Background
Gout is a common inflammatory arthritis characterised by deposition of monosodium urate (MSU) crystals in joints and other tissues [
1]. Hyperuricaemia is a key checkpoint in MSU crystallisation and the clinical presentation of gout [
1]. Previous genome-wide association studies (GWAS) have identified 28 single nucleotide polymorphisms (SNPs) associated with serum urate concentration [
2]. Some of these SNPs encode renal and/or gut urate transport-related proteins and associate with the risk of gout [
2‐
5]. Others contribute to hyperuricaemia via regulation of purine synthesis and glucose metabolism pathways [
2].
Long-established dietary factors for hyperuricaemia and gout risk include red meat, seafood, and alcohol [
6,
7]. In addition, multiple studies have reported that increased coffee intake is associated with reduced serum urate [
8‐
12] and risk of developing gout [
13,
14]. This is the case for both caffeinated and de-caffeinated coffee. This association has been attributed to several potential mechanisms including improved insulin resistance [
15‐
21] and caffeine-mediated inhibition of xanthine oxidase [
8].
SNPs associated with habitual coffee consumption have also been identified by GWAS [
22]. Of note, the GWAS of coffee consumption identified a number of genes which overlap with the genes identified in the separate hyperuricaemia GWAS [
2,
22]. Specifically, alleles in several genes associated with hyperuricaemia (
GCKR,
ABCG2,
MLXIPL, and
CYP1A2 [
2]) are also associated with decreased habitual coffee intake [
22]. These four alleles are the sole alleles currently known to be associated with both serum urate, gout risk, and habitual coffee consumption from existing genome-wide analysis studies. Regional association plots suggest that the signals for both urate and habitual coffee consumption are very similar for the four loci (Additional file
1: Figure S1). The aim of this study was to determine whether the lead urate-associated SNPs at the four coffee and serum urate associated loci influence the risk of gout through their effects on coffee consumption.
Methods
Participants
This research was conducted using the UK Biobank Resource [
23,
24] (approval number 12611). Data for the UK Biobank were gathered over 2006–2010 from people between the ages of 40 and 69 years old. The North West Multi-Centre Research Ethics Committee granted ethical approval for UK Biobank participants. All participants provided written informed consent.
Inclusion criteria for this analysis were European ethnicity (white, British, Irish, any other white background) and genome-wide genotypes available. Exclusion criteria were self-reported sex mismatch with genetic sex, genotyping quality-control failure, and related individuals. Gout cases were ascertained in the UK Biobank using a validated case-definition of “self-reported gout or urate-lowering therapy (ULT) use” [
25].
Genotyping and SNP selection
UK Biobank samples were genotyped using an Affymetrix Axiom array (820,967 markers) and imputed to ~ 73.3 M SNPs using SHAPEIT3 and IMPUTE2, with a combined UK10K and 1000 Genomes reference panel.
Details of the lead coffee-associated SNP at each locus reported by Cornelis et al. (2015) [
22] were extracted from the publication. The equivalent information was extracted from the serum urate GWAS publication for overlapping loci [
2] and aligned to that of the coffee GWAS. For both
GCKR and
CYP1A2, the lead SNP reported in both studies was identical. For
ABCG2 and
MLIXPL, the linkage disequilibrium (LD) between the lead urate-associated and coffee-associated SNP was calculated using the European 1000 Genomes phase 3 (September 2014) and UK Biobank data with PLINK v1.90 [
26,
27]. The
ABCG2 SNPs were in high LD (
r21000Genomes = 0.94;
r2UKBiobank = 0.99), whilst the
MLIXPL SNPs were in moderate LD (
r21000Genomes = 0.58;
r2UKBiobank = 0.57) (Table
1). Because of the adequate LD, the four urate-associated SNPs,
GCKR (
rs1260326),
ABCG2 (
rs2231142),
MLXIPL (
rs1178977), and
CYP1A2 (
rs2472297), were employed in the analysis (Table
1).
Table 1
SNPs associated with serum urate levels and habitual coffee intake in previous GWAS using participants of European ancestry
Coffee GWAS [20] | SNP | rs1260326 | rs1481012 | rs7800944 | rs2472297 |
Chr: Position (B37) | 2:27730940 | 4:89039082 | 7:73035857 | 15:75027880 |
Effect allele/other | T/C | A/G | T/C | T/C |
Effect allele freq. | 0.41 | 0.89 | 0.72 | 0.24 |
Beta (cups/day) | −0.04 | 0.06 | −0.05 | 0.15 |
SE | 0.01 | 0.01 | 0.01 | 0.01 |
P
| 1.06 × 10−7 | 1.13 × 10−6 | 7.82 × 10−9 | 6.45 × 10−47 |
Urate GWAS [2] | SNP | rs1260326 | rs2231142 | rs1178977 | rs2472297 |
Chr: Position (B37) | 2:27730940 | 4:89271347 | 7:72494985 | 15:75027880 |
Effect allele/other | T/C | G/T | A/G | T/C |
Effect allele freq. | 0.41 | 0.89 | 0.81 | 0.24 |
Beta (mg/dl) | 0.07 | −0.22 | 0.05 | −0.03 |
SE | 0.01 | 0.01 | 0.01 | 0.01 |
P
| 1.20 × 10−44 | 1.00 × 10−134 | 1.20 × 10−12 | 3.85 × 10−6 |
LD | Euro r2 | Same SNP | 0.94 | 0.58 | Same SNP |
| rs2231142: T | rs1178977: G | |
| rs1481012: G | rs7800944: C | |
UKBB r2 | Same SNP | 0.99 | 0.57 | Same SNP |
| rs2231142: T | rs1178977: G | |
| rs1481012: G | rs7800944: C | |
Coffee and food-frequency intake
At the time of recruitment, data were collected on usual coffee consumption in addition to other foodstuffs [
28]. Coffee intake was determined by participants’ answers to the question “How many cups of coffee do you drink each day (include de-caffeinated coffee)?” Values were numerical in exact cups per day (including 0) or under 1 cup per day. Any non-zero value was determined as “any” coffee consumption. Other food intakes were determined by responses to questions about daily or weekly food frequency [
28] (Additional file
2: Table S1 for codes).
Statistical analysis
Multivariate linear and logistic regression analysis was completed using SPSS version 24 (IBM, New York, USA). All models were adjusted for age, sex, body mass index (BMI), hypertension, kidney disease, diabetes mellitus, and reported intake of cups of tea, fruit, vegetables, meat, fish, bread, cereal, and cheese. Beer and spirits were included as additional variables in separate analyses due to high levels of missing data, with only 70.4% of participants having information on beer/cider intake and 70.2% of participants with information on spirit intake. The effect of each SNP was analysed using two models, a dominant model (for the presence of least one urate-raising allele) and a recessive model (homozygosity for the urate-raising allele). Experiment-wide significance was set at P < 0.0125 after Bonferroni correction for multiple testing (0.05 divided by four SNPs). For gout, multivariate logistic regression analysis with gout as the dependent variable and coffee intake or SNP status as the independent variables was completed. For coffee intake, multivariate linear regression analysis was completed with coffee intake (cups per day) as the dependent variable and SNP status as the independent variables. All regression analyses were fully adjusted as described above.
The PROCESS macro v2.16.3 for SPSS was used to construct a mediation pathway to gout risk. Model 4 (used for simple mediation models exploring the relationship between a single independent variable, single mediator, and single dependent variable as presented in Additional file
3: Figure S2) was used from the PROCESS macro with 1000 bootstraps. This aimed to quantify the direct and indirect relationships between SNPs, coffee intake, and gout. Only SNPs associated with both gout and coffee consumption in the UK Biobank data (
GCKR and
ABCG2) were included in the mediation analysis; this aimed to establish whether the effects of
GCKR/
ABCG2 on gout risk were mediated through coffee intake. Direct and indirect standardised effect estimates were calculated using multivariate-adjusted linear and logistic regression analysis as described above. All effect estimates were adjusted for the same potential confounding variables as described above. Bootstrapping was used to determine whether the indirect effect of the SNP on gout risk through coffee intake was significant.
Discussion
This study provides further evidence that coffee intake is inversely associated with gout. This association was observed in those who consumed any coffee versus none, with evidence of a dose-effect response for the number of cups consumed per day. Although the urate-raising GCKR and ABCG2 alleles were associated with both lower coffee consumption and higher risk of gout, mediation analysis demonstrated that these SNPs largely influence gout risk directly, rather than indirectly through their dual effect on coffee consumption.
In our analysis of the UK Biobank, all of the loci from previous GWAS of habitual coffee consumption [
22] tested here (
ABCG2,
GKCR,
MLXIPL, and
CYP1A2) were replicated, strengthening evidence for the genetic basis of this trait.
CYP1A2 and its product, cytochrome P450, have been demonstrated to have a central role in the metabolism of caffeine [
29]. Indeed,
CYP1A2 is one of the strongest known loci for coffee consumption, both in this report and in other studies [
30]. Our results are consistent with a recent GWAS meta-analysis that reported genome-wide significant associations of the
CYP1A2 locus and nominal association of the
GCKR and
ABCG2 loci with plasma caffeine levels [
31].
Despite the association of
GCKR and
ABCG2 with both gout and reduced coffee intake, our mediation analysis indicates that the dominant mechanism for
GCKR and
ABCG2 on gout risk is not through coffee consumption. The urate-raising alleles of these SNPs are associated with reduced coffee intake; this suggests that implementing a dietary intervention such as increased coffee consumption may be more difficult in those genetically pre-disposed towards gout as urate levels would be raised. Analysis of modifiable exposures associated with gout risk alleles is an avenue for further study as GWAS for various dietary factors become more common. Indeed, a recent UK Biobank study identified several SNPs associated with alcohol consumption, including
GCKR rs1260326 [
32]. This is especially interesting given the recently reported interaction between
GCKR rs780094 (in very strong linkage disequilibrium with
rs1260326), alcohol, and gout risk [
33].
We did not use Mendelian randomisation to directly address the question of the causality of coffee consumption in regulating serum urate levels and the risk of gout. This is because the majority of the coffee consumption-associated genetic variants also associate with serum urate levels making the variants unsuitable as Mendelian randomisation instrumental variables. However, we note that the
AHR (aryl hydrocarbon receptor) locus, with an approximately equal effect size on coffee consumption as
CYP1A2 [
22], does not associate with serum urate levels [
2]. Consistent with our mediation analysis, this does not support a direct causal role of coffee consumption in regulating serum urate levels. Very recently, a conventional Mendelian randomisation study [
34] provided support for a causal role of coffee consumption in reducing the risk of gout. However, the authors of this Mendelian randomisation also observed in the separate gout case-control data set [
2] that genetic variation in
AHR, one of the strongest genetic effects on habitual coffee consumption [
22], did not contribute to the protective effect for gout. One of the fundamental assumptions of Mendelian randomisation is that the SNPs used for the instrumental variables are unrelated to the outcome [
35]. Although statistical methods exist for evaluating the effect of pleiotropy in Mendelian randomisation studies, the fact that coffee- and urate-associated loci overlap limits the application of Mendelian randomisation to assessing a possible causal role of coffee in gout.
This study has several limitations. Only three of the four urate-associated SNPs used in this study were identical to (or in very high LD with) the SNPs reported in the previous coffee GWAS. However, we did observe an association between coffee consumption and all four urate-associated SNPs tested in this analysis. The study population is restricted to those of white European ancestry, thereby limiting the generalisability to other populations. There may also be some bias in reported dietary intakes. Furthermore, widely used food-frequency questionnaires lack specificity for some foodstuffs [
36]. Any effect of this is likely minimised due to the UK Biobank’s extremely comprehensive data collection gathered through multiple modalities. Similarly, the non-specific nature of data collection for general use rather than specific research questions may also help minimise the effects of recall bias. Key strengths of this study include the large sample size and wide access to multiple data sources such as self-reported illness, medications, hospital records, and dietary data.
Conclusions
In summary, this work further supports the hypothesis that coffee consumption is protective for gout risk. The exact mechanism of this protective effect remains unclear. Although several SNPs associate with both lower total coffee consumption and higher gout risk, mediation analysis indicates that these SNPs have direct effects on gout risk rather than indirect effects mediated by coffee consumption. The coffee- and urate-associated loci could influence coffee consumption and urate levels, respectively, through separate biological mechanisms.