Abstract
Background Elevated serum urate has been associated with an increased risk of cardiovascular disease (CVD), but it is not known whether this relationship is causal, non-linear or sex-specific. Basic science and clinical trial data have supported an effect of serum urate on systolic blood pressure (SBP), which may in turn mediate an effect on CVD risk.
Methods Using recently available data from the Million Veterans Program and UK Biobank, we applied improved genetic instruments for serum urate in one and two-sample Mendelian randomization (MR) analyses to investigate for effects on risk of coronary heart disease (CHD), peripheral artery disease (PAD), and stroke. Potential mediation through SBP was explored, as well as evidence of non-linear and sex-specific effects. We performed a number of sensitivity analyses using different MR methodologies and populations to investigate consistency of our findings.
Results Higher genetically predicted serum urate was associated with an increased risk of CHD (odds ratio [OR] per 1-standard deviation increase in genetically predicted urate 1.19, 95% confidence interval [CI] 1.10-1.30, P=4×10−5), PAD (OR 1.12, 95% CI 1.03-1.21, P=9×10−3), and stroke (OR 1.11, 95% CI 1.05-1.18, P=2×10−4). Similar estimates were produced when applying MR methods that make distinct assumptions, or when considering different populations. SBP was estimated to mediate 29%, 44% and 45% of the effect of urate on CHD, PAD and stroke respectively. There was no evidence of non-linear or sex-specific effects of genetically predicted urate on CVD risk.
Conclusions Using contemporary data and MR methods, we provide support for an effect of serum urate on CVD risk that may partly be mediated through SBP. High-quality trials are necessary to provide definitive evidence on the cardiovascular benefit of urate lowering.
Background
Urate is a breakdown product of purine metabolism. Its raised levels have been associated with a number of adverse health outcomes including gout, hypertension and cardiovascular disease (CVD)1. Efficacious treatments such as allopurinol and probenecid are available for lowering serum urate in the prevention and treatment of gout. While observational studies consistently support associations of urate with diseases other than gout2, it remains unclear whether these represent causal effects. The relationship between serum urate, obesity, diet and other cardiovascular risk factors raises considerable potential for confounding and reverse causation3.
Pre-clinical studies support a causal role for urate in hypertension4. Randomized clinical trial data have shown that both allopurinol and probenecid reduce systolic blood pressure (SBP) in hyperuricemic adolescents1, 5. As elevated blood pressure is a risk factor for CVD, it is important to clarify any role of SBP in mediating an effect of urate on cardiovascular outcomes. Recent trials have highlighted potential non-linear effects of urate lowering6, and numerous observational studies have suggested that the relationship between serum urate and CVD is stronger in women than men7-9. The presence of non-linear or sex-specific relationships between urate and cardiovascular risk therefore also require further exploration.
The genetic determinants of serum urate levels have been increasingly well-characterized3. This has made it possible to identify better instruments for Mendelian randomization (MR) analyses investigating the effect of genetically predicted serum urate on cardiovascular outcomes than in previous efforts10-16. The use of variants randomly allocated at conception to proxy the effect of modifying serum urate means that MR is less susceptible to the environmental confounding, measurement error and reverse causation bias that can limit causal inference in traditional epidemiological approaches. Previous MR studies have relied on much smaller genome-wide association studies (GWASs) for serum urate than are currently available3, 10-17, and furthermore did not incorporate recently developed mediation analysis and non-linear MR methods18-20.
Using the largest available GWASs of serum urate and the increased number of instruments available3, 17, we update previous MR analyses investigating the effect of serum urate on coronary heart disease (CHD) and stroke with improved precision, and now also investigate peripheral arterial disease (PAD) as an outcome. MR methods that are robust to pleiotropic effects of the genetic variants employed as instruments for urate are applied, with formal mediation analyses integrated to investigate for any role of SBP in mediating the effect of urate on risk of CVD outcomes. Finally, we explore for potential non-linear and sex-specific effects of urate on CVD outcomes. Our aim was to perform a comprehensive investigation into the effect of genetically predicted serum urate on cardiovascular outcomes, and thus provide crucial insight to guide the design of future clinical trials.
Methods
Genetic association estimates
In two-sample MR, genetic association estimates for serum urate were obtained by using PLINK software to meta-analyze summary data from GWAS analyses of 110,347 European-ancestry individuals and 343,836 White British UK Biobank participants respectively17, 21, 22. Urate estimates are provided per 1-standard deviation (SD) increase throughout, which corresponds to 80.3µmol/L. For consideration of SBP as a mediator in the two-sample multivariable MR, genetic association estimates were obtained from a GWAS of 317,195 White British UK Biobank participants, where SBP was measured using automated readings with correction made for any anti-hypertensive drug use by adding 10mmHg to the measured reading23. In contrast, when investigating SBP as an outcome, genetic association estimates were obtained from the International Consortium for Blood Pressure (ICBP) GWAS analysis of 287,245 European-ancestry individuals (excluding UK Biobank participants)24. A different population was considered when studying SBP as an outcome to avoid overlap with the UK Biobank participants used to obtain genetic association estimates for urate, as this can bias MR estimates25. The ICBP study adjusted for body mass index (BMI) when obtaining genetic association estimates for SBP, and hence these data were not meta-analyzed with UK Biobank SBP genetic association estimates in the study of SBP as a mediator, as such BMI adjustment can also bias MR estimates26. SBP estimates are provided per 1-SD increase, which corresponds to 18.6mmHg. Genetic association estimates for CHD were obtained from the CARDIoGRAMplusC4D Consortium 1000G multi-ethnic GWAS (77% European-ancestry) of 60,801 cases and 123,504 controls, with a broad and inclusive definition of CHD applied27. Genetic association estimates for PAD were obtained from the Million Veterans Program (MVP) multi-ethnic (72% European-ancestry) GWAS of 31,307 cases and 211,753 controls, with case definitions made using hospital diagnosis and procedure codes28. Genetic association estimates for stroke were obtained from the MEGASTROKE multi-ethnic (86% European-ancestry) GWAS of 67,162 cases and 454,450 controls29, with the stroke definition including both ischemic and hemorrhagic etiologies.
One-sample MR analysis was performed in the UK Biobank, a prospective cohort study of approximately 500,000 participants recruited between 2006 and 2010 from 22 assessment centers across the UK30. Self-reported, physical, biochemical and genetic information was collected and participants were linked retrospectively and prospectively to their hospital episode statistics as previously described30. Analysis was restricted to self-reported White British participants. In order to increase statistical power while also maintaining consistency with the disease outcomes considered in the two-sample MR, CHD, PAD and stroke cases were pooled as CVD events. International Classification of Diseases version 10 codes were used to identify cases (Supplementary Table 1). To obtain estimates for the association between the urate GRS and serum urate in UK Biobank participants, the score was regressed in a linear model against measured levels, with adjustment made for age, age2, sex, recruitment center and the first four principal components. To obtain genetic association estimates for the urate GRS with risk of incident CVD, a Cox proportional hazards regression was used, with adjustment similarly made for age, age2, sex, recruitment center and the first four principal components. Cases where death occurred prior to any incident CVD event were censored. Genetic association estimates of the urate GRS with SBP were obtained by performing a linear regression against SBP with the same adjustments as above. SBP was calculated as the mean of the two automated readings taken on the initial assessment.
Genetic variants used as instruments
Genetic instruments for the two-sample MR were identified as single-nucleotide polymorphisms (SNPs) that were associated with urate (or SBP, in mediation analyses) at genome-wide significance (P<5×10−8) and were in pair-wise linkage disequilibrium (LD) with R2<0.001. Clumping was performed using the TwoSampleMR package of R31. For univariable MR, instrument strength was estimated using the F-statistic, with variance in the exposure explained assessed using the R2 value11.
A weighted genetic risk score (GRS) for serum urate was used as the instrument in the one-sample MR analysis. This was constructed using SNPs with minor allele frequency >0.05 that were associated with urate at P<5×10−8 in the summary data from a GWAS performed in 288,666 European-ancestry individuals, clumped to a pairwise LD R2<0.001 using the TwoSampleMR package of R3, 31. The UK Biobank population used to select instruments in the two-sample MR analysis was not included here to avoid overlap with the UK Biobank cohort used to obtain genetic association estimates with outcomes25. Similarly, the population used to derive the weighted GRS for urate in the one-sample analysis was not also included when deriving instruments for the two-sample analysis because this would have resulted in overlap of the MVP population that was also used to obtain genetic association estimates for PAD3, 25, 28. The GRS in the one-sample MR was weighted for the association of the constituent variants with serum urate, and for each individual participant was measured as the sum of the imputed number of alleles present multiplied by their variant’s weight.
Statistical analysis
In all analyses, SNPs were aligned by their effect alleles and no additional consideration was made for palindromic variants. Two-sample MR analyses were performed to investigate the effect of genetically predicted serum urate on CHD, PAD, stroke and SBP respectively. A Bonferroni threshold (P<0.01) that corrected for multiple testing related to the four outcomes was used to ascertain statistical significance in the main analysis. Inverse-variance weighted MR was used in the main analysis, with the simple median32, contamination-mixture method33, PRESSO34 and multivariable MR35 (only for CVD outcomes) sensitivity analyses used to explore the robustness of the findings to potential pleiotropy of the genetic variants. Given the previously demonstrated overlap in the genetic determinants of urate with other metabolic traits3, the multivariable MR sensitivity analysis adjusted for genetic associations of the instruments with BMI, estimated glomerular filtration rate, type 2 diabetes mellitus, serum low-density lipoprotein cholesterol, serum high-density lipoprotein cholesterol and serum triglycerides together in the same model. Such multivariable MR was not performed when considering SBP as an outcome, due to population overlap with the cohorts used to obtain genetic association estimates for the metabolic exposures25. It was expected that any direct effects of the instruments on the considered cardiovascular outcomes might be related to their effect on serum urate3. Thus, the simple median estimator was preferred to its precision weighted counterpart and the MR-Egger method was avoided, as these methods are unreliable in such scenarios32, 36. In MR mediation analyses, multivariable MR was applied in the two-sample setting to adjust for the genetic association of the instruments with SBP, and network MR was used to estimate the proportion of the total effect of urate on each cardiovascular outcome that is mediated through SBP37. Standard errors were estimated using the propagation of error method. Further details on these analyses are provided in the Supplementary Methods.
To obtain MR estimates in the one-sample analysis, the ratio method was used, with first order weights for estimating standard errors38. To investigate for any discrepancy in the effects of urate on CVD risk and SBP respectively based on observed urate levels (non-linear effects) and sex (sex-specific effects), the cohort was stratified based on these traits and MR analysis was performed for each stratum separately. Rather than stratify on measured urate itself however, the cohort was divided into four quantiles based on residual urate, which is the measured urate minus the genetically raised urate as determined by each individual’s weighted GRS. Such stratification by residual urate is necessary, as stratifying on measured urate directly would introduce collider bias if urate is on the causal pathway from the genetic variants to the outcomes18. For consideration of sex-specific effects, the cohort was simply divided by self-reported sex. Exposure and outcome genetic association estimates for each stratum were calculated separately. The Cochran’s Q test was used to investigate heterogeneity between strata based on residual urate and sex respectively, with P<0.05 denoting statistical significance.
Ethical approval, data availability and reporting
All data used in this work are either publicly accessible or available on request from their original studies, which obtained appropriate patient consent and ethical approval. The UK Biobank data were accessed through application 236. All results generated in this work are presented in either the manuscript or its supplementary files. This study has been reported as per the STROBE-MR guidelines (Supplementary Checklist)39. The study protocol and details were not pre-registered.
Results
Two-sample Mendelian randomization
The instruments for urate that were used in the two-sample MR are presented in Supplementary Table 2. The main IVW MR showed that higher genetically predicted serum urate levels were associated with an increased risk of CHD, with odds ratio (OR) per 1-SD increase in genetically predicted urate 1.19, 95% confidence interval (CI) 1.10-1.30, P=4×10−5. Consistent results were obtained in MR sensitivity analyses (Figure 1). Higher genetically predicted serum urate was also associated with increased risk of both PAD (OR 1.12, 95% CI 1.03-1.21, P=9×10−3) and stroke (OR 1.11, 95% CI 1.05-1.18, P=2×10−4) in the main IVW analysis, with similar results again obtained in MR sensitivity analyses that are more robust to the inclusion of pleiotropic variants (Figure 1). For the multivariable MR adjusting for genetic confounding through BMI, estimated glomerular filtration rate, type 2 diabetes and lipid traits, direct effects of these exposures on risk of the respective CVD outcomes are presented in Supplementary Table 3. Considering SBP, the main IVW and sensitivity MR analyses all provided supporting evidence of a causal effect of serum urate (IVW estimate in SD units per 1-SD increase in genetically predicted urate 0.09, 95% CI 0.05-0.12, P=6×10−7; Figure 2). Scatter plots depicting the association of the instruments variants with serum urate and the respective outcomes are presented in Supplementary Figures 1-4.
Performing multivariable MR to adjust for genetically predicted SBP showed attenuation of the urate effect estimates for the CVD outcomes as compared to the main IVW univariable MR (Figure 3), supporting that part of the effect of urate on these outcomes is mediated through SBP. Network MR mediation analysis quantified this as 29% (95% CI 9%-48%) for CHD, 44% (95% CI 5%-83%) for PAD and 45% (95% CI 14%-76%) for stroke. For CHD there remained evidence of a direct effect of urate even after adjusting for SBP (OR 1.13, 95% CI 1.03-1.23, P=0.01). In contrast for PAD and stroke, although the estimate for the direct effect of urate that is not mediated through SBP was positive, the confidence interval crossed the null and the results were not statistically significant (PAD OR 1.08, 95% CI 1.00-1.17, P=0.07; stroke OR 1.06, 95% CI 0.99-1.12, P=0.10). Direct effects of SBP on the outcomes after adjusting for genetically predicted serum urate are presented in Supplementary Table 4.
One-sample Mendelian randomization
Baseline characteristics for the UK Biobank population considered in the one-sample MR analysis are detailed in Supplementary Table 5. A total of 373,167 participants were included in analyses after accounting for missing data (Supplementary Table 6). Included participants had similar characteristics to the UK Biobank population more generally (Supplementary Table 5). The first participant was recruited on 19 December 2006 and outcomes data were available up to 31 March 2017. A total of 29,903 incident CVD events (26,280 CHD, 420 PAD, and 3,203 stroke) were identified in the analysis, with a mean follow up time of 2,928 days. The 107 SNPs included in the urate GRS for the one-sample MR and their respective weights are detailed in Supplementary Figure 7. Considering the whole population, the OR of CVD per 1-SD increase in genetically predicted urate was 1.13 (95% CI 1.07-1.19, P=1×10−6) and the SD change in SBP per 1-SD increase in genetically predicted urate was 0.06 (95% CI 0.05-0.07, P<1×10−10).
After splitting the cohort into four quantiles based on residual urate (observed urate corrected for genetic predisposition), MR estimates for the causal effect of genetically predicted urate on CVD risk were similar between quantiles (Figure 4), with Cochran’s Q test for heterogeneity P=0.62. In contrast, there was statistically significant evidence of heterogeneity between the four quantiles when considering SBP as an outcome (Cochran’s Q P=0.02), with the quantile that had lowest residual urate levels having a higher estimate, and 95% CIs that did not overlap with those for the quantile with highest residual urate levels (Figure 4). When stratifying by sex, there was no evidence of any discrepancy in the effect of genetically predicted urate levels on CVD risk (Cochran’s Q P=0.37), but there was evidence of a greater effect on SBP for women as compared to men (Cochran’s Q P=0.01) (Figure 4). Supplementary Table 7 details stratum specific characteristics and results.
Discussion
We used a comprehensive framework of MR methodologies to perform detailed investigation into the association of genetically predicted serum urate with CVD outcomes, and replicated findings in the independent UK Biobank population. The two-sample MR analyses identified consistent evidence of an effect of higher genetically predicted serum urate levels and risk of CHD, PAD, stroke and SBP. Performing multivariable MR to adjust for genetic association with SBP attenuated the estimates for the CVD outcomes to support that at least some of the effect of urate may be mediated through raised blood pressure. One-sample MR in the UK Biobank using individual participant data did not provide evidence to support non-linear or sex-specific effects of urate on CVD risk, although there was some suggestion that a fixed increase in genetically predicted urate may raise SBP more in those with lower urate levels, and in women as compared to men.
Hyperuricemia has been postulated to cause endothelial dysfunction by increasing oxidative stress1. Consistent with our findings, this could directly increase risk of CHD, PAD and stroke through effects on the vascular endothelium, as well as indirectly by raising blood pressure via impairment of renal function1. In an animal model where hyperuricemia was induced using a uricase inhibitor, hypertension followed after three weeks while controls remained normotensive4. A systematic review and meta-analysis of 81 trials including 10,684 participants showed a significant decrease in total cardiovascular events with allopurinol use, although the reduction in major adverse cardiovascular events and mortality did not reach statistical significance6. However, the majority of trial evidence was of low or moderate quality6. The results of the Febuxostat and Allopurinol Streamlined Trial (FAST) and the Allopurinol and cardiovascular outcomes in patients with ischemic heart disease (ALL-HEART) trials will help provide further insight on potential re-purposing of these existing agents for CVD prevention40, 41. Allopurinol could represent an inexpensive, safe and well-tolerated drug for reducing cardiovascular risk42.
Our current study has several strengths. Some previous MR analyses, including from our group12, 13, have suggested that there may not be any causal effect of genetically predicted serum urate on CVD outcomes. These findings may have been due to any effect of genetically predicted urate on CVD risk being concealed by pleotropic effects on other metabolic traits3, as well as insufficient statistical power. In the current study, we were not overly reliant on MR methods that are vulnerable to violations of the instrument-strength independent of direct effect (‘InSIDE’) assumption, such as MR-Egger36, as the genetic determinants of serum urate have been shown to share their etiology with other cardiometabolic risk factors, thus rendering this likely violated3. Instead, we used the contamination-mixture, simple median and MR-PRESSO methods as sensitivity analyses, which typically perform better than MR-Egger43. Our present analysis also incorporated more powerful instruments for serum urate than were available previously. The variants used as instruments in our one-sample MR analysis of UK Biobank participants explained approximately 7.7% of the variance in serum urate3, in contrast to the 5.3% explained when selecting instrument variants from the previous largest published GWAS of serum urate21. Furthermore, to our knowledge this is the first MR study to investigate for non-linear and sex-specific effects of genetically predicted serum urate on CVD risk, as well as potential mediation through SBP. The absence of evidence to support a greater effect of changes in genetically predicted urate levels on CVD outcomes in those with higher baseline urate levels supports that relative risk reduction may be similar across the clinically observed range of urate levels (although absolute risk reduction could differ). The consistent findings for CVD risk reduction in both men and women can similarly inform trial inclusion and exclusion criteria.
Our work also has limitations. The MR approach makes a series of modelling assumptions, and in particular requires that the genetic variants used as instruments do not affect the considered outcomes through pathways that are independent of urate. While this can never be completely excluded, we performed a range of MR sensitivity analyses that make distinct assumptions on the presence of pleiotropic variants and found consistent estimates. MR estimates also represent the lifelong cumulative effect of genetically predicted variation in serum urate levels, and may overestimate the consequence of a clinical intervention aimed at altering urate levels in adult life44. Finally, we investigated SBP in mediating the effect of genetically predicted urate on CVD outcomes, rather than diastolic blood pressure. These two traits are highly genetically and phenotypically correlated24, and it follows that a similar mediating role may be found for diastolic blood pressure, as has been highlighted previously45.
To summarize, we have found consistent evidence for an effect of genetically predicted serum urate levels on risk of CHD, PAD, stroke and SBP. Our analyses suggest that urate may be causally associated with increased cardiovascular risk via both blood pressure dependent and independent mechanisms, with no evidence of non-linear or sex-specific effects. High-quality trial data are now necessary to provide definitive evidence on the cardiovascular benefit of urate lowering, with some large-scale studies already underway40, 41.
Contributors
DG, JD, ET and IT designed the study. DG and XL had full access to the data and performed the analysis. All authors interpreted the results. DG, JD and IT drafted the manuscript. All authors critically revised the manuscript for intellectual content. All authors approved the submitted version and are accountable for the integrity of the work.
Data Availability
All data used in this work are either publicly accessible or available on request from their original studies, which obtained appropriate patient consent and ethical approval. The UK Biobank data were accessed through application 236. All results generated in this work are presented in either the manuscript or its supplementary files.
Sources of Funding
This work was supported by funding from the US Department of Veterans Affairs Office of Research and Development, Million Veteran Program Grant MVP003 (I01-BX003362). This publication does not represent the views of the Department of Veterans Affairs of the US Government. DG is funded by the Wellcome 4i Clinical PhD Program at Imperial College London. VK is funded by the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant (721567). PE acknowledges support from the Medical Research Council (MR/S019669/1), the National Institute for Health Research Imperial Biomedical Research Centre, Imperial College London (RDF03), the UK Dementia Research Institute (DRI) at Imperial College London funded by UK DRI Ltd (funded by Medical Research Council, Alzheimer’s Society, Alzheimer’s Research UK), and Health Data Research (HDR) UK London funded by HDR UK Ltd (funded by a consortium led by the Medical Research Council 1004231). SMD was supported by the Department of Veterans Affairs Office of Research and Development (IK2-CX001780). ET is funded by a Cancer Research UK Career Development Fellowship (C31250/A22804). The MEGASTROKE project received funding from sources specified at http://www.megastroke.org/acknowledgments.html. Details of all MEGASTROKE authors are available at http://www.megastroke.org/authors.html.
Disclosures
JD and TQ have received charitable research income to conduct clinical trials of allopurinol use in people with stroke. The remaining authors have no conflicts of interest to declare.
Acknowledgements
The authors acknowledge the contributors of the data used in this work: CARDIoGRAMplusC4D, CKDGen, DIAGRAM, Genetic Investigation of ANthropometric Traits, Global Lipids Genetics Consortium, International Consortium for Blood Pressure, MEGASTROKE, Million Veterans Program and UK Biobank.