Causal relationships between gut microbiota and programmed cell death protein 1/programmed cell death-ligand 1: A bidirectional Mendelian randomization study

Abstract

Background: Multiple clinical studies have indicated that the gut microbiota influences the effects of immune checkpoint blockade (ICB) therapy comprising PD-1/PD-L1 inhibitors, but the causal relationship is unclear. Because of numerous confounders, many microbes related to PD-1/PD-L1 have not been identified. This study aimed to determine the causal relationship between the microbiota and PD-1/PD-L1 and identify possible biomarkers for ICB therapy.

Method: We used bidirectional two-sample Mendelian randomization with two different thresholds to explore the potential causal relationship between the microbiota and PD-1/PD-L1 and species-level microbiota GWAS to verify the result.

Result: In the primary forward analysis, genus_Holdemanella showed a negative correlation with PD-1 [βIVW = -0.25; 95% CI (-0.43 to -0.07); P_FDR = 0.028] and genus_Prevotella9 showed a positive correlation with PD-1 [βIVW = 0.2; 95% CI (0.1 to 0.4); P_FDR = 0.027]; order_Rhodospirillales [βIVW = 0.2; 95% CI (0.1 to 0.4); P_FDR = 0.044], family_Rhodospirillaceae [βIVW = 0.2; 95% CI (0 to 0.4); P_FDR = 0.032], genus_Ruminococcaceae_UCG005 [βIVW = 0.29; 95% CI (0.08 to 0.5); P_FDR = 0.028], genus_Ruminococcus_gnavus_group [βIVW = 0.22; 95% CI (0.05 to 0.4); P_FDR = 0.029], and genus_Coprococcus_2 [βIVW = 0.4; 95% CI (0.1 to 0.6); P_FDR = 0.018] were positively correlated with PD-L1; and phylum_Firmicutes [βIVW = -0.3; 95% CI (-0.4 to -0.1); P_FDR = 0.031], family_ClostridialesvadinBB60group [βIVW = -0.31; 95% CI (-0.5 to -0.11), P_FDR = 0.008], family_Ruminococcaceae [βIVW = -0.33; 95% CI (-0.58 to -0.07); P_FDR = 0.049], and genus_Ruminococcaceae_UCG014 [βIVW = -0.35; 95% CI (-0.57 to -0.13); P_FDR = 0.006] were negatively correlated with PD-L1. The one significant species in further analysis was species_Parabacteroides_unclassified [βIVW = 0.2; 95% CI (0-0.4); P_FDR = 0.029]. Heterogeneity (P > 0.05) and pleiotropy (P > 0.05) analyses confirmed the robustness of the MR results.

Keywords: PD-1/PD-L1; bidirectional Mendelian randomization; causality; gut microbiota; pQTL.

Figure 1

Workflow of this MR analysis. SNPs, single-nucleotide polymorphisms; MR, Mendelian randomization; LD, linkage disequilibrium; eQTL, expression quantitative trait loci; pQTL, protein quantitative trait loci.

Figure 2

Forest plot of causal relationships estimated and sensitivity analysis for genus-level microbes and PD-1/PD-L1, the significant result (P_FDR <0.05) by the IVW method in forward two-sample MR analysis (includes two thresholds). The words in bold type indicate significant results. CI, confidence interval; F, F-statistics; R², the genetic variants for instrument; IVW, inverse variance weighted.

Figure 3

The significant (P_FDR <0.05) and robust results (Family_ClostridialesvadinBB60group and Genus_RuminococcaceaeUCG014) in forward MR analysis with two different thresholds. Scatter plot of microbe-related SNP effects on PD-L1, with the slope of each line corresponding to the estimated MR effect per method. Vertical and horizontal black lines around each point show the 95% confidence interval for each polymorphism exposure association and polymorphism outcome association. Forest plot lists single and combined (IVW and MR egger) SNP MR-estimated effect sizes; the effect estimates represent the β for PD-L1 per one-s.d. increase in mean microbes. The one-sided leave-one-out and symmetric funnel plots meant that the results were stable without outliers. Family_Clostridiales_vadin_BB60_group: (A) Forest plot at the loose threshold. (B) MR scatter at the loose threshold. (C) Leave one out at the loose threshold. (D) Funnel plot at the loose threshold. (E) MR scatter at the strict threshold. Genus_Ruminococcaceae_UCG014: (F) Forest plot at the loose threshold. (G) MR scatter at the loose threshold. (H) Leave one out at the loose threshold. (I) Funnel plot at the loose threshold. (J) MR scatter at the strict threshold.

Figure 4

In the species-level bidirectional two-sample MR analysis, microbial features were prefixed with species(s). The selection of species-level microbes is based on the significant result of genus-level microbes. The MR estimates and 95% CI values are shown in the plot. The point of the plot indicates the P value of IVW, and red indicates significance (P_FDR <0.05). The “+” and “-” in the legend indicate the direction of the estimate effect (beta).

Figure 5

When sample 1(exposure) and sample 2(outcome) are used for causal estimates in MR inference, three assumptions must be satisfied (11). ① Relevance assumption: the genetic variations are highly related to the exposure, ② independence assumption: the genetic variants are not associated with any putative confounder of the association between exposure and result, and ③ exclusion restriction: the variants do not alter the outcome independently of exposure.