Participants
The European sample set comprised 1,699 participants with gout and 14,350 controls (2,422 asymptomatic HU, and 11,928 NU). The NZ Polynesian sample set (individuals of NZ and Cook Island Māori, Samoan, Tongan, Niuean and Tokelauan ancestry) comprised 912 participants with gout, and 696 controls (202 HU and 494 NU) (Additional File 1). Hyperuricemia was defined, for both sexes, as serum urate ³0.42 mmol/L (7 mg/dL).
All people with gout fulfilled the 1977 American Rheumatism Association gout classification criteria (33). Gout cases were recruited from New Zealand (979 Europeans, 912 Polynesians), Australia and Europe (720 Europeans). The 14,350 European controls (all self-reported as not having physician-diagnosed gout, not having kidney disease and not taking urate-lowering medication) were obtained from five sources: 452 individuals recruited from New Zealand, 6,970 participants from the Atherosclerosis Risk in Communities (ARIC) study, 2,689 participants from the Framingham Heart Study (FHS), 1,492 participants from the Coronary Artery Risk Development in Young Adults (CARDIA) study, and 2,747 participants from the Cardiovascular Health Study (CHS). Phenotypes from baseline exams were used for all studies with the exception of CARDIA, where phenotypes from exam six were used. The 696 Polynesian controls were recruited from New Zealand using the same exclusion criteria.
Ancestry for controls from ARIC, FHS, CARDIA and CHS was used as provided in these datasets. Ancestry for the NZ, Australian and European gout cases and the NZ controls was classified based on principal components computed from genome-wide SNP data. The NZ Polynesian sample set was divided into Eastern and Western Polynesian ancestral groups based on principal components from genome-wide genotype data (34). Eastern Polynesian comprised of Cook Island and NZ Māori (543 cases and 462 controls), while Western Polynesian comprised of Samoa, Tonga, Tuvalu, Niue and Tokelau (369 cases and 234 controls) (35). A separate NZ Māori sample set of 124 cases and 50 controls is included within the Eastern Polynesian sample set data presented above. These participants were recruited in collaboration with Ngāti Porou Hauora Charitable Trust from the Ngāti Porou rohe (tribal territory) located in the East Coast (Tairāwhiti) region of the North Island of New Zealand.
Genotyping
Three SNPs were examined: rs2231142, rs10011796 (both ABCG2) and rs11942223 (SLC2A9). Genotypes for the NZ gout cases and controls, and for the European and Australian gout cases were determined using either a) Taqman® assays (rs2231142: C_15854163_70; rs10011796: C_9510320_10; rs11942223: C_1216479_10; Applied Biosystems, Foster City, USA) on a Lightcycler® 480 machine (Roche Applied Science, IN, USA), or b) Illumina Infinium CoreExome v24 bead chips processed at the University of Queensland (Centre for Clinical Genomics). Bead chip genotypes were auto-clustered using GenomeStudio v2011.1 software (Illumina, San Diego, USA). The Illumina GenomeStudio best practice guidelines and quality control protocols of Guo et al. (36) were applied to these auto-clustered genotypes to ensure final genotype calls were of the highest possible quality. SNP rs7442295 was identified as being in complete linkage disequilibrium with rs11942223 in European and East Asian populations using LDLink (37), and was used as a proxy on the bead chip for rs11942223. 2,940 samples were genotyped on both platforms for all 3 SNPs, with genotype concordance exceeding 99.8%.
Publicly-available genotype data for the ARIC, FHS, CARDIA and CHS participants were used. These genotypes were generated by the Affymetrix 6.0 platform (ARIC and CARDIA datasets), the combined Affymetrix 50K and 500K platform (the FHS dataset), and the Illumina Human CNV370v1 platform (the CHS dataset). Rs2231142 was genotyped on the Affymetrix 50K and 500K platforms and imputed using IMPUTE2 using the 1000 Genomes Phase I haplotype reference panel in the ARIC, CARDIA and CHS sample sets. Similarly, rs10011796 (ABCG2) and rs11942223 (SLC2A9) were imputed in all four sample sets.
Statistical analysis
All analyses were performed using R v3.1.2 using RStudio 0.98. Deviation from Hardy-Weinberg equilibrium within each group (gout, HU controls and NU controls) was tested separately using the Haldane Exact test for Hardy-Weinberg equilibrium (a=0.01) for each population and each SNP. Allelic association testing of ABCG2 rs2231142 and rs10011796, and SLC2A9 rs11942223 for comparison, with gout and HU was performed using logistic regression. We analysed gout vs. all controls, NU controls vs. HU controls, and gout vs. HU controls. The SNPs were also tested for association with serum urate concentration at recruitment in combined NU and HU controls using linear regression. Association of each of the three variants with self-reported number of gout flares over a 12-month period was examined using linear regression. Analyses were adjusted by age, sex (except male-only analyses) and principal components (PCs) 1 to 10 (Polynesian sample sets only). For association testing of gout in the presence of HU, analyses were additionally adjusted by highest recorded serum urate concentration. All Polynesian meta-analyses were performed in R using meta v4.8-2 (38), and an inverse-variance fixed effect model unless otherwise stated.
Given the low linkage disequilibrium between ABCG2 rs2231142 and rs10011796 in European and East Asian populations (r2=0.087 and 0.12, respectively), genotypes were categorised as risk allele-positive or -negative for both SNPs, and association of rs2231142-rs10011796 genotype combinations were tested for gout vs. all controls, and gout vs. HU controls using logistic regression. Non-additive interaction between the two SNPs in determining the risk of gout was also examined, including an interaction term between rs2231142 and rs10011796 in the multivariate-adjusted logistic regression model.