Evaluation of the causal effects of blood metabolites on irritable bowel syndrome: Mendelian randomization

For MR analysis, three assumptions need to be met. These include the association assumption: a robust strong correlation exists between genetic variation (Z) and exposure factor (X); the independence assumption: genetic variation (Z) is independent of the confounding factors (U) influencing the “exposure factor (X) - outcome (Y)” relationship; the exclusion restriction assumption: genetic variation can only influence the outcome through the exposure factors, and cannot affect the outcome via other pathways [13]. This study utilized two sample MR and Mendelian randomization pleiotropy residual sum and outlier (MR-PRESSO) software packages, both implemented in the R software (4.2.3) (Fig. 1).

Genetic information on blood metabolites was obtained from the Metabolomics Genome-Wide Association Study (GWAS) server (https://​metabolomics.​helmholtz-muenchen.​de/​gwas/​). Currently, this is the most thorough documentation available for the genetic loci of blood metabolites, made possible by a genome-wide association scan and advanced metabolic analysis performed by Shin et al. who identified ~ 2.1 million SNPs linked to 486 metabolites associated with human genetic variation [14]. The study included 7824 European citizens, including 6056 participants from the UK Twin Study and 1768 from the German KORA F4 Study. Of the 486 metabolites analyzed, 107 were categorized as unidentified because their chemical properties were ambiguous. Furthermore, 309 metabolites were chemically verified and were distributed across eight major metabolic categories: amino acids, carbohydrates, cofactors, vitamins, energy, lipids, nucleotides, peptides, and exogenous metabolism [15].

The GWAS summary data for IBS was sourced from the GWAS Catalog (https://​www.​ebi.​ac.​uk/​gwas/​) with accession number GCST90016564. This data originates from a large-scale IBS meta-analysis by Eijsbouts and colleagues published in Nature Genetics [16]. The sample includes 40,548 participants of European descent from the UK Biobank, meeting IBS diagnostic criteria based on DHQ Rome III symptom data, self-reported past IBS medical diagnosis, or electronic medical records [16].

Given the limited number of SNPs identified through stringent filtering, we relaxed the significance threshold to P < 1 × 10^-5, without altering other selection criteria. Subsequently, we designated linkage disequilibrium (LD) r2 <  0.01 within a 500 KB range, a standard commonly adopted in similar studies [16, 17]. In reverse causation, considering that the GWAS data for IBS possess a sufficient number of significant SNPs, to ensure that the instrumental variables in the model have adequate strength and thereby avoid potential weak instrument bias and violations of the exclusion restriction, we have chosen a genome-wide significance threshold (P < 5 × 10^− 8) as the criterion for including SNPs, and LD parameters set to r^2 = 0.001 and kb = 10,000 as conditions for removing LD. A rule of thumb is that the strength of the IVs, as measured by the F-statistic, should be at least 10. When the F-statistic is below 10, the estimates of causal effects can be severely biased [18]. However, a strict threshold may lead to an underestimation of genetic effects by excluding relevant IVs. Using an F-statistic threshold of F = 10 provides a balance, retaining robust instruments while minimizing the risk of weak instrument bias. Therefore, we calculated the R² and F values for each SNP and excluded SNPs with an F value less than 10. Palindromic SNPs with intermediate-effect allele frequencies were excluded. Finally, the preserved SNPs were used for MR analysis.

The statistical power was calculated using an online tool (https://​shiny.​cnsgenomic.​org/​mRnd/​) [19, 20]. This tool calculates power using the asymptotic theory to detect causal effects inferred from IVs. The Type I error rate was set at 0.05, and power was computed using the R² of IV, the proportion of cases with outcomes, and odds ratios (OR) obtained through IVW analysis.

Following the above screening criteria, 486 blood metabolites were considered exposure variables. IVW random effects were used to analyze the causal relationship between blood metabolites and IBS. Assuming no horizontal pleiotropy for all SNPs, IVW estimates were derived from a summary study of Wald ratios for all genetic variations. Under this premise, IVW provides the most accurate assessment of causal effects [21]. We used the weighted median (WM) and MR-Egger methods for additional analyses to strengthen the dependability of the results and ensure their accuracy [22]. Four analytical methods were used for the sensitivity analysis: MR-PRESSO, MR-Egger intercept, Cochran-Q test, and leave-one-out analysis (LOO). MR-PRESSO is a recently developed MR method designed to detect and correct abnormal horizontal pleiotropy values, thereby providing accurate estimates [23]. The MR-Egger intercept was calculated to assess the horizontal pleiotropy better and reduce the risk of bias from faulty instrumental variables (IVs) and horizontal pleiotropy [24]. Cochran’s Q test was used to assess heterogeneity among SNPs [25]. LOO was performed to evaluate the influence of each variable on causal estimates by sequentially removing one SNP from the analysis to ascertain if a single SNP predominantly influenced the findings [26].

Employing the False Discovery Rate (FDR) method, the p-values of the IVW analysis results for 486 metabolites were corrected. Metabolites with an FDR less than 0.05 were considered to have a relatively convincing causal relationship [27]. In contrast, metabolites with an FDR greater than 0.05 but a p-value less than 0.05 were deemed to have a nominally significant causal relationship and were identified as potential risk factors.

Despite setting strict screening criteria and conducting a sensitivity analysis to ensure data reliability, we repeated the IVW analysis in another IBS cohort using GWAS data of IBS cases released by the FinnGen consortium R9 (https://​r9.​finngen.​fi/​). Specifically, these data were obtained from a meta-analysis of GWAS conducted on 9323 Finnish patients with IBS and 301,931 controls. For this investigation, we utilized GWAS data with the accession number GCST012879, and for the replication, we used GWAS data from the FinnGen collaboration. Through a meta-analysis of the two MR analyses, we ultimately determined the blood metabolites that were responsible for the causative effects on IBS. The Review Manager (version 5.3) random-effects IVW model was used to perform the meta-analysis. This experiment was performed according to the protocol described by Cai et al. [28].

Genetic correlation is a key population parameter that describes the shared genetic architecture of complex traits and diseases [29]. Due to the genetic correlation between traits, MR estimates might violate causal effects [30]. To ensure that the causal effects are not confounded by the genetics of exposure and outcome, we used linkage disequilibrium score regression (LDSC) to examine the genetic correlation between selected metabolites and IBS.

Despite conducting an array of sensitivity analyses to assess the horizontal pleiotropy of the MR results and identify any SNPs that contravened the MR assumptions, some residual confounding SNPs may persist. We examined the metabolites in the IVs using the Phenoscanner V2 website (http://​www.​phenoscanner.​medschl.​cam.​ac) to evaluate the association of each SNP with known IBS risk factors, including psychological factors, such as anxiety [31]. To explore whether IBS has any causal effect on the identified important blood metabolites, we used IBS-related SNPs as IVs for reverse MR analysis (i.e., IBS as the exposure and identified blood metabolites as the outcome for MR analysis) [10].

We used GWAS data for IBS from another cohort downloaded from the FinnGen Alliance R9 and repeated the above MR analysis. We used GWAS data for IBS from another cohort downloaded from FinnGen Alliance R9 and replicated the above-mentioned MR analyses. We found that the replication results for Methionine, 1-palmitoylglycerophosphocholine, and Valine showed considerable differences, while Stearate, Arginine, 1-palmitoylglycerol, and 1-palmitoylglycerophosphoinositol yielded similar results, although they did not reach statistical significance. Meta-analysis was performed using random effects in Review Manager software (version 5.3, [32]. Our results showed similar outcomes. A previous meta-analysis has identified four metabolites associated with IBS. Methionine, 1-palmitoylglycerophosphocholine, and valine were excluded because of insignificant results (Figs. 4 and 5).

According to the results of the genetic correlation analysis, evidence of a genetic correlation between stearate and others in IBS is weak. This indicates that the shared genetic components did not confound the MR results (Table 2).

For each SNP of the identified 20 metabolites, we searched for phenotypes associated with exposure and the resulting variables. In the IVs of arginine and N-acetyl aspartic acid, we found that rs837763 and rs11030392 seemed to be related to mental state, but this did not affect our results. After excluding these confounding IVs, the results of arginine were still significant (odds ratio [OR]:1.41, 95% confidence interval [CI]:1.10–1.80), and the final meta-analysis results were significant (OR:1.36, 95% CI:1.08–1.70). In the reverse MR analysis, none of the 20 known metabolites showed a reverse causal relationship with IBS. Only the unknown metabolites X-11438, X-11521, X-14374, and X-13741 may have a reverse causal relationship with IBS (Fig. 6).