2 or larger. Genomic controls (Devlin et al., 2001) for the case-control phenotype were calculated with R-2.5.0 (http://cran.r-project.org) on a genome-wide level in the MARS GWAS sample. In addition, population stratification was tested with EIGENSTRAT implemented in EIGENSOFT (Price et al., 2006)

(http://genepath.med.harvard.edu/∼reich/EIGENSTRAT.htm). Neither the genomic control method (λ = 1.023, see Figure S1) nor EIGENSTRAT analysis gave any indication for population stratification. The LD pattern and haplotype block delineation were determined by applying Haploview 4.0 (http://www.broad.mit.edu/mpg/haploview) (Barrett et al., 2005). Blocks were defined using the confidence interval method described KU-55933 chemical structure by Gabriel et al. (Gabriel et al., 2002). Pairwise LD measures (r2 and D’) were calculated in the 366 healthy controls of the

GWAS sample and in 284 controls of the African-American sample for the eight most associated SNPs on chr12.21.31 (see Figure 2). German controls were also compared to the HapMap CEU population (CEPH sample consisting of Utah residents with Vorinostat ancestry from northern and western Europe, n = 60, http://www.hapmap.org) (Frazer et al., 2007). No deviation in LD could be observed in this comparison (data not shown). Genome-wide case-control analyses were conducted by applying the WG-Permer software (http://www.mpipsykl.mpg.de/wg-permer/). For post-hoc analyses, applications in R-2.5.0 (http://cran.r-project.org) and SPSS for Windows (releases 16, SPSS, Chicago, IL, USA) were used. SNPs with genotype distributions deviating from HWE at a significance level of 10−5 or 0.05 with a call rate below 98% or 95% in the GWAS or German replication sample, respectively, and SNPs with a MAF below 5% were excluded from statistical analysis. Autosomal SNPs were tested for association with unipolar depressive disorder in a case-control design 3-mercaptopyruvate sulfurtransferase using Chi-square test statistics under allelic and both alternative recessive-dominant modes of inheritance. The level of significance was set to 5% (family-wise error rate). Nominal p values were corrected for

multiple comparisons by the permutation-based minimum p method proposed by Westfall and Young (Westfall and Young, 1993 and Westfall et al., 2001) under 104 permutations over the three performed genetic models and all SNPs tested per study. Empirical and nominal p values for all reported associations did not deviate from each other. Moreover, sample demographic statistics and post-hoc tests on age, gender, and German origin, life events, recurrence of MD, age at onset, number of previous depressive episodes, first-degree family history of MD, and lifetime attempted suicide status were performed by logistic regression analysis and ANCOVA. P values including these covariates did not differ from those of the Chi square test statistics for all reported associations.