Depression is among the most prevalent mental disorders, characterised by extensive clinical heterogeneity. Previous genome-wide association studies (GWAS) have primarily employed either a case–control design or summed depressive symptom scores to examine the genetic underpinnings of depression. These approaches, however, may obscure the heterogeneity inherent in depressive phenotypes by aggregating distinct symptom dimensions into a single construct. Analysing the genetic liability of depression by using homogeneous symptoms as measures could address this gap and might enhance the power to detect genetic signals.
We standardised depression by conducting symptom-level GWAS on each of nine depressive symptoms, predominantly using data from the UK Biobank and All of US cohort. By integrating data across multiple time points and multiple cohorts, we have conducted the largest meta-analysis for depressive symptoms to date (total N > 750,000), which substantially enhanced statistical power and enabled the detection of hundreds of risk-associated loci. Summary statistics of participants of European ancestry were subsequently used to estimate each symptom’s SNP-based heritability and genetic correlations both among symptoms and with external traits via linkage disequilibrium score regression (LDSC). We observed consistently high genetic correlations across depressive symptoms, but the genetic correlations with external variables largely varied in magnitude. To elucidate the genetic factor structure underlying depression, we applied genomic structural equation modelling (GSEM), which revealed three distinct factors of genetic liability underlying the symptoms, highlighting extensive heterogeneity in their genetic underpinnings. Gene- and gene-set–based enrichment analyses were performed to identify and functionally annotate risk genes. Collectively, our findings showed that the heterogeneous presentations of depression extend beyond the phenotypic level, revealing distinct genetic underpinnings across depressive symptom domains.