602. Neurobiological subtyping of major depression identifies distinct cardiometabolic risk profiles in UK Biobank

Abstract

Abstract Background Major depressive disorder (MDD) is a clinically and biologically heterogeneous condition associated with a high rate of somatic morbidity and mortality, largely driven by cardiometabolic comorbidities [1]. Elucidating the mechanisms underlying this elevated risk remains challenging, as MDD spans wide inter-individual variability across neuroanatomical, microstructural, functional, and genetic domains [2]. Data-driven stratification approaches may therefore identify biologically meaningful subtypes with differential vulnerability to medical complications, and inform personalized prevention and treatment strategies [3]. Aims & Objectives We integrated multimodal neuroimaging and polygenic scores (PGSs) with unsupervised and supervised machine-learning techniques to (i) derive neurobiologically informed subgroups of individuals with lifetime MDD and (ii) identify subtype-specific biological features associated with cardiovascular diseases and diabetes risk. Method The study included 3,887 UK Biobank participants with lifetime MDD who underwent standardized multimodal neuroimaging and genotyping. Neuroimaging features encompassed cortical thickness, surface area, grey-matter volumes, diffusion tensor metrics (fractional anisotropy, axial and radial diffusivity), resting-state functional connectivity strengths, and task-related activation during emotional processing. PGSs for psychiatric, metabolic, and inflammatory traits were computed using LDpred2-auto [4]. Consensus clustering [5] was applied to identify homogeneous patient subgroups. K-means clustering was iterated 1,000 times for clustering solutions ranging from k=2 to k=5, with each iteration sampling 80% of participants. The optimal cluster number was determined by internal validity indices, while statistical significance was assessed via 10,000 Monte Carlo simulations. Between-cluster differences in biological and clinical variables were tested using the Wilcoxon rank-sum and χ2 tests, correcting for False Discovery Rate. Within each cluster, nested cross-validated elastic-net models predicted cardiovascular disease and diabetes, accounting for class imbalance, computing model significance via 1,000 permutations, and estimating feature stability through 5,000 non-parametric bootstrap samplings. Results Two statistically significant clusters emerged (Cluster0: N=2,030; Cluster1: N=1,857; p<0.001). One subgroup (Cluster1) exhibited widespread cortical thinning (q<0.001), reduced cortical and subcortical volumes (q<0.001), diffuse white-matter microstructural abnormalities (q<0.001), diminished resting-state connectivity (q<0.001) and emotional-task activation (q<0.05), and elevated PGSs for MIF (q=0.047) and RANTES (q=0.046), with respect to the other. Clinically, the same cluster showed a higher prevalence of cardiovascular diseases (q=0.013) and diabetes (q=0.021), alongside poorer motor (q=0.015) and processing (q=0.030) speed performance. Elastic-net models revealed distinct cardiometabolic risk signatures across clusters (q=0.009-0.047). While genetic liability for type 2 diabetes contributed to diabetes risk in both subgroups, white-matter diffusivity measures were most influential in the less neurobiologically compromised cluster (Cluster0), whereas grey-matter structural metrics predominated in the more affected subgroup (Cluster1). Cardiovascular disease prediction involved multimodal features in both clusters, but genetic and neuroimaging contributions were broader and stronger in the more impaired subtype (Cluster1). Discussion & Conclusions This large-scale, multimodal, data-driven analysis identified two neurobiologically distinct MDD subtypes with distinct cardiometabolic vulnerability profiles. The subtype characterized by more marked structural, microstructural, functional, and inflammatory-genetic alterations showed increased cardiometabolic burden and cognitive slowing, aligning with immuno-metabolic models of depression. Importantly, the biological pathways linking MDD to cardiometabolic disease differed across subtypes, underscoring the value of stratified approaches for advancing precision psychiatry and targeting preventive and therapeutic interventions.