The Causal Role of Gut Microbiota in Pulmonary Embolism and the Mediating Effects of Circulating Metabolites: A Mendelian Randomization Study.

Y. G. contributed to study design, was responsible for statistical analysis, data interpretation, and wrote the article. X. J. contributed to data curation, statistical analysis, and participated in article revision. Y. L. contributed to data processing, literature review, and article revision. H. H. contributed to data curation, assisted in statistical analysis, and participated in article revision. S. W. contributed to data analysis, assisted in result interpretation, and participated in article revision. Y. H. contributed to article revision. H. Z. provided statistical advice and revised the article. Z. Z. offered clinical interpretation, participated in article review and revision. W. S. provided methodological guidance, was responsible for conceptualization, study design, project supervision, and article revision. P. Y. was responsible for conceptualization, project supervision, funding acquisition, and article revision. All authors have reviewed and approved the final article.

Not applicable. This study utilized publicly available summary‐level data from genome‐wide association studies. All original studies contributing to these consortium datasets had obtained ethical approval and participant consent from their respective institutional review boards. No separate ethical approval was required for the current analysis.

This study conducted a two‐step MR analysis to investigate the causal relationship between gut microbiota and PE relationship and the potential mediating role of plasma metabolites. First, the causal effect of each gut microbiota genus on each candidate metabolite (β 1 ) was evaluated. Next, the causal effect of each metabolite on PE (β 2 ) was estimated, and the total effect (β) of gut microbiota on PE was also estimated. The mediation (indirect) effect was calculated as the product of the two coefficients (β 1 × β 2 ). The direct effect was calculated as the total effect minus the indirect effect (β ‐ β 1 β 2 ). The proportion of the total effect mediated by the metabolite was quantified as (β 1 β 2 /β) [ 14 ] (Figure  1 ). Schematic diagram of the study design. This study employed a comprehensive two‐sample Mendelian randomization design, leveraging SNPs as instrumental variables from large‐scale genome‐wide association studies of gut microbiota, blood metabolites, and PE to first establish causality and then elucidate the mediating role of specific metabolites using a two‐step MR mediation framework. SNPs were used in this study as instrumental variables (IVs) to explore the causal relationship between gut microbiota and PE. The analysis complied with the basic principles and three core assumptions of MR [ 15 ]: (1) a significant correlation existed between the IVs and the exposure factors; (2) IVs were not associated with other confounding factors that influenced both the exposure factors and outcome variable; (3) IVs did not directly affect the outcome variable and influenced it only through the exposure factors. GWAS data of the gut microbiota were from the MiBioGen consortium, which published a GWAS analysis on the composition of gut microbiota [ 16 ]. This dataset included 18,340 participants across 24 cohorts, with a majority being of European descent. The outcome variable of interest, PE, was derived from GWAS summary data of the latest R10 version released by the FinnGen consortium (Release 10, available at https://www.finngen.fi/en ), which included 10,046 cases and 401,128 controls. The data on circulating metabolites were from a large‐scale genome‐wide meta‐analysis involving 136,016 participants across 33 cohorts [ 17 ]. This MR analysis included 196 bacterial taxa, which comprised 9 phyla, 16 classes, 20 orders, 32 families, and 119 genera, excluding 15 unspecified gut microbiota groups. SNPs associated with gut microbiota were selected with a significance threshold of P < 1 × 10 −5 , optimizing the balance between statistical stringency and the strength of genetic instruments, in line with established methodological practices. Additionally, an allelic distance greater than 10,000 kb was set, and the threshold parameter r 2  < 0.001 was applied to reduce the impact of linkage disequilibrium. The F statistic was used to assess the strength of the association between SNPs and gut microbiota, where an F value of less than 10 indicated a weak correlation between the SNPs and the exposure factors. The inverse variance weighting (IVW) method was utilized as the primary analytical approach for MR owing to its high statistical power. Furthermore, MR‐Egger regression and the weighted median (WM) method were additionally employed to verify the robustness of the findings. To further validate the stability and reliability of the results, sensitivity analyzes were conducted, including horizontal pleiotropy testing, heterogeneity testing, and leave‐one‐out analysis. MR‐Egger regression was employed to test for horizontal pleiotropy, where a significant intercept in the analysis would indicate the presence of horizontal pleiotropy. The MR pleiotropy residual sum and outlier test (MR‐PRESSO) was used to identify outliers, and significant outliers were removed to reduce the impact of horizontal pleiotropy. Heterogeneity testing was performed using Cochran's Q statistic, with a significant p ‐value (< 0.05) indicating the presence of heterogeneity. Leave‐one‐out analysis was employed as a sensitivity analysis to assess the impact of individual SNPs on the IVW results. In this study, a reverse MR analysis was conducted using gut microbiota that were causally related to PE. The aim was to eliminate the interference of reverse causality and ensure the accuracy of our findings. This study utilized R version 4.3.2 and the “TwoSampleMR” package to conduct the MR analysis. The results are presented as odds ratios (ORs) along with their 95% confidence intervals (CIs). A p ‐value of < 0.05 was considered as the statistical significance threshold. A p ‐value < 0.05 was considered nominally significant. To account for multiple testing across the 196 gut microbial taxa and 233 circulating metabolites, we applied two complementary correction methods: the Benjamini‑Hochberg false discovery rate (FDR) procedure (q < 0.05 considered statistically significant) and the Bonferroni correction ( p  < 0.05/196 or p  < 0.05/233).

We first performed MR analysis to explore potential causal associations between gut microbiota and PE. Using a nominal significance threshold ( p  < 0.05), we identified 12 microbial taxa with suggestive associations (Figure  2 ). IVW results showed that the Subdoligranulum, Lentisphaerae, and Erysipelatoclostridium nominally increased the risk of PE, whereas LachnospiraceaeUCG010, FamilyXIII, Streptococcaceae, Collinsella, DefluviitaleaceaeUCG011, Firmicutes, Sutterella, Defluviitaleaceae, and Intestinimonas were negatively associated with PE. However, to account for multiple testing across 196 microbial features, we applied both the Benjamini‐Hochberg (FDR) and Bonferroni corrections. After FDR correction, none of these associations remained statistically significant (q > 0.05; Supporting Information Table  S1 ). Similarly, no association survived Bonferroni correction. These nominal findings should be interpreted as exploratory and hypothesis‐generating, not as robust causal evidence. Forest plot of the significant associations between 12 gut microbial taxa and PE. Significant positive and negative associations between gut microbiota and PE are shown, based on IVW analysis. CI, confidence interval. According to MR‐Egger regression intercept analysis, genetic pleiotropy did not bias the results, and MR‐PRESSO analysis demonstrated that no horizontal pleiotropy in the MR analysis ( p  > 0.05). The Cochran's Q tests showed no significant heterogeneity ( p  > 0.05). The overall effect of the genus Subdoligranulum, phylum Lentisphaerae, and genus Erysipelatoclostridium on PE are shown in scatter plots (Figure  3A–C ). Scatter plot of the associations between bacterial traits and PE. Trends in the causal associations between A. genus Subdoligranulum and PE; B. phylum Lentisphaerae and PE; C. genus Erysipelatoclostridium and PE. The results of “leave‐one‐out” analysis proved that the MR analysis turned out to be reliable (Figure  4A–C ). Next, a reverse MR analysis was performed to determine whether PE affected specific gut microbiota. The analysis showed that no reverse causal relationship between gut microbiota and PE (Supporting Information Table  S2 ). Additional detailed results are shown in Supporting Information Figures  S1 – 2 . Leave‐one‐out plot of the associations between bacterial traits and PE. MR leave−one−out sensitivity analysis for A. genus. Subdoligranulum and PE; B. phylum. Lentisphaerae and PE C. genus. Erysipelatoclostridium and PE. Employing the IVW method as the primary analytical approach, we investigated the potential causal effects of circulating metabolites on PE. The analysis identified 26 metabolites exhibiting suggestive evidence of a causal relationship with PE risk. These metabolites spanned several lipid classes and metabolic pathways, including triglycerides, cholesterol constituents, phospholipids, fatty acids, among others. A comprehensive summary of these causal estimates for all significant metabolites is presented in Figure  5 . After Bonferroni correction for 233 tests, Triglycerides in small HDL and Ratio of omega‐3 fatty acids to total fatty acids remained statistically significant. Following Benjamini‐Hochberg FDR correction (q < 0.05), 6 metabolites showed suggestive associations (Supporting Information Table  S3 ). Sensitivity analyzes showed no evidence of horizontal pleiotropy for all metabolites that served as mediators, both the MR‐Egger intercept test ( p  > 0.05) and the MR‐PRESSO global test ( p  > 0.05) supported this conclusion, with the exception of one non‐mediator metabolite for which the global test gave p  = 0.03 (Supporting Information Table  S4 ). In a few cases, the MR‐PRESSO global p ‐value could not be computed because of an insufficient number of instrumental variables, indicated as “NA” or “Not enough instrumental variables” in Supporting Information Table  S4 . Forest plot of the significant associations between 26 circulating metabolites and PE. Significant positive and negative associations between circulating metabolites and PE are shown, based on IVW analysis. In the next MR analysis, we analyzed the 12 significant gut bacterial taxa above with metabolites (Figure  6 ). These gut bacteria were be causally related to the following categories of metabolites: the first category included associations with lipoproteins, such as triglycerides to total lipids ratio in medium HDL (high‐density lipoprotein); the second category included associations with lipids, such as ratio of 226 docosahexaenoic acid to total fatty acids; the third category included associations with amino acids, such as histidine levels. Heatmap of associations between gut microbiota and circulating metabolites. Colors represent odds ratios (OR) for each pairwise association, with blue (OR  1), scaled around a neutral value of 1 (white). Asterisks denote statistical significance: * p  < 0.05, ** p  < 0.01. All associations are nominal (exploratory). Although the total effects of the gut microbiota on PE did not survive multiple testing correction, we performed exploratory mediation analyzes to identify potential metabolic pathways that might underlie the observed nominal associations. Two‐step MR analysis revealed several significant mediating pathways in the causal relationship between gut microbiota and PE. Specifically, the causal association of Defluviitaleaceae on PE risk was significantly mediated by histidine levels (indirect effect: 0.0178, 95% CI = 0.0027 ~ 0.032, p  = 0.021). Similarly, the influence of Lentisphaerae on PE risk was partly mediated through its effect on triglyceride metabolism within small HDL particles (indirect effect: 0.0067, 95% CI = 0.0016 ~ 0.0118, p  = 0.011). Furthermore, the effects of Erysipelatoclostridium were mediated by the ratio of 226 docosahexaenoic acid to total fatty acids (indirect effect: −0.0060, 95% CI = − 0.0114 ~ − 0.0006, p  = 0.029) (Supporting Information Table  S5 ).

This study represents a comprehensive MR investigation to explore the potential causal relationships between gut microbiota, circulating metabolites, and PE. Our findings, although exploratory, provide preliminary evidence supporting a role of the gut‐vascular axis in thromboembolic disease. By leveraging genetic instruments, we have provided new evidence supporting the role of specific gut microbial features and metabolite profiles in the pathogenesis of PE, highlighting the potential of the gut‐pulmonary vascular axis for understanding thromboembolic disease. Our analysis identified 12 bacterial taxa with putative causal associations on PE risk, including members of the phyla Lentisphaerae and Firmicutes, families such as Defluviitaleaceae and Streptococcaceae, and genera like Intestinimonas and Subdoligranulum. These findings aligned with existing literature suggesting that gut microbiota composition influences vascular health [ 18 ]. In patients with chronic thromboembolic pulmonary hypertension (CTEPH), a devastating sequela of PE, a dysregulation of the gut microbiome is observed, characterized by a reduction in microbial diversity and alterations in specific bacterial taxa [ 19 ]. Notably, several taxa identified as protective in our study, such as Firmicutes and Defluviitaleaceae, have been previously associated with anti‐inflammatory properties, partly attributable to their capacity to produce SCFAs like butyrate [ 20 ]. Butyrate has been shown to enhance endothelial integrity and suppress pro‐inflammatory pathways, thereby potentially reducing thrombotic risk [ 21 ]. Similarly, butyrate has been shown to ameliorate pulmonary hypertension by modulating the immune response and inhibiting smooth muscle cell proliferation [ 19 ]. In contrast, Erysipelatoclostridium was nominally associated with increased PE risk. Although its specific metabolic functions in humans remain incompletely characterized, related members of the family Erysipelotrichaceae have been implicated in bile acid metabolism and lipid regulation. Dysregulation of bile acid profiles can influence coagulation and endothelial function, potentially linking gut microbiota to thrombotic risk. However, direct evidence linking Erysipelatoclostridium to these pathways is lacking. Future studies are needed to clarify its role in PE pathogenesis. Therefore, this study identified the causal role of specific gut microbiota in PE, laying the groundwork for the following mediation analyzes to unravel the underlying metabolic pathways. A particularly innovative aspect of our study was the integration of metabolomic data within an MR mediation framework. We identified several circulating metabolites, including histidine, glycine, lipoprotein lipid fractions, and polyunsaturated fatty acids, that not only exhibited causal effects on PE but also mediated the influence of gut microbiota. The mediating roles of these metabolites suggest that the gut microbiota may contribute to PE susceptibility largely through modulating host metabolic pathways. This is consistent with the concept of the gut microbiome as an “endocrine organ” capable of generating and modifying bioactive molecules that enter systemic circulation and influence distal physiological processes, including coagulation and endothelial function [ 22 ]. Gut microbiota‐derived metabolites such as TMAO have been shown to exacerbate pulmonary hypertension by activating macrophages to express inflammatory cytokines and inducing endothelial dysfunction [ 19 ]. In a previous study, we have found significant alterations in amino acid metabolism in CTEPH patients, particular in chronic thromboembolic tissue with atherosclerosis‐like lesions [ 23 ]. In our study, Defluviitaleaceae appeared to exert a protective effect against PE partly through modulating histidine levels. Elevated levels of histidine have been associated with pro‐thrombotic states [ 24 ]. However, the precise mechanisms by which histidine, small HDL triglycerides, and the docosahexaenoic acid ratio influence thromboembolism remain incompletely understood. Histidine may affect platelet function via histamine metabolism, as suggested by previous studies on histidine‐rich glycoprotein [ 25 ]. Regarding lipid mediators, while large HDL particles are generally considered cardioprotective, the role of triglycerides within small HDL subfractions is less established; it is possible that this metric reflects altered lipoprotein remodeling or inflammatory states rather than a direct protective effect [ 26 ]. DHA is a precursor to resolvins and protectins, which have anti‐platelet and anti‐inflammatory effects [ 27 ]. A higher DHA ratio would therefore increase these protective mediators, consistent with our observed inverse mediation effect. Direct experimental evidence linking these specific metabolites to coagulation cascades is currently lacking, and our mediation findings should be viewed as hypothesis‐generating for future mechanistic studies. Regarding the directionality of lipid findings, the observation that Lentisphaerae influences PE risk through triglycerides in small HDL does not directly contradict the known cardioprotective role of total HDL cholesterol. A recent study demonstrated that larger HDL particle size is associated with lower VTE risk, whereas smaller HDL particle size increases risk in a dose‐dependent manner, suggesting that HDL subfraction quality may modulate thrombotic risk [ 28 ]. Therefore, increased triglyceride content within small HDL might not reflect a beneficial effect. Similarly, our findings revealed the role of gut microbiota in modulating lipids and lipoprotein‐related pathways to influence PE risk. Our findings that gut microbiota influence PE risk through lipid‐related metabolites are further substantiated by our recent GWAS in the Han Chinese population, which identified a functional variant in the fatty acid transport gene FABP2 as a novel risk locus for PE [ 29 ]. Notably, FABP2 is primarily expressed in the intestinal epithelium, where it facilitates the absorption of dietary fatty acids. This intestinal‐specific expression pattern positions FABP2 as a key molecular bridge, potentially mediating the influence of gut luminal content, including the microbiota, on systemic lipid profiles, thereby contributing to the pro‐thrombotic milieu that increases PE risk. In our study, microbial taxa including Lentisphaerae and Subdoligranulum influenced PE risk through alterations in triglyceride metabolism within HDL subfractions. These findings were consistent with existing knowledge regarding the cardioprotective functions of HDL, not only in reverse cholesterol transport but also in anti‐inflammatory and antioxidant activities [ 19 ]. Our findings are further substantiated by human pathological evidence from the pulmonary arterial lesions of CTEPH patients, which frequently exhibit co‐localization of atherosclerosis‐like plaques and organized thrombi, highlighting dysregulated lipid metabolism and thrombus chronicization as intertwined processes in PE progression [ 30 ]. Collectively, these studies highlight the importance of lipid metabolic pathways in thromboembolic events, and our study extended this to PE by showing that gut microbiota‐modulated metabolites mediated these effects. From a clinical perspective, our findings open new avenues for PE prevention and intervention. Targeting the gut microbiota through dietary modifications, prebiotics, or probiotics may represent a novel strategy to modulate metabolite profiles and reduce thrombotic risk [ 19 ]. Moreover, future studies may investigate the potential of the metabolites identified herein as biomarkers for identifying high‐risk individuals or monitoring responses to interventions. This phenomenon, in which nominal associations do not remain significant after multiple‐testing correction, is not unique to our study. Recently, multiple microbiome MR studies involving diseases such as rheumatoid arthrities [ 31 ], hepatic encephalopathy [ 32 ], ARDS [ 33 ], endometriosis [ 34 ], and multiple system atrophy [ 35 ]have reported similar results. Collectively, these findings consistently indicate that in microbiome MR studies, nominal associations tend to be attenuated under stringent correction, and our results are consistent with this previously documented pattern. In all these cases, the authors concluded that their findings provided hypothesis‐generating evidence and identified candidate taxa for future mechanistic research. Furthermore, the conservative nature of Bonferroni correction in high‐dimensional microbiome data, where many taxa are biologically correlated has been noted, with some studies adopting modified FDR methods based on the effective number of independent tests to balance type I and type II errors [ 36 ]. These precedents support the notion that, in microbiome‐wide MR studies, nominal associations with robust sensitivity analyzes and consistent effect directions can provide valuable hypothesis‐generating signals, even when they do not survive the most stringent multiple‐testing corrections. Several limitations should be acknowledged. First, population heterogeneity may affect the generalizability of our findings. The gut microbiota GWAS data are derived from multiple cohorts predominantly of European descent but not exclusively Finnish, whereas the PE outcome data are from the Finnish population. Although both are broadly European, subtle differences in ancestry, diet, or environmental factors could influence gut microbiota composition and genetic architecture. If the instrumental SNPs have different allele frequencies or linkage disequilibrium patterns across these populations, the MR estimates might be biased. Our sensitivity analyzes (MR‑Egger, leave‑one‑out) did not indicate substantial horizontal pleiotropy, but replication in independent, ancestrally matched cohorts is still necessary to confirm our findings. Second, the multiple‐testing correction methods applied Bonferroni and FDR are inherently conservative, particularly in the context of microbiome data, where many taxa are biologically correlated and not independent. This may increase the risk of type II errors, potentially overlooking weak but genuine signals. The use of more flexible correction strategies could be considered in future studies to better balance the trade‐off between type I and type II errors.

In conclusion, our mediation analyzes identified histidine, triglycerides in small HDL, and the docosahexaenoic acid ratio as potential metabolic mediators linking gut microbiota to PE risk, providing testable hypotheses for future functional and mechanistic studies. Although direct microbial effects did not survive multiple‐testing correction, the consistent direction of effects and robust sensitivity analyzes support the exploration of the gut‐vascular axis as a promising area for PE research. These findings underscore the importance of lipid subfraction and amino acid metabolism in thromboembolic risk and highlight the need for replication in independent cohorts and mechanistic investigation.

Pulmonary embolism (PE) is a prevalent and life‐threatening disorder characterized by the obstruction of blood flow in the pulmonary arteries, most commonly due to the migration of a thrombus from the deep veins of the lower extremities [ 1 ]. Globally, the incidence of PE ranges from 60 to 120 cases per 100,000 individuals annually. In the United States alone, it accounts for between 60,000 and 100,000 deaths each year, underscoring its substantial burden on public health [ 2 ]. The pathogenesis of PE involves multifactorial and multistep pathological processes, influenced by an interplay of genetic, environmental, and acquired risk factors. Recent discussions on PE management have highlighted ongoing challenges in prophylaxis and treatment [ 3 ]. In recent years, the conceptual framework of the “gut‐lung axis” has gained traction, with studies revealing correlations between gut microbial dysbiosis and multiple respiratory conditions, including chronic obstructive pulmonary disease and asthma [ 4 ]. As a complex ecosystem, the gut microbiota maintains intimate interactions with host health, and its dysbiosis has been broadly associated with various diseases. Emerging evidence has increasingly indicated the potential involvement of the gut microbiota in the pathogenesis of PE [ 5 ]. Dysregulation of the microbiota–metabolite axis (dysbiosis) may trigger a shift in the host metabolic environment, leading to a prothrombotic state that predisposes to or exacerbates PE. Of particular interest is the potential mechanism by which gut microbiota influences PE susceptibility through modulation of circulating metabolites. The gut microbiota actively releases a diverse repertoire of bioactive metabolites, many of which gain access to the host's systemic circulation [ 6 ]. These metabolites may act as crucial mediators of the gut‐lung axis, potentially linking distal microbial communities to pulmonary vascular pathophysiology. Microbial‐derived metabolites such as trimethylamine N‐oxide (TMAO), short‐chain fatty acids (SCFAs), and bile acids have been shown to exert profound effects on the host [ 7 ]. Alterations in the blood levels of microbiota‐dependent metabolites can directly affect platelet function, coagulation cascades, and endothelial integrity—thereby modulating core mechanisms of thrombus formation and stability [ 8 ]. For example, TMAO enhances platelet hyperreactivity and amplifies pro‐thrombotic signaling pathways [ 9 ]. In contrast, SCFAs like butyrate exhibit anti‐inflammatory and endothelium‐protective effects, suggesting a potential protective role against thrombus formation [ 10 ]. Investigating the cause‐mediator‐effect relationship between gut microbiota, metabolites, and PE may not only deepen our understanding of the pathogenesis of PE but also provide new microbiome‐ or metabolite‐targeted strategies for PE prevention and treatment. Mendelian randomization (MR) is a statistical method that uses genetic variation to assess whether the association between risk factors and outcomes is causal [ 11 ]. It leverages the random allocation of single‐nucleotide polymorphisms (SNPs) at conception and the irreversibility of genetics to reduce the influence of reverse causation and potential confounding factors in epidemiological studies [ 12 ]. This study performed a comprehensive MR analysis using summary statistics from the most up‐to‐date large‐scale genome‐wide association studies (GWAS) of gut microbiota, blood metabolites, and PE to dissect the associations between them [ 13 ].

The authors declare no conflicts of interests.

Supporting File 1 Supporting File 2 Supporting File 3 Supporting File 4 Supporting File 5 Supporting File 6