Analysis of the pulmonary microbiome in ARDS patients using bronchoalveolar lavage fluid metagenomic next-generation sequencing: a retrospective observational study.

Abstract

Background: Acute respiratory distress syndrome (ARDS) exhibits significant clinical heterogeneity, with inflammatory subphenotypes (hypoinflammatory and hyperinflammatory) representing a key axis for precision medicine. The role of the pulmonary microbiome in these subphenotypes remains poorly understood.

Methods: This retrospective study enrolled 159 ARDS patients. Using a validated machine-learning classifier, patients were stratified into hypoinflammatory (n=92) and hyperinflammatory (n=67) groups. Bronchoalveolar lavage fluid (BALF) was analyzed by metagenomic next-generation sequencing (mNGS) and conventional microbiological testing (CMT). Clinical characteristics, pathogen profiles, and pulmonary microbiome composition were compared between groups.

Results: Patients in the hyperinflammatory phenotype had more severe disease, with significantly higher in-hospital mortality (65.7% vs. 30.4%, P < 0.001) and 28-day mortality (53.7% vs. 22.8%, P < 0.001). mNGS demonstrated superior diagnostic performance, identifying pathogens in 18.2% of cases that were negative by conventional microbiological testing (CMT), whereas CMT alone detected pathogens in only 2.5% of mNGS-negative cases. mNGS showed significant advantages in viral detection (74.5% vs. 28.2%, P < 0.001) and in the identification of mixed infections (74.5% vs. 41.1%, P < 0.001). Acinetobacter baumannii was the most prevalent species in both phenotypes; however, the hyperinflammatory phenotype was enriched for Klebsiella pneumoniae, Legionella pneumophila, and influenza A (H1N1), whereas Stenotrophomonas maltophilia and herpesviruses were more prevalent in the hypoinflammatory phenotype. Species richness was significantly reduced in the hyperinflammatory phenotype (Chao1, P < 0.001; ACE, P = 0.001). In adjusted analyses, this association remained virtually unchanged after adjustment for ARDS etiological category and pulmonary vs. extrapulmonary ARDS, and remained significant in the fully adjusted model additionally accounting for age, sex, and immunosuppression (Chao1: P = 0.002; ACE: P = 0.003). Evenness indices (Shannon/Simpson) and overall community structure (β-diversity; PERMANOVA, P = 0.156) did not differ between phenotypes, suggesting a selective depletion of rare, low-abundance taxa rather than a global restructuring of the pulmonary microbiota. Linear discriminant analysis effect size (LEfSe) identified differentially abundant taxa of exploratory significance: the hyperinflammatory phenotype was enriched for Bifidobacterium dentium and Gemella sanguinis, whereas the hypoinflammatory phenotype was enriched for commensals such as Streptococcus mitis. Network analysis revealed well-defined positive and negative correlations between pathogens and commensal taxa.

Conclusions: The hyperinflammatory phenotype of ARDS is characterized by greater clinical severity and a distinct pulmonary microbiome signature. Metagenomic next-generation sequencing (mNGS) substantially outperforms conventional methods for etiological diagnosis. LEfSe analysis identified differentially enriched taxa, with the hyperinflammatory phenotype enriched for microorganisms that typically colonize the oral cavity or gut (e.g., Bifidobacterium dentium), suggesting potential microbial translocation along the oral-lung or gut-lung axis. These findings provide a novel microbiome dimension for the precision subphenotyping of ARDS.