Common variants at ten genomic loci influence hemoglobin

Genotyping, imputation and quality control Study samples are described in Table 1. Study specific parameters and pre-imputation filters are specified in Table S1. Each study applied similar criteria for data calling. Prior to imputation, the criteria applied for exclusion of SNPs were: (i) minor allele frequency (MAF) <0.01, (ii) Hardy-Weinberg equilibrium P <10-4 or 10-6 and (iii) call-rate <0.90 or 0.95. Criteria applied for exclusion of samples were: (i) call-rate <0.95 or <0.97, (ii) sex mismatch between genotypes and reported sex, and (iii) outliers as assessed by population structure analysis. Imputation of additional autosomal SNPs from the HapMap CEU (1) reference panel was performed using the software MACH (2) or IMPUTE (3). As standard for imputation, we excluded sex chromosome-linked SNPs from analyses given the difficulty of accurately imputing non-autosomal SNPs and the poor overlap of X-chromosome SNPs across different platforms. SNPs were also excluded if the cohort-specific imputation quality was particularly poor (observed-over-expected variance ratio (r2.hat) <0.3 if MACH was used for imputation, or proper-info <0.4 if IMPUTE was used) or if MAF < 0.01. In total, up to 2.5 million genotyped or imputed autosomal SNPs were considered for meta-analysis. We only report on individual SNPs imputed or genotyped in ≥6,000 participants. Statistical methods for primary analyses In each cohort we fitted a linear regression model using measured HbA 1C (%) as the dependent variable to evaluate the additive effect of genotyped and imputed SNPs. The model was adjusted for age, sex and/or study site and family structure (Table S1). The association was tested taking genotype and imputation uncertainty into account, using a missing data likelihood test as implemented in SNPTEST (3) or by using allele dosages in the linear regression model as implemented in ProbABEL (4) or MACH2QTL (2) for unrelated samples or in Merlin (5) or using a linear mixed effects model implemented in the lmekin function of the R kinship package for family-based studies. Regression estimates for each SNP were combined across studies in a meta-analysis using a fixed effect inverse-variance approach, as implemented in METAL The individual cohort analysis results were corrected prior to performing the meta-analysis for residual inflation of the test statistic using the genomic control method if the lambda coefficient was > 1.0. Heterogeneity was assessed using the standard chi-square test implemented in METAL, Cochran's Q statistic and the I 2 statistics (6). Conditional analyses We used conditional analyses to infer whether …