SEPARATING DEPRESSION FROM ADHD: IMPROVING GENOMIC LOCUS DISCOVERY AND SPECIFITY IN DSM-DEFINED MAJOR DEPRESSIVE DISORDER USING GENOMIC STRUCTURAL EQUATION MODELLING

Background GWAS of major depressive disorder (MDD) increasingly rely on minimal phenotyping (e.g., EHR codes, single-item questionnaires), potentially diluting diagnostic specificity. The Identical Depression Phenotype (IDP) consortium meta-analysed harmonised DSM-based MDD assessments to enhance aetiological specificity. Notably, EHR-based MDD shows strong genetic correlations with ADHD that are absent in DSM-defined MDD, suggesting broad phenotyping captures substantial non-MDD liability. Here, we applied Genomic SEM to jointly model DSM-defined MDD alongside broadly defined phenotypes, separating MDD-specific signal from ADHD and other non-specific genetic variance, and systematically evaluated locus-level signal changes between input and output summary statistics. Methods We performed LD clumping on pre- and post-Genomic SEM summary statistics across seven MDD phenotype configurations, including IDP DSM, 23andMe, PGC MDD, and EHR-based definitions. A power-aware attenuation framework classified locus loss as genuine when statistical power exceeded 80% and signal decreased by ≥2 orders of magnitude (−log₁₀p). Results The DSM-defined IDP phenotype showed 80 pre-GSEM loci reduced to 36 post-GSEM, with 22 novel loci emerging and mild attenuation (mean 2.58 −log₁₀p units; 18 of 22 well-powered loci lost). By contrast, the EHR-based phenotype was heavily attenuated (163→34 loci; mean 5.15 units; 64/68 well-powered loci lost), as was the full PGC MDD (699→134; mean 5.83; 309/319 lost). The largest combined model (785→194 loci; 105 novel) lost 332 of 345 well-powered loci (mean 5.44). Across all configurations, broadly defined phenotypes showed consistently greater attenuation than DSM-defined phenotypes. PGS derived from the IDP GWAS explained more variance in African (1.69%) and South Asian (1.73%) ancestry samples than PGS from the larger 23andMe and PGC MDD GWAS, despite substantially smaller discovery sample size. Conclusions DSM-based MDD phenotyping preserves genetic signal through multivariate modelling, whereas broadly defined phenotypes — particularly EHR-based definitions contaminated by ADHD-related liability — show substantial attenuation consistent with capture of non-specific variance. This framework provides a principled approach for separating depression-specific genetic architecture from comorbid psychiatric liability.