398 (95% CI 1.254–1.557) with a combined two-sided p value of 1.41e-09 (Figure 3). Considering only the replication studies (thus excluding MARS), we have an estimate of 1.315 for the OR (95% BKM120 supplier CI 1.172–1.477) with a two-sided p value of 3.19e-06. While the association with major depression and depressive symptoms thus was consistent in samples across different ethnicities, this did not hold true for incident late-life depression. The association did not replicate in the Rotterdam

study (n = 3512) (Hofman et al., 2007), where subjects older than 55 years of age and free of dementia were screened with the CES-D at baseline and two follow-up time points (Luijendijk et al., 2008). A case-control analysis was selleck performed by comparing subjects who developed depressive disorders and depressive syndromes at follow up time points (n = 438) with individuals without clinically relevant depressive symptoms (n = 3074) (mean age ± standard deviation [SD] of cases: 72.7 ± 7.4 years and controls: 73.9 ± 8.3 years). None of the investigated SNPs reached significant association. The average age of 72 years at which the index depressive episode in the Rotterdam sample was diagnosed is substantially older

than the average age in the combined German discovery sample and recurrent depression sample (50.4 ± 13.9 years), the Dutch ERF sample (48.7 ± 15.0 years), and the African-American sample (39.3 ± 13.7 years). In fact, in the other samples significant associations with rs1545843 were only observed

when individuals ≤ 55 years were selected but not in the older age group. A series of studies indicate that late-life depression is pathophysiologically distinct from earlier onset MD being more strongly related to vascular disease and future cognitive impairment (Alexopoulos, 2006). In summary, as shown in the forest plot in Figure 3, the initial GWAS revealed a SNP (rs1545843) on chr12q21.31 to be associated with MD with experiment-wide significance. This could be confirmed in a meta-analysis across six additional independent samples, including one sample of African-American heritage, with the recessive model of all rs1545843 reaching genome-wide significance. The associated SNP lies within a region of SNPs in moderate LD that span a gene desert of about 450 kb in size on 12q21.31 mapping neither to any annotated gene nor to predicted human mRNAs with the exception of some small human expressed sequence tags (EST, Figure 1A and Table S1). The closest RefSeq annotated gene is SLC6A15 (NM_182767), which ends 287 kb further distal to the region of association. It belongs to the solute carrier 6 (SLC6) gene family and codes for a sodium-dependent branched-chain amino acid transporter ( Bröer, 2006). SLC6A15 gene expression is highest in the human brain as well as the brain of other vertebrate species ( UniGene, 2009 and Allen Brain Atlas, 2008).