3.1 Instrumental variable selection
We identified 195 bacterial traits and 2559 SNPs at the class, family, genus, order, and phylum levels (of which 15 genomes were excluded because the information was unknown and one genome was excluded because there were no eligible SNPs). Detailed GM and the associated SNPs are in Additional file 2: Table S1. The F-statistics for all variables are greater than 10, indicating no weak instrumental variable bias. We identified 1820 SNPs associated with 91 inflammatory proteins, and the F-statistics for all variables were greater than 10(Additional file 3: Table S2).
3.2 UVMR results of GM and inflammatory proteins on cirrhosis
Figure 2 shows the effect of changes in abundance of the 192 GM taxa on the risk of cirrhosis. A total of 7 gut microorganisms (including one order, three families, and three genera) were significantly associated with cirrhosis(Additional file 4: Table S3, Fig. 3). Details of the 89 SNPs for the seven gut microbes are shown in Additional file 5: Table S4. The causal relationship between GM and cirrhosis was also analyzed using IVW, MR-Egger, WME, weighted, and Simple models (Additional file 6: Table S5). The above seven intestinal flora met the IVW analysis with a p-value of less than 0.05 as well as a horizontal pleiotropy test with a p-value of greater than 0.05, and the IVW was in the same direction as the effect size (β) calculated by MR-Egger(Fig. 3). Genetic prediction of 3 GM (family Clostridiaceae1, family Clostridiales vadin BB60 and genus Ruminococcus torques) is associated with reduced risk of cirrhosis. The family Clostridiaceae1 (OR = 0.787, 95%CI = 0.631 ~ 0.981, P = 0.033), family Clostridiales vadin BB60 (OR = 0.818, 95%CI = 0.701 ~ 0.954, P = 0.011), genus Ruminococcus torques (OR = 0.711, 95%CI = 0.536 ~ 0.943, P = 0.018). There are 4 GM that can increase cirrhosis risk: class Melainabacteria (OR = 1.166, 95%CI = 1.006 ~ 1.353 P = 0.042), family Lachnospiraceae (OR = 1.290, 95%CI = 1.044 ~ 1.593 P = 0.018), genus Eubacterium nodatum (OR = 1.121, 95%CI = 1.000 ~ 1.257 P = 0.050), genus Eubacterium ruminantium (OR = 1.162, 95%CI = 1.031~ 1.309 P = 0.014).
Figure 4 shows the effect of 91 circulating inflammatory protein changes on the risk of cirrhosis. A total of 11 inflammatory proteins were significantly associated with cirrhosis(Additional file 7: Table S6). Details of the 205 SNPs for the 11 inflammatory proteins are shown in Additional file 8: Table S7. The causal relationship between inflammatory proteins and cirrhosis was also analyzed using IVW, MR-Egger, WME, weighted, and Simple models (Additional file 9: Table S8). We discarded the T-cell surface glycoprotein CD5 (CD5) because the p-value of the multiplicity test for it was more significant than 0.05. Thus 10 inflammatory proteins were causally associated with cirrhotic episodes at the genetic level (Figure 4). 5 inflammatory proteins negatively correlated with cirrhosis: C-C motif chemokine 4 (CCL4) (OR = 0.923, 95%CI = 0.859 ~ 0.992, P =0.030), CD40L receptor (CD40LR) (OR = 0.915, 95%CI = 0.841 ~ 0.996, P =0.041), Leukemia inhibitory factor receptor (LIFR) (OR = 0.851, 95%CI = 0.760 ~ 0.952, P =0.005), Monocyte chemoatractant protein-3 (MCP-3) (OR = 0.866, 95%CI = 0.764 ~ 0.980, P =0.023), Tumor necrosis factor ligand superfamily member 12 (TNFSF12) (OR = 0.828, 95%CI = 0.733 ~ 0.937, P =0.003). 5 inflammatory proteins positively correlated with cirrhosis: C-X-C motif chemokine 10 (CXCL10) (OR = 1.138, 95%CI = 1.011 ~ 1.281, P =0.032), Fms-related tyrosine kinase 3 ligand (FLT3) (OR = 1.206, 95%CI = 1.084 ~ 1.342, P =0.001), Interleukin-10 receptor subunit alpha (IL10RA) (OR = 1.185, 95%CI = 1.018 ~ 1.380, P =0.029), Interleukin-1-alpha (IL-1α) (OR = 1.220, 95%CI = 1.017 ~ 1.463, P =0.032), Osteoprotegerin (OPG) (OR = 1.252, 95%CI = 1.026 ~ 1.527, P =0.027).
3.3 Sensitivity analysis
According to the MR-Egger regression intercept method and Cochran's Q test, there was no horizontal pleiotropy and heterogeneity of GM significantly associated with cirrhosis; detailed test results can be found in Additional file 10: Table S9 and Additional file 11: Table S10. We discarded T-cell surface glycoprotein CD5 (CD5) based on the MR-Egger regression intercept method that showed horizontal pleiotropy among the inflammatory proteins significantly associated with cirrhosis. The remaining ten inflammatory proteins did not show horizontal pleiotropy(Additional file 12: Table 11). Heterogeneity of Osteoprotegerin (OPG) among the inflammatory proteins significantly associated with cirrhosis according to Cochran's Q test was demonstrated, so we used a randomized IVW approach to explain the causal relationship. The remaining ten inflammatory proteins were not heterogeneous(Additional file 13: Table 12). Significantly associated GM and inflammatory proteins were visualized with cirrhosis as the outcome. The results of the forest, leave-one-out, funnel, and scatter plots were generally consistent with these results(Additional file 1).
3.4 Reverse Mendelian randomization analysis
In the inverse MR analysis of GM to cirrhosis, we set the significance level to P < 1×10-5 and the parameter for removal of linkage disequilibrium (LD) to R2 = 0.001, kb = 10,000. Among the 7 GM, the genus Eubacterium ruminantium showed reverse causality with cirrhosis, so we discarded it(The p-value in the IVW analysis was less than 0.05 and in the same direction as the MR-Egger effect size). Detailed results of the MR analysis can be found in Additional file 14: Table 13. We followed the same parametric criteria as the forward MR analysis for reverse MR causality analysis of 10 inflammatory proteins to cirrhosis, and the results showed no reverse causality(Additional file 15: Table 14).
3.5 MVMR results
The GM and inflammatory proteins obtained from the first two steps in Fig. 1 were used as exposures and endpoints for MR analysis of GM to inflammatory proteins(Fig. 1, Step 3). We analyzed the causal relationship between GM and inflammatory proteins described above and obtained eight groups of inflammatory proteins with possible mediating roles(Additional file 16: Table 15). Class Melainabacteria and TNFSF12, LIFR and MCP-3 all showed positive correlations(OR=1.104, 95%CI =1.019~ 1.196, P=0.016)、(OR=1.096, 95%CI =1.005~ 1.195, P=0.038)、(OR=1.101, 95%CI =1.004~ 1.210, P=0.041). Family Clostridiaceae1 and IL10RA, CCL4 and MCP-3 all showed negative correlation(OR=0.838, 95%CI =0.711~ 0.987, P=0.035)、(OR=0.866, 95%CI =0.760~ 0.987, P=0.032)、(OR=0.857, 95%CI =0.742~ 0.990, P=0.037). Family Lachnospiraceae and IL-1α are negatively correlated(OR=0.882, 95%CI =0.786~ 0.990, P=0.033). Positive correlation between Genus Ruminococcus torques and TNFSF12(OR=1.176 95%CI =1.001~ 1.383, P=0.049).
We sequentially included eight pairs of GM and inflammatory proteins associated with cirrhosis in mediation analyses using the MVMR method. We calculated the indirect effects (b') and proportions (c × b'/a) mediated by these inflammatory proteins. As shown in Table 2, After adjusting for Genus Ruminococcus torques, TNFSF12 still showed a significant correlation with cirrhosis, in contrast to Genus Ruminococcus torques, whose effect on cirrhosis became non-significant under multivariate adjustment(P=0.097). Therefore, we suggest that TNFSF12 affects the correlation between Genus Ruminococcus torques and cirrhosis and that TNFSF12 may mediate the Genus Ruminococcus torques to cirrhosis pathway. The mediation percentage was 6.8% (P=0.023)