We selected genetic variants associated with serum total cholesterol (TC, n = 187365), triglyceride (TG, n = 177861), HDL cholesterol (HDL-C, n = 187167), LDL cholesterol (LDL-C, n = 173082), apolipoprotein A1 (ApoA1, n = 20687) and apolipoprotein B (ApoB, n = 20690) levels and then extracted the corresponding effect sizes for CKD using the largest GWAS summary-level dataset (8, 9). The data source of this study is based on re-analyzing previously published GWAS; therefore, there is no ethical approval. CKD data (n = 117165) were acquired from the CKDGen consortium (n = 117165) (10). CKD was defined as an eGFR based on serum creatinine (eGFRcrea) lower than 60 mL/min/1.73 m2. All datasets were obtained from large-scale randomized double-blind trials and population cohort studies based on European descent. Gender, age and body mass index (BMI) should be corrected in regression models of serum lipid GWAS (8, 9), and age and gender were also adjusted in CKDGen.
Because of the potential population stratification in our selected datasets, the genome control of each sample is applied to correct for inconsistent test statistics.
In our current analysis, the IVs provided by genetic variants should contain three assumptions as in our previous research (11): (1) IVs must be strongly associated with exposure; (2) IVs should be without any association with known confounders; and (3) the IVs we selected must be conditionally independent of exposure (serum lipid levels), outcomes (CKD) and confounders. If the IVs contained the second and third assumptions, it may be regarded as independent from pleiotropy.
Initially, IVs should be strongly correlated with exposure (serum lipid level). Then, the P-value that we selected should be < 5 × 10 − 8 in the relevant GWAS dataset to ensure the close association between IVs and exposures. After that step, to ensure independence among selected IVs, PLINK 1.90(12) was performed to calculate the pairwise linkage disequilibrium (LD). If the r2 was greater than 0.001, these SNPs were excluded from our research.
We selected these IVs must be conditionally independent of outcome (CKD), considering the correlated traits of exposures (serum lipid level), and independent of any known confounders. For selected IVs, only exposure factors (serum lipid level) and no other pathways or confounding factors can affect the outcome (CKD). This finding is consistent with the previous two assumptions (13). First, we made the corresponding effect estimates of these variables on CKD. We should choose the proxy SNPs that are highly correlated (r2 was greater than 0.8) based on the SNP Annotation and Proxy (SNAP) search system for substitution when the selected SNPs cannot be used in CKD (14). Next, MR-Egger regression was performed to calculate the horizontal pleiotropic (15). Afterwards, we removed any palindromic SNPs for which the minor allele frequency (MAF) was greater than 0.3 to ensure that the influence of the SNPs on the exposures (serum lipid level) corresponded to the same allele as their influence on CKD (16). Subsequently, we employed the GWAS Catalog to check for the associations between selected IVs and to adjust for potential confounding. In addition, we calculated the F statistic with a web application (https://sb452.shinyapps.io/overlap/) to examine the association of selected IVs with the exposures (17).
MR-Egger regression was employed to calculate the horizontal pleiotropic pathway between IVs and CKD, independent of serum lipid level (15). As an effective method to detect bias in publication meta-analysis, MR-Egger regression was derived from Egger regression. The method can be expressed through the equation αi = βγi + β0. In this equation, different letters indicate different meanings. αi represented the effect between IVs and CKD; γi was employed to estimate the effect between serum lipid level and IVs; slope β denoted the estimated causal effect of exposure (serum lipid level) on outcome (CKD); and intercept β0 represented the estimated average value of horizontal pleiotropic. When the P-value of the intercept was greater than 0.05, no horizontal pleiotropy could be found. In addition, the slope can also be defined as the estimated pleiotropy-corrected causal effect. However, if the SNPs we selected in this analysis do not account for most of the differences in exposure, then there is a lack of evaluation of this estimate (15).
In our current study, inverse variance weighted (IVW) was used as the key method to calculate the causal effect between serum lipid level and CKD for TSMR analysis (18). The causal effect β was estimated and shown as wi (αi /γi). In this equation, i refers to the IVs, αi represents the association effect of IVs on CKD, γi defines the association effect of IVs on serum lipid level, and wi represents the weights of the causal effect of serum lipid level on CKD.
We employed various methods to calculate follow-up sensitivity, including maximum likelihood, MR Egg, weight median, penalized weight median, simple mode, weight mode and robust adjusted profile score (RAPS) (19). Compared with IVW, these methods have greater robustness to individual genetics with strongly outlying causal estimates and would generate a consistent estimate of the causal effect when valid IVs exceed 50% (20). Then, leave-one-out sensitivity analysis was performed to screen out whether the correlation was out of relationship to be affected by a single SNP. Subsequently, we employed TSMR analysis again, leaving out each SNP, in turn, and the overall analysis including all SNPs was shown for comparison(21). All of the analysis was implemented by the “TwoSampleMR” package in the R software environment.