Genome-wide association analyses identify distinct genetic architectures for early-onset and late-onset depression

Abstract

Major depressive disorder (MDD) is a common and heterogeneous disorder of complex etiology. Studying more homogeneous groups stratified according to clinical characteristics, such as age of onset, can improve the identification of the underlying genetic causes and lead to more targeted treatment strategies. We leveraged Nordic biobanks with longitudinal health registries to investigate differences in the genetic architectures of early-onset (eoMDD; n = 46,708 cases) and late-onset (loMDD; n = 37,168 cases) MDD. We identified 12 genomic loci for eoMDD and two for loMDD. Overall, the two MDD subtypes correlated moderately (genetic correlation, r_g = 0.58) and differed in their genetic correlations with related traits. These findings suggest that eoMDD and loMDD have partially distinct genetic signatures, with a specific developmental brain signature for eoMDD. Importantly, we demonstrate that polygenic risk scores (PRS) for eoMDD predict suicide attempts within the first 10 years after the initial diagnosis: the absolute risk for suicide attempt was 26% in the top PRS decile, compared to 12% and 20% in the bottom decile and the intermediate group, respectively. Taken together, our findings can inform precision psychiatry approaches for MDD.

Fig. 1. Study design and analysis overview.

EMR, electronic medical record.

Fig. 2. MDD subtype GWAS meta-analysis in the Nordic cohorts and SNP heritability.

a,b, Mirrored Manhattan plots of the GWAS results from the combined Nordic countries, using inverse-variance-weighted meta-analysis with a genome-wide significance threshold of P < 5 × 10⁻⁸ (denoted by the dashed horizontal line), for eoMDD (a) and loMDD (b). c, Enrichment of open chromatin marks in the eoMDD and loMDD GWAS. The dashed line indicates the one-sided Bonferroni-corrected P threshold, set at P = 0.05/102 = 0.0005. d, Single-nucleotide polymorphism (SNP)-based heritability $\left(h_{\mathrm{SNP}}^2\right)$ across a range of population prevalence estimates, with the labeled dashed lines indicating the point estimate of the population prevalence. e, Genetic overlap between broad MDD, eoMDD and loMDD, with the upper triangle containing the standard errors and the diagonal containing estimates of polygenicity from SBayesS. * For loMDD, the algorithm did not converge, making the polygenicity estimate unreliable, although it was consistently higher than for eoMDD across successful runs.

Fig. 3. Genetic correlations.

a, Genetic correlations from linkage disequilibrium (LD) score regression (LDSC) comparing eoMDD (based on GWAS n = 153,532) and loMDD (GWAS n = 158,588) with key health outcomes and related psychiatric disorders (GWAS n in Supplementary Table 15). The error bars represent the 95% CIs. P values for the associations are reported in Supplementary Table 8. b, Results from genomic SEM linear regression, where the outcomes are regressed on eoMDD (GWAS n = 153,532) while controlling for overlap with loMDD (GWAS n = 158,588), and vice versa. The single asterisks denotes a significant difference between eoMDD and loMDD at P < 0.05. The error bars represent the 95% CIs.

Fig. 4. Causal relationships between eoMDD, loMDD and health outcomes.

a, Results of the MR analyses with MDD as the exposure, with the IVW statistic as the MR effect size estimate (ordered according to the eoMDD effect size). The error bars represent the 95% CIs. b, Effects with MDD as the outcome (ordered according to the effect size in a). Included in both a and b are the number of instrument SNPs used in the analysis (‘SNPs’), the percentage variance explained by the instrument SNPs in the exposure (‘% R²’) and the instrument strength (‘F’). We only tested plausible relationships (in line with time ordering), with improbable relationships, for example, suicide death as a risk factor for MDD, excluded from the figure. If the MR-Egger sensitivity analysis had a significant intercept (triangles), this indicates that the MR estimate was pleiotropic and should not be interpreted. Estimates with an asterisk were significant after correction for multiple testing (exact P values are reported in Supplementary Table 10). The error bars represent the 95% CIs.

Fig. 5. Associations of PRS for eoMDD and loMDD with clinical outcomes and suicide.

a, Meta-analyzed associations between PRS and MDD outcomes in the Nordic cohorts, using LOO GWAS summary statistics of eoMDD and loMDD. The error bars represent the 95% CIs. b, Selected mental-health-related and suicide-related ICD-10 group associated with eoMDD and loMDD from the PRS PheWAS. The error bars represent the 95% CIs. c, Number of unique and shared associations with eoMDD and loMDD from the PheWAS. d, Mean symptom scores of self-reported suicidality (Paykel Suicide Scale) stratified according to the top 10%, middle 80% and bottom 10% eoMDD PRS. The error bars represent the 95% CIs. e, Cumulative incidence of suicide attempt over a 10-year period since the first eoMDD diagnosis, stratified according to the top, middle and bottom eoMDD PRS. The gray shading indicates the 95% CIs.