Metagenomic next-generation sequencing of bronchoalveolar lavage fluid from children with severe pneumonia in pediatric intensive care unit

Pneumonia remains a leading cause of infectious mortality globally. Accurate detection and differentiation of true pathogens from commensals are essential to guide therapy. BALF is a suitable sample for pathogen detection in respiratory infections, and metagenomic next-generation sequencing enables culture-independent, sensitive pathogen detection and comprehensive microbial profiling. Prior adult studies demonstrated mNGS utility on BALF for identifying pathogens missed by conventional methods and influencing management. Pediatric data are limited but suggest mNGS can detect candidate pathogens and support surveillance. This study asked: What is the microbiome composition of BALF in children with severe pneumonia in the PICU, and how do microbial features correlate with clinical indicators (inflammation, immune cell subsets, severity, outcomes)? The purpose was to use mNGS to explore potential pathogens and microecological features associated with disease severity, immunodeficiency, sepsis, and mortality in this population.

Previous work has established BALF as an informative matrix for respiratory pathogen detection. mNGS protocols for BALF (e.g., sequential lysis or mechanical disruption) can recover bacterial and viral nucleic acids and have diagnosed unexpected infections, altering management in adults. Studies have shown higher sensitivity of mNGS versus culture or smear/PCR, and that BALF mNGS sensitivity may exceed that of transbronchial biopsy while specificity may differ. Comparisons between tracheal aspirates and mini-BALF in severe pneumonia showed similarities in bacterial pneumonia. Pediatric reports (e.g., mNGS on 10 children with respiratory failure) detected significant bacterial/viral reads in most cases and identified pathogens missed by conventional testing, including complete viral genomes for surveillance. Viral reactivation in ICUs (e.g., Herpesviridae) has been linked to worse outcomes. Fungal respiratory pathogens (Aspergillus, Pneumocystis) contribute to ICU morbidity/mortality and are prominent in coinfections during severe influenza. Protozoan coinfections can modulate respiratory infection risk and inflammation, though detection in BALF is rare and typically requires targeted validation. These data support comprehensive, unbiased mNGS of BALF to capture bacterial, viral, fungal, and protozoan signatures and link them to clinical phenotypes.

Design and setting: Retrospective study at the Pediatric Intensive Care Unit of the Children’s Hospital of Fudan University (Shanghai, China) from February 2018 to February 2020. Participants: Children meeting WHO-based criteria for severe pneumonia. Primary criteria included invasive mechanical ventilation, fluid-refractory shock, urgent need for noninvasive positive-pressure ventilation, hypoxemia requiring FiO2 above general care limits. Secondary criteria included tachypnea, PaO2/FiO2 <250, multilobar infiltrates, PEWS >6, altered mental status, hypotension, effusion, comorbidities (immunosuppression/deficiency), unexplained metabolic acidosis. Exclusions: age <28 days or >18 years, noninfectious pulmonary conditions (e.g., congenital heart disease, pulmonary edema, asthma, upper airway obstruction, cystic fibrosis), bronchoscopy contraindications, severe cardiopulmonary dysfunction, coagulopathy. BALF collection: Under sedation and topical anesthesia, pediatric flexible fiberoptic bronchoscopy targeted the most affected lobe (or right middle lobe). Warm saline (1 mL/kg, max 20 mL per fraction) was instilled; at least 40% recovery by suction at ~50–100 mmHg. Clinical laboratory testing: Venous blood (2 mL) collected for serum LPS, PCT, CRP, IL-6 (details in Supplementary Appendix). Lymphocyte subsets (CD3+, CD4+, CD8+, CD19+, CD16/CD56+) analyzed by flow cytometry using 6-color BD Multitest reagents. Conventional microbiology: BALF cultured on blood, chocolate, and MacConkey agar (5–10% CO2, 35°C, 24–48 h); fungi at 28°C and 35°C for 5–7 days. Suspect colonies identified by MALDI-TOF MS. Drug susceptibility testing performed per local standards; optochin testing for Streptococcus pneumoniae; quality control with ATCC strains at Shanghai Clinical Laboratory Centre. Common respiratory viruses (RSV, adenovirus, influenza A/B, parainfluenza, hMPV) detected by immunofluorescence per manufacturer. mNGS laboratory workflow: mNGS performed on 126 BALF samples: predominantly DNA-level (101/126, 80.1%) and a subset RNA-level (25/126, 19.8%) guided by clinical suspicion, following previously published protocols (Xu 2020; Wang 2021; Yan 2021; Zhou 2016, 2019). Each sample generated on average 2.6 ± 1.4 × 10^7 clean reads. Bioinformatics and ecological metrics: Reads normalized to reads per million (rpm). Microbial ecological diversity computed per sample: total abundance (sum rpm), number of species, alpha diversity (Chao1, Shannon) using R (vegan package). Pathogen determination followed prior criteria (Yan 2021) integrating relative abundance and outlier z-scores across the cohort. Correlation analyses used Pearson/Spearman via cor.test(), linear modeling via lm(), visualization via ggplot2; significance by Wilcoxon rank-sum tests. Some variables capped for visualization (e.g., rpm <0.01 set to log10 −2; z-score >6 capped at 6). Clinical correlations: Microbial features were correlated with serum inflammatory markers (LPS, CRP, PCT, IL-6), leukocyte and lymphocyte subsets (WBC, CD3+, CD4+, CD8+, NK), pneumonia category (CAP vs HAP), sepsis, immunodeficiency, oxygenation index (OI; severity), and outcomes (death/abandoned treatment).

Cohort and baseline: 126 PICU children (84 CAP, 42 HAP). Mean peak temperature 39.0°C ± 0.95; heart rate 160 ± 19 bpm; hospital stay 37 ± 29 days; PICU stay 30 ± 29 days. Thirty-day analysis: 24 deaths and 16 treatment abandonments (29.4%). Comorbid/complications: respiratory failure 86.5% (109/126), severe pneumonia 24.6% (31/126), sepsis 24.6% (31/126), septic shock 5.5% (7/126). Mean serum CRP 67 ± 56 mg/L; PCT 8.9 ± 19 ng/mL; LPS 1.0 ± 7.6 pg/mL; IL-6 390 ± 850 pg/mL. WBC 18 ± 12 ×10^9/L; CD3+ 1.2 ± 1.3 ×10^9/L; CD4+ 6.8 ± 8.7 ×10^8/L; CD8+ 5.0 ± 5.9 ×10^8/L; NK 1.4 ± 1.8 ×10^8/L. Conventional methods detected pathogens in 81 patients. Pathogen detection by mNGS: 101 DNA-level (80.1%), 25 RNA-level (19.8%). Average 2.6 ± 1.4 ×10^7 clean reads/sample. Detected nucleic acids from 88 potentially pathogenic bacteria (notably multiple Streptococcus spp., Staphylococcus spp., and Acinetobacter spp.), 13 viruses (including EBV, HCMV, HHV-6B, human mastadenovirus B, human polyomavirus 4), 39 fungi (including Aspergillus spp., Pneumocystis jirovecii), and 27 protozoans (including Plasmodium spp., Trypanosoma spp.). Correlations with inflammatory markers: Higher BALF bacterial diversity/abundance positively correlated with serum LPS; bacterial species count positively correlated with PCT. Specific associations: Pseudomonas aeruginosa abundance positively correlated with LPS, CRP, PCT; Staphylococcus aureus, S. hominis, S. saprophyticus, Haemophilus influenzae, P. aeruginosa, Mycoplasma salivarium positively correlated with PCT. Acinetobacter baumannii and A. nosocomialis positively correlated with IL-6. Several Streptococcus spp. (e.g., S. pseudopneumoniae, S. agalactiae, S. oralis, S. mitis) negatively correlated with IL-6. Correlations with leukocytes/lymphocytes: Total WBC negatively correlated with BALF abundance of several bacteria (e.g., P. aeruginosa, Achromobacter xylosoxidans), herpesviruses (HHV-6B, EBV), and fungi (P. jirovecii, Aspergillus terreus). BALF abundances of S. aureus, S. haemolyticus, S. warneri negatively correlated with blood CD4+, CD8+, and NK cell counts. CAP vs HAP: HAP had higher A. baumannii and Streptococcus mitis, and lower Schaalia odontolytica. HAP also showed higher abundance of potentially pathogenic fungus Diplodia corticola and parasite Plasmopara halstedii. Severity and outcomes: Greater fungal and protozoal richness (Chao1) and diversity (Shannon) in deceased vs non-deceased patients. Escherichia coli abundance higher in deceased vs non-deceased; Klebsiella pneumoniae and Corynebacterium segmentosum higher in abandoned treatment vs others. Pneumocystis jirovecii abundance highest in deceased group. Oxygenation index (OI) positively correlated with abundances of bacteria (E. coli, K. pneumoniae, Streptococcus agalactiae, S. aureus), fungi (Aspergillus fumigatus, Rhizopus microsporus), and viruses (human mastadenovirus B, Torque teno virus 29); overall viral abundance showed a weak positive correlation with OI. Sepsis and immunodeficiency: Patients with sepsis had significantly higher fungal and protozoal richness/diversity and species counts. Immunodeficient patients had higher total number of bacterial species, higher BALF Elizabethkingia anophelis abundance, and higher HCMV (Human betaherpesvirus 5) abundance. Viral activation signals: Herpesviruses (EBV, HHV-6B) abundances were negatively correlated with WBC, and HCMV abundance was higher in immunodeficient patients, suggesting possible reactivation in critically ill children.

This study demonstrates that mNGS of BALF in critically ill pediatric patients can comprehensively characterize microbial communities and identify potential pathogens across bacteria, viruses, fungi, and protozoa. The microbial ecological features correlated with clinical phenotypes: higher bacterial diversity associated with systemic inflammation (LPS, PCT), while specific pathogens (e.g., Pseudomonas aeruginosa, Acinetobacter baumannii) aligned with markers of inflammation and with known epidemiology in PICU and hospital-acquired settings. Differences between CAP and HAP profiles (e.g., elevated A. baumannii in HAP) support clinical relevance. Viral findings suggest that herpesviruses and other viruses may reactivate or contribute to lung disease in immunodeficient hosts, with overall viral burden relating to hypoxemic severity (OI). Fungal and protozoal diversity and specific pathogens (e.g., Pneumocystis jirovecii, Aspergillus fumigatus) were associated with sepsis, higher severity, and mortality, emphasizing the importance of detecting eukaryotic pathogens in this population. These correlations support the utility of mNGS for hypothesis generation and potential clinical guidance, though causality cannot be inferred from correlations and careful interpretation is required given host status, colonization, and ICU exposures.

mNGS of BALF from children with severe pneumonia in the PICU identified potentially pathogenic bacterial infections and revealed meaningful associations between microbial ecology and clinical indicators. Increased bacterial diversity and abundance correlated with serum inflammatory markers; specific bacterial abundances correlated with inflammatory markers and lymphocyte alterations. Potential viral coinfections were linked to immunodeficiency and greater disease severity, suggesting possible viral reactivation. Increased diversity of potentially pathogenic fungi was associated with death and sepsis, and certain fungal and bacterial pathogens correlated with worse oxygenation. These findings support mNGS as a comprehensive diagnostic adjunct for pathogen detection and risk stratification in critically ill children. Future work should include rapid turnaround mNGS integrated into clinical decision-making, multicenter validation, targeted confirmatory testing (including 18S rRNA sequencing for protozoa), and expanded assays capturing both DNA and RNA viruses simultaneously.

- Retrospective single-center design with evolving mNGS turnaround (often several days during 2018–2020), limiting real-time clinical decision-making based on mNGS. - BALF volume constraints precluded measurement of inflammatory markers directly in BALF; blood markers were used instead. - Pathogen calls from mNGS require further clinical microbiology validation (culture, serology/antibody testing); limited confirmatory testing reported. - mNGS workflow did not simultaneously capture DNA and RNA viruses for all samples; only 25/126 were tested at the RNA level based on clinical judgment, risking missed RNA virus detection. - Detection of protozoal nucleic acids was based on limited reads from large genomes; presence requires confirmation (e.g., 18S rRNA sequencing) and may reflect low-level or incidental signals. - Due to limited detection of some specific pathogens, comprehensive multivariable regression for individual taxa was not performed; correlation analyses of diversity indices may underestimate complex relationships. - Generalizability requires multicenter validation; potential confounding from ICU interventions, prior antibiotics, and immunosuppression.