Biological heterogeneity of melancholic, atypical, anxious, and psychotic depression.

Major depressive disorder is characterised by clinical and biological heterogeneity, yet evidence concerning melancholic, atypical, anxious, and psychotic depression is fragmented across biological systems and further complicated by heterogeneous phenotype definitions. This critical narrative synthesis integrated findings from studies using DSM criteria, the CORE measure, latent class analysis, and dimensional instruments across five domains: HPA-axis regulation, inflammation, neural circuitry and neuromodulatory systems, genetic liability, and circadian regulation. Across studies, biological associations varied by phenotype and domain; no phenotype was defined by a single abnormality. The most consistently supported components were marked HPA-axis hyperfunction in psychotic depression; hypercortisolaemia or impaired feedback in melancholic depression; immunometabolic activation in reversed-neurovegetative atypical profiles; and lower fractional anisotropy across fronto-limbic and association tracts in anxious depression. Support for other phenotype-associated components was narrower or less consistent, with many findings remaining preliminary, indirect, or operationalisation-dependent. Together, these domain-specific findings provide the evidence base for an integrative, hypothesis-generating model. Because the underlying data were derived largely from separate cohorts, the model synthesises phenotype-level evidence rather than demonstrating within-person configurations. Within this framework, the phenotypes are conceptualised as evidence-informed prototypes in a continuous multidimensional biological space; their positions, boundaries, and degrees of overlap remain to be determined in multimodal cohorts. The model predicts that mixed and intermediate presentations will show systematic combinations of alterations across these domains. Heuristic rather than taxonomic, the model organises heterogeneous evidence and frames questions for longitudinal and biomarker-stratified research without requiring a new clinical classification.