GWAS, genome-wide association studies; NHS, Nurses’ Health Study. Figure 3 Manhattan plot of the GWAS meta-analysis in four NHS substudies (N = 6989). GWAS, genome-wide association

studies; NHS, Nurses’ Health Study. NHS-GWAS-PS FRAX597 analyses The genome-wide PS similarly explained a small fraction of variance in the long-term average depression score (Table 3). Using the most liberal threshold of P < 0.5 to select SNPs in the training set, the genome-wide PS was associated with the depression score in the testing set (P = 0.004), Inhibitors,research,lifescience,medical but explained only 0.1% of the variance. The maximum percentage of variance explained was achieved with slightly more conservative P-value thresholds for SNP selection (at P < 0.3), in which the genome-wide PS explained 0.2% of the variance (P = 0.003). When restricted to nonoverlapping Ptraining threshold Inhibitors,research,lifescience,medical ranges, the SNPs with the most significant association were those with Ptraining

between 0.1 and 0.2; this group alone comprised nearly 9900 SNPs, but explained 0.1% of phenotype variation (Table 3). Table 3 Meta-analysis of percentage Inhibitors,research,lifescience,medical of variance explained in depression phenotype in NHS by the genome-wide agnostic polygenic scores in the leave-one-substudy-out analysis (N = 6989). GAIN-MDD-PS and PGC-MDD-PS analyses Regardless of the P-value threshold chosen, the GAIN-MDD-PS was not significantly associated Inhibitors,research,lifescience,medical with either the continuous or dichotomized depression phenotype in

the NHS sample (Table 4). The maximal proportion explained by genome-wide PS comparing women at the extremes of the phenotype was higher than that in the full sample (0.4% vs. 0.1%); however, it was not statistically significant, likely due to the reduction in sample size when using only individuals with extreme values of the phenotype (n = 2920) (data not shown). Table 4 Meta-analysis of percentage of variance explained in depression phenotype in NHS by the genetic risk scores using external GAIN-MDD sample as the training set (N = 6989). When applying the agnostic PS Inhibitors,research,lifescience,medical from a nine-study meta-analysis of PGC-MDD, the genome-wide risk scores derived from SNPs with less stringent Ptraining threshold were significantly associated with the continuous long-term depressive score, but they only explained at most 0.1% of variance in phenotype. The Nagelkerke’s R2 was also at most 0.1% when the depression phenotype was modeled dichotomously without over the statistical significance (Table 5). Table 5 Meta-analysis of percentage of variance explained in depression symptoms in NHS by the genetic risk scores using external PGC-MDD sample as the training set (N = 6989). Candidate-PS analyses Three individual SNPs (rs36011, rs1417584, and rs6917735) showed nominally significant associations at α threshold of 0.05, but none remained significant after Bonferroni correction.