A comprehensive analysis of the association of common variants of ABCG2 with gout

The objective of the present study was to determine whether there was an association between single nucleotide polymorphisms (SNPs) in ABCG2 and gout. We recruited 333 participants including 210 patients with gout and 123 controls and genotyped 45 SNPs in both cohorts. We found that 24 SNPs in ABCG2 are susceptibility loci associated with gout. Haplotype analysis revealed five blocks across the ABCG2 locus were associated with an increased risk of gout with odds ratios (ORs) from 2.59–3.17 (all P < 0.0001). A novel finding in the present study was the identification of rs3114018 in block 3 and its association with increased gout risk. We found that the rs2231142T allele in block 2 and the rs3114018C-rs3109823T (C-T) risk haplotype in block 3 conferred the greatest evidence of association to gout risk (P = 1.19 × 10−12 and P = 9.20 × 10−11, respectively). Our study provides an improved understanding of ABCG2 variations in patients with gout and, as shown by haplotype analysis, that ABCG2 may have a role in gout susceptibility.

Fine mapping of ABCG2 and haplotype analysis. We calculated linkage disequilibrium (LD) coefficients and constructed haplotypes using Haploview version 4.2 (Mark Daly's Laboratory, Massachusetts Institute of Technology/Harvard Broad Institute, Cambridge, MA, USA) 24 . For haplotype construction, genotype data from both case and control groups were used to estimate intermarker LD by measuring pairwise D′ and r 2 , and to define LD blocks 24, 25 . Statistical analysis. Categorical variables were expressed as percentages and were analyzed by chi-square (χ 2 ) test or Fisher's exact test, as appropriate. Continuous variables were expressed as mean ± SD. All P values in this study were two sided, and P < 0.05 was considered statistically significant. SNP frequencies were tested for departure from HWE using an exact test in control subjects. Allele and genotype frequencies for each SNP were compared between patient and control cohorts using the χ 2 test. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated using logistic-regression analysis. In addition to obtaining nominal P values, empirical P values were generated by running 10,000 permutations using the Max (T) permutation procedure implemented in PLINK v1.07 26 . In addition, we applied Bonferroni correction and set the significance threshold for these analyses at α = 1.1 × 10 −3 , which corresponds to a stringent Bonferroni correction for testing 45 independent markers. All statistical analyses were performed using SPSS 20.0 (IBM Corp., Armonk, NY, USA). Each marker was tested for association using PLINK v1.07 (http://pngu.mgh.harvard.edu/purcell/plink/) 26 . Haploview (v4.2) was used for assessing LD patterns and haplotype association statistics 24 . Haplotype blocks were determined using the algorithm of Gabriel et al. 27 . An omnibus (or global) test of haplotype association was performed using PLINK. ORs and 95% CIs for haplotype-specific risks were calculated using VassarStats (http://vassarstats.net/).

Results
Characteristics of study subjects. Our study consisted of 333 participants of which 210 were patients with gout and 123 were controls. Detailed information of study participants is shown in Table 1. The mean age of affected individuals was 52.4 ± 12.9 years (range 20-85 years) with a male-to-female ratio of 201:9 (approximately 22.3:1), while the mean age of controls was 51.9 ± 11.83 years (range 27-81 years) with a male-to-female ratio of 107:16 (approximately 6.7:1) ( Table 1). As gout primarily affects males, fewer females than males participated in this study. However, we found that there was no significant difference between cohorts in terms of age distribution (P = 0.588).
In addition, following adjustments for Bonferroni correction for testing 45 independent tests (α < 0.0011), we  Table 2. Characteristics of the polymorphisms in ABCG2 and risk of gout. OR: odds ratio, CI: confidence interval.
found that all SNPs with the exception of eight SNPs (rs45557042, rs11935697, rs3109824, rs2725256, rs2231138, rs3114017, rs2725254, and rs1564481) remained significantly associated with gout. We found that the inclusion of age and gender as covariates in logistic regression models did not substantially change the significance of the observed associations (data not shown).
Linkage disequilibrium plot and haplotype analysis. Using Haploview v4.2, we generated an LD plot of the 45 genotyped SNPs in ABCG2 in our affected cohort (Fig. 1), and found that more than half of the SNPs tested in the present study were highly correlated with each other (r 2 ≥ 0.80), of which 220 out of 990 (22.2%) pairs revealed perfect linkage disequilibrium (D′ = 1). Moreover, the haplotype block structure spanning ABCG2 as derived by Haploview is shown in Fig. 1b with haplotype frequencies shown in both Fig. 1b and in Table 3.
We found that there are five haplotype blocks across the ABCG2 locus. Based on the model of Gabriel et al. 26 , we identified a total of 25 common haplotypes in these five blocks (Table 3), which span approximately 16 kb, 33 kb, <1 kb, 15 kb, and 1 kb derived from 5, 11, 3, 4, and 2 SNPs, respectively (Fig. 1a). Haploview predicted 34 possible connections of haplotypes for recombination between blocks at a frequency >1%. In addition, calculating the frequency of recombination between blocks as a value of a multiallelic D′ coefficient, we found values of 0.85 between blocks 1 and 2, 0.91 between blocks 2 and 3, 0.84 between blocks 3 and 4, and 0.97 between blocks 4 and 5 ( Fig. 1b and Table 3). Among the identified 25 common haplotypes, five haplotypes were associated with an increased risk for gout (all permuted P < 0.0001) and seven haplotypes were associated with a decreased risk for gout ( Table 3). The five haplotypes associated with an increased risk of gout following correction with  Table 3). The haplotype that conferred the greatest risk was the rs3114018C-rs3109823T haplotype (C-T) in block 3, with a haplotype frequency of 0.816 in affected individuals and 0.585 in controls (P = 9.20 × 10 −11 ; OR = 3.16; 95% CI = 2.21-4.50; Table 3).

Discussion
Gout is an increasing global health problem caused by multiple genetic and environmental factors. In recent years, many variants in a growing number of genes involved in the pathogenesis of gout and hyperuricemia have been identified 2, 9 . ABCG2 is located at a gout-susceptibility locus on chromosome 4q22, which was previously identified in several genome-wide linkage studies of gout 7,12,20 . ABCG2, which is also known as breast cancer resistance protein (BCRP), is a high-capacity urate exporter, the dysfunction of which increases the risk of gout and hyperuricemia 8,28 . ABCG2 mediates renal urate secretion as a urate efflux transporter in the brush-border membrane on the luminal surface of kidney proximal tubule cells 2,8,9 . In addition, ABCG2 is expressed at high levels in the intestine and liver 27 and functions as an efflux transporter for many drugs and molecule substrates, including anticancer agents, antibiotics, antivirals, HMG-CoA reductase inhibitors, flavonoids, allopurinol, and uric acid [28][29][30][31][32][33][34] .
Using haplotype analyses, we found five blocks of LD that were significantly associated with gout. Moreover, an LD plot of ABCG2 demonstrated extensive correlation among 45 SNPs. Based on measures of r 2 , perfect linkage (r 2 = 1) was detected in 220 out of the 990 pairs (D′ = 1) and strong LD (1 > |D′| ≥ 0.8) was detected in more than half of variant pairs. A novel finding in the present study was the identification of rs3114018 in block 3 and its association with increased gout risk. In addition, the minor T allele of rs2231142 in the second block of ABCG2 was associated with an increased risk of gout (OR = 3.29; 95% CI = 2.36-4.60), a finding that was similarly reported in previous studies in other populations 7, 10-12, 17, 21, 33, 35-37 .
Two independent functional studies of ABCG2 found that the Q141K (rs2231142) polymorphism occurs in a highly conserved region of the gene and is a loss-of-function mutation 8,33 . These studies found that the rs2231142 risk allele resulted in a urate secretory transporter with a 53% reduced ability to transport urate in Xenopus oocytes 8 and HEK293 membrane vesicles 33 . Moreover, Abcg2-knockout mice had increased serum uric  Table 3. Haplotypes in the haplotype blocks spanning the ABCG2 locus. OR: odds ratio, CI: confidence interval. *SNPs are as numbered in Fig. 1a  acid levels and renal urate excretion, and decreased intestinal urate excretion 28 . Furthermore, Woodward et al. 37 demonstrated the utility of using small molecules to correct the Q141K defect in expression and function as a potential therapeutic approach for hyperuricemia and gout. The association between the rs2231142T allele and the risk of gout has been replicated in many diverse study populations including Caucasian 7, 11, 12, 33 , African 7 , Japanese 20, 33, 35 , Mexican-American 12 , Native American 12 , Han Chinese 10,17,36 , and New Zealand Pacific Island ancestry 11 . These findings indicate that ABCG2 may have specific and important functions in the pathology of gout. However, an association between rs2231142 and gout has not been found in Maori populations 11 and some studies in the Chinese population 21,38 . The reason for this discrepancy is not known, but the difference may be because of either differences in gene structure or sampling bias 13 . Furthermore, an additional confounding factor is that the etiology of gout is linked to various genetic and environmental factors such as lifestyle and diet 1,5,35,39 . However, the baseline socioeconomic status and diet habit were not available in the database, so we were unable to perform the analysis.
In this study, we thoroughly captured common genetic variation across ABCB2 and performed a comprehensive evaluation of common SNPs at ABCB2 associated with gout risk. Using haplotype analysis, we found five haplotype blocks that were associated with an increased risk of gout: block 1 with an OR of 2.67 (95% CI = 1.87-3.81), block 2 with an OR of 2.59 (95% CI = 1.81-3.69), block 3 with an OR of 3.16 (95% CI = 2.21-4.50), block 4 with an OR of 2.76 (95% CI = 1.89-4.04), and block 5 with an OR of 2.94 (95% CI = 2.05-4.22; all P < 0.0001) ( Table 3). Our results, combined with those from previous studies, suggest that genetic variation in ABCG2 may influence gout susceptibility in the Han Chinese population. Consistent with the genetic susceptibility identified in patients with gout in several other populations, we observed that the minor allele of rs2231142 was associated with an increased risk for gout 7, 10-12, 17, 20, 33, 35-37 , while we found other SNPs in the present study that may confer a protective effect on susceptibility to gout. This finding is consistent with the hypothesis of two functional polymorphisms near the SNPs evaluated in this study, one that increases the risk of developing gout whereas the second confers a protective effect 40 . In addition, considering that the genomic regions of the five SNP haplotype blocks are characterized by high LD, we postulate that such SNPs are likely to tag any hitherto unidentified common SNPs in the candidate gene. For example, two recent studies from northwest China 21, 38 found a significant difference in mean serum urate levels between a novel SNP, rs3114018, in ABCG2 and gout risk, which is consistent with the findings of the present study. In addition to rs2231142 in block 2, the greatest evidence of association in the present study was between the C-T risk haplotype of rs3114018 and rs3109823 in block 3. To the best of our knowledge, the relationships demonstrated in the present study between SNPs in blocks 1, 3, 4, and 5 with gout, such as the novel SNP rs3114018 in block 3 have not been previously observed until recently 21,38 . Of note, rs2231137 is located in the same block with rs2231142 (block 2), resulting in a V12M substitution (p = 1.4 × 10 −9 ). Our findings and previous studies 10,17 indicated that V12M substitution was associated with a decreased risk of hyperuricemia and/or gout. However, in vitro functional assays showed that V12M substitution did not result in any changes in protein expression and risk to phenotypes such as serum urate levels and gout 33,41 . Further studies are required to elucidate the functional contributions of these novel SNPs in these genomic regions or blocks that confer increased risk for gout.
The present study had the following limitations. First, although we could identify genetic associations with gout, we could not elucidate the underlying causal mechanisms. Nonetheless, our findings with rs2231142 and rs3114018 are consistent with those of studies of other populations, which highlight their robustness and support for a role in gout. Second, considering the marked difference in SNP minor allele frequencies among populations, ethnic differences may exist, which would confound the identification of genetic risk factors for gout 2,9,42 . Future studies should incorporate larger sample sizes to verify present findings across more populations. Finally, the biological functions of other SNPs in ABCG2 have not been fully characterized, and therefore, the findings from the present study require functional confirmation by future expression studies.
In conclusion, this large-scale thorough evaluation of SNPs has identified common genetic variants in ABCG2 that are associated with gout risk. None of the tested SNPs, with the exception of rs2231142, which were identified as significant in this study were listed among the most significant results of three recently conducted GWAS on gout 7,12,20 . In addition to rs2231142, haplotype analysis of polymorphisms in ABCG2 revealed SNP-derived haplotypes associated with gout risk. Further identification of the functional and causal variant(s) in ABCG2 will lead to a better understanding of the mechanism underlying the development of gout pathologies.