GWA analyses were performed by each of the participating studies. BMI was regressed on age, age2, and study site (if needed) to obtain residuals, separately by sex and case-control status, if needed. Residuals were inverse-normally transformed to obtain a standard normal distribution with a mean of 0 and a SD of 1. For studies with unrelated subjects, each SNP was tested for additive association with BMI by regressing the transformed residuals on the number of copies of the SNP effect allele, adjusting for population structure as measured by the first ten eigenvectors calculated for each study. Analyses were stratified by sex and case-control status (if needed). For studies that included related individuals, family-based association tests were conducted that take into consideration the genetic relationships among the individuals. Study-specific lambda values ranged from 0.95 to 1.08 (Supplementary Table 2). We applied genomic control (GC) in the Stage 1 analysis (i.e. divided by the median of all χ2 statistics for each study) to eliminate any remaining over dispersion before combining the GWAS in the meta-analysis. In Stage 1, we conducted a fixed
