If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Rheumatology and Immunology, Zhejiang Provincial People's Hospital, People's Hospital of Hangzhou Medical College, Hangzhou, 310014, China
Department of Rheumatology and Immunology, Zhejiang Provincial People's Hospital, People's Hospital of Hangzhou Medical College, Hangzhou, 310014, China
Address correspondence and reprint requests to: Y. Li, School of Public Health, Hangzhou Medical College, 481 Binwen Road, 310053, Hangzhou, Zhejiang, China. Tel.: 571-87692815.
The bidirectional association between osteoarthritis (OA) and urate levels and gout, though well-documented, is inconclusive. This Mendelian randomization (MR) study aims to examine the bidirectional causality between OA and urate levels as well as gout.
Methods
We used summary statistics data for serum urate levels from 288,649 CKDGen participants and gout from 69,374 Global Urate Genetics Consortium participants. The summary statistics data for OA were obtained from genome-wide association studies including up to 826,690 participants of mainly European ancestry. MR was performed using established analytical methods including the Wald ratio, inverse variance weighted (IVW), weighted median (WM) and MR-Egger.
Results
Genetically determined urate levels [IVW odds ratio (OR) = 0.99, 95% confidence interval (CI) = 0.96, 1.02, P = 0.484] and gout (Wald ratio OR = 1.00, 95% CI = 0.98, 1.02, P = 0.908) were not associated with the risk of total OA. In site-specific OA analyses, there was no causal effect of urate levels on knee, hip, spine, thumb and hand OA, and no evidence was provided that gout increased the risk of OA at any site. In the reverse MR analyses, we found no causal effect of total OA on urate levels (IVW Beta = −0.011, 95% CI = −0.095, 0.074, P = 0.807) or gout (IVW OR = 1.05, 95% CI = 0.66, 1.68, P = 0.839). A null effect of site-specific OA was also observed.
Conclusion
Our MR study supports no bidirectional causal effect of urate levels and gout on total and site-specific OA.
. Thus, it is essential to explore the etiology of OA intended to prevent OA development. Emerging studies have been reported that urate is an independent risk factor of OA
Uric acid is the end product of purine metabolism. It exists in the blood in the form of urate. Hyperuricemia is characterized by an abnormally high serum urate levels (>6.8 mg/dL)
Although considerable evidence supported the correlation of urate levels and gout in OA, the causal direction of urate levels and gout to OA is still ambiguous. Some observational studies suggested that OA may lead to increased serum urate levels, which increases the risk of gout
. Conflicting evidence mentioned above may be due to the fact that traditional epidemiological methods are susceptible to confounding factors and reverse causality
. As the genetic variants of an individual cannot be modified after conception, using genetic variants as instrumental variables (IVs) can greatly increase the reliability of the results. In addition, based on the availability of publicly genetic data, this method can be widely applied.
In this two-sample MR analysis, we aimed to demonstrate whether there is a bidirectional causal relationship between OA and serum urate levels as well as gout.
Methods
Study design overview
The design of bidirectional MR study is overviewed in Fig. 1. In briefly, the causal effects of urate levels and gout on OA were estimated first, then the causal effects of OA on urate levels and gout were evaluated. Genetic variants can be treated as IVs only when they meet the three stringent assumptions as follow
. Firstly, genetic variants are highly correlated with exposure. Secondly, genetic variants are not associated with confounders (e.g., body mass index, gender and old). Lastly, genetic variants cannot be directly involved in outcome but via the exposure pathway. We utilized summary statistic datasets from recent meta-analyses of genome-wide association studies (GWASs) of urate, gout and OA.
Fig. 1The design of bidirectional Mendelian randomization (MR) study. The '×' means that genetic variants are not associated with confounders or cannot be directly involved in outcome but via the exposure pathway. The '√' means that genetic variants are highly correlated with exposure. Solid paths are significant; dashed paths should not exist in the MR study. SNP: single nucleotide polymorphism; OA, osteoarthritis.
Data sources and SNP selection for serum urate levels and gout
We selected the summary statistics data of serum urate levels from a published meta-analysis of GWAS involving approximately 288,649 participants of European ancestry
, and we used a publicly available summary statistics data of gout from another independent GWAS which included 69,374 individuals (2,115 cases vs 67,259 controls)
. The sample characteristics of study population were described in Supplementary Table S1. 123 variants associated with serum urate levels at genome-wide significance (P < 5 × 10−8) were reported by the GWAS. In the meantime, a linkage disequilibrium (LD) test was performed on these nucleotide polymorphisms (SNPs) to clump SNPs for independence. The genetic variants for pairs in LD (r2 > 0.001) were pruned. Two independent (r2 < 0.001) variants associated with gout at genome-wide significance were extracted from the GWAS. Then we used these SNPs to match in the summary data of SNP-outcome (OA) association estimates and SNPs not available in the outcome GWAS were removed or replaced by proxy SNPs with strong LD (r2 > 0.8). Furthermore, based on sample size of the exposure dataset, number of IVs and genetic variance, R2 and the F-statistics were calculated to assess the strength of IVs
. Finally, we searched all of the SNPs associated with exposure as well as their proxies in the Phenoscanner database (http://www.phenoscanner.medschl.cam.ac.uk/) to identify whether there were SNPs associated with confounding factors (P < 1 × 10−5). We manually removed these SNPs to avoid possible pleiotropic effects. Details of instrument SNPs for urate and gout are shown in Supplementary Tables S2–S3.
Data sources and SNP selection for total and site-specific OA
Summary statistics data on OA and its specific sites, knee, hip, spine, thumb and hand OA, were obtained from the latest publicly available GWAS of OA with up to 826,690 Genetics of Osteoarthritis (GO) Consortium participants (177,517 OA patients)
. OA was defined by GO based on either self-reported status, hospital diagnosed, ICD10 codes or radiographic as defined by the TREAT-OA consortium. In the GWAS analysis, variants were categorized according to which OA site was most significantly associated with. To minimize the bias from sample overlap, we used strong instruments (e.g., F-statistic much greater than 10 for the instrument–exposure association)
. Finally, we obtained the variants associated with total, knee, hip, spine, thumb and hand OA at genome-wide significance (P < 1.3 × 10−8) and independence (r2 < 0.001) level, respectively. Similarly, we removed the SNPs whose proxy SNPs were not available in the outcome GWAS. SNPs were checked by Phenoscanner database. Details of instrument SNPs are listed in Supplementary Tables S4–S5.
MR analysis
In this study, TwoSampleMR package with R software was used for MR analyses
. We employed the inverse-variance-weighted (IVW) method as our primary MR analysis approach which estimates causal effects of genetically predicted exposure on outcome through weighted regression of SNP-specific Wald ratios (i.e., beta outcome/beta exposure). Several sensitivity analyses were conducted, including the weighted median (WM) method
. MR-Egger regression method effectively tests the null causal hypothesis and gives a consistent estimate to causality, even if no genetic variants are valid IVs
. The MR-Egger regression method is robust to horizontal pleiotropy. We quantified the level of heterogeneity by Cochran Q statistics and I2 statistics
. Larger I2 values show increasing heterogeneity. Besides, “leave-one-out analysis” by removal of every single SNP at turn could make sure the reliability of the results. We reported the effect estimates in beta where the outcome was continuous (i.e., levels of urate) and converted to odds ratios (ORs) where the outcome was binary (i.e., gout and OA status). In the study, ORs were interpreted as odds for outcome per unit increase in exposure. For levels of urate, the unit was mg/dL.
In views of the issue of multiple testing, main results had statistical significance at P values < 0.002 (0.05/24) after a Bonferroni correction. All statistical analyses were two-sided.
Results
Causal effect of serum urate levels and gout on OA
2 SNPs could not be obtained for their proxy SNPs on the online platform SNiPA (https://snipa.helmholtz-muenchen.de/snipa3/) and were removed. After searching the remaining urate-associated SNPs in the Phenoscanner database, we found that 24 SNPs were associated with traditional risk factors of OA (e.g., body mass index, body fat percentage and/or bone mineral density). Lastly, 69 LD-independent genetic variants were taken as IVs for urate (Fig. 2). In a similar way, rs1481012 was associated with body mass index and removed. The remaining genetic variants were taken as IVs for gout. F-statistics for IVs of urate and gout were above the threshold of 10, indicating that the IVs were strong instruments and thus reducing the bias of IVs estimates.
Fig. 2Forest plot of the causal effects of urate associated SNPs on osteoarthritis (OA) Figure showed the Mendelian randomization estimated effects sizes for (A) urate-total OA, (B) urate-knee OA, (C) urate-hip OA, (D) urate-spine OA, (E) urate-thumb OA and (F) urate-hand OA. Data are expressed as beta values with 95% CI.
As shown in Table I, heterogeneity was detected by Cochran's Q test (P < 0.05). Therefore, we used inverse variance weighted (IVW) method in a random-effects model. Genetically elevated serum urate levels were not causally associated with total OA [OR = 0.99, 95% confidence interval (CI) = 0.96, 1.02, P = 0.484] in IVW model. The results of MR-Egger and WM model were consistent with IVW. And MR-Egger analysis did not suggest any directional pleiotropy for the IVs (P for intercept = 0.395). Using the Wald ratio method, genetically predicted gout had no causal effects on total OA (OR = 1.00, 95% CI = 0.98, 1.02, P = 0.908).
Table IMR results of the causal effect of urate and gout on OA and its other phenotypes
In site-specific OA analysis, we observed that genetically determined levels of serum urate were not causally associated with knee OA (IVW odds ratio (OR) = 1.00, 95% CI = 0.96, 1.05, P = 0.827), hip OA (IVW OR = 1.00, 95% CI = 0.95, 1.06, P = 0.879), spine OA (IVW OR = 0.96, 95% CI = 0.92, 1.01, P = 0.109), thumb OA (IVW OR = 1.02, 95% CI = 0.93, 1.11, P = 0.681), and hand OA (IVW OR = 1.01, 95% CI = 0.95, 1.08, P = 0.722). Cochran's Q test did not detect the heterogeneity only in spine OA. And MR-Egger analysis did not suggest any directional pleiotropy for the IVs (P for intercept = 0.427 for knee OA, P for intercept = 0.386 for hip OA, P for intercept = 0.122 for spine OA, P for intercept = 0.856 for thumb OA, and P for intercept = 0.589 for hand OA). Using the Wald ratio method, genetically predicted gout had no causal effects on knee OA (OR = 1.01, 95% CI = 0.98–1.05, P = 0.445), hip OA (OR = 1.01, 95% CI = 0.97, 1.06, P = 0.574), spine OA (OR = 0.98, 95% CI = 0.93, 1.03, P = 0.418), thumb OA (OR = 0.98, 95% CI = 0.90, 1.07, P = 0.703), and hand OA (OR = 0.99, 95% CI = 0.94, 1.06, P = 0.859).
Causal effect of OA on serum urate levels and gout
6 SNPs that could not be matched in the GWAS of urate and could not be obtained for their proxy on the online platform SNiPA were removed. After searching in the Phenoscanner database, 18 SNPs were found to be correlated with confounding factors and were removed. Lastly, 12 SNPs for total OA, 14 SNPs for knee OA, 28 SNPs for hip OA, 4 SNPs for thumb OA, and 7 SNPs for hand OA were extracted for the association with urate through LD tests, respectively (Fig. 3). Similarly, 10 SNPs for total OA, 12 SNPs for knee OA, 22 SNPs for hip OA, 3 SNPs for thumb OA, and 6 SNPs for hand OA were used for the association with gout, respectively (Fig. 4). F-statistics for IVs of OA were all above the threshold of 10.
Fig. 3Forest plots for association of osteoarthritis (OA) with urate. Figure showed the Mendelian randomization estimated effects sizes for (A) total OA-urate, (B) knee OA-urate, (C) hip OA-urate, (D) thumb OA-urate, (E) hand OA-urate. Data are expressed as beta values with 95% CI.
Fig. 4Forest plots for association of osteoarthritis (OA) with gout. Figure showed the Mendelian randomization estimated effects sizes for (A) total OA-gout, (B) knee OA-gout, (C) hip OA-gout, (D) thumb OA-gout and (E) hand OA-gout. Data are expressed as beta values with 95% CI.
As displayed in Table II, heterogeneity for the associations between the selected IVs of the knee OA and hip OA and urate was observed (P < 0.05). So IVW in random-effects model was used among them. Using the IVW method in a random-effects model, neither knee OA (IVW Beta = −0.015, 95% CI = −0.086, 0.056, P = 0.672) nor hip OA (IVW Beta = −0.001, 95% CI = −0.035, 0.033, P = 0.966) was causally associated with urate. No heterogeneity was detected by Cochran's Q test for total OA (P = 0.247), thumb OA (P = 0.053), and hand OA (P = 0.549). Total OA (IVW Beta = −0.011, 95% CI = −0.095, 0.074, P = 0.807), thumb OA (IVW Beta = 0.006, 95% CI = −0.053, 0.065, P = 0.844), and hand OA (IVW Beta = 0.009, 95% CI = −0.026, 0.045, P = 0.601) were not related to urate. These findings were similar across other MR estimates. MR-Egger analysis did not suggest any directional pleiotropy for the IVs (P for intercept = 0.206 for total OA, P for intercept = 0.517 for knee OA, P for intercept = 0.736 for hip OA, P for intercept = 0.570 for thumb OA, and P for intercept = 0.205 for hand OA, respectively).
Table IIMR results of the causal effect of OA and its other phenotypes on urate and gout
In addition, total and site-specific OA had no causal effect on risk of gout (IVW OR = 1.05, 95% CI = 0.66, 1.68, P = 0.839 for total OA, IVW OR = 0.92, 95% CI = 0.59, 1.44, P = 0.712 for knee OA, IVW OR = 1.08, 95% CI = 0.89, 1.32, P = 0.437 for hip OA, IVW OR = 1.05, 95% CI = 0.71, 1.55, P = 0.820 for thumb OA, and IVW OR = 0.96, 95% CI = 0.69, 1.34, P = 0.813 for hand OA). Using other MR estimate methods, we obtained the similar findings. No heterogeneity was detected by Cochran's Q test. MR-Egger analysis did not suggest any directional pleiotropy for the IVs (P for intercept = 0.947 for total OA, P for intercept = 0.722 for knee OA, P for intercept = 0.433 for hip OA, P for intercept = 0.933 for thumb OA, and P for intercept = 0.966 for hand OA).
Scatter plots, funnel plots and leave-one-out sensitivity analysis are presented in Supplementary Figs. S1–S3. The results of leave-one-out sensitivity analysis showed that there were SNPs with potential effect on the pooled results, suggesting the need for careful interpretation of the results.
Discussion
This bidirectional MR analysis provides no evidence in favor of the causal association of urate levels and gout with total and site-specific OA. In the opposite direction, there is also no causal effect of total and site-specific OA on urate levels or the risk of gout.
To the best of our knowledge, this is the first MR study to identify the bidirectional causal relationship between multiple OA phenotypes and urate levels as well as gout. A previous study have provided no evidence that uric acid level and gout had causal effect on the increased risk of OA (P = 0.482), but included a small sample size and was in only one direction
. Another phenome-wide MR study used polygenic risk scores for serum urate levels in the UK Biobank, indicating that genetically determined high urate levels were not associated with OA (P = 0.382), but did not include stratified analysis by various OA sites
. In previous studies, high levels of serum urate might be a risk factor for OA development. A case–control study reported that subjects with gout were more likely to have OA than those without gout (P = 0.017)
Serum uric acid levels and multiple health outcomes: umbrella review of evidence from observational studies, randomised controlled trials, and Mendelian randomisation studies.
. Actually, the observational study is typically affected by confounding factors or reverse causality and MR study can avoid these.
Similarly, the causality of total and site-specific OA with urate and gout is still inconclusive. Some studies reported that OA may stimulate urate production. One article suggested that cartilage at the joints affected by OA might provide a nucleation surface for urate crystallization and deposition
. Moreover, OA itself is associated with the IL-1β response gene, which determines IL-1β production. IL-1β is actively involved in inflammatory responses
. However, using three different MR estimate methods (IVW, WM, and MR-Egger), we found no causality of total and site-specific OA with urate or gout. It is possible that there is a gap between in vivo and in vitro animal experiments and the human situation
This study has several limitations. First, the number of SNPs involved in our study is relatively small, especially gout and spine OA. Thus, a larger GWAS with more SNPs as instruments is needed to replicate MR studies to improve power of testing association. Second, the effect of most genetic variants is moderate as these SNPs identified have a small variance
. It means that more samples and a more reliable instrument are needed for more precise detection of the causal association between the exposures and outcomes. Third, only participants of European descent are included in the study, and instruments identified in European populations are not transferable to non-European populations. Further MR studies of other ethnic groups are needed to verify the causality. Fourth, we do not stratify the causal association between OA and urate or gout by gender or obesity; although some studies suggested that gender or obesity may affect the causality. What's more, bidirectional MR assumes that the causal association occurs in a single direction. Due to the complexity of biological systems, feedback loops between the exposure and outcome may exist, which may make the result inaccurate
Nevertheless, this study has some advantages. Firstly, we included the variants across six OA phenotypes from an up-to-date meta-analysis and extracted the IVs of urate from the largest GWAS. Secondly, by using MR method, it is easy to obtain a large amount of outcome genetic data from publicly genetic dataset. And the meta-analysis of summary statistics is as efficient (in terms of statistical power) as collecting individual-level data across some smaller studies. Finally, the overlap of the study population is prevented by selecting samples from three different databases, namely, OA samples from GO Consortium, urate samples from CKDGen Consortium and gout samples from GUGC.
In conclusion, the causal effect of serum urate levels and gout on OA was not found by our MR analysis, which indicated that the increase in serum urate levels may not mean an increased OA risk in European individuals. In reverse analysis, our evidence also did not support the causal association between OA and serum urate levels and gout.
Contributions
Dingwan Chen and Huiqing Xu mainly designed and performed analysis, verifying data and wrote the manuscript; Lingling Sun, Yasong Li and Tianle Wang performed experiments and analysis; Yingjun Li supervised the entire project. All authors have read, provided critical feedback on intellectual content and approved the final manuscript.
Conflict of interest
The authors have declared no conflicts of interest.
Funding sources
No specific funding was received from any bodies in the public, commercial or not-for-profit sectors to carry out the work described in this article.
Ethics approval
The manuscript does not contain clinical studies or patient data.
Data availability statement
The data underlying this article are available in the article and in its online supplementary material.
Acknowledgements
The authors thank all the participants and researchers for their participation in this MR study. All authors interpreted the data, critically revised the manuscript for important intellectual content, approved the final version of the manuscript, and agreed to be responsible for all aspects of the work. The corresponding author attests that all listed co-authors meet authorship criteria and that no others meeting the criteria have been omitted.
Appendix A. Supplementary data
The following is the Supplementary data to this article:
Serum uric acid levels and multiple health outcomes: umbrella review of evidence from observational studies, randomised controlled trials, and Mendelian randomisation studies.
00986-9&id=gr1.jpg){kind=link}
00986-9&id=gr2.jpg){kind=link}
00986-9&id=gr3.jpg){kind=link}
00986-9&id=gr4.jpg){kind=link}