Causal effect of lipids on gout
We chose 92, 48, 58 and 53 lipid-specific instrumental variables for HDL, LDL, TC and TG when evaluating their associations with gout. Together, these instrumental variables accounted for a total of 5.7%, 3.1%, 2.6% and 4.3% phenotypic variances for HDL, LDL, TC and TG, respectively. The F statistics for all these SNPs were greater than 10 (Supplementary Tables S1-S4), suggesting that weak instrument bias was potentially null in our analysis. The Cochran's Q test [55, 56] showed that there was little evidence of instrumental heterogeneity for HDL (p = 0.945), TC (p = 0.179) or TG (p = 0.379). However, an instrumental heterogeneity was observed for LDL at the marginal significance level of 0.05 (p = 0.047). Therefore, we employed the fixed-effects IVW method for HDL, TC and TG, and applied the random-effect IVW method for LDL when estimating the causal effects on gout.
Among the four lipids we identified that only HDL was associated with gout (Fig. 2A-C and Fig. 3). Specifically, the estimated odds ratio (OR) per standard deviation (SD) increase of HDL (~ 12.26 mg/dL) on gout was 0.75 (95% CI 0.62 ~ 0.91, p = 3.31E-3) and was statistically significant after Bonferroni correction. The OR per SD increase of LDL (~ 30.25 mg/dL), TC (~ 36.32 mg/dL) or TG (~ 112.33 mg/dL) for gout was 0.81 (95% CI 0.61 ~ 1.07, p = 0.079), 0.98 (95% CI 0.90 ~ 1.06, p = 0.604) and 1.16 (95% CI 0.93 ~ 1.45, p = 0.186) (Fig. 3 and Supplementary Table S5), respectively.
Causal effect of lipids on serum urate
We generated 87, 45, 57 and 49 lipid-specific instrumental variables for HDL, LDL, TC and TG when evaluating their associations with serum urate. These instrumental variables explicated a total of 5.0%, 3.0%, 2.5% and 3.7% phenotypic variances for HDL, LDL, TC and TG, respectively (Supplementary Tables S6-S9). Thereafter, we used the random-effects IVW method to estimate the causal effects due to the heterogeneity observed for all the four sets of instruments (the p values of the Cochran's Q test were 5.37E-10, 6.95E-04, 3.41E-06 and 2.67E-03 for HDL, LDL, TC and TG, respectively). HDL is negatively associated with serum urate after Bonferroni correction (estimated causal effect = -0.09, 95% CI -0.12 ~ -0.05, p = 7.00E-04) (Fig. 2D-F and Fig. 3). In addition, a positive association existed between TG and serum urate (estimated causal effect = 0.10, 95% CI 0.06 ~ 0.14, p = 9.87E-05) (Fig. 2G-I and Fig. 3). However, no association was observed between LDL/TC and serum urate (Supplementary Table S5 and Fig. 3).
Sensitivity analyses to validate the estimated causal effects
We subsequently performed sensitivity analyses to validate the causal association observed above. Herein, we relegated the results of serum urate to Supplementary Results and focused only on gout in the following paragraphs. Due to the insignificant association between LDL/TC and gout/serum urate, we only summarized their results in Fig. 3 and Supplementary Figures S1-S3, but did not pursue any of these two sets of traits further.
With the maximum likelihood method, the estimated OR per SD increase of HDL on gout was 0.75 (95% CI: 0.62 ~ 0.91, p = 3.31E-03), in line with the IVW estimate; while the median-based method identified a consistently inverse but insignificant association (OR = 0.69, 95% CI 0.51 ~ 0.94, p = 0.019). The LOO analysis showed that none of the instruments of HDL could individually affect the substantial causal effect estimates (Supplementary Figure S4). However, Fig. 2A delineated that two instrumental variables of HDL with large effect sizes on gout seemed to be potential outliers (i.e. rs2566091 in gene ALDH1A2 and rs9989419 in gene AC012181.1). However, they did not exert any substantial impact on the estimated causal effect of HDL on gout. Specifically, after removal, the OR was 0.74 (95% CI 0.60 ~ 0.91, p = 4.18E-03), similar to that obtained with the instruments in total (Fig. 2B). Furthermore, the funnel plot of HDL for gout exhibited a symmetric pattern around the causal effect point estimates (Fig. 2C), offering little evidence for horizontal pleiotropy. The MR-Egger regression also precluded the possibility of pleiotropy of instrument variables (the intercept = -0.013, 95% CI -0.028 ~ 0.003, p = 0.113). Excluding 28 instruments associated with other traits/diseases (Supplementary Table S10) did not substantially alter our estimation of causal effect (OR = 0.77; 95%CI 0.62 ~ 0.95, p = 0.016). We further performed a multivariable Mendelian randomization analysis to estimate the association between HDL and gout while adjusting for LDL, TC and TG. The result showed HDL a de facto inverse relationship with gout even after controlling for the residual lipid traits (Supplementary Table S11), implying the independent causal role of HDL in the risk of gout.
In the reverse causal analysis, we kept one instrumental variable for gout in estimation of its causal effect on HDL, and 24 (or 22) instrumental variable of serum urate when estimating its effect on HDL (or TG) (Supplementary Tables S12-S14). With the fixed-effects IVW method, the causal effect of gout on HDL was estimated to be -0.019 (95% CI: -0.018 ~ 3.20E-03, p = 0.173), and the causal effects of serum urate on HDL and TG were estimated to be -0.033 (95% CI -0.065 ~ -7.31E-04, p = 0.045) and 0.016 (95% CI: -0.016 ~ 0.048, p = 0.330), respectively. All of estimates were insignificant in case of multiple hypothesis testing. Therefore, this analysis excluded the probability of reverse causation from gout/serum urate to lipids, indicating that dyslipidemia (e.g. decreased HDL and/or increased TG) was a causal factor rather than a clinical manifestation of gout or serum urate.
Results of the mediation analysis
To estimate the possible mediation of serum urate in the progression from dyslipidemia and gout, we combined both the instrumental variables of lipids and serum urate, and reserved 116, 65, 76 and 73 lipid-specific instrumental variables for HDL, LDL, TC and TG in the mediation analysis (Supplementary Tables S15-S19). Notably, the following reported effects were corrected for the same scales (Supplement Methods).
In the mediation analysis we observed that the summation of the mediation effect and the direct effect was approximately equal to the total effect (Table 2), indicating that our correction strategy used for those effects was justifiable. The total effect of HDL on gout was estimated to be -0.154 (95% CI: -0.256 ~ -0.051, p = 3.31E-3; Fig. 3) and the direct effect of HDL on gout was estimated to be -0.106 (95% CI: -0.189 ~ -0.022, p = 1.35E-02). More importantly, the mediation effect of HDL on gout was estimated at -0.020 (95% CI: -0.033 ~ -0.008, p = 1.67E-03), which accounted for about 13.0% (= 0.020/0.154) of the total effect, indicating that serum urate was a promising mediator between HDL and gout, and that HDL could indirectly affect the risk of gout via serum urate in addition to the direct impact. Likewise, the total effect of TG on gout was estimated at 0.082 (95% CI: -0.039 ~ 0.202, p = 1.86E-01) and the direct effect of TG on gout was estimated at 0.048 (95% CI: -0.057 ~ 0.152, p = 3.71E-01). Furthermore, the mediation effect of TG on gout was estimated at 0.023 (95% CI: 0.010 ~ 0.037, p = 8.39E-04), accounting for about 28.0% (= 0.023/0.082) of the total effect. TG also served as a mediator between TG and gout and could indirectly increase the risk of gout via serum urate. However, the mediation effects of serum urate between LDL/TC and gout were insignificant (Table 2).
Table 2
Mediation analysis of lipids on gout with serum urate concentrations
| Mediation effect (95% CI) | Direct effect (95% CI) | Total effect (95% CI) |
Sobel | Bootstrap | Proportion (%) |
HDL | -0.020 (-0.033 ~ -0.008) | -0.020 (-0.034 ~ -0.008) | 13.0 | -0.106 (-0.189 ~-0.022) | -0.154 (-0.256 ~ -0.051) |
LDL | -0.010 (-0.023 ~ 0.004) | -0.010 (-0.024 ~ 0.002) | 8.56 | -0.090 (-0.212 ~ 0.031) | -0.117 (-0.246 ~ 0.013) |
TC | -0.007 (-0.021 ~ 0.007) | -0.007 (-0.022 ~ 0.007) | 58.3 | 0.039 (-0.007 ~ 0.084) | -0.012 (-0.058 ~ 0.033) |
TG | 0.023 (0.010 ~ 0.037) | 0.023 (0.011 ~ 0.037) | 28.0 | 0.048 (-0.057 ~ 0.152) | 0.082 (-0.039 ~ 0.202) |
All the effects were corrected for the same scales. |
HDL, High-density lipoprotein cholesterol; LDL, Low-density lipoprotein cholesterol; TC, Total cholesterol; TG, Triglyceride |