Mendelian randomization analysis associates increased serum urate, due to genetic variation in uric acid transporters, with improved renal function

Abstract

Increased serum urate predicts chronic kidney disease independent of other risk factors. The use of xanthine oxidase inhibitors coincides with improved renal function. Whether this is due to reduced serum urate or reduced production of oxidants by xanthine oxidase or another physiological mechanism remains unresolved. Here we applied Mendelian randomization, a statistical genetics approach allowing disentangling of cause and effect in the presence of potential confounding, to determine whether lowering of serum urate by genetic modulation of renal excretion benefits renal function using data from 7979 patients of the Atherosclerosis Risk in Communities and Framingham Heart studies. Mendelian randomization by the two-stage least squares method was done with serum urate as the exposure, a uric acid transporter genetic risk score as instrumental variable, and estimated glomerular filtration rate and serum creatinine as the outcomes. Increased genetic risk score was associated with significantly improved renal function in men but not in women. Analysis of individual genetic variants showed the effect size associated with serum urate did not correlate with that associated with renal function in the Mendelian randomization model. This is consistent with the possibility that the physiological action of these genetic variants in raising serum urate correlates directly with improved renal function. Further studies are required to understand the mechanism of the potential renal function protection mediated by xanthine oxidase inhibitors.

The cause–effect relationship between hyperuricemia and renal function is unclear. Epidemiological studies demonstrate that elevated serum urate (SU) is a modest predictor of chronic kidney disease, independent of other classical risk factors.^1,2 The relationship between elevated SU and reduced renal function is supported by clinical intervention studies where urate-lowering treatment, predominantly with the xanthine oxidase inhibitor allopurinol, improves renal function and slows progression of chronic kidney disease.³ In animals, hyperuricemia induces renal inflammation, preglomerular arterial disease and glomerular hypertension,^4–6 nephropathies that can be ameliorated and reversed by xanthine oxidase inhibition.⁶ However, the clinical significance and independence of these observations remain to be clarified.⁷ With regard to the epidemiological evidence, the observed relationship may be confounded by factors that have not been identified and measured and that may be extensively interrelated.⁸ The beneficial effects of allopurinol on renal function in human clinical trials and experimental studies could be mediated via effects of its active metabolite, oxypurinol, on inhibition of production of oxidants by xanthine oxidase⁹ and subsequent improvement of endothelial function,¹⁰ additional or alternative to inhibition of urate production. Disentangling and further understanding the basis of the SU and renal function relationship and the mechanism of protection of renal function by xanthine oxidase inhibition is important given the increasing clinical trial focus on the use of urate-lowering drugs for renal function.^11–14

Genetic variants in renal (SLC2A9, SLC17A1, SLC22A11, SLC22A12, ABCG2) and gut (ABCG2) uric acid transporter variants collectively explain, depending on sex, 3–5% of the variance in SU levels¹⁵ and are associated with gout.^15–18 Variants in 18 other genes have also recently been associated with SU levels, with two pathways identified—glucose metabolism and inhibins-activins signaling pathways.¹⁹ Genetic variants in these additional pathways account for <2% of the variance in SU concentrations.¹⁹ The glucose metabolism pathway may affect urate levels through metabolism of fructose, which is known to elevate SU levels.²⁰ How the inhibins–activins pathway influences SU is unclear.

Mendelian randomization (also known as instrumental variables analysis) is a statistical genetics approach that allows the disentangling of cause and effect in the presence of potential confounding. This technique has previously been used to demonstrate that increased SU is a consequence and not a cause of adiposity,^21,22 that elevated SU is causative of hypertension²³ and that elevated SU is not causal of type 2 diabetes²⁴ or metabolic syndrome.²⁵ This analysis assumes that inherited genetic risk variants for one phenotype are naturally randomized at conception with respect to a second phenotype, and can therefore be used as instrumental variables to control confounding, and are not influenced by health outcomes, so that associations are not affected by reverse causality.^26,27 A Mendelian randomization approach has previously been applied to the relationship between SU and renal function (estimated glomerular filteration rate (eGFR)), with no association between a urate genetic risk score and eGFR.¹⁵ However, this was done by simply regressing the genetic risk score (the instrumental variable) against eGFR. An alternative method, two-stage least squares,²⁸ regresses the instrumental variable (e.g. genetic risk score for SU) against the outcome variable (e.g. eGFR) using changes in measure (e.g. SU) directly attributable to the instrumental genetic risk score variable. This approach would be expected to be less confounded than the straightforward regression approach. Here we used Mendelian randomization by the two-stage least squares approach to study the relationship between SU and renal function.

RESULTS

Using a genetic risk score (Table 1) as an instrumental variable for SU, we conducted Mendelian randomization using the two-stage least squares approach for renal function with the uric acid transporter genetic risk score as instrumental variable (Table 2). Two-stage least squares is a form of Mendelian randomization, as is a standard genetic association study between a marker and phenotype, with the advantage that the two-stage least squares approach provides a quantitative effect of exposure (urate) on outcome (renal function). Mean SU, serum creatinine (SCr) and eGFR according to SU genetic risk score is shown in Supplementary Table S1 online. As genetic risk score increased, SCr decreased considerably more in males (SCr ranged from 103.05 to 84.66 μmol/l) than in females (range 80.44 to 79.56 μmol/l). There was evidence for a causal role for SU in determining SCr and eGFR in males (Table 2; β = 45.06 μmol/l decrease in SCr and 39.26 ml/min per 1.73 m² increase in eGFR from each unit increase in SU attributable to the genetic risk score, P = 0.020 and 0.045, respectively). However, the direction of effect was inverse to that predicted from the SU vs. eGFR and SCr regression analyses (Table 2; Durbin–Hausman P<1×10⁻⁴). Thus, using a uric acid transporter genetic risk score as an instrumental variable, there was evidence that increased SU caused by genetic variation in uric acid transporters improved renal function in males.

Table 1.

Association between uric acid transporter genetic risk score and SU

	SU transporter genetic risk score
	F-statistica	R2a
ARIC
Overall	114.81	0.0215
Men	63.05	0.0256
Women	99.25	0.0335
FHS
Overall	73.32	0.0259
Men	27.56	0.0215
Women	94.73	0.0593
Combined
Overall	190.74	0.0231
Men	90.6	0.0242
Women	185.14	0.0409

Abbreviations: ARIC, Atherosclerosis Risk in Communities; FHS, Framingham Heart Study; SU, serum urate.

Single-nucleotide polymorphisms rs11942223, rs2231142, rs1183201, rs2078267, and rs3825018 were used for both ARIC and FHS for SLC2A9, ABCG2, SLC17A1, SLC22A11, and SLC22A12, respectively.

^aF-statistic represents the strength, and R² the percent variance in SU explained, of the association between the genetic risk score and SU.

Table 2.

Mendelian randomization analysis of SU against eGFR/SCr using uric acid transporter genetic risk score as instrumental variable

	Ordinary least square regression			Two-stage least square			DH P
	βa	SEb	P	βa	SE	P
SCr
All
Crudec	87.20	1.70	4.01E – 305	− 13.15	13.49	0.33	<0.0001
Adjustedd	37.14	2.04	1.42E – 72	− 19.23	10.76	0.07	<0.0001
Males
Crude	35.75	2.76	1.18E – 37	− 46.52	19.90	0.019	< 0.0001
Adjusted	38.32	3.08	7.33E – 35	− 45.06	19.42	0.020	< 0.0001
Females
Crude	32.31	2.44	2.22E – 39	− 3.04	12.37	0.81	0.0029
Adjusted	37.14	2.74	6.31E – 41	1.05	12.50	0.93	0.0025
eGFR
All
Crude	− 11.20	1.79	4.05E – 10	14.75	12.08	0.22	0.10
Adjusted	− 37.04	2.33	4.01E – 56	12.20	12.09	0.31	< 0.0001
Males
Crude	− 30.60	2.95	6.03E – 25	37.11	20.49	0.07	0.001
Adjusted	− 35.04	3.16	3.89E – 28	39.26	19.54	0.045	< 0.0001
Females
Crude	− 34.61	3.17	2.56E – 27	− 0.84	16.04	0.96	0.11
Adjusted	− 36.73	3.45	4.33E – 26	− 6.34	15.65	0.69	0.11

Abbreviations: DH, Durbin-Hausman; eGFR, estimated glomerular filteration rate; SCr, serum creatinine.

The left side is the standard linear (ordinary least square) regression between the explained variables (serum urate and SCr/eGFR) and the right side is the two-stage least squares analysis.

^aβ represents the change in SCr (μmol/l) attributed to a unit change in serum urate in the linear regression (on the left) and the change in SCr (μmol/l) and eGFR (ml/min per 1.73 m²) caused by a unit change in serum urate attributed to the instrumental variable in the two-stage least squares analysis (on the right).

^bs.e.

^cAdjusted by the sample set.

^dAdjusted by the sample set, sex (for overall), body mass index, systolic blood pressure, diastolic blood pressure, age and first two eigenvalues of genome-wide single-nucleotide polymorphism principal components analysis.

We next used linear regression to further characterize the association of the uric acid transporter genetic risk score with renal function in sex-specific SU quartiles (Table 3; Supplementary Table S1 online). This revealed correlation of an increase in uric acid transporter genetic risk score with improved renal function in hyperuricemic men and women (SCr; β = −1.057, P = 5×10⁻⁴ and β = −0.630, P = 0.009, respectively).

Table 3.

Linear regression analysis of uric acid transporter instrumental variable with SCr/eGFR as the dependent variables within SU quartiles

	SCr						eGFR
	Males			Females			Males			Females
	βa	SE	P	β	SE	P	β	SE	P	β	SE	P
Q1b
ARIC
Unadj.	−0.613	0.353	0.083	−0.105	0.242	0.663	0.510	0.297	0.086	0.166	0.230	0.469
Adj.c	−0.582	0.363	0.109	0.074	0.244	0.760	0.496	0.302	0.101	−0.058	0.236	0.806
FHS
Unadj.	−0.143	0.392	0.716	−0.126	0.314	0.688	0.331	0.646	0.609	0.444	0.670	0.508
Adj.	−0.110	0.425	0.796	0.055	0.344	0.874	0.206	0.695	0.768	0.120	0.682	0.860
Comb.
Unadj.	−0.495	0.262	0.059	−0.185	0.196	0.345	0.515	0.312	0.099	0.325	0.316	0.304
Adj.	−0.465	0.274	0.090	−0.091	0.205	0.659	0.464	0.315	0.141	0.212	0.321	0.509
Q2
ARIC
Unadj.	−0.769	0.319	0.016	−0.113	0.258	0.662	0.575	0.262	0.028	0.109	0.234	0.641
Adj.	−0.839	0.331	0.011	−0.182	0.267	0.496	0.637	0.266	0.017	0.099	0.240	0.678
FHS
Unadj.	−0.605	0.416	0.146	−0.074	0.332	0.825	0.455	0.650	0.484	0.134	0.641	0.834
Adj.	−0.431	0.438	0.326	−0.312	0.360	0.386	0.454	0.665	0.495	0.689	0.685	0.315
Comb.
Unadj.	−0.630	0.256	0.014	−0.090	0.198	0.649	0.398	0.272	0.144	0.068	0.273	0.804
Adj.	−0.679	0.267	0.011	−0.186	0.207	0.369	0.518	0.272	0.058	0.089	0.282	0.752
Q3
ARIC
Unadj.	−0.402	0.341	0.239	−0.701	0.264	0.008	0.267	0.252	0.289	0.569	0.229	0.013
Adj.	−0.455	0.363	0.211	−0.540	0.273	0.048	0.376	0.264	0.155	0.402	0.233	0.085
FHS
Unadj.	−0.038	0.423	0.929	−0.012	0.328	0.971	−0.402	0.654	0.539	−0.194	0.669	0.772
Adj.	−0.283	0.465	0.543	−0.057	0.347	0.869	0.305	0.699	0.663	−0.153	0.703	0.828
Comb.
Unadj.	−0.329	0.266	0.217	−0.317	0.207	0.126	0.139	0.271	0.609	0.266	0.244	0.277
Adj.	−0.413	0.287	0.150	−0.262	0.215	0.223	0.321	0.280	0.252	0.184	0.245	0.452
Q4
ARIC
Unadj.	−1.006	0.374	0.007	−0.522	0.303	0.086	0.577	0.272	0.034	0.506	0.244	0.038
Adj.	−1.023	0.388	0.009	−0.590	0.311	0.058	0.621	0.277	0.025	0.525	0.247	0.034
FHS
Unadj.	−1.029	0.431	0.018	−0.309	0.348	0.376	1.709	0.667	0.011	0.163	0.680	0.811
Adj.	−1.139	0.454	0.013	−0.571	0.365	0.118	1.822	0.674	0.007	0.862	0.684	0.208
Comb.
Unadj.	−0.999	0.287	0.001	−0.562	0.233	0.016	0.956	0.292	0.001	0.460	0.258	0.075
Adj.	−1.057	0.301	<0.001	−0.630	0.242	0.009	0.996	0.291	0.001	0.540	0.254	0.034

Abbreviations: Adj., adjusted; Comb., combined; Unadj., unadjusted.

^aβ represents the change in SCr (μmol/l) or eGFR (ml/min per 1.73 m²) resulting from each unit increase in serum urate instrumental variable.

^bQuartile ranges are presented in Supplementary Table S3 online.

^cAdjusted by sample set (in combined), body mass index, systolic blood pressure, diastolic blood pressure, age and first two eigenvalues of genome-wide single-nucleotide polymorphism principal components analysis.

The use of an instrumental variable in Mendelian randomization requires that it fulfills three assumptions.²⁶ First, the uric acid transporter genetic risk score is strongly associated with SU, with F-statistics being considerably greater than 10 and explaining >2% of the variance in SU (Table 1), indicating an adequate instrumental variable.²⁹ Second, that the instrumental variable is independent of the factors that confound the association of SU and renal function. This can be partly verified by testing the instrumental variable for association by linear regression with the measured confounders adjusted for here (Supplementary Table S2 online)—there was no evidence for association of the uric acid transporter instrumental variable with any of the confounders tested. Third, that the uric acid transporter instrumental variable has an effect on renal function solely via SU (i.e. there are no pleiotropic effects). This was verified to a limited extent by adjusting the Mendelian randomization analysis with covariates (age, sex, systolic and diastolic blood pressure, body mass index) that could potentially influence the association between SU and renal function. However, it is not possible to fully verify that the third assumption is satisfied given the likelihood that other pleiotropic effects exist (refer Discussion). To check for a role of population stratification, we also included the first two eigenvalues from genome-wide single-nucleotide polymorphism (SNP) principal components analysis as covariates.

Given that the five uric transporters used in the genetic risk score either exchange uric acid for other metabolites (SLC2A9, SLC22A11, SLC22A12, SLC17A1) or deplete adenosine triphosphate (ABCG2), it is possible that the instrumental variable we used violates the third assumption via the biochemical effect of uric acid transporter activity on renal function. To investigate this further, we tested the individual SNPs comprising the genetic risk score for association with renal function in males by two-stage least squares Mendelian randomization (Table 4)—if the effect on improved renal function was mediated directly by urate, then the variants with the strongest effects on SU level (SLC2A9 in particular) would be expected to have the strongest effect on renal function. In support of the possibility that the genetic risk score violates the third assumption of Mendelian randomization, we observed that the variant with the strongest effect in protection of renal function was within SLC22A11 (OAT4) (Table 4; β = −296.4, P = 0.08, P_{Durbin – Hausman} <1×10⁻⁴), which explained only a small proportion of the variation in SU (0.21%). Conversely, the SLC2A9 variant with the strongest influence on SU levels (explaining 1.94% of the variance in males, 9.2-fold greater than SLC22A11) had a considerably weaker influence on renal function (β = 14.9, P = 0.50, P_{Durbin – Hausman} = 0.01). The SU-increasing alleles within SLC2A9, ABCG2, SLC22A11, and SLC22A12 all had negative β-values in two-stage least square regression (≤ −14.9), with SLC2A9 and ABCG2 having significant Durbin – Hausman P values. In contrast, the SLC17A1 β-value was positive (12.21, P_{Durbin–Hausman} = 0.63).

Table 4.

Mendelian randomization analysis of SU against SCr using single genetic variants in males in the combined ARIC and FHS cohorts

	F-statistica	R2a	βb	SE	P	DH P
rs11942223 (SLC2A9)	73.34	0.0194	−14.88	22.20	0.503	0.011
rs2231142 (ABCG2)	50.08	0.0133	−21.49	24.67	0.384	0.010
rs2078267 (SLC22A11)	7.71	0.0021	−296.4	170.9	0.083	<0.0001
rs1183201 (SLC17A1)	6.93	0.0019	12.21	54.82	0.824	0.63
rs3825018 (SLC22A12)	5.21	0.0014	−85.23	80.04	0.287	0.06

Abbreviations: ARIC, Atherosclerosis Risk in Communities; FHS, Framingham Heart studies; SCr, serum creatinine; SU, serum urate.

All analyses adjusted by sample set (in combined), body mass index, systolic blood pressure, diastolic blood pressure, age and first two eigenvalues of genome-wide single-nucleotide polymorphism principal components analysis.

^aF-statistic and R² represent the association of individual single-nucleotide polymorphisms with SU.

^bβ results from the two-stage least squares analysis and corresponds to the change in SCr (μmol/l) per risk allele.

DISCUSSION

Using the statistical genetics approach of Mendelian randomization we have provided evidence that the uric acid transporter genetic risk score positively associates with improved renal function in European Caucasian males. However, as discussed below, the data are consistent with the hypothesis that the activity of the uric acid transporters, rather than SU itself, is protective of renal function. Subjects self-reporting renal disease or gout, or taking anti-hypertensive medication, were excluded from the study. There was an inverse causal relationship of the uric acid transporter instrumental variable in men in an opposing direction to that of the relationship between SU and SCr (P = 0.020), with a highly significant Durbin–Hausman P value (Table 2). Our data are consistent with those of McKeigue et al., who used a Bayesian-based Mendelian randomization method and SLC2A9 genotype as an instrumental variable to show that raised SU may protect against metabolic syndrome.²⁵ Our findings will require replication in other data sets. The P value for the two-stage least squares analysis in males was 0.020, which has to be considered in the context of the multiple testing (sex stratification, P_c = 0.040), which increases the possibility of type 1 error. However, P = 5×10⁻⁴ in quartile 4 SU males (Table 3) is robust to multiple testing (sex stratification and four quartiles analyzed, P_c = 0.004).

The finding that the two-stage least squares β-values in men are in an opposite direction to the SU vs. renal function relationship (i.e. the significant Durbin–Hausman P values) (Table 2) provides the evidence for inverse causality. Our instrumental variable explained only a small proportion of variance in SU (2.4% in men and 4.1% in women), meaning that other unidentified factors would be expected to confound the relationship between SU and renal function observed in conventional linear regression. A similar phenomenon was observed in the relationship between C-reactive protein and metabolic risk factors in British women where conventional multivariate-adjusted analysis shows a positive linear relationship.³⁰ However, use of Mendelian randomization by two-stage least squares, with C-reactive protein haplotypes as instrumental variables, demonstrated that increased C-reactive protein was associated with a reduced homeostatis model assessment for insulin resistance, body mass index, and triglyceride levels, as demonstrated by significant Durbin–Hausman statistics. This provides evidence that circulating levels of C-reactive protein are not causal of insulin resistance and increased body mass index and triglycerides, with the conventional association being influenced by residual confounding and/or reverse causality.

Determining the pathophysiological role of urate in metabolic disease is complicated by the disconnection between epidemiological evidence and current understanding of the physiological roles of urate. Whilst urate is a direct cause of pathological disorders owing to the deposit of monosodium urate crystals and consequent detrimental effect on health, much of this evidence is difficult to reconcile with the increasing evidence for beneficial effects of urate, most conspicuously its antioxidant effect. In the case of kidney disease, the situation is paradoxical. A possible role for urate in improving renal function, as intimated by the statistical genetics Mendelian randomization approach used here, conflicts with the current evidence. It is well established that urate can cause acute nephropathy as a result of precipitation within the renal tubules producing a chronic inflammatory response and kidney damage. Experimentally, induction of mild hyperuricemia in rats with the use of the uricase inhibitor, oxonic acid, induced vasoconstriction and glomerular hypertension.⁶ Also, in animals, experimentally raising and lowering the SU correlates with reduced and improved renal function, respectively.⁴ Clinical intervention studies indicate that lowering SU levels by treatment with allopurinol improves renal function.^11–14 However, it is not possible to separate out other possible effects of allopurinol, for example by inhibiting the free radical-producing effects of xanthine oxidase⁹ or direct interference of the active metabolite oxypurinol with the activity of renal uric acid transporters³¹ and a resultant direct influence on SU or beneficial effect on renal tubule biochemistry and function. Retrospective evaluation of renal function in chronic kidney disease stage-3 patients revealed that those treated with benzbromarone did not have improved renal function,³² whereas administration of rasburicase (recombinant urate oxidase) reduced the SU and improved renal function in a small elderly sample.³³ In chronic heart failure, allopurinol improves endothelial function by reducing vascular oxidative stress and not by reducing SU.¹⁰ Considering the effects of urate on the oxidative status of cells and tissues, this appears to be context dependent. Urate is an established antioxidant, neutralizing a significant proportion of serum free radicals.³⁴ However, intracellular urate can act as a pro-oxidant.⁹ Because the Mendelian randomization instrumental variable used here was derived from genetic variants associated with serum (extracellular) urate, it is possible to speculate that the antioxidant role of urate can protect renal function. However, as discussed below, our data are more consistent with an alternative explanation.

The possibility of pathways distinct from SU levels influencing improved renal function was supported by the observation that the individual genetic variants of the genetic risk score with the strongest effect on SU did not have the strongest beneficial effect on renal function (Table 4). This is consistent with the possibility that the physiological action of these genetic variants in raising SU correlates directly with improved kidney function, perhaps influencing tubule biochemistry by exchanging uric acid for other metabolites and cofactors. SLC2A9 exchanges hexose sugars for uric acid, ABCG2 is an adenosine triphosphate-dependent secretor of uric acid, URAT1 reabsorbs luminal uric acid in exchange for intracellular organic and inorganic anions (e.g. Cl⁻, lactate), OAT4 exchanges organic anions (including uric acid) for dicarboxylates, and NPT1 is a Cl⁻-dependent polyspecific organic anion transporter (reviewed in Anzai and Endou³⁵). This possibility would be consistent with two important observations in our data: that the protective effect of increased genetic risk score on renal function is the strongest in the quartile with the highest SU levels (Table 3) and the observation that the uric acid transporter genetic risk score positively associates with improved renal function only in males by two-stage least squares analysis (Table 2). Fractional excretion of uric acid is lower (i.e. a lower transporter excretory capacity) in men than in women^36,37 and the highest SU quartile would be enriched for uric acid under-excretors, consistent with the hypothesis that renal SU control rather than SU levels per se is protecting renal function. Data on fractional excretion of uric acid are not available in the Atherosclerosis Risk in Communities (ARIC) or Framingham Heart Study (FHS) cohorts, which precludes testing of this hypothesis. A Mendelian randomization study with fractional excretion of uric acid as exposure (instead of SU) would therefore be very informative. If the activity of uric acid transporters were a factor in the improved renal function associated with increased genetic risk score, then the uric acid transporter genetic risk score would violate the third assumption for an instrumental variable used in Mendelian randomization, namely that any effect on renal function would be mediated only by the direct effects of urate.

Overall, the relationship between SU and renal function is likely to be very complex, involving beneficial and detrimental pathways and not necessarily mediated directly through urate, involving direct effects (perhaps in both directions) and input of non-genetic factors possibly also interacting with genetic factors. Our data are consistent with the possibility that the activity of renal uric acid transporters in raising SU is beneficial to renal function. Therefore, future research aimed towards understanding the relationship between SU genetic risk score and renal function will require the use of large well-phenotyped cohorts that include robust dietary information and data on fractional excretion of uric acid, to include in Mendelian randomization models. Our findings suggest that the relationship of urate with renal function is complex and that the molecular mechanism(s) whereby allopurinol clinically improves renal function requires further examination.

MATERIALS AND METHODS

Subjects

Subjects were included from the ARIC and FHS longitudinal cohorts and comprised 5237 (2456 males and 2907 females) and 2742 (generation 3 only; 1258 males and 1514 females) individuals, respectively, of European Caucasian ancestry (Supplementary Table S4 online). Individuals taking antihypertensive medication and who self-reported physician-diagnosed kidney disease or gout were excluded from the analysis. Data from Exam 1 were used for ARIC (1987–1989) and from Exam 1 (2002–2005) for FHS Gen3. In the ARIC and FHS cohorts, a modified Jaffe method was used to measure SCr levels. DART creatinine reagent (Coulter Electronics, Hialeah, FL) was used and analyzed by the Dacos Chemistry Analyser (Coulter Electronics). eGFR was calculated using the Modification of Diet in Renal Disease formula: eGFR (ml/min per 1.73 m²) = 32,788×SCr^−1.154×Age^−0.203×(0.742 if female). In ARIC, urate was measured using the uricase oxidation method. Dri-STAT Uric Acid-W Endpoint Reagent (Beckman Instruments, Carlsbad, CA) was used and analyzed by the Dacos Chemistry Analyser (Coulter Electronics). In FHS, urate was measured by a carbonate-based method as previously described.³⁸ The research procedures were in accordance with the ethical standards of the institutional review boards relevant to the ARIC and FHS.

Statistical analysis

All analyses were done using STATA version 8.0 (StataCorp, College Station, TX). The Mendelian randomization approach by two-stage least squares regression was done as previously described²² using the genetic risk score instrumental variable selected as described below. In the two-stage least squares regression, SU was first regressed by ordinary least square on the genetic instrumental variable. The response under examination was then regressed on the changes in measure (fitted values) associated with the genetic risk score. For example, to test a causal effect of SU in SCr levels, the change in SU resulting from the SU genetic risk score instrument in step 1 was regressed against the SCr in step 2. Provided that the assumptions underlying Mendelian randomization are not violated,²⁶ the regression coefficient obtained in the second stage can be interpreted as being the causal effect of the ‘explained’ variable on the response of interest. The estimates derived from the regression between the explained variables (SU and eGFR) and the two-stage least squares regression were then compared using the Durbin–Hausman test.³⁹ The Durbin–Hausman test, which examines the null hypothesis that the ordinary least square estimator of the relationship between the two explained variables (SU and SCr/eGFR) yields consistent estimates when compared with the instrumental variable analysis, indicates if the instrumental variable effect on the response can be attributed to the explained variable (with a P value reflecting similar association in magnitude and direction to that with the explained variable) or if the two-stage least squares regression is not accurately predicting the relationship with the explained variable that is explained by unknown confounder(s), or if there is a causal relationship in a direction opposite to the relationship with the explained variable.

Histograms and residual plots were examined to determine whether normative transformations need to be performed on SU and eGFR. As none of the histograms were skewed and residual values were evenly distributed above and below zero, transformations were not required. The two-stage least squares analyses were conducted using the ivreg function on STATA 8.0, where the exposure represented the endogenous variable, outcome the dependent variable, and the appropriate genetic risk score the instrumental variable. Normal linear regression analyses were carried out using the regress function on STATA 8.0. P value <0.05 was regarded as nominally significant. The F-statistic was calculated in the first stage of regression. To address the possibility of confounding by population stratification, eigenvalues corresponding to the first two principal components generated from genome-wide SNP data (Affymetrix 6.0 and 5.0 platforms for ARIC and FHS, respectively) using SMARTPCA⁴⁰ were added as covariates to the analysis.

Genetic instrumental variables

For the uric acid transporter genetic risk score, SU-increasing alleles were those as reported previously.^15,19 Although a weighted genetic risk score does have merit, substantial bias can result if the data under analysis are included in those used to derive the weighting for each associated allele (Stephen Burgess, personal communication). Therefore, we used an allele-counting genetic risk score and each SNP was coded 0–2 based on the number of alleles that associated with the increase in SU and scores were combined, ranging from 0 to 10. SNPs used in the uric acid transporter instrumental variable were rs11942223 (SLC2A9), rs2231142 (ABCG2), rs1183201 (SLC17A1), rs2078267 (SLC22A11), and rs3825018 (SLC22A12) for both ARIC and FHS. Largely because the uric acid transporter genetic risk score was an adequate instrumental variable (Table 1) we did not use SNPs from the glycolysis or inhibins–activins pathways.¹⁹ Another reason to not use SNPs in the SU glycolysis pathway was because of concerns that they would violate one assumption of Mendelian randomization, namely that the instrumental variable is independent of the factors that confound the association between SU and renal function.²⁶ For example, the GCKR gene, associated with SU levels and involved in glycolysis,¹⁹ has previously been reported as associated with body mass index.⁴¹ For two-stage least squares analysis, although there is not yet a standard approach for estimating power, estimations can be done using repetitive data set simulations. Published simulations⁴² indicate that a sample set of 8270 individuals with one instrumental variable is well powered, where R² is 0.01–0.05 (similar to our instrumental variable (Table 1)); when effect size (β_XY) is ⩾0.3, power is 100%.

Supplementary Material

Supplemental Tables

Table S1. Serum urate, serum creatinine and eGFR according to uric acid transporter genetic risk score in the combined ARIC and FHS sample sets.

Table S2. Analysis of association of SU genetic risk score instrument variable with tested confounders.

Table S3. Serum urate quartile ranges (mmol/l).

Table S4. Summary characteristics of the ARIC and FHS data sets.