*Sequence variants that remain significant after adjustment for previously reported sequence variants (online supplemental table 1). Bold indicates candidate causal genes (summarised in figure 2).

†We performed a meta-analysis using logistic regression analysis assuming a multiplicative model, reporting OR and two-sided p values adjusted for year of birth, sex and origin (Iceland) or the first 20 principal components (other countries). Variants were split into five classes based on their genome annotation and significance threshold based on the number of variants in each class. The adjusted significance thresholds are 1.3×10^–7 for variants with high impact (splice donor, splice acceptor, stop gained, frameshift, stop lost, initiator codon), 2.6×10^–8 for variants with moderate impact (missense, splice region, stop retained, inframe indels), 2.4×10^–9 for low-impact variants (synonymous, 5’ UTR, 3’ UTR, upstream and downstream), 1.2×10^–9 for other low-impact variants in DNase I hypersensitivity sites (intronic, intergenic, regulatory-region) and 5.92×10^–10 for all other variants not in DNase I hypersensitivity sites. Primary signal at each locus (1 Mb) was selected based on conditional association analysis of all variants at each locus, using Bonferroni corrected p values (0.05×P/class-specific p value threshold). We report the coding signal when two markers are equivalent after conditional analysis. Secondary signals are sequence variants that remained GWAS significant after adjustment for the lead signal and other independent (secondary) signals at the locus. When different but correlated variants are lead in RA overall and seropositive RA, the seropositive RA signal is presented here. See further in online supplemental tables 2 and 3.

GWAS, genome-wide association study; Phet, a p value for test of heterogeneity between the effects in seropositive and seronegative RA subsets; RA, rheumatoid arthritis.