SNP genotype data from the UK cohort of the IMSGC and WTCCC2 MS GWAS study4 were used to localize primary association signals. SNP and sample exclusion lists were as in the original study. SNPs that were not in Hardy-Weinberg equilibrium (HWE) within either control cohort (P<0.001) were also excluded. Inferred haplotypes for 379 samples of European ancestry (EUR) from the March 2012 release of the 1000 Genomes Project were used for analysis of haplotype structure and imputation of untyped variants. Only SNPs typed successfully in case and control cohorts were used for imputation with IMPUTE Version 2Ref12, removing SNPs whose minor allele frequencies (MAF) differed by more than 0.1 between control cohorts. Association analysis with a logistic model accounting for uncertainty in genotype imputation, including gender as a covariate, was carried out using SNPTEST28, on SNPs with MAF>0.01 that had an imputation information score greater than 0.9. We used a Bayesian, additive test with the score method to measure evidence for association and we report the Bayes factor comparing the models of association and no association. We assumed the default
