with IMPUTE2 (Howie et al., 2009). The latter programs do not allow for the inclusion of indels, and therefore all simple and complex indels (n = 8,751) were excluded prior to genotype pre-phasing and imputation. Finally, ancestry estimations were calculated from variants with a MAF ≥ 0.01 using principal components analysis (Price et al., 2006), as implemented in the Genome-wide Complex Trait Analysis software (Yang et al., 2011). The resulting ancestry estimates were used as covariates in subsequent association tests to control for possible population substructure.
