We also included GWAS data from SAGE (n=4077) (dbGaP study id phs000092.v1.p1) (11, 12), which were genotyped by the Illumina Human 1M array covering ∼1 M total SNPs. SAGE genotypes were compared with the HapMap3 reference datasets, comprised of CEU, YRI, and CHB populations, and separated into EA and AA populations by using the first PC scores in principal components analysis with Eigensoft. The raw genotypes were imputed with IMPUTE2 (13). The 1000 Genomes phase 1 reference panel (6) containing phased haplotypes for 1092 individuals was used for imputation. Dosage genotypes were transformed into hard call genotypes with a custom perl script for heterozygous and homozygous genotypes, i.e., dosage probability for heterozygous genotypes were between 0.8 and 1.4 and homozygous genotypes of reference and alternative alleles ranged from 0 to 0.4 and 1.6 to 2, respectively, with others assigned as missing. The QC criteria for the individual genotyping missing rate <5%, MAF >1% and missing call frequency <5% were also utilized to filter GWAS data. After QC, ∼14 M variants were retained for meta-analysis with Yale-Penn 1-3.
