Even though the direction of the association of rs1545843 with depression and depressive symptoms was the same in all samples with non-geriatric depression, the effect sizes were heterogeneous, with a much larger effect in the discovery sample (OR=2.8 for the recessive model) as compared to the other samples with odds ratios ranging from 1.18 to 1.61. The strong association and low p-value in the discovery sample is probably due to the “winners’ curse”, but this phenomenon has also been observed for other, now established, disease loci. For example, the association of a SNP in the FGFR2 gene with breast cancer was much stronger in the rather small discovery sample than in any of the subsequent replication samples. However, the direction of the association was consistent and reached a p-value of 2e-76 in close to 30,000 cases and controls (Easton et al., 2007). This indicates that heterogeneous effect sizes with overestimation of the effect in a small discovery sample may still be in agreement with a true signal. In addition to the genome-wide significance (Dudbridge and Gusnanto, 2008) observed in our
