outcomes with high cross-cohort consistency (discovery: r=0.24–0.58; validation: r=0.27–0.67; all p