Genotyping was performed based on buccal or blood DNA samples collected in different research projects (see e.g., Willemsen et al. 2010). Imputation was performed based on the 1000G GIANT phase1 panel as a first reference set, and on the GONL version 4 as a second reference set (see Supplementary Methods for details). As best guess genotypes (computed using Beagle, Browning and Yu 2009) were used in the analyses, we applied stringent post imputation quality thresholds on the imputation quality measure (i.e., we retained only SNPs with an imputation quality score above 0.8) and for the Hardy–Weinberg equilibrium test (α = 1 × 10−4). Both the imputation quality and Hardy–Weinberg equilibrium (i.e., based on the summed genotype probability counts) were assessed in the phenotyped sample using SNPTEST (Marchini, 2007). The GoNL- and the 1000G-based imputed datasets contained ~6 million well imputed SNPs (i.e., with a mean imputation quality score above 0.96 in both datasets). The association and survival analyses were carried-out by varying the reference panel used for imputation, while including the same phenotyped sample (i.e., 6744 and 5148 participants, respectively).
