Lung Microbiota is Associated with Bacterial Detection and with Clinical Improvements in Severe Community-Acquired Pneumonia Patients

Background: Few studies have described the key features of the lung microbiota in patients with severe community-acquired pneumonia (SCAP). We conducted the study to identify the association between the lung microbiota on admission and the clinical prognosis in SCAP patients. Methods: The consecutive SCAP patients admitted from intensive care unit (ICU) were enrolled prospectively. 16S rRNA gene sequencing was applied to bronchoalveolar lavage uid (BALF) collected within 48 hours. The clinical information was recorded during the stay of hospitalization. The primary endpoint was the rate of clinical improvements dened as a decrease of 2 categories and above on a 7-category ordinal scale within 14 days following bronchoscopy. Results: Sixty-seven patients were included. Multivariable Permutational multivariate analysis of variance (PERMANOVA) found that bacterial detection had the strongest independent relationship with the lung microbiota (R 2 =0.033; p=0.018), followed by acute kidney injury (AKI R 2 =0.032; p=0.011) and plasma MIP-1beta level (R 2 =0.027; p=0.044). Random forest identied that families Prevotellaceae, Moraxellaceae, Staphylococcaceae were the biomarkers related to bacterial detection. In the patients with positive bacteriology results, the mean relative abundance of families Prevotellaceae and Actinomycetaceae decreased while families Moraxellaceae, Staphylococcaceae and Streptococcaceae increased. Multivariable Cox regression showed that the increase in alpha-diversity and the abundance of families Prevotellaceae and Actinomycetaceae were associated with clinical improvements. Conclusions: The bacterial detection and patients’ intrinsic factors were associated with the lung microbiota. The increased alpha diversity and the enrichment of families Prevotellaceae and Actinomycetaceae in the lung microbiota were associated with clinical improvements.


Background
Patients with severe community acquired pneumonia (SCAP) are usually complicated with hypoxemia, acute kidney injury (AKI), sepsis and required intensive care unit (ICU) care [1,2]. Although the broadspectrum antibiotic and advanced oxygen support are used commonly in the therapeutic regimen, the death in SCAP patients remains as high as 36% [3]. Next generation sequencing (NGS) technology has revealed that the human lung contains complex and dynamic microbiota community [4]. Recently, several studies have suggested that the lung microbiota is associated with the respiratory infection and disease outcomes. The murine model proves that the intact microbiota contributes to the protection against bacterial pathogens with GM-CSF signaling [5]. The sputum microbial composition is found to be associated with length of stay and ICU admission in the children hospitalized for CAP [6]. Besides, the enrichment of speci c taxon in the airway microbiota, such as Enterobacteriaceae, might be related to high plasma in ammatory cytokine level and the development of acute respiratory distress syndrome (ARDS), and be predictive of fewer ventilator-free days in patients with critical illness [7,8]. However, few studies have described the key features of the lung microbiota in patients with SCAP.
A range of factors may contribute to the complexity of the lung microbiota in SCAP patients. The overgrowth of the invasive pneumonia-associated pathogens in patients' lower respiratory tract may cause a decline in the lung microbiota diversity and the microbiota might even be dominated by single species [9]. Based on the new conceptual models of respiratory microbiology proposed by Robert P Dickson, the lung microbiota is determined by three factors: microbial immigration, elimination, and the relative reproduction rates of the members which are affected by the regional growth conditions such as oxygen tension, pH, alveolar ventilation and temperature in the lung [10,11]. Critical illness alters the internal environment and pathophysiology of the respiratory tract of the patients. Laryngeal dysfunction, supine positioning, aspiration or confusion, the ICU management including antibiotic use, mechanical ventilation and vasopressors have impacts on the balance of the immigration and elimination of the airway microorganisms [12]. Collectively, these clinical parameters have the potential to change the lung microbial composition of the SCAP patients.
Therefore, we conducted the study to explore the clinical factors that may be associated with the lung microbiota in patients with SCAP and identify the important taxa which may predict the clinical prognosis.

Study populations
All consecutive severe pneumonia cases admitted from ICU between March 2018 and March 2019 were registered prospectively. The inclusion criteria were (1) age ≥ 18 years old; (2) diagnosed with SCAP according to the 2007 Infectious Disease Society of America/American Thoracic Society guidelines (Table S1) [13]; (3) The time from onset of illness ≤ 7 days. The patients whose time of disease less than 14 days while condition exacerbation within 7 days before admission to ICU were also included ( Fig. 1).
Patients were excluded if they had one of the following criteria: (1) a history of hospitalization within 14 days before illness onset; (2) bronchoscopy couldn't be performed within 48 hours after admission; (3) peripheral blood specimens were not available within 24 hours; (4) pregnancy or breastfeeding; (5) had an alternative diagnosis at the end of the study, including lung cancer, pulmonary tuberculosis or pulmonary embolism.

Sample Collection
The bronchoscopy was performed at bedside within 48 hours after the patients admitted to ICU using the standard clinical protocol. The bronchoscope was inserted through the nose or orotracheal tube and the bronchoalveolar lavage uid (BALF) samples were collected. About 15 ml specimen was immediately sent to the microbiology laboratory for routine bacterial, fungal and viral examinations, and the remaining BALF was stored at -80℃ until further processing. Blood samples were obtained within 24 hours of ICU arrival. They were centrifuged and the plasma stored at -80℃.

Data Collection
The following data were collected using a standard case report form: demographic data, underlying diseases, the time of illness onset, clinical symptoms, laboratory ndings, microbiology results, radiographic data, antimicrobial use, glucocorticoid use, mechanical ventilation use and so on. The included patients were followed up until they were discharged or died.

Cytokine Measurement
The Plasma IL-4, IL-6, IL-8, MIP-1beta, VEGF-A and MMP-9 of the SCAP patients were detected using a magnetic bead-based multiplex immunoassay and read on a Bio-Plex 200 suspension array system (Bio-Rad, Hercules, CA, USA) according to the manufacturer's instructions.

De nition
The primary endpoint was the rate of clinical improvements, which were de ned as a decrease of 2 categories and above on a 7-category ordinal scale within 14 days following bronchoscopy. The ordinal scales which were also used in our in uenza and Covid-19 studies consisted of the following categories: 1, not hospitalized with resumption of normal activities; 2, not hospitalized, but unable to resume normal activities; 3, hospitalization, not requiring supplemental oxygen; 4, hospitalization, requiring supplemental oxygen; 5, hospitalization, requiring nasal high-ow oxygen therapy and/or noninvasive mechanical ventilation; 6, hospitalization, requiring extracorporeal membrane oxygenation and/or invasive mechanical ventilation; 7, death [14][15][16]. Immunocompromised status in our study was de ned as the patients having a history of cancer with neutropenia (absolute neutrophil count < 0.5 × 109/L), hematological malignancies, solid malignancies receiving chemotherapy during the previous 3 months, solid organ or bone-marrow transplant, active graft-versus-host disease, bronchiolitis obliterans, human immunode ciency virus infection, immunoglobulin de ciency, using immunosuppressive agents, or current treatment with systemic corticosteroids (≥ 20 mg of prednisone per day or its equivalent) for > 30 continuous days before illness onset [17]. The presence of ARDS on admission were diagnosed according to the Berlin de nition [18]. For the nonventilated subjects who had a history of acute respiratory failure within 7 days because of a known respiratory events and bilateral pulmonary in ltration on chest x-ray with PaO2/FiO2 below 300 mmHg, ARDS was also considered. Sepsis, septic shock and acute kidney injury were diagnosed based on the third international consensus and Kidney Disease Improving Global Guidelines (KDIGO) clinical practice guidelines, respectively [19,20]. Acute cardiac insu ciency was diagnosed by cardiologists based on clinical vitals, laboratory ndings and echocardiography. Pneumonia severity was assessed by CURB-65, APACHE-II and PSI risk class. The order of performing bronchoscopy is based the time sequence of signing the bronchoscopy informed consent.

Pathogen Detection
All the specimens for the microbiology diagnosis were collected within 48 hours after admission. Viral etiology was considered positive if the respiratory virus was detected in sputum, endotracheal aspirates (Zhijiang, Shanghai, China); (5) bacteria with moderate to heavy growth (> 3 + growth) in quali ed sputum or ETA, or quanti ed culture in BALF of ≥ 104 CFU/mL [17]. The diagnosis of IFD was based on the revision and update of the consensus de nitions of invasive fungal disease [21].

Statistical analysis
The software VSEARCH (version 2.7.1) and USEARCH version (10.0) were used to process the sequencing data. Reads were denoised into Zero-radius Operational Taxonomic Units (ZOTUs) with UNOISE3. After removal of ZOTUs identi ed as contaminants with decontam package or observed in the controls (Table  S2) and whose relative abundance was less than 0.01%, a total of 490 ZOTUs were analyzed. Statistical analysis was performed in R version 3.6.2 via the Rstudio interface. PERMANOVA (vegan R-package) based on Bray-Curtis distance was performed to assess the association between the clinical factors and the lung microbiota. A random forest learning approach (randomForest R-package) was used to identify the clinical factors-associated taxon. Wilcoxon rank sum test test and generalized linear models (GLM) were performed to compare the relative abundance of the species. Multivariable-adjusted Cox regression (adjusted for sample season, plasma IL-8 level, CURB-65, APACHEII scores, presence of shock at sampling, oxygen index on admission, creatinine level and microbiology results) was performed to assess the association between lung microbiota and clinical improvements.

Bacteriology Results In uenced Lung Microbial Taxa
Prevotellaceae family played the most important role in distinguishing bacterial detection positive from negative samples, followed by the pathogenic taxa, such as, families Moraxellaceae, Staphylococcaceae and Streptococcaceae (Fig. 3a). In the patients with positive bacteriology results, the mean relative abundance of families Prevotellaceae, Microbacteriaceae and Actinomycetaceae decreased while families Moraxellaceae, Staphylococcaceae and Streptococcaceae increased ( Fig. 3a-b), which were in line with their high isolation rate of typical bacterial culture (Table 2). 16S rRNA gene sequencing was able to produce the same results as culture methods in 61% of patients (Table 2). However, there was no signi cant difference in the relative abundance of family Pseudomonadaceae between the two groups patients (Fig. 3a). The samples with bacteria detected had lower alpha diversity while the difference wasn't signi cant (Fig. 3d). a: The patients had more than 2 days on ventilator were de ned as "Yes" and less than 2 days or those never received mechanical ventilation were de ned as "No". b: Two patients had positive urinary antigen for Streptococcus pneumoniae with negative detection in bronchoalveolar lavage uid (BALF).
c: The relative abundance of the most abundant ZOTU was shown.
The bold font represented that there were different results between bacterial detection and sequencing method.

Lung Microbiota Predicted Clinical Improvements
The lung microbial community composition of the patients on admission wasn't associated with the clinical improvement (PERMANOVA; Bray-Curtis distance; R 2 = 0.011; P = 0.76). Compared with the patients whose richness of baseline lung microbiota was higher than 200, the multivariable-adjusted Hazard Ratio (HR) for the favorable prognosis in the lower Richness group patients was 0.17 and 0.16, respectively (95% CI: 0.04-0.71, p = 0.02; 95% CI: 0.03-0.88, p = 0.04 Fig. 4a, Table 3). The increased Shannon index predicted a faster decrease of 2 categories which was analyzed continuously (adjusted HR 1.54, 95% CI: 1.08-2.20, p = 0.02) or by organizing their value into ranges (adjusted HR 0.08, 95% CI: 0.01-0.51, p = 0.007 Table 3, Fig. 4b, Table 3). Although the bacterial detection wasn't associated with clinical improvements, we found that every 1% increase in the relative abundance of families Prevotellaceae and Actinomycetaceae in the lung microbiota of the patients raised the probability for clinical improvements by 13% and 12% respectively (95% CI: 1.02-1.25, p = 0.02; 95% CI: 1.03-1.21, p = 0.007 Table 3, Fig. 4c-d). However, families Moraxellaceae, Staphylococcaceae and Streptococcaceae were not associated with clinical improvements (Table 3, Fig. 4c-d). The proportions of patients with ARDS, septic shock, CURB-65 ≥ 3, PSI ≥ IV were similar irrespective of the different level of alpha diversity or relative abundance of taxa (Table S4). c: We turned the continuous variables into categorical variables by organizing their value into ranges. Firstly, according to the value of the variable, 67 samples were put in order. Secondly, we divided them into 10 groups evenly. Finally, we combined the neighboring groups that had the similar probability for the event happened.
d: 1% referred to that the probability for clinical improvements of every 1% increase in the relative abundance of taxa.

Discussion
The core ndings of our study were that bacterial detection had the strongest independent relationship with the lung microbiota composition on patients' admission, followed by AKI and plasma MIP-1beta level. The increased alpha diversity and the enrichment of families Prevotellaceae and Actinomycetaceae in the lung microbiota were associated with clinical improvements.
In this study, the positive bacterial detection on admission was found to be the most important factor related to patients' lung microbiota. The bacteria that could be identi ed by culture-based methods might have a higher load in the lower respiratory tract [22]. Therefore, 16S rRNA gene sequencing revealed that the lung microbiota dominated by the corresponding taxa in the patients with positive bacteriology results (such as Acinetobacter baumannii and Staphylococcus aureus). Although the results of 16S rRNA and culture-based methods were consistent in terms of dominant bacteria, the results related to Pseudomonas aeruginosa differed. The relative abundance of family Pseudomonadaceae wasn't signi cantly different between the bacteriology-positive and bacteriology-negative groups despite the high isolation rate (33.33%) of Pseudomonas aeruginosa. P. aeruginosa is one of the main nosocomial pathogens in ICU. The risk factors for P. aeruginosa acquisition consisted of antibiotics pressure, mechanical ventilation use, length of hospitalization, cross-transmission among patients and medical staffs and environmental factors, such as contamination of water tap [23][24][25]. All of our patients had received antimicrobial treatment and more than half were exposed to mechanical ventilation before bronchoscopy. They might have similar opportunity to obtain P. aeruginosa colonization under the ICU status. Although 16S rRNA gene sequencing had a low accuracy for pathogen identi cation, it could capture all the bacterial information, including fastidious bacteria in the respiratory tract, which might explain the different results between culture-dependent and culture-independent analysis in our study [26,27].
After adjusted for the microbiology results and severity of illness, multivariate analysis showed that clinical improvements were associated with increased relative abundance of family Prevotellaceae and Actinomycetaceae. It was found that relative abundance of Prevotella was inversely associated with airway in ammation in cystic brosis patients [28]. An increase in the abundance of Prevotella in nose/throat microbiota was related to lower susceptibility to in uenza A infection [29]. Another set of researches showed that the enrichment of lung microbiota with Prevotella enhanced the level of BALF in ammatory cytokines, and was associated with the development of asthma in children and ARDS in severe patients [7,30,31]. The conclusions appeared to be inconsistent. However, the reduction in abundance of Prevotella was observed at the initial stage of chronic pulmonary in ammation in mice [32]. The causal relationship between Prevotella and the diseases couldn't be clearly inferred from these association studies [33]. Besides, families Prevotellaceae and Actinomycetaceae usually resided in healthy the respiratory tract or oral cavity [30,34,35]. Earlier study reported that the commensal bacteria (commensal Prevotella spp and Actinomyces spp) could induce weak activation of human dendritic cells compared with pathogenic species (Haemophillus spp. and Moraxella spp) [36]. While co-culture experiments observed that Prevotella spp. were able to inhibit Haemophillus in uenzae-induced IL-12p70 in dendritic cells [36]. Conclusively, we thought that Prevotellaceae and Actinomycetaceae served as commensal colonization and might promote the recovery of the lung from infection in our study.
The antibiotic use was also proved to decrease the airway microbial diversity in COPD patients [37].
However, the type of antibiotic use before sampling was not associated with the lung microbiota in our study, consistent with the result of a study in intubation patients [9]. As all of our patients had received antibiotics before bronchoscopy, we couldn't exclude the possibility of an association between the antibiotics and the lung microbiota. Besides, the lung microbiota of our patients wasn't clustered by the presence of ARDS or septic shock at sampling. Unlike the report in two previous published studies, the relative abundance of family Enterobacteriaceae had no association with the severity of pneumonia or clinical outcomes [7,8]

Competing interests
The authors declare that they have no competing interests.  Study ow chart *The patients whose time of disease less than 14 days while condition exacerbation within 7 days before admission to ICU were also included. Abbreviation: SCAP, severe community acquired pneumonia; ICU, intensive care unit; AECOPD, acute exacerbation of chronic obstructive pulmonary disease.