Machine learning reveals two neuroanatomical subtypes of adolescent depression with distinct white matter, functional, and cognitive profiles.

BACKGROUND: Major depressive disorder (MDD) is a clinically and neurobiologically heterogeneous disorder typically emerging in adolescence. Delineating MDD subtypes with distinguishing neurobiological substrates and clinical profiles is a promising strategy for guiding precise diagnosis and treatments. METHODS: In this study, we included 152 adolescents with first-episode MDD and 140 demographically matched healthy controls. All participants underwent brain imaging scans and comprehensive neurocognitive assessments. Based on brain morphometric features of gray matter and subcortical volume, the Heterogeneity through Discriminative Analysis (HYDRA) was conducted to identify potential MDD subtypes. We then compared clinical characteristics, morphometric features, white matter microstructure, functional connectivity, and neurocognitive performance between the subgroups and healthy controls to elucidate the profiles of identified subtypes. RESULT: Two robust distinct neuroanatomical subtypes were identified. Subtype 1 showed widespread cortical and subcortical gray matter reduction, manifesting as decreased cortical thickness, cortical volume, surface area, and subcortical volume, along with poorer white matter integrity. Subtype 2 exhibited preserved gray matter morphology and less pronounced white matter disruptions. Moreover, both subtypes demonstrated hypo- and hyperconnectivity among the visual, sensorimotor, default mode, and attention networks, and the differences were more pronounced in subtype 2. Furthermore, both subtypes showed significant cognitive impairments, and subtype 2 specifically performed worse in reward related emotion processing. CONCLUSION: This study identified two robust subtypes of adolescent MDD with divergent brain morphology, white matter, functional connectivity and cognitive profiles, reflecting the clinical and neurobiological heterogeneity inherent in this disorder. These findings hold promises for subtype-based approaches to advance diagnosis and treatment.