Exploring the causal relationship between acute respiratory distress syndrome and gut microbiota: Unveiling the gut-lung axis through a large-scale Mendelian randomization study

Abstract

Observational studies have linked gut microbiota (GM) to acute respiratory distress syndrome (ARDS), but the cause-and-effect relationship is yet to be determined. We acquired the latest genome-wide association study (GWAS) summary statistics on GM taxa from the German (N = 8956), Dutch (N = 7738), and MibioGen (N = 18,340) consortium GWAS catalogs, as well as ARDS (N_case = 431, N_control = 4,93,301) GWAS summary statistics from the FinnGen consortium. We employed Mendelian randomization (MR) analysis using inverse-variance weighted, MR-Egger, weighted mode, and weighted median methods to investigate the causal link between GM and ARDS. We used a Bonferroni correction to account for statistical bias resulting from repeated comparisons in order to control for false positive outcomes during multiple hypothesis testing. Meta-analyses were conducted to enhance the statistical power of the MR analysis. GM had no statistically significant impact on ARDS with Bonferroni correction. Actinobacteria, Proteobacteria, Bifidobacteriales, Bifidobacteriaceae, Rikenellaceae, Dorea, Streptococcus, Bilophila wadsworthia, Escherichia unclassified, OTU99_17 (Parabacteroides), TestASV_7 (Bacteroides), TestASV_26 (Phascolarctobacterium), and TestASV_43 (Parasutterella) are notable GM taxa with low uncorrected MR inverse-variance weighted P-values, potentially indicating a reduced risk of ARDS. Bifidobacterium longum, OTU99_30 (Parasutterella), and TestASV_16 (Bacteroides) are potentially linked to an increased risk of ARDS. Meta-analyses based on 3 GWAS summary statistics suggested that Streptococcus was potentially indicating a reduced risk of ARDS (odds ratio 0.610; 95% confidence interval 0.430–0.870; P = .006). These associations were proven to be robust through analyses of sensitivity, heterogeneity, and horizontal pleiotropy. From a genetic perspective, our findings suggest a potential relationship between the GM and ARDS rather than definitive causality. Given possible confounding and methodological constraints, the results should be interpreted with caution. Additional studies are needed to elucidate underlying mechanisms and clarify microbial interactions within the GM.

Key points

• •Question: Is the gut microbiota causally related to acute respiratory distress syndrome (ARDS)?
• •Findings: From a genetic perspective, our study suggests a causal link between gut microbiota and ARDS development. Meta-analyses based on 3 GWAS summary statistics suggested that Streptococcus was potentially indicating a reduced risk of ARDS. Additional studies are required to elucidate the mechanisms behind these effects and to comprehend the interactions within the gut microbiota.
• •Meaning: This Mendelian randomization study supported a causal relationship between gut microbiota and ARDS. More research is needed to determine the precise processes driving these impacts on the incidence of ARDS and the interactions within the gut microbiota.

1. Introduction

Acute respiratory distress syndrome (ARDS) can result from either primary or secondary acute lung injury and is characterized by a strong inflammatory response and cytokine activation.^[1] Around 10% of patients in intensive care units globally are estimated to suffer from ARDS.^[2–7] Mortality rates for ARDS patients range from 17% to 39%, according to several studies.^[8,9] Gas exchange dysfunction is a critical aspect of the pathogenesis and pathophysiology of ARDS, contributing to increased mortality in these patients. Severe inflammatory responses and cytokine activation cause damage to the alveolar wall, leading to pulmonary congestion.^[10] Consequently, immune response modulation and the maintenance of alveolar function homeostasis are 2 key components of current therapeutic strategies for ARDS patients.

The term “gut microbiota (GM)” refers to the diverse community of bacteria, fungi, and viruses that reside in the human digestive system. These microbes are crucial in modulating immune responses and inflammation in the lungs.^[11,12] Disease-induced gut dysbiosis, characterized by a reduction in beneficial GM metabolites, compromise of the gut barrier, and increased intestinal permeability, can allow bacterial components, such as lipopolysaccharides and pathogen-associated molecular patterns, or even entire bacteria, to enter the lymphatic or bloodstream. This process contributes to oxidative stress, systemic inflammation,^[13] and may accelerate the development of respiratory illnesses.^[14] Gut dysbiosis heightens lung and intestinal inflammation, impairs the ability to control inflammation, and affects immune cell function, increasing the expression of inflammatory factors and exacerbating lung damage during infections.^[15,16]

There might be a connection between ARDS and GM. ARDS and GM may be connected, according to observational clinical research, and variations in the microbiota may correlate with the severity of the disease.^[17] In patients suffering from severe SARS-CoV-2 infections, for instance, the GM is enriched with clostridia, such as Clostridium hathewayi and Clostridium ramosum.^[18] In individuals with severe acute pancreatitis and ARDS, there is a reduced abundance of Bifidobacterium but an increased presence of Enterobacteriaceae, Shigella, Proteobacteria, and Klebsiella pneumoniae.^[19] Furthermore, a potentially protective factor against ARDS could be a rich and diverse gut microbiome.^[20] Supplementing with symbiotic microbial metabolites has the potential to enhance treatment strategies for severe COVID-19 cases.^[20] Certain microorganisms, such as Eubacterium rectale, Clostridium butyricum, Clostridium leptum, and Faecalibacterium prausnitzii, exhibit altered abundance in COVID-19 patients with ARDS.^[21,22] However, it is challenging to adequately depict the direct relationship between ARDS and GM due to the limitations of observational clinical research, including an excessive number of influencing factors, inadequate sample representation, and limitations of fecal samples.

Mendelian randomization (MR) often employs single nucleotide polymorphisms (SNPs) as instrumental variables (IVs) to infer causal relationships in disease pathogenesis research. Multiple MR studies have demonstrated significant associations between GM and diseases such as major depressive disorder, schizophrenia, and Alzheimer’s disease.^[23] While there is evidence suggesting that ARDS development is linked to GM dysbiosis, the direct causal link between the 2 remains uncertain. We performed a MR and meta-analysis to investigate the causal relationship between GM and ARDS, utilizing extensive genome-wide association study (GWAS) summary data on GM taxa from 3 primary databases. This study seeks to clarify the connection between GM and ARDS using MR techniques, providing new insights for future research by utilizing the proven reliability of MR studies in determining causality.

2. Methods

2.1. Study design

To determine the causal relationships between variations in gut microbial abundance and the risk of ARDS, we employed causal inference (MR analysis). In summary, the MR analysis examined GM as the exposure and ARDS as the outcome (Fig. 1). Table S1, Supplemental Digital Content, https://links.lww.com/MD/Q478 shows the STROBE-MR checklist for this study.^[24,25]

Figure 1.

Outline of the analytical procedure for examining the causal link between gut microbiota and acute respiratory distress syndrome via bidirectional Mendelian randomization. The Mendelian randomization study must meet 3 assumptions in the above illustration: (1) there must be a robust association between the instrument and the exposure; (2) the genetic variant is not associated with confounders of the exposure–outcome association; and (3) the instrument affects the outcome solely through the exposure. The gut microbiota served as exposure, and acute respiratory distress syndrome served as the outcome. The single nucleotide polymorphisms obtained from the gut microbiota were considered as instrumental variables. ARDS = acute respiratory distress syndrome, MR = Mendelian randomization, SNP = single nucleotide polymorphism.

This study utilized GWAS summary data sourced from publicly accessible databases. Summary data on GM were sourced from 3 datasets: the German consortium, the Dutch Microbiome Project, and the International MibioGen consortium.

The German consortium dataset includes 8956 participants from 5 independent cohorts across biobanks in northern, northeastern, and southern Germany, specifically in Kiel, Mecklenburg-Western Pomerania, and Augsburg.^[26] After post-filtering for microbial composition, 233 univariate features were identified for abundance-based analysis, categorized as 4 at the phylum level, 8 at the class level, 6 at the order level, 10 at the family level, 29 at the genus level, and 65, 62, and 49 at the 97% OTU, 99% OTU, and ASV levels, respectively.

The Dutch Microbiome Project, which examines the gut microbiome’s composition and function in 8208 individuals, is the primary source of data for the Dutch population. Analysis of summary GWAS data from 7738 subjects revealed 207 gut microbial components, including 5 phyla, 10 classes, 13 orders, 26 families, 48 genera, and 105 species.^[27]

The MibioGen dataset includes data from 18,340 participants across 24 cohorts in countries like Canada, the United States, Belgium, Denmark, Finland, Germany, the Netherlands, Sweden, the United Kingdom, Israel, and South Korea, with approximately 78% being of European descent.^[28] MibioGen comprises data on 211 taxa, with 196 categorized into 9 phyla, 16 classes, 20 orders, 32 families, and 119 genera, after excluding 15 unidentified families and genera.

The FinnGen study, a large-scale genomics project involving over 5,00,000 Finnish biobank samples, investigated genetic data on ARDS by correlating genetic variations with health information to examine disease mechanisms and predispositions. The project involves collaboration among research organizations, biobanks, and International industry partners.^[29] GWAS results from FinnGen round 12 were released on November 4, 2024, including 5,00,348 individuals (2,82,064 females and 2,18,284 males), with 2,13,11,644 variants and 2502 disease phenotypes. ARDS patients were identified using ICD10 (International classification of diseases) and ICD9 (International classification of diseases) diagnosis codes, and 4,93,301 controls and 431 ARDS patients were selected for the analysis (Table 1).

Table 1.

Description of genome-wide association study data related to study.

Trait	Consortium	Year	Ancestry	Sample size	Number of taxas	Data availability	PMID
Gut microbiome	MiBiGen	2021	European	18,340 participants	196	MiBiGen (https://mibiogen.gcc.rug.nl/)	33462485
	Germany	2021	German	8956 participants	233	NHGRI-EBI GWAS catalog (https://www.ebi.ac.uk/gwas), from accession GCST90011301to GCST90011730	33462482
	Dutch	2022	Dutch	7738 participants	207	NHGRI-EBI GWAS Catalog (https://www.ebi.ac.uk/gwas/) from accession GCST90027446 to GCST90027585	35879415
ARDS	FinnGen	2024	European	4,93,732 participants (431 cases, 4,93,301 controls)	-	FinnGen (https://r12.risteys.finngen.fi/) for J10_ARDS	36653562

ARDS = acute respiratory distress syndrome, GWAS = genome-wide association study, SNP = single nucleotide polymorphisms.

The risk of sample overlap is minimal since the GWAS summary data originate from populations of European ancestry. Ethical approval was obtained for all original studies, and as the datasets used were publicly available, further ethical approval was not required for this analysis.

2.2. Instrument variables selection

To ensure the validity and precision of the causal relationships between GM and ARDS, we applied quality control procedures to select suitable SNPs as IVs. To eliminate confounding SNPs in the exposure, we excluded SNPs associated with potential confounders. The remaining SNPs were used as Ivs for MR analysis.

We selected SNPs associated with the exposure at a threshold of P < 5 × 10⁻⁵ due to the scarcity of SNPs at genome-wide significance.^[30,31] This relaxed threshold balances the number of available IVs and their credibility, with sensitivity analyses conducted to ensure robust results.

The selected SNPs’ linkage disequilibrium (LD) was evaluated using a clumping procedure, referencing European data from the 1000 genomes project. A strict LD threshold of R² <0.01 was enforced within a 10,000 kb window. Proxy SNPs (r² > 0.8) were used when a SNP was unavailable in the outcome dataset, utilizing the LDproxy tool (https://ldlink.nci.nih.gov/).

Palindromic SNPs, such as G/C or A/T alleles, were left out to prevent complications with strand orientation or allele coding. Each SNP was examined for prior associations with confounders using the LDtrait tool to prevent potential confounding (https://ldlink.nih.gov/?tab=home).^[32]

The F-statistic, indicating instrument strength, was determined using the formula $F={\beta }_i^2/{\mathrm{S}\mathrm{E}}_i^2$, where β_i represents the effect size on exposure and SE_i denotes the standard error. A higher F-statistic indicates a stronger instrument. A value exceeding 20 indicates a low risk of weak instrument bias.^[33,34]

We utilized Mendelian randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) and MR-Egger regression tests to assess possible horizontal pleiotropy.^[35,36] The MR-PRESSO global test yielded a P-value indicating overall horizontal pleiotropy, and the MR-PRESSO outlier test detected pleiotropic SNPs. SNPs with significant outlier P-values were removed, and the analysis was repeated until the global test P-value exceeded .05, indicating non-significance.

The power of the study was estimated using the Burgess method, which considers the sample size, case-control ratio, and genetic instrument variance explained for exposure.^[37,38] Statistical power was assessed using an online tool, with a type I error rate of 0.05% (https://sb452.shinyapps.io/power/).

2.3. MR analysis

We performed a MR study to evaluate the causal link between GM and ARDS. The Wald ratio test was utiluzed to features involving a single independent variable.^[39] For features with multiple IVs, we utiluzed 4 methods: weighted mode, MR-Egger regression, inverse-variance weighted (IVW), and weighted median estimator test.^[36,40–43] The main analysis utilized the IVW method, with 3 additional methods providing complementary tests, given IVW’s greater power under specific conditions.^[39] Causal relationships were considered suggestive if P-values were below nominal significance (P < .05).^[44]

2.4. Sensitivity analysis

We applied the Bonferroni correction to adjust significance levels for multiple tests, taking into account the number of taxa at each classification tier.^[45] In the MibioGen dataset, the significance thresholds for multiple tests are set at 5.6 × 10⁻³ for phylum, 3.1 × 10⁻³ for class, 2.5 × 10⁻³ for order, 1.6 × 10⁻³ for family, and 4.2 × 10⁻⁴ for genus. In the Dutch Microbiome Project, thresholds are 1 × 10⁻² for phylum, 5.0 × 10⁻³ for class, 3.8 × 10⁻³ for order, 1.9 × 10⁻³ for family, 1.0 × 10⁻³ for genus, and 4.8 × 10⁻⁴ for species. In the German consortium dataset, thresholds are 1.3 × 10⁻² for phylum, 6.3 × 10⁻³ for class, 8.3 × 10⁻³ for order, 5.0 × 10⁻³ for family, 1.7 × 10⁻³ for genus, 7.7 × 10⁻⁴ for 97% OUT, 8.1 × 10⁻⁴ for 99% OUT, and 1.0 × 10⁻³ for ASV.

Heterogeneity was assessed using Cochran’s Q statistic. A Q value with P < .05 suggests study heterogeneity.^[42]

A leave-one-out analysis was conducted to determine if any single SNP accounted for the causal signal by evaluating the variance explained in both the exposure and outcome. The causal relationship was considered reliable if the independent variables explained more variance in the exposure than in the outcome.^[46]

2.5. Meta-analyses

Meta-analyses were conducted to enhance the statistical power of the MR analysis.^[47] To investigate the genetic relationships among all consortia, a meta-analysis was conducted for the GM taxa for which the MR IVW P-value was <.05. At least 2 consortia reporting nonoverlapping data were needed for a GM taxon to be examined and reported in our meta-analyses. The association between GM taxa and ARDS was assessed using the odds ratio and 95% confidence interval. Meta-analyses were performed utilizing the R package meta (v7.0-0).^[48] Heterogeneity was assessed using Cochran’s Q and I² statistics. A fixed-effects model was utilized when heterogeneity was not significant (P > .10 and I² < 50%).^[49,50]

Statistical analyses were conducted using RStudio version 4.4.0. The R packages utilized were TwoSampleMR (v0.6.3), MR (v0.8.0), and MR-PRESSO (v1.0).^[35,51] The analysis was executed in June 2024.

3. Results

3.1. Exclusion of GM taxa

To eliminate confounding SNPs in the exposure, we excluded 91 SNPs associated with potential confounders (e.g., other factors related to ARDS development). The remaining SNPs were used as IVs for MR analysis (Table S2, Supplemental Digital Content, https://links.lww.com/MD/Q478). We excluded GM taxa with fewer than 3 Ivs from our primary analysis. These included the genus Blautia, genus Lachnospira, genus Erysipelotrichaceae UCG003, and family Christensenellaceae, all of which were part of the MiBioGen consortium. The Dutch consortium data excluded genus Desulfovibrio, genus Holdemania and species Eggerthella unclassified, species Lachnospiraceae bacterium 1_1_57FAA, species Dorea formicigenerans, species Streptococcus salivarius. Causal associations were then analyzed between 506 known GM taxa from the Dutch, German, and MiBioGen consortia. The taxa linked to ARDS included 9 phyla, 16 classes, 20 orders, 41 families, 144 genera, 100 species, and 65, 62, and 49 taxa identified at the 97% OTU, 99% OTU, and ASV levels, respectively (Fig. 2).

Figure 2.

Flow chart for gut microbiota taxa based on the 3 international consortia. ASV = amplicon sequence variant, IVs = instrumental variables, OUT = operational taxonomic unit.

3.2. Causal effects of GM on ARDS

We identified 2083 IVs related to 192 GM taxa from the MiBioGen dataset. This was done through LD analysis, removing IVs directly linked to ARDS risk factors and excluding palindromic SNPs (Table S3, Supplemental Digital Content, https://links.lww.com/MD/Q478). The uncorrected IVW analysis indicated that Phylum Actinobacteria, order Bifidobacteriales, family Bifidobacteriaceae, and genus Dorea might be associated with an decreased risk of ARDS (Fig. 3 and Table S4, Supplemental Digital Content, https://links.lww.com/MD/Q478).

Figure 3.

Causal analysis of gut microbiota on acute respiratory distress syndrome based on Mendelian randomization analyses. (A) The causal effect (beta) of the results with significant inverse-variance weighting P-value in either consortium. The symbol “*” indicates the significance of the MR methods (*, P < .05; **, P < .01). The gradient brown–yellow indicates positive association or causal effect, while the gradient blue indicates a negative association or causal effect. (B) From top to bottom, the results of inverse-variance weighted correspond to the MibioGen consortium, the Dutch consortium, and the German consortium. The gut microbial taxa with causal relationships in the MibioGen consortium are represented by the red CI line; those in the Dutch consortium are represented by the blue CI line; and those in the German consortium are represented by the orange CI line. ARDS = acute respiratory distress syndrome, CI = confidence interval, MR = Mendelian randomization, nSNP = number of single nucleotide polymorphisms, OR = odds ratio.

Analyzing GM data from the Dutch consortium, we identified 1785 Ivs linked to 200 GM taxa after conducting LD analysis and excluding palindromic SNPs and IVs associated with ARDS risk factors (Table S3, Supplemental Digital Content, https://links.lww.com/MD/Q478). Uncorrected IVW analysis suggested that a lower risk of ARDS may be suggestively linked to phylum Proteobacteria, genus Streptococcus, species B wadsworthia, and species E unclassified, while an increased risk may be linked to species B longum (Fig. 3 and Table S4, Supplemental Digital Content, https://links.lww.com/MD/Q478).

Using data from the German consortium, we identified 2662 IVs linked to 232 different GM taxa, following similar LD analysis procedures (Table S3, Supplemental Digital Content, https://links.lww.com/MD/Q478). The uncorrected IVW analysis suggests that family Rikenellaceae, OTU99_17 (Parabacteroides), TestASV_7 (Bacteroides), TestASV_26 (Phascolarctobacterium), and TestASV_43 (Parasutterella) may be associated with a lower risk of ARDS, whereas OTU99_30 (Parasutterella), and TestASV_16 (Bacteroides) may be associated with a higher risk (Fig. 3 and Table S4, Supplemental Digital Content, https://links.lww.com/MD/Q478).

3.3. Meta-analysis results

We identified 7 gut microbial taxa for which MR results were present in at least 2 consortia: phylum Actinobacteria, phylum Proteobacteria, order Bifidobacteriales, family Bifidobacteriaceae, family Rikenellaceae, and genus Streptococcus. The meta-analysis indicates a potential association between a reduced risk of ARDS and the relative abundance of Streptococcus (odds ratio 0.610; 95% confidence interval 0.430–0.870; P = .006), as illustrated in Table S6, Supplemental Digital Content, https://links.lww.com/MD/Q478 and Figure 4.

Figure 4.

Meta-analyses results for causal analysis of gut microbiota on acute respiratory distress syndrome based on Mendelian randomization analyses. Estimates are presented as odds ratios (OR) and 95% confidence interval (CI). ARDS = acute respiratory distress syndrome, CI = confidence interval, OR = odds ratio.

3.4. Power analysis

The MR power analysis in this study indicates that with the selected genetic instruments and sample size, the power reached approximately 80%, adequately detecting moderate to strong causal effects. However, further power improvements could be achieved by increasing the sample size or using more robust genetic instruments (Table S7, Supplemental Digital Content, https://links.lww.com/MD/Q478).

3.5. Sensitivity analysis

MR-Egger, weighted mode, and weighted median methods yielded comparable causal estimates in both magnitude and direction. The MR-Egger regression intercept method indicated no horizontal pleiotropy for GM in ARDS (P > .05). The MR-PRESSO analysis showed no outliers. Furthermore, no significant heterogeneity was observed in the Cochran Q statistics (P > .05; see Tables 2 and S5, Supplemental Digital Content, https://links.lww.com/MD/Q478). The leave-one-out sensitivity analysis indicated that removing any individual SNP did not significantly alter the causal effect (Figs. 5 and S1, Supplemental Digital Content, https://links.lww.com/MD/Q479). In summary, the association was robust according to analyses for sensitivity, heterogeneity, and horizontal pleiotropy (Figs. S2–S4, Supplemental Digital Content, https://links.lww.com/MD/Q479).

Table 2.

Heterogeneity and directional pleiotropy detection.

Exposure source	Outcome source	Exposure	Outcome	Heterogeneity (Cochrane’s Q)		Pleiotropy
				IVW P-value	MR-Egger P-value	Egger intercept	P-value (MR-Egger)	PRESSO RSSobs	P-value RSSobs
MiBioGen	FinnGen	Phylum Actinobacteria	ARDS	.209	.170	−0.058	.648	19.433	.235
MiBioGen	FinnGen	Order Bifidobacteriales	ARDS	.530	.452	0.038	.079	9.762	.557
MiBioGen	FinnGen	Family Bifidobacteriaceae	ARDS	.530	.452	0.038	.648	9.762	.564
MiBioGen	FinnGen	Genus Dorea	ARDS	.122	.132	−0.100	.349	16.533	.188
Dutch	FinnGen	Phylum Proteobacteria	ARDS	.754	.667	−0.017	.862	7.715	.779
Dutch	FinnGen	Genus Streptococcus	ARDS	.422	.344	0.086	.653	10.522	.433
Dutch	FinnGen	Species Bifidobacterium longum	ARDS	.761	.676	0.025	.837	7.717	.784
Dutch	FinnGen	Species Bilophila wadsworthia	ARDS	.459	.362	−0.060	.615	6.018	.549
Dutch	FinnGen	Species Escherichia unclassified	ARDS	.924	.876	0.042	.755	3.374	.934
German	FinnGen	Family Rikenellaceae	ARDS	.659	.782	0.164	.149	11.885	.679
German	FinnGen	OTU99_17 (Parabacteroides)	ARDS	.969	.974	−0.060	.431	4.358	.947
German	FinnGen	OTU99_30 (Parasutterella)	ARDS	.997	.994	−0.019	.776	3.268	.998
German	FinnGen	TestASV_7 (Bacteroides)	ARDS	.857	.798	0.028	.737	6.086	.857
German	FinnGen	TestASV_16 (Bacteroides)	ARDS	.849	.813	−0.038	.622	9.828	.870
German	FinnGen	TestASV_26 (Phascolarctobacterium)	ARDS	.967	.950	0.013	.913	8.960	.973
German	FinnGen	TestASV_43 (Parasutterella)	ARDS	.471	.407	−0.017	.796	18.842	.483

Cochran’s Q-test for heterogeneity detection, the Mendelian randomization (MR)-Egger test and Mendelian randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) global test for directional pleiotropy detection.

ARDS = acute respiratory distress syndrome, IVW = inverse-variance weighting, MR = Mendelian randomization, MR-PRESSO = Mendelian randomization Pleiotropy RESidual Sum and Outlier, RSSobs = observed residual sum of squares.

Figure 5.

Leave-one-out analysis, funnel plot, scatter plot, and forest plot for genus Streptococcus on ARDS. (A) Leave-one-out analysis for genus Streptococcus on ARDS. The error bar represents the 95% confidence interval with the method of IVW. (B) Funnel plot for genus Streptococcus on ARDS. (C) Scatter plot for phylum Actinobacteria on ARDS. (D) Forest plot for genus Streptococcus on ARDS. ARDS = acute respiratory distress syndrome, IVW = inverse variance weighted.

4. Discussion

We used MR analysis to identify a potential causal relationship between ARDS and genetic susceptibility to specific GM taxa using extensive genetic data from more than 4,90,000 Europeans. We identified a suggestive causal link between ARDS and genetic susceptibility to specific GM. The 3 largest GWAS on GM provided robustly linked gene variants. Across different consortia, we identified 16 GM taxa with potential causal associations with ARDS. Six of these GM taxa were present in different consortiums, and 3 out of 6 causal associations were found at the family and genus levels. The meta-analysis suggests that the decreased morbidity of ARDS may be associated with genetic susceptibility to the genus Streptococcus. These results suggest that these GM taxa may serve as potential biomarkers or therapeutic targets for ARDS, providing valuable insights into the genetic relationship between ARDS and GM taxa that may influence public health initiatives aimed at reducing ARDS incidence.

Our study suggests that a higher relative abundance of Streptococcus may serve as a protective factor against ARDS, as inferred from the causal impact of GM on the condition. Currently, over 170 taxa in the Streptococcus genus are known, including described species, synonyms, and unpublished species.^[52] The Streptococcus species are categorized into 7 groups: Pyogenic, Anginosus, Bovis, Mutans, Mitis, Salivarius, and Hyovaginalis.^[53–57] The relationship between Streptococcus and ARDS is not well understood. Although several Streptococcus species, such as S pyogenes, S pneumoniae, S equi, and S agalactiae, are pathogenic, others like S thermophilus are either beneficial or nonpathogenic to humans.^[58,59]

The gut-lung axis plays a synergistic role in maintaining immune and lung physiological function. Research has established the gut-lung axis, highlighting interactions between GM and respiratory diseases.^[60] Dysbiosis in the lungs disrupts immune function and alters the composition of the lung microbiota.^[61–64] Maintaining balance in the gut-lung axis is essential for preserving healthy immune function. Certain GM can influence innate lymphoid cells in the lungs, promoting the secretion of cytokine IL-33 for host defense.^[65] Through ACE2 receptors, the gut-lung axis coordinates immune responses, impacting both the GM and extrapulmonary symptoms in ARDS patients, particularly in COVID-19-related cases.^[66]

An imbalance in the gut-lung axis exacerbates immunological responses and accelerates ARDS progression. Gut dysbiosis impairs the intestinal barrier, increases intestinal permeability, and exacerbates inflammation, which worsens both gut and lung inflammation in ARDS.^[13–15] By correcting gut dysbiosis, we can mitigate lung injury, control immune responses, reduce inflammation, and increase the production of anti-inflammatory factors, thus improving ARDS outcomes.^[16,67]

A thorough exploration of the gut-lung axis regulatory mechanisms in ARDS may unveil new therapeutic strategies and drug targets for disease prevention and management. This would not only facilitate more accurate and effective treatment of ARDS but also reduce the financial burden on both patients and society, ultimately improving patients’ overall quality of life.

In a recent MR study, Ma et al explored the causal impact of ARDS on GM taxa and identified 8 GM taxa potentially linked to ARDS.^[68] Their study exclusively used GM data from the 2021 MiBioGen consortium. In contrast, our analysis incorporated the latest 2022 GM data from the Dutch and German consortia in addition to the MiBioGen dataset. By combining findings from all 3 consortia and conducting meta-analyses, we identified several novel potential causal associations, thereby expanding the current understanding of the gut-lung axis in ARDS. These methodological differences, particularly in data sources and analytical scope, may account for the discrepancies in the identified bacterial taxa between the 2 studies.

This study is the first to utilize MR comprehensively to evaluate the causal relationship between GM and ARDS using 3 human GM GWAS databases. It is the first longitudinal microbiome study preceding ARDS, offering valuable insights into the effects of GM on ARDS in the absence of large-scale clinical randomized controlled trials. Sensitivity analyses using MR-PRESSO and MR-Egger confirmed the results’ robustness, detecting no significant pleiotropic effects. A meta-analysis combining 3 independent consortia further strengthened the conclusions.

Several limitations must be acknowledged. First, factors like sex, lifestyle, medication, and sample collection time can influence GM composition, potentially reducing the genetic variance attributable to microbiota.^[69] Second, we used more relaxed thresholds for IV screening, as few IVs met the stringent P < 5 × 10⁻⁸ criterion. Third, the study’s focus on a population with primarily European ancestry may limit the generalizability of the findings to other ethnic groups. Fourth, despite examining GM across 3 databases, the diversity and heterogeneity of the microbiota may still constrain the findings. Finally, the variability introduced by 16S rRNA sequencing in GWAS summary datasets may impact the validity of our microbiota data results. More advanced methods like metagenomic sequencing could provide a more accurate picture and warrant further large-scale studies.

Our analyses rely on summary-level datasets for GM and ARDS, which may be affected by unmeasured or residual confounding. Potential influences include demographics, lifestyle and nutrition, prior and concurrent medications, preexisting comorbidities, ICU care exposures, environmental factors, and population structure. Methodological constraints such as limited taxonomic resolution, cross-cohort batch effects, weak instruments and horizontal pleiotropy in genetic analyses, and possible phenotype misclassification may further bias effect estimates. Accordingly, the evidence should be viewed as suggestive rather than definitive. Future work that integrates longitudinal clinical covariates, high-resolution multi-omics, and experimental validation is needed to delineate causality more clearly.

Within these constraints, the implicated taxa and pathways may serve as candidate biomarkers for risk stratification and may indicate modifiable therapeutic avenues along the gut-lung axis. Microbially derived metabolites, including short-chain fatty acids, bile acid derivatives, and tryptophan AhR ligands, could modulate alveolar macrophage activation, epithelial and endothelial barrier integrity, and neutrophil recruitment through pattern recognition and inflammasome pathways including TLR, NF-κB, and NLRP3. If confirmed, microbiota-driven host interactions could help identify patients at increased risk of ARDS progression or treatment failure and could inform prevention or adjunctive therapy. These implications remain hypothesis generating and require confirmation in prospective cohorts and mechanistic studies.

Research in the future should aim to uncover the pathways that link GM with ARDS. Large-scale, multicenter, stratified RCTs focusing on high-risk ARDS patients are needed to confirm these findings. Probiotics, known for their immunomodulatory, anti-inflammatory, and antibacterial effects, may offer therapeutic potential for ARDS. In animal studies, they have been shown to strengthen the immune system. Probiotics may also significantly affect immune response modulation, intestinal secretion and motility, as well as exhibit antibacterial and targeted anti-inflammatory properties. Furthermore, they interact with dietary components to produce beneficial metabolic effects. According to evidence from evidence-based medicine, probiotics have demonstrated clear advantages in preventing nosocomial infections and ventilator-associated pneumonia in critically ill patients.^[70,71] Given their potential, probiotics should be carefully evaluated for their ability to alleviate ARDS. Conclusive evidence requires high-quality human clinical RCTs.

5. Conclusion

In summary, our results suggest a potential relationship between specific gut microbial features and ARDS. Streptococcus may be suggestively linked to a decreased risk of ARDS. Given unmeasured confounding and methodological limitations, these findings should be interpreted cautiously and do not establish definitive causality. The implicated taxa and pathways may represent candidate biomarkers or therapeutic targets along the gut-lung axis, pending rigorous validation. Future longitudinal and experimental work is warranted to clarify directionality and clinical utility.

Acknowledgments

We want to acknowledge the participants and investigators of the FinnGen study. We also want to acknowledge the other studies referenced in the present analysis for providing pubilc datasets.

Author contributions

Data curation: Hui Peng, Fangjie Lu, Gaofeng Zhang, Nana Xu, Xunxun Chen, Xikun Gao, Cong Li.

Funding acquisition: Fangjie Lu, Gaofeng Zhang, Xikun Gao.

Investigation: Gaofeng Zhang, Nana Xu, Xikun Gao, Cong Li.

Methodology: Cong Li.

Supervision: XG and CL jointly supervised this work.

Project administration: Cong Li.

Writing – original draft: Hui Peng, Fangjie Lu.

Writing – review & editing: Cong Li.