The study protocol was reviewed and approved by the Ethics Review Committee of Jiangmen Central Hospital (No. 2019-15). Written informed consent was obtained from the patients or their legal representatives before the collection of bronchoalveolar lavage fluid (BALF) samples by bronchoalveolar lavage using fiberoptic bronchoscopy.

A retrospective analysis was conducted on patients with ARDS caused by sepsis who were admitted to the intensive care unit (ICU) of Jiangmen Central Hospital from January 2018 to June 2021. The inclusion criteria were as follows: 1) the diagnosis of ARDS met the Berlin 2012 definition (Force et al., 2012), 2) the etiology of ARDS was sepsis, 3) the age was greater than 18 years, and 4) the clinical profile was complete. The exclusion criteria were as follows: 1) ARDS caused by non-infectious factors, 2) age <18 years, and 3) an incomplete clinical profile. The patients were divided into two groups according to their initial infection status: intrapulmonary infection-induced ARDS (ARDSp group) and extrapulmonary infection-induced ARDS (ARDSexp group). The ARDSp group included 111 patients, the ARDSexp group included 45 patients, and the control group included 28 patients, including patients with mild pulmonary infection and non-ARDS, all of whom had a good prognosis and did not return to the ICU within 90 days of being transferred out of the ICU.

All patients with sepsis were treated according to the sepsis guidelines (Rhodes et al., 2017) and empirical anti-infective therapy in conjunction with clinical indicators of infection and imaging information. Patients with ARDS were mechanically ventilated according to the ARDS ventilation guidelines (Bein et al., 2016; Griffiths et al., 2019), and the anti-infective regimen was adjusted according to the patient’s inflammatory indicators, imaging data, and microbiological tests.

All patients were intubated and mechanically ventilated in the ICU, and BALF specimens were obtained using a fiberoptic bronchoscope (Chinese Thoracic Society, 2017). Baseline specimens were collected within 24 h of ARDS diagnosis in the ICU before antibiotic administration. Some patients were treated for 7 days, and post-treatment specimens were retained. The baseline and post-treatment specimens were sent to the laboratory for pathogenic culture. The remaining specimens were also sent to the clinical experimental center for DNA extraction and stored at -20°C for research purposes. All laboratory consumables were purchased from Guangzhou Jet Bio-Filtration Co., Ltd. (China). The final specimens were sent to Guangdong Longsee Biomedical Co., Ltd. for metagenomic sequencing, including Bacterial nucleic acid purification, DNA library preparation, high-throughput sequencing, bioinformatics analysis, and pathogenic data interpretation (Miao et al., 2018).

As shown in Figure 1, patients in all groups were stratified according to their prognosis. Those who improved in the ICU with treatment and ventilation and could be successfully transferred out of the ICU within 7 days were defined as the “survival group,” while those who did not benefit from ICU care and died of organ failure were defined as the “dead group.” Based on different etiologies with different prognoses, 57 cases were classified as the ARDSp-survival group, 54 cases as the ARDSp-dead group, 20 cases as the ARDSexp-survival group, and 25 cases as the ARDSexp-dead group. We analyzed the similarities and differences in the microbiome between the different groups and searched for microbial markers associated with prognosis using metagenomic DNA sequencing of the BALF collected from these patients.

Thirty patients with complete data in the ARDSp group, including pre-treatment and post-treatment data, were selected and divided into the pre-treatment group (ARDSp-preT group) and post-treatment group (ARDSp-poT group). The ARDSp-poT-survival (post-treatment) group was defined as the post-treatment-survival group, with 15 cases, based on the improvement after ICU care allowing the cessation of ventilator use and survival within 7 days of transfer out of ICU transfer. Conversely, the ARDSp-poT-dead group was defined as a failure of treatment in the ICU and death due to organ failure (15 cases). The control group included 28 cases. The similarities and differences in the microbiome between the different groups were also analyzed.

In total, 2,728 microorganisms were sequenced in this study. The MetaPhlAn database was used for the taxonomic assignment (Truong et al., 2015). As the RPM values of different microorganisms varied considerably, the microorganisms were divided into pathogenic and background microorganisms and were analyzed separately. Pathogenic microorganisms had significantly higher RPM values than the background microorganisms. We extracted common nosocomial infection pathogenic microorganisms, including common bacteria, fungi, viruses, and specific pathogens, based on 2019 CHINET surveillance data (Hu et al., 2020) and common pathogenic microorganisms in sequencing laboratory testing, and defined 57 microorganisms as pathogenic microorganisms (Supplementary Table S1). The remaining microorganisms were considered background microorganisms, with a total of 1,040 at the genus level.

We divided the microorganisms into pathogenic and background microorganisms, and counted the number of positive and negative sequencing results for each group separately. Among pathogenic microorganisms, those with RPM values ≥1 were defined as positive and those with RPM values <1 as negative; among background microorganisms, those with RPM values >0 were defined as positive and those with RPM values = 0 as negative.

To analyze the positive rate of the microbiota, we used the chi-square test or Fisher’s test and presented the data using a histogram. To analyze the abundance of the microbiota, we log2-processed the RPM values from the sequencing results, compared the median values using a paired t-test, and presented the data using a heatmap. Analyses were performed using GraphPad 9.3 or R4.1.2. Statistical significance was set as p < 0.05. Cox and cluster analyses were performed using the SPSS software (version 26.0) and multiple experiment viewer, respectively. Principal component analysis (PCA) was performed using GraphPad prism 9.3 and the Adonis analysis was performed on the Omicshare platform. The Shannon–Wiener index was calculated as follows: H = ∑ ( p i ) ( ln ⁡ p i ) .

A total of 156 patients with sepsis-induced ARDS were selected for this study and were divided into two groups according to their initial infection status: intrapulmonary infection-induced ARDS (ARDSp, n = 111), and extrapulmonary infection-induced ARDS groups (ARDSexp, n = 45), and a control group (n = 28, Figure 1). No significant statistical difference in age and sex was found between the ARDS and control groups (Table 1). The ARDS group was divided into the ARDSp group and the ARDSexp group for comparison, but there was no significant statistical difference in the basic characteristics of the two groups. (Table 2). Pre-treatment laboratory tests showed that the ARDSexp group had significantly different results compared with the ARDSp group, including PCT (p < 0.001), PLT (p = 0.011), Scr (p = 0.041), T. Bil (p = 0.014), Lac (p < 0.001), pre-treatment APACHE II (p < 0.001), and the SOFA score (p < 0.001; Table 3). In addition, a higher number of patients in the ARDSexp group required treatment with vasoactive drugs (p < 0.001) and CRRT (p < 0.001) than in the ARDSp group, with no statistical difference in the 90-day all-cause mortality (Table 4). These findings suggest that patients in the ARDSexp group were sicker than those in the ARDSp group; however, there was no significant difference in the prognosis between the two groups.

ARDS may be caused by several pathogenic microorganisms, including bacteria, fungi, viruses, and specific pathogens. We first compared the composition and abundance of pathogenic microorganisms among the control, ARDSp, and ARDSexp groups. Shannon’s diversity index results suggested that the ARDSp group had less microbiome diversity compared with the ARDSexp and control groups (Supplementary Figure 7A). Further analysis revealed that the ARDSp group had a higher positive rate for Escherichia coli, Staphylococcus haemolyticus, and fungi than the control group (Figure 2A); the ARDSexp group had a higher positive rate for Escherichia coli than both groups (Figure 2B). No statistically significant difference in the pathogenic microorganisms was found between the ARDSp and ARDSexp groups (Supplementary Figure 1A).

Furthermore, we performed PCA on background microorganisms among the control, ARDSp, and ARDSexp groups. As shown in the graph (Figure 2C), the control group had the most concentrated community composition, whereas both the ARDSp and ARDSexp groups had a larger number of different microorganism species than the control group. This indicates that the community composition of the background microorganisms was relatively consistent in the control group, and there were a few abnormally increased background microorganisms in the ARDSp and ARDSexp groups. In a subsequent comparison between the ARDSp, ARDSexp, and control groups, we found that the ARDSp group was characterized by a reduction in both positivity and abundance compared with the control group (Figure 2D; Supplementary Figure 1B), suggesting that the ARDSp microbiome was characterized by a reduction in diversity. Therefore, we speculate that an increase in pathogenic microorganisms inhibits the growth of the “normal” respiratory microbiome. Interestingly, increased positivity and an abundance of intestinal microbes such as Shigella in ARDSp were observed. In contrast, the ARDSexp group presented a predominant increase in the positivity of microbes including Comamonas, Klebsiella, Staphylococcus, Ruminococcus, Tyzzerella, Odoribacter, Lachnospiraceae, and Erwinia, implying that the ARDSexp microbiome was characterized by an increased diversity of the microbiome (Figure 2E; Supplementary Figure 1C), most of which are of intestinal origin. In addition, the ARDSexp group had a higher positive rate and abundance of microbes than the ARDSp group such as Cutibacterium (Figure 2F; Supplementary Figure 1D), which were also mainly of intestinal origin.

We divided the ARDSp group into ARDSp-survival and ARDSp-dead groups according to different prognoses and performed PCA together with the control group (Figure 3A). PCA results showed that there was a relatively large amount of taxonomic overlap between the control and ARDSp-survival groups, with more obviously isolated specimens in the ARDSp-dead group. This suggests that the control and ARDSp-survival groups had a relatively similar microbial community composition, whereas the ARDSp-dead group had several abnormally increased background microorganisms.

The pathogenic microorganisms and background microbiome that increased or decreased simultaneously in the ARDSp-dead group compared with the control and ARDSp-survival groups are probably related to prognosis. Therefore, we compared the composition and abundance of microorganisms among these three groups. The Shannon index result suggested that the ARDSp group had less diversity than the control group; however, there was no statistically significant difference in the microbial diversity between the ARDSp-survival and ARDSp-dead groups (Supplementary Figure 2B). In addition, the ARDSp-dead group had a significantly higher positive rate for pathogenic microorganisms including total pathogens, total fungi, Staphylococcus haemolyticus, and Escherichia coli than the control group (Supplementary Figure 2B). Compared with the ARDSp-survival group, the ARDSp-dead group had a significantly higher positive rate for Pseudomonas aeruginosa (Supplementary Figure 2C).

Moreover, the ARDSp-survival and ARDSp-dead groups showed a predominant decrease in the positivity and abundance of background microorganisms compared with the control group (Figures 3B–D; Supplementary Figure 2E,F). Nevertheless, no prognosis-related microbes were identified among pathogenic or background microorganisms (Figures 3E,F).

We divided the ARDSexp group into ARDSexp-survival and ARDSexp-dead groups according to different prognoses and performed PCA together with the control group (Figure 4A). The PCA results revealed that there was a relatively large taxonomic overlap between the control and ARDSexp-survival groups, whereas the ARDSexp-dead group had more distinct isolated specimens. This suggests that the control and ARDSexp-survival groups had a similar microbial community composition, whereas the ARDSexp-dead group had several abnormally increased background microorganisms.

Therefore, we compared the differences in the composition and abundance of microorganisms among the three groups. The Shannon index result indicated a similar diversity between the ARDSexp-survival, ARDSexp-dead, and control groups (Supplementary Figure 7C). The ARDSexp-dead group had significantly higher positive rates for Escherichia coli and Haemophilus influenzae than the control and ARDSexp-survival groups, respectively (Supplementary Figure 3A,B). However, no pathogenic microbial markers that might be associated with prognosis were identified.

In addition, the ARDSexp-survival and ARDSexp-dead groups showed a predominant increase in the positivity and abundance of background microorganisms compared with the control group (Figures 4B–D and Supplementary Figure 3E,F). In contrast, the ARDSp group showed a decrease in these microorganisms. Notably, there was a simultaneous increase in the microorganisms of the ARDSexp-dead group including Bifidobacterium, Bilophila, Mediterranea, Anaerostipes, Bacillus, Dorea, and Collinsella (Figure 4E), when compared with the ARDSexp-survival and control groups. Further prognostic survival analysis was conducted and the results indicated that these seven microbes were strongly associated with poor prognosis (Figures 5A–G). In contrast, the Cox univariate analysis showed that increased pre-treatment APACHE II and SOFA scores and increased Bifidobacterium, Bilophila, Mediterranea, Bacillus, Dorea, and Collinsella may be risk factors for ARDS (Figure 5H). Furthermore, the Cox multivariate analysis also indicated that the increase in Bilophila was most likely associated with mortality in patients with ARDS (Figure 5I).

To explore the changes in the lung microbiome before and after treatment in the ARDSp group, 30 cases with complete pre-and post-treatment comparative data were selected for analysis and screened for pathogenic organisms associated with death (potential risk factors for death) and markers associated with survival (potential protective factors for survival). Alpha diversity analysis showed that the ARDSp-preT group had lesser microbial diversity than the control group. There was a minor but not statistically significant increase in the microbial diversity after treatment compared with pre-treatment (Supplementary Figure 7D).

We conducted a cluster analysis to determine differences in pathogenic microorganisms and found that Escherichia coli, Staphylococcus aureus, and Candida albicans were significantly increased in the ARDSp-poT-dead group, whereas Acinetobacter baumannii and Acinetobacter nosocomialis were increased or unchanged, mainly in the ARDSp-poT-survival group (Figure 6A). An analysis of the microbial composition and abundance revealed increasing trends in positivity (p = 0.014, p = 0.080, and p = 0.002, respectively) and abundance (p = 0.524, p = 0.015, and p = 0.001, respectively) for Escherichia coli, Staphylococcus aureus, and Candida albicans in the ARDSp-poT-dead group (Figures 6B,D). The microorganisms that showed a significant increasing trend in positivity in the ARDSp-poT-dead group included Klebsiella pneumoniae, Escherichia coli, and Candida albicans (Supplementary Figure 4A). There was also a decreasing trend in the positivity and abundance of Acinetobacter baumannii and Acinetobacter nosocomialis; however, the result was not statistically significant (Figures 6C,E).

In addition, the PCA showed that the ARDSp-poT-survival and control groups were relatively concentrated in background microorganism composition, and the ARDSp-poT-dead and ARDSp-preT groups had significantly distinct background microorganisms. These results indicated that the microbial community composition of the control and ARDSp-poT-survival groups were relatively similar, whereas the ARDSp-poT-dead and ARDSp-preT groups had several unusual background microorganisms (Figure 7A).

We speculated that the microorganisms that were increased in the ARDSp-poT-survival group compared with the ARDSp-preT group (Supplementary Figure 6E), especially those decreased in the ARDSp-poT-dead group compared with the ARDSp-poT-survival group and the control group (Figure 7B; Supplementary Figure 6C), were potential ARDSp protective factors. Notably, the most common microorganism was Hydrobacter (Figure 7D); Hydrobacter had the highest positivity or abundance values in the control group and decreased as prognosis deteriorated or increased as prognosis improved (Figures 7E,F). In contrast, we speculated that the microorganisms that were increased in the ARDSp-poT-dead group compared with the ARDSp-preT group (Supplementary Figure 6A), in the ARDSp-poT-dead group compared with the ARDSp-poT-survival group (Figure 7B), and an increase of in the ARDSp-poT-dead group compared with the control group (Supplementary Figure 6C) were potential risk factors for death in patients with ARDSp. The common microorganisms identified simultaneously were Cryptococcus, Escherichia, and Lachnospiraceae (Figure 7C), which showed the least difference in either positivity or abundance in the control group. These three microorganisms increased as the prognosis worsened or decreased as the prognosis improved (Supplementary Figure 4D-I).