The participants in this study were children treated in the pediatric intensive care unit (PICU) of the Children’s Hospital of Fudan University from February 2018 to February 2020. The inclusion criterion was a diagnosis of severe pneumonia according to the diagnostic criteria established by the World Health Organization (Bradley et al., 2011; Chisti et al., 2015; Williams et al., 2016). The primary diagnostic criteria were invasive mechanical ventilation, fluid-refractory shock, an urgent need for noninvasive positive-pressure ventilation, and hypoxemia requiring an FiO₂ greater than the inhalation concentration or flow rate feasible within general care. The secondary criteria were an increased respiratory rate, PaO₂/FiO₂ ratio of < 250, multilobar infiltration, Pediatric Early Warning Score of > 6, altered mental status, hypotension, presence of effusion, comorbidities (e.g. immunosuppression, immunodeficiency), and unexplained metabolic acidosis. The exclusion criteria for this study were an age of < 28 days or > 18 years; noninfectious factors such as congenital heart disease, pulmonary edema, asthma, upper airway obstruction, or pulmonary cystic fibrosis; contraindications to fiberoptic bronchoscopy; severe cardiopulmonary dysfunction; and coagulopathy.

With reference to the recommendations of the European Respiratory Society (de Blic et al., 2000), sedation and topical anesthesia were administered, and the patient’s lungs were lavaged using an age-appropriate pediatric flexible fiberoptic bronchoscope. The more severely diseased region in patients with diffuse lung disease or the right middle lobe was selected based on radiological findings or evidence from bronchoscopy. Warm saline (1 mL/kg body weight, maximum 20 mL per fraction) was dripped into the selected lung lobes, and at least 40% of the fluid was recovered by mechanical suction using a pressure of approximately 50 to 100 mmHg.

Venous blood (2 mL) was collected and stored at room temperature with heparin anticoagulation to complete the assay for serum lipopolysaccharide (LPS), procalcitonin (PCT), C-reactive protein (CRP), interleukin 6 (IL-6), and other indices. The specific processes are detailed in the Supplementary Appendix .

Venous blood (2 mL) was collected, and the peripheral blood lymphocyte subsets were first labeled with fluorescently labeled monoclonal antibodies. B cells (CD19⁺), total T cells (CD3⁺), CD8⁺ T cells (CD3⁺CD8⁺), CD4⁺ T cells (CD3⁺CD4⁺), and natural killer cells (CD16⁺CD56⁺). Cells were stained in 6-color TBNK Reagent (BD Multitest) that included CD3-FITC, CD16/CD56-PE, CD45-PerCP-Cy5.5, CD4-PE-Cy7, CD19-APC, and CD8-APC-Cy7. Flow cytometry (BD Biosciences, San Jose, CA, USA) was then performed to complete the lymphocyte subpopulation analysis.

We performed metagenomic sequencing with reference to our previously published studies (Wang et al., 2019; Xu et al., 2020; Wang et al., 2021; Yan et al., 2021; Zhou et al., 2016; Zhou et al., 2019), as described in the Supplementary Appendix .

Pearson correlation coefficients and Spearman correlation coefficients for each microbial ecological diversity index and clinical phenotype were first calculated by the cor.test() function of R software (version 3.6.5). Next, the lm() function of R software (version 3.6.5) was used to perform a linear fit of the microbial ecological diversity index to the clinical phenotypes, was then the ggplot2 package of R software was used to draw the scatterplot. Statistically significant p-values were obtained by the Wilcoxon rank sum test function (wilcox.test) of R software.

This study involved 126 children in the PICU, including 84 with community-acquired pneumonia (CAP) and 42 with hospital-acquired pneumonia (HAP). Their mean maximum body temperature was 39°C ± 0.95°C, mean heart rate was 160 ± 19 beats per minute, mean total duration of hospitalization was 37 ± 29 days, and mean number of days in the PICU was 30 ± 29 days. The 30-day mortality analysis revealed 24 deaths and 16 cases of abandoned treatment, accounting for 29.4% of all patients. In total, 86.5% (109/126) of patients had combined respiratory failure, 24.6% (31/126) had combined severe pneumonia, 24.6% (31/126) had combined sepsis, 5.5% (7/126) had septic shock, one had encephalitis, and four were being hospitalized after cord blood stem cell transplantation.

The mean serum CRP concentration was 67 ± 56 mg/L, PCT concentration was 8.9 ± 19 ng/mL, LPS concentration was 1.0 ± 7.6 pg/mL, and IL-6 concentration was 390 ± 850 pg/mL. The white blood cell (WBC) count was 18 ± 12 × 10⁹ cells/L, CD3⁺ lymphocyte count was 1.2 ± 1.3 × 10⁹ cells/L, CD4⁺ T cell count was 6.8 ± 8.7 × 10⁸ cells/L, CD8⁺ T cell count was 5.0 ± 5.9 × 10⁸ cells/L, and NK cell count was 1.4 ± 1.8 × 10⁸ cells/L. Pathogenic microorganisms were detected in 81 patients based on conventional clinical microbiological testing methods such as blood and other cultures.

mNGS was performed on 126 samples [at the DNA level in most (101/126, 80.1%) and at the RNA level in a smaller number (25/126, 19.8%)]. Each sample sequenced yielded 2.6 ± 1.4 × 10⁷ clean sequences. The potentially pathogenic microorganisms in the samples were further identified by considering both the relative abundance of the microorganisms in the samples to be tested and the outlier levels (z-score) in all samples, following the method used in our previous study (Yan et al., 2021). The relative abundance and z-score distribution of the pathogenic microorganisms obtained are shown in Figure 1 .

Nucleic acid sequences of 88 pathogenic bacteria were detected, including 9 species of Streptococcus (Streptococcus agalactiae, Streptococcus australis, Streptococcus milleri, Streptococcus mitis, Streptococcus oralis, Streptococcus parasanguinis, Streptococcus pneumoniae, Streptococcus pseudopneumoniae, and Streptococcus salivarius), 7 species of Staphylococcus (Staphylococcus aureus, Staphylococcus capitis, Staphylococcus epidermidis, Staphylococcus haemolyticus, Staphylococcus hominis, Staphylococcus saprophyticus, and Staphylococcus warneri), and 3 species of Acinetobacter (Acinetobacter baumannii, Acinetobacter johnsonii, and Acinetobacter nosocomialis). The top 25 bacteria are listed in Figure 2 . In total, 13 viral nucleic acid sequences were detected, including 4 human herpesviruses [Human alphaherpesvirus 1, Human betaherpesvirus 5 (also known as human cytomegalovirus), Human betaherpesvirus 6B, and Human gammaherpesvirus 4 (also known as Epstein–Barr virus)], Human mastadenovirus B, and Human polyomavirus 4. All viruses are shown in Supplementary Figure 1 . Nucleic acid sequences were detected for 39 pathogenic fungi, including 4 Aspergillus species (Aspergillus fischeri, Aspergillus fumigatus, Aspergillus mulundensis, and Aspergillus nidulans) and Pneumocystis jirovecii; the other fungi are shown in Supplementary Figure 2 . Nucleic acid sequences were detected for 27 protozoans, including 10 protozoa of the genus Plasmodium (Plasmodium falciparum, Plasmodium gaboni, Plasmodium gallinaceum, Plasmodium gonderi, Plasmodium knowlesi, Plasmodium malariae, Plasmodium reichenowi, Plasmodium relictum, Plasmodium sp, and Plasmodium yoelii) and 2 protozoa of the genus Trypanosoma (Trypanosoma cruzi and Trypanosoma theileri). The others are specified in Supplementary Figure 3 .

We found that the serum LPS concentration was correlated with the pathogenic bacterial abundance in BALF. Notably, most of the 12 serum LPS-associated bacteria identified were positively correlated with the serum LPS concentration, including Acinetobacter baumannii, Neisseria meningitidis, Neisseria subflava, Pseudomonas aeruginosa, Stenotrophomonas maltophilia, Mycoplasma salivarium, and several others; the full results are shown in Figure 3A . We also found that the serum CRP concentration was correlated with the pathogenic bacterial abundance in BALF, with Pseudomonas aeruginosa and Corynebacterium striatum positively correlated with the serum CRP concentration, as shown in Figure 3B . Moreover, the serum PCT concentration was correlated with the abundance of pathogenic bacteria in BALF; the abundance of Staphylococcus aureus, Staphylococcus hominis, Staphylococcus saprophyticus, Haemophilus influenzae, Pseudomonas aeruginosa, and Mycoplasma salivarium was positively correlated with the serum PCT concentrations, as shown in Figure 3C . Likewise, we found that serum IL-6 concentration was correlated with the abundance of pathogenic bacteria in BALF, with positive correlations noted for Acinetobacter baumannii and Acinetobacter nosocomialis. Notably, we found that certain Streptococcus spp. Pathogenic bacteria in BALF, such as Streptococcus pseudopneumoniae, Streptococcus agalactiae, Streptococcus oralis, and Streptococcus mitis, were negatively correlated with the serum IL-6 concentration, as shown in Figure 3D .

The data showed that the total WBC count was negatively correlated with the abundance of pathogenic bacteria in BALF, such as Pseudomonas aeruginosa, Prevotella jejuni, Streptococcus australis, Achromobacter xylosoxidans, Sphingopyxis fribergensis, Actinomyces pacaensis, Actinomyces naeslundii, Herbaspirillum huttiense, and others ( Supplementary Figure 4A ). We also found that the total WBC count was negatively correlated with the abundance of Human betaherpesvirus 6B and Human gammaherpesvirus 4 in BALF ( Supplementary Figure 4B ) and with the abundance of fungi in BALF such as Pneumocystis jirovecii and Aspergillus terreus ( Supplementary Figure 4C ).

Total T cell count was correlated with the abundance of pathogenic bacteria in BALF such as Staphylococcus aureus, Staphylococcus haemolyticus, and Staphylococcus warneri ( Supplementary Figure 5A ). Notably, we found that the abundance of Staphylococcus aureus, Staphylococcus haemolyticus, and Staphylococcus warneri in BALF was negatively correlated with the counts of blood CD4⁺ T cell, CD8⁺ T cell, and NK cell ( Supplementary Figures 5B–D ).

We performed a differential analysis of the pathogenic microbial composition of BALF in patients with CAP and HAP and found 3 pathogenic bacteria that were present at significantly different levels. There were significantly greater amounts of Acinetobacter baumannii and Streptococcus miti in patients with HAP than that in patients with CAP, while the amount of Schaalia_odontolytica was significantly lower in the patients with HAP than that in the patients with CAP. We also found a significantly higher amount of the potentially pathogenic fungus Diplodia corticola and the potentially pathogenic parasite Plasmopara halstedii in the BALF of patients with HAP than of patients with CAP ( Figure 4 ).

We assessed the correlation between lethal phenotypes (death or abandonment of treatment) and the ecological diversity of pathogenic microorganisms in BALF. The results showed that the fungal and protozoa richness (Chao1 index) was significantly higher in BALF from deceased than non-deceased patients ( Figure 5A ); the fungal and protozoal diversity (Shannon index) was also significantly higher in BALF from deceased than non-deceased patients ( Figures 5B, C ). We also assessed the correlation between lethal phenotypes (death or abandonment of treatment) and the pathogenic microbial composition of the BALF. The results showed that Escherichia coli abundance was significantly higher in the BALF from deceased than non-deceased patients. Notably, we found significantly higher abundance of Klebsiella pneumoniae and Corynebacterium segmentosum in the BALF of patients in the abandoned treatment group than in the deceased/non-deceased group ( Figure 5D ). Additionally, there was a significantly higher abundance of Pneumocystis jirovecii in the BALF of deceased patients than in both the non-deceased and abandoned treatment groups ( Figure 5E ).

OI reflects the severity of disease in patients with severe pneumonia. The data showed that OI was associated with increased abundance of pathogenic bacteria in BALF, including Escherichia coli, Klebsiella pneumoniae, Streptococcus agalactiae, and Staphylococcus aureus ( Figure 6A ). Moreover, The OI was positively correlated with an increased abundance of the fungi Aspergillus fumigatus and Rhizopus microspores in BALF ( Figure 6B ) and with an increased abundance of the viruses Human mastadenovirus B and Torque teno virus 29 in BALF ( Figure 6C ). In addition, there was a weak positive correlation between OI and the overall viral abundance ( Figure 6D ).

Among patients with severe pneumonia, the BALF fungal and parasite species richness (Chao1 index), Shannon diversity, and number of species were significantly higher in patients with than without sepsis ( Supplementary Figure 6 ). Additionally, the total number of bacterial species in BALF was significantly higher in patients with than without immunodeficiency ( Supplementary Figure 7A ). Moreover, the abundance of Elizabethkingia anophelis in BALF was significantly higher in patients with than without immunodeficiency ( Supplementary Figure 7B ). Notably, we found that the abundance of Human betaherpesvirus 5 (i.e., human cytomegalovirus) in BALF was significantly higher in patients with than without immunodeficiency ( Supplementary Figure 7C ).

In the present study, we found that the BALF bacterial diversity indices, including the Chao1 diversity index and overall abundance, were positively correlated with the serum LPS concentration; the bacterial species count was positively correlated with the serum PCT concentration; and that the Pseudomonas aeruginosa abundance in BALF was positively correlated with the serum LPS, CRP, and PCT concentrations and negatively correlated with the whole blood WBC count and CD4⁺ T cell count.

Pseudomonas aeruginosa is a conditionally pathogenic bacterium widely present in nature, human skin, intestine, and respiratory tract; studies have shown that P. aeruginosa is an important causative agent of hospital-acquired pneumonia (HAP) and ventilator-associated pneumonia (VAP) (Pang et al., 2019); notably, P. aeruginosa is prone to cause intranasal and ventilator-associated pneumonia in immunocompromised and hospitalized patients (Killough et al., 2022). Consistent with this, we found that P. aeruginosa abundance in BALF was negatively correlated with whole blood WBC counts and CD4⁺ T cell counts; also, in this study, we found that the number of bacterial species detected in BALF was elevated in the immunodeficient group relative to the non-immunodeficient group ( Supplementary Figure 7A ). In short, this implies that the rise in the abundance of P. aeruginosa may be associated with immunocompromised or impaired immunity in patients. Pseudomonas aeruginosa is the most common multidrug-resistant Gram-negative bacterium and a common cause of ventilator-associated pneumonia in patients in ICUs. Ventilator-associated pneumonia caused by Pseudomonas aeruginosa infection is characterized by high morbidity and mortality (Zhuo et al., 2008; Valencia and Torres, 2009). Experimental models suggest that ventilator-induced lung injury is associated with increased bronchial vascular permeability in BALF, higher cell counts and protein concentrations, and increased infiltration of inflammatory cells into lung tissue (Wolters et al., 2009). Cytokines are small proteins that communicate through intercellular signaling and can be considered immunomodulators of immune and inflammatory responses (Spelman et al., 2006). Human studies have shown that cytokine/chemokine release and leukocyte recruitment contribute to ventilator-associated lung injury (Halbertsma et al., 2005) and that IL-6, an inflammatory marker of ventilator-associated lung injury, may contribute to alveolar barrier dysfunction in patients with acute respiratory distress syndrome (Wolters et al., 2009). IL-6 is an important pro-inflammatory cytokine that is released during infection or tissue injury; IL-6 is mainly expressed by natural immune cells such as monocyte macrophages, but can also be produced by Th2 cells, vascular endothelial cells, and fibroblasts (Kang and Kishimoto, 2021). In the present study, we observed a correlation between serum IL-6 levels and the abundance of specific pathogenic microorganisms in BALF ( Figure 3D ). Tsay et al. found that, like mechanical ventilation, intranasal perfusion with P. aeruginosa also induced IL-6 expression in the lungs (Tsay et al., 2016). We did not find a correlation between P. aeruginosa abundance and serum IL-6 levels in BALF and could not identify the source of serum IL-6.

Acinetobacter baumannii, a Gram-negative aerobic bacterium, is listed by the World Health Organization as posing the greatest threat to human health and as being in urgent need of new antibiotics (World health organization, 2017; Lee et al., 2017). In recent years, isolates of Acinetobacter baumannii have been recovered from a variety of extra-hospital sources (e.g., vegetables, water treatment plants, and fish and shrimp farms) in addition to known natural habitats (soil and moist environments), and the wide range of sources of this bacterium might explain the occurrence of community-acquired infections (Eveillard et al., 2013). Acinetobacter baumannii is a serious pathogen involved in hospital-acquired and community-acquired infections (Dijkshoorn et al., 2007). Cases of CAP caused by Acinetobacter baumannii are rare in Europe and the United States (Serota et al., 2018); instead, most cases of CAP caused by Acinetobacter baumannii occur in countries with tropical or subtropical climates. Acinetobacter baumannii is an emerging pathogen in the Asia-Pacific region, with a high prevalence in Hong Kong, Singapore, Taiwan, Korea, and Australia (Kim et al., 2018). In the present study, we found that the abundance of Acinetobacter baumannii in BALF of children in the PICU was positively correlated with the serum IL-6 and LPS concentrations. Notably, the abundance of Acinetobacter baumannii in BALF was significantly higher in patients with HAP than CAP; this may be related to the fact that Acinetobacter baumannii is more common in hospitals than in the community. The primary target of Acinetobacter baumannii is patients in ICUs, and the use of antibiotics has led to outbreaks and epidemics of infections caused by multidrug-resistant Acinetobacter baumannii in hospitals (Dijkshoorn et al., 2007; Ramirez et al., 2020).

Respiratory viral infections are an important cause of pneumonia in children. Viral bronchiolitis is the most common cause of respiratory failure in young children requiring invasive ventilation, and coinfections such as bacterial and fungal infections may prolong and complicate hospital stays in the PICU (Wiegers et al., 2019; Li et al., 2020b). We performed mNGS on BALF from children in the PICU and found that viral nucleic acid sequences were present; in addition, the Chao1 diversity index of viruses was positively correlated with the maximum body temperature on admission and negatively correlated with the CD3⁺ lymphocyte count. As a microbial ecology index, the viral Shannon diversity index reflects both the number of viruses and their abundance. In calculating the Shannon diversity index of BALF samples, the included viruses contain both eukaryotic viruses and prokaryotic viruses such as phages, and usually the number of phage species is more abundant than eukaryotic viruses. We observed that the Shannon diversity index of viruses was negatively correlated with the counts of tatol T cell, CD4⁺ T cell, CD8⁺ T cell and NK cell, suggesting whether certain viruses may have a protective role, such as phages. Considering that phages usually act as commensal microorganisms and not as pathogenic microorganisms; therefore, they have not been analyzed and discussed here.

The number of ICU admissions for viral infections is rising. The Herpesviridae family, including Human gammaherpesvirus 4 (i.e., Epstein–Barr virus) and Cytomegalovirus, can be reactivated in ICU patients, and such viral activation is associated with morbidity and mortality (Cantan et al., 2019). Consistent with this, we found that the abundance of Human betaherpesvirus 6B and Human gammaherpesvirus 4 was negatively correlated with the total WBC count. Additionally, the abundance of Human betaherpesvirus 5 (i.e., human cytomegalovirus) in the BALF of patients with severe pneumonia was significantly higher in those with than without immunodeficiency, suggesting a possible correlation between immune system dysregulation and viral activation. We also found that the severity of pneumonia (i.e., the OI) was positively correlated with the overall viral abundance in the BALF; notably, we found that the OI was positively correlated with an increase in the abundance of the viruses Human mastadenovirus B and Torque teno virus 29 in BALF. In summary, these findings suggest a correlation between the viral load in BALF and the severity of pneumonia.

Fungi are another important cause of respiratory infections (Schmidt et al., 2018). We found significantly higher species richness (Chao1 index), Shannon diversity, and number of species of pathogenic fungi in the BALF of patients with than without sepsis in the present study. Aspergillus, Pneumocystis, and Cryptococcus are the key fungal pathogens known to cause respiratory infections (Hayes and Denning, 2013). Pneumocystis jirovecii causes pneumocystis pneumonia, a life-threatening disease in the ICU with a high mortality rate, particularly in immunocompromised patients (Salzer et al., 2018; Schmidt et al., 2018). Schmidt et al. (Schmidt et al., 2018) found that the overall hospital mortality rate was 25.4% and increased to 58.0% if ICU admission was required. In line with this, we found a significantly higher abundance of Pneumocystis jirovecii in the BALF of deceased patients than in both non-deceased patients and patients for whom treatment was abandoned.

Patients in the ICU are more likely to have a history of coinfection with bacteria and fungi than are patients in the general ward (Schauwvlieghe et al., 2018). Beumer et al. (Beumer et al., 2019) found that the incidence of coinfection with bacteria and fungi in the lungs of patients with influenza was 55.6% among ICU inpatients, which was significantly higher than in patients in the normal ward (20.1%), and Aspergillus fumigatus and Pneumocystis jirovecii were the predominant coinfecting fungi (Beumer et al., 2019). In line with this, we found that an increase in the abundance of the fungal pathogen Aspergillus fumigatus in BALF was positively correlated with the severity (i.e., the OI) of patients with severe pneumonia, while an increase in the abundance of the bacterial pathogens Klebsiella pneumoniae, Streptococcus agalactiae, and Staphylococcus aureus in BALF of children in the PICU was also positively correlated with the severity (i.e., the OI) of patients with severe pneumonia.

Protozoan coinfection increases the risk of pneumonia in children (D'Acremont et al., 2014; Edwards et al., 2015). We found significantly higher protozoan species richness (Chao1 index), Shannon diversity, and number of species in the BALF of patients with than without sepsis in the present study; we also detected nucleic acid sequences of Plasmodium and Trypanosoma in the BALF of children in the PICU. Notably, protozoa have a large genome, and the mNGS results only cover a very small part of their genome. Although in-depth validation is needed, our results suggest the possibility of protozoan coinfection.

This study aimed to gain a comprehensive understanding of the microbiome composition in the BALF of children with severe pneumonia in the PICU. Although uncommon, protozoa such as parasites may also represent the “potential” pathogenic microbial composition in BALF (Van den Steen et al., 2010; Deroost et al., 2013; Wihokhoen et al., 2016; Kumar et al., 2022). Therefore, after metagenomic sequencing, the sequenced reads were aligned to bacterial, viral, and fungal databases, as well as to protozoan databases, and the “potential” protozoan composition was obtained. Considering that the protozoan genome size is larger than that of bacteria, viruses, and fungi, it is difficult to determine whether protozoa actually exist with a limited number of reads, but we just try to understand that the nucleic acid sequence of the parasite was detected. In the next step, we need to verify the existence of protozoa by 18S rRNA sequencing technology.

Considering that specific pathogenic microorganisms were detected only in a small number of BALF samples; therefore, we did not perform a correlation/regression analysis of the abundance of specific pathogenic microorganisms and clinical indicators, or a multiple linear regression analysis. Instead, we performed a logistic linear regression of the microecological diversity indicators with the clinical indicators, and the results are shown in Figure 6D , where it can be seen that OI was positively correlated with the abundance of pathogenic bacteria in BALF. Indeed, the correlation is weaker, but reflects a trend.

All patients included in this study were from pediatric intensive care units; usually, DNA viruses, bacteria, fungi, and mycoplasma are the predominant pathogenic microorganisms associated with the respiratory tract in children, with RNA viruses accounting for a relatively small proportion. The mNGS used in this study cannot take into account both DNA and RNA viruses. According to the clinical empirical judgment of two specialized physicians, 25 of 126 patients were inclined to RNA virus infection, in order to obtain the test results in the shortest time and save clinical costs. We only tested the above 25 cases at the RNA level and 121 cases at the DNA level, and the test results were also consistent with the empirical judgment results. Of course, the mNGS technique in this study cannot take into account both DNA and RNA viruses, and there may be missed detection. Nowadays, our team has developed the mNGS pathogen capture technique based on mNGS, and the development of this technique will make up for this deficiency and provide a new direction of choice for pathogenic diagnosis in line with clinical thinking.

This study had some limitations. The inflammatory marker testing using BALF samples can provide direct insight into the impact of different pathogenic infections on pneumonia outcomes. However, since the volume of BALF obtained from children is limited; therefore, BALF samples were not used for inflammatory marker assays in addition to pathogenic assays; as an alternative strategy, we performed inflammatory marker assays on blood samples. The pathogenic nucleic acid sequences we identified in BALF in the PICU require in-depth validation, including completion by clinical microbiological methods such as BALF culture and antibody testing. Further, multicenter scale-up validation is also needed for an in-depth assessment of mNGS results relevance to clinical symptoms in children with severe pneumonia in the PICU.