Bronchoalveolar Lavage Microbiota in Critically Ill Patients with Aspiration Pneumonia: An Exploratory Comparison with Community-Acquired Pneumonia and Analysis of Gemella Detection and Mortality

Abstract

Background

Aspiration pneumonia (AP) is a common and severe infection among critically ill older adults, but its lower respiratory tract microbiota and clinical associations remain incompletely characterized.

Methods

This single-center retrospective exploratory study included 72 intensive care unit patients, comprising 49 patients with AP and 23 with community-acquired pneumonia (CAP). Bronchoalveolar lavage fluid (BALF) microbiota were profiled using metagenomic next-generation sequencing (mNGS). Alpha and beta diversity, taxonomic composition, differential abundance, and associations between bacterial taxa and clinical or peripheral immune markers were analyzed. Within the AP cohort, the association between Gemella detection and in-hospital mortality was evaluated using Firth’s penalized logistic regression.

Results

AP patients had higher Shannon and Simpson diversity than CAP patients. Overall community composition did not differ significantly between the groups based on Bray–Curtis dissimilarity and PERMANOVA (R²=0.022, P=0.066). Streptococcus and Rothia showed nominally higher relative abundances in AP, whereas Acinetobacter and Stenotrophomonas showed nominally lower relative abundances. Only Acinetobacter remained more abundant in CAP after FDR correction (FDR-adjusted P=0.032). Nominal correlations between bacterial taxa and clinical or immune variables were not significant after FDR correction. Gemella was detected in 9 AP patients, including 8 of 26 non-survivors and 1 of 23 survivors (P=0.026). In the unadjusted Firth model, Gemella detection was associated with mortality (OR=6.892, 95% CI 1.360–69.062; P=0.018), but the association was attenuated and non-significant after adjustment for APACHE II score (OR=4.708, 95% CI 0.859–48.395; P=0.076).

Conclusion

AP was associated with higher alpha diversity. Selected oral-associated taxonomic patterns and their immune-cell percentages correlations were exploratory and were not confirmed after multiple-testing correction. The association between Gemella detection and mortality was imprecise and sensitive to sparse data and model specification, requiring validation in larger prospective studies.

Introduction

Aspiration Pneumonia (AP) is primarily defined as a lower respiratory tract infection caused by the inhalation of oropharyngeal or gastric contents, with a high incidence among elderly individuals, patients with dysphagia, or those with neurological disorders.^1,2 The occurrence rate of AP accounts for 5–15% of all pneumonia cases in hospitalized patients, while this proportion may rise to as high as 40% in bedridden or hospitalized elderly patients.^3–5 Furthermore, AP is closely associated with elevated in-hospital mortality and an increased risk of complications.^6,7 Notably, previous studies have found the in-hospital mortality rate for AP can reach 20–30%, substantially exceeding that of other pneumonia subtypes.⁸ The high incidence and mortality rates of AP mean that it has become an important public health issue on a global scale.^1,3

While the lungs were long considered sterile, extensive evidence now supports the presence of a resident microbial community even in healthy individuals.⁹ In recent years, multiple studies have revealed that alterations in the diversity and composition of the lower respiratory tract microbiota play a significant role in the pathogenesis and progression of respiratory diseases. For instance, in patients with community-acquired pneumonia (CAP) and ventilator-associated pneumonia (VAP), a marked reduction in respiratory microbiota diversity has been observed, which is closely associated with disease severity and clinical outcomes.^10,11 In diseases such as chronic obstructive pulmonary disease (COPD) and acute respiratory distress syndrome (ARDS), the predominant colonization of specific pathogenic bacteria and microbial dysbiosis have been demonstrated to be associated with exacerbated inflammatory responses and worsened clinical outcomes.^12,13

Metagenomic next-generation sequencing (mNGS) provides broad microbial coverage and higher taxonomic resolution than targeted 16S rRNA sequencing and is therefore useful for characterizing microbial communities in lower respiratory tract specimens.^14–17 Previous studies have described microbiota changes in AP using 16S rRNA sequencing,^13,18 but BALF-based mNGS studies in AP remain limited. In particular, the relationships between BALF microbial profiles, peripheral inflammatory or immune variables, and clinical outcomes have not been fully characterized. Further investigation is therefore needed to identify microbiota patterns associated with AP and to determine whether these patterns are related to clinical or immunological features.

Critically ill patients with CAP requiring Intensive Care Unit (ICU) care were selected as the control because they were evaluated in the same clinical setting and underwent the same BALF collection and mNGS procedures as the AP group. This comparison was intended to reduce variation related to ICU care, critical illness, sampling procedures, and sequencing workflow, while allowing exploration of microbial patterns associated with clinically diagnosed aspiration. However, the CAP group should not be interpreted as representative of CAP in the general community, because ICU exposure, prior antimicrobial treatment, and severe illness may influence the respiratory microbiota. The primary objective of this study was to compare BALF microbial diversity and taxonomic composition between critically ill patients with AP and those with CAP. The secondary objectives were to explore associations between differentially abundant bacterial taxa and inflammatory or peripheral immune markers and, within the AP cohort, to perform a hypothesis-generating analysis of the association between Gemella detection and in-hospital mortality. Because of the observational and retrospective design, all identified relationships were interpreted as associative rather than causal.

Materials and Methods

Patient Enrollment

We retrospectively enrolled 72 patients admitted to ICU of Jinshan Hospital of Fudan University between 27 September 2021 and 23 August 2024, including 49 patients with AP and 23 patients with CAP. Inclusion criteria were: (1) Age ≥ 18 years; (2) Expected ICU stay > 48 hours; (3) Diagnosis consistent with AP or CAP; (4) BALF samples collected for mNGS testing during the initial diagnostic evaluation; and (5) Complete clinical data available. Exclusion criteria were: (1) Concurrent infections in other organ systems; (2) Congenital immunodeficiency or long-term use of systemic glucocorticoids or immunosuppressants; and (3) Prolonged antibiotic therapy (≥ 7 days) prior to BALF sampling.

Given the retrospective ICU-based design, AP and CAP were defined according to the treating physicians’ diagnosis at ICU admission. AP was diagnosed based on pulmonary inflammatory findings together with overt aspiration, strongly suspected aspiration, or confirmed dysphagia.¹⁹ The specific criteria were as follows: (1) Presence of clear aspiration risk factors (eg, post-stroke dysphagia, impaired consciousness, gastroesophageal reflux, post-general anesthesia), with aspiration pneumonia caused by acute aspiration occurring in the hospital; (2) Acute onset with new or progressive pulmonary infiltrates on chest imaging; (3) Meeting the clinical criteria for pneumonia (eg, fever, leukocytosis, purulent sputum, etc). CAP was diagnosed according to the 2019 ATS/IDSA guideline.²⁰ It was defined as a new or progressive pulmonary infiltrate on chest imaging, together with at least one compatible clinical feature of lower respiratory tract infection, including cough, dyspnea, sputum production, fever, or hypothermia. Microbiological results were recorded when available but were not required for inclusion, because prior antimicrobial exposure could reduce microbiological yield.

The study was performed in accordance with the Declaration of Helsinki and was approved by the ethics committee of the Jinshan Hospital of Fudan University (approval ID: JIEC 2024-S99). Written informed consent was obtained from all participants or their legally authorized representatives.

Specimen Collection

BALF was obtained from 72 participants. BALF samples were collected within 0–2 days of transfer to the ICU. All patients had received β-lactam (n=47) or quinolone (n=25) antibiotics before BALF sampling. The median duration of antibiotic exposure before sampling was 10 h (IQR, 8–16 h). All subjects signed informed consent forms. Prior to bronchoscopy, the subjects’ medical histories were collected and routine preoperative examinations were performed. BALF samples were collected from patients according to standard procedures.²¹ Following topical anesthesia of the patient’s throat with lidocaine, a fiberoptic bronchoscope was introduced. Through the bronchoscope, the lungs were repeatedly lavaged with sterile saline at room temperature (20 mL per aliquot, totaling 100 mL). The recovery rate of BALF samples exceeded 40%. The lavage fluid was collected via negative-pressure suction, and a 5 mL aliquot was extracted from the recovered solution and transferred into a nuclease-free DNA sampling tube. The BALF sample was transported within 24 hours to the sequencing company (Dinfectome Inc., Nanjing, China) for mNGS examination under refrigeration (4–8 °C).

Data Collection

The demographic characteristics and laboratory test results of the subjects were collected on the day of BALF sample acquisition. The demographic information included age, sex, and underlying diseases. Laboratory data encompassed complete blood count, inflammatory markers, blood gas analysis, and lymphocyte subsets.

Nucleic Acid Extraction and Metagenomic Next-Generation Sequencing

BALF DNA was extracted using the TIANamp Magnetic DNA Kit (TIANGEN, China) according to the manufacturer’s protocols. The quantity and quality of DNA were assessed using Qubit 2.0 Fluorometers and Nanodrop 8000 spectrophotometers (Thermo Fisher Scientific, USA), respectively. Library construction was performed according to the instructions of Hieff NGS C130P2 OnePot II DNA Library Prep Kit for MGI (Yeasen Biotech, China). The Agilent 2100 Bioanalyzer system (Agilent, Santa Clara, CA, USA) and the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific Inc., Waltham, MA, USA) were used to assess the DNA library fragment sizes and concentrations. Finally, we completed sequencing in the single-end 50 bp sequencing mode using DIFSEQ-200 (Dinfectome, Nanjing, China). No-template controls (NTCs) were included during nucleic acid extraction, library preparation, and sequencing to monitor potential reagent or environmental contamination.

Bioinformatics Analysis

We used an in-house bioinformatics pipeline for taxonomic profiling and pathogen identification. Quality control was performed with fastp to remove adapter sequences, duplicate reads, and low-quality/short reads (< 36 bp). Host-derived sequences were filtered by alignment to the human reference genome (hs37d5) using Bowtie2 (v2.4.2). The remaining non-human reads were then analyzed using Kraken2 (v2.1.1) for taxonomic classification against a custom microbial genome database (downloaded from GenBank release 238), followed by abundance estimation with Bracken (v2.6.2). The database included bacterial, fungal, viral, and parasites genomes (http://ftp.ncbi.nlm.nih.gov/genomes/genbank/). We used the following criteria when interpreting the results of mNGS: For Mycobacterium, Nocardia, and Legionella pneumophila, a positive result was defined when a species-specific read count of 1 or more was detected by mNGS. For bacteria (excluding Mycobacterium, Nocardia, and Legionella pneumophila), fungi, viruses, and parasites, a positive result was defined when a species had at least three non-overlapping reads detected by mNGS. Pathogens detected in the negative NTCs were excluded from the analysis, except when the read counts in the samples were at least 10-fold higher than those observed in the NTCs.

The 72 BALF samples yielded an average of 25.04 ± 10.07 M sequencing reads per sample, and the mean proportion of reads unmapped to the human reference genome was 11.13 ± 10.03%. Samples in which fewer than five microorganisms were detected in the final analysis were defined as low-biomass samples and were excluded from the microbial community diversity analysis.

Alpha diversity was assessed using the Shannon and Simpson indices calculated from the taxonomic profile of each sample. Beta diversity was evaluated using Bray-Curtis dissimilarity and visualized by principal coordinate analysis (PCoA). Differences in overall microbial community composition between two groups were formally evaluated by permutational multivariate analysis of variance (PERMANOVA) using the R package “vegan”. Differences in the relative abundances of bacterial genera and species between two groups were evaluated using the Mann–Whitney U-test. Taxa with a mean relative abundance greater than 0.1% and a prevalence greater than 10% across all samples were included in the differential abundance analysis. To account for multiple comparisons, P values were adjusted using the Benjamini-Hochberg false discovery rate (FDR) procedure. Spearman correlation analyses were performed in the overall cohort and separately within the AP subgroup to assess associations between clinical variables and the relative abundances of bacterial taxa using the cor.test function in R. P values from the correlation analyses were adjusted for multiple comparisons using the Benjamini-Hochberg FDR procedure. FDR adjusted P<0.05 was considered statistically significant. Results with P<0.05 but FDR adjusted P≥0.05 were considered exploratory.

Statistical Analysis

Categorical variables were expressed as percentages (%), and statistical comparisons were performed using the chi-square test or Fisher’s exact test, as appropriate. Continuous variables were reported as mean ± standard deviation (SD) or median (interquartile range, IQR), and compared using the Student’s t test or Mann–Whitney U-test. Clinical and baseline comparisons were performed using SPSS version 27.0. A two-sided P<0.05 was considered statistically significant.

All regression and sensitivity analyses were performed in AP group using the “logistf” package in R. Given the high proportion of zero values in Gemella abundance data, Gemella detection status was treated as a binary variable (detected versus undetected; detection defined as abundance > 0). Because Gemella was detected in only nine AP patients, including eight non-survivors and one survivor, Firth’s penalized logistic regression with profile penalized-likelihood confidence intervals was used to reduce small-sample and sparse-data bias. Gemella detection was the exposure of interest. The unadjusted model included Gemella detection alone, whereas the primary adjusted model included Gemella detection and Acute Physiology and Chronic Health Evaluation II (APACHE II) score. An alternative Firth model adjusted for age was performed as a sensitivity analysis. Conventional maximum-likelihood logistic regression models were fitted using the “logistf” package with Firth correction disabled (firth = FALSE). A leave-one-out influence analysis was also conducted by sequentially excluding each patient and refitting the APACHE II-adjusted Firth model. Odds ratios (ORs), 95% confidence intervals (CIs), and two-sided P values were reported.

Results

Participants and General Characteristics

This study enrolled 49 patients with AP and 23 patients with CAP. The clinical characteristics and baseline demographics data are presented in Table 1.

Table 1.

Characteristics and Clinic Parameters of the AP and CAP Patients

Characteristics	AP (n=49)	CAP (n=23)	P-value
Age	77.76±9.29	75.83±8.35	0.54
Gender, n, (% male)	39(79.59)	17(73.91)	0.76
Comorbidities
Hypertension, n (%)	30(61.22)	18(78.26)	0.19
Diabetes, n (%)	11(22.45)	5(21.74)	1.00
Coronary heart disease, n (%)	9(18.37)	4(17.39)	1.00
Cerebral infarction, n (%)	16(32.65)	4(17.39)	0.26
Parkinson’s disease, n (%)	4(8.16)	2(8.70)	1.00
Atrial fibrillation, n (%)	6(12.24)	3(13.04)	1.00
Chronic obstructive pulmonary disease, n (%)	6(12.24)	2(8.70)	1.00
Laboratory findings
WBC (×109/L)	10.3(6.2,14.4)	8.4(6.6,13.3)	0.646
Neutrophils (×109/L)	8.5(5.2,13.1)	7.5(5.5,11.3)	0.583
Monocyte (×109/L)	0.4(0.2,0.7)	0.5(0.1,0.8)	0.957
Lymphocytes (×109/L)	0.6(0.4,1.0)	0.6(0.2,1.0)	0.349
CRP (mg/L)	49.8(20.9,112.2)	97.4(39.5,176.3)	0.085
PCT (mg/L)	1.1(0.5,3.8)	2.0(0.2,9.0)	0.966
pH	7.3(7.2,7.4)	7.4(7.3,7.5)	0.05
PaO2 (kpa)	14.0(10.1,18.9)	10.8(7.9,19.4)	0.19
PaCO2 (kpa)	5.8(4.8,8.2)	5.4(4.1,6.8)	0.251
Lactate (mmol/L)	2.1(1.4,3.0)	2.3(1.4,3.1)	0.668
CD3+%	57.4(56.7,58.7)	58.7(57.9,61.7)	0.023*
CD4+%	31.8(30.1,32.2)	21.9(5.5,26.8)	0.001**
CD8+%	23.0(17.6,23.0)	23.1(21.7,24.1)	0.056
CD19+%	17.9(13.7,17.9)	18.4(12.3,23.8)	0.048*
NK%	23.7(16.0,23.7)	19.4(10.6,19.4)	0.019*
Antibiotic exposure before BALF sampling
β-lactams, n (%)	35(71.4)	12(52.2)	0.110
Quinolones, n (%)	14(28.6)	11(47.8)
Antibiotic exposure time (h)	12(8,18)	9(8,10)	0.025*

Notes: Data are presented as n (%) for categorical variables and as median (IQR) or mean ± SD for continuous variables, as appropriate. P values were calculated using the chi-square test or Fisher’s exact test for categorical variables and the Mann–Whitney U-test for continuous variables. *p < 0.05; **p < 0.01.

Comparative analysis of baseline characteristics revealed no significant differences in age, gender, distribution of underlying diseases, complete blood count, C-reactive protein (CRP), procalcitonin (PCT), or arterial blood gas analysis between the two groups. However, the AP group exhibited significantly higher ratios of CD4⁺ and NK cells compared to the CAP group (P=0.001, P=0.019), while the CD3⁺ and CD19⁺ cell ratio was significantly lower in the AP group (P =0.023, P=0.048). All patients in both groups had a history of antibiotic exposure prior to BALF collection, including β-lactams and quinolones. The duration of antibiotic exposure before BALF collection was significantly longer in the AP group than in the CAP group (median 12 vs 9 h, P=0.025).

The Composition of the Lower Respiratory Tract Microbiota in AP and CAP Patients

We first characterized the overall composition of the lower respiratory tract microbiota in both AP and CAP patients. Differences emerged at the genus and species levels (Figure 1A and B).

Figure 1.

Microbial composition at genus and species level in CAP and AP groups. (A) Stacked bar chart showing the top 20 genera by relative abundance in the two groups. (B) Stacked bar chart showing the top 20 species by relative abundance in the two groups.

At the genus level, the CAP group exhibited a community strongly dominated by Acinetobacter, which accounted for an average relative abundance of 32.88%, followed by Klebsiella (9.96%) and Stenotrophomonas (8.89%). In contrast, the AP group showed a more evenly distributed profile of dominant genera. Corynebacterium was the most abundant genus in AP patients (11.65%), followed by Klebsiella (9.08%), Pseudomonas (8.90%), and the oral-associated bacterium Rothia (6.72%). Notably, the dominant genus in CAP, Acinetobacter, constituted only 3.85% of the microbiota in the AP group.

These distinct patterns were also evident at the species level. In the AP group, the top five most abundant species were Corynebacterium striatum (11.92%), Pseudomonas aeruginosa (9.32%), Klebsiella pneumoniae (7.70%), Rothia mucilaginosa (7.08%), and Enterococcus faecium (6.12%). In the CAP group, the profile was led by Klebsiella pneumoniae (15.14%) and Corynebacterium striatum (11.67%), followed by Pseudomonas aeruginosa (6.31%) and Staphylococcus haemolyticus (6.24%).

To further assess the community overlap, we performed a Venn diagram analysis. At the genus level, Venn diagram analysis showed partial overlap of detected genera and species between AP and CAP groups. The CAP group harbored a larger number of unique genera (90) compared to the AP group (51) (Figure S1A). At the species level, 77 species were shared, while the AP group contained more unique species (178) than the CAP group (161) (Figure S1B).

Microbiota Differences Between AP and CAP Patients

To characterize the microbial distinctions between AP and CAP, we analyzed the diversity and composition of the lower respiratory tract microbiome in both patient cohorts.

First, we assessed the within-sample microbial richness and diversity using alpha diversity indices. Although no significant differences were observed in the Chao1 and Ace indices (Figures 2A and B), the Shannon index (P=0.008) and Simpson index (P=0.0068) revealed that the diversity of the lower respiratory tract microbial community in the AP group was significantly higher than that in the CAP group (Figure 2C and D). Subsequently, beta diversity was utilized to compare the overall community structures between the two groups through principal component analysis (PCA) and principal coordinate analysis (PCoA). Overall microbial community structure was not significantly different between the two groups based on the original data matrix (PERMANOVA, R²=0.022, P=0.064) and Bray‑Curtis dissimilarity (PERMANOVA, R²=0.022, P=0.066) (Figure 2E and F).

Figure 2.

Diversity of lower respiratory tract microbiome among CAP and AP groups. (A) Alpha diversity analysis based on Chao1 index. (B) Alpha diversity analysis based on ACE index. (C) Alpha diversity analysis based on Shannon index. (D) Alpha diversity analysis based on Simpson index. Beta diversity of lower respiratory tract microbiome in BALF samples from AP and CAP groups. (E) Principal component analysis (PCA) based on the original data matrix. (F) Principal coordinate analysis (PCoA) based on Bray–Curtis dissimilarity.

To identify the specific taxa responsible for these differences, we performed a differential abundance analysis. At the genus level, the AP group showed nominally higher relative abundances of the oral-associated genera Streptococcus (P=0.032, FDR-adjusted P=0.22) and Rothia (P=0.039, FDR-adjusted P=0.22), whereas Acinetobacter (P=0.002, FDR-adjusted P=0.032) and Stenotrophomonas (P=0.044, FDR-adjusted P=0.22) showed nominally lower relative abundances in the AP group than in the CAP group (Figure 3A). After Benjamini-Hochberg false discovery rate correction, only Acinetobacter remained significantly enriched in the CAP group (Figure S2A). At the species level, Veillonella parvula (P=0.048, FDR-adjusted P=0.28) and Staphylococcus haemolyticus (P=0.015, FDR-adjusted P=0.28) showed nominally higher relative abundances in the AP and CAP groups, respectively (Figure 3B). However, neither difference remained statistically significant after Benjamini-Hochberg FDR correction (Figure S2B).

Figure 3.

Comparison of relative microbial abundance of bacteria between CAP and AP groups before FDR correction. (A) Differential relative abundance of the top 20 most abundant genus. (B) Differential relative abundance of the top 20 most abundant species. Statistical significance was evaluated using the Wilcoxon rank-sum test. Asterisks (*) indicate P < 0.05 prior to FDR correction. *p-value < 0.05; ** p-value <0.01. Only Acinetobacter remained significantly enriched in the CAP group after FDR correction.

Correlation of Differential Microbiota with Clinical and Immunological Indices

To investigate the clinical relevance of the observed microbial alterations, Spearman correlation analysis was performed to evaluate the associations between the lower respiratory tract microbiota and clinical parameters in patients, with a focus on taxa showing different relative abundances between the AP and CAP groups.

At the genus level, Streptococcus and Rothia, which were nominally enriched in the AP group, exhibited nominally significant associations with selected inflammatory and immune markers. Specifically, Streptococcus abundance showed a nominal positive correlation with lactate (r=0.31, P=0.008), PCT (r=0.28, P=0.021), and NK% (r=0.25, P=0.034), and a nominal negative correlation with CD3⁺% (r=−0.30, P=0.011) and CD8⁺% (r=−0.27, P=0.024). Similarly, Rothia abundance was nominally positively correlated with lactate (r=0.37, P=0.002), and nominally negatively correlated with LYMPH count (r=−0.25, P=0.033) and blood pH (r=−0.24, P=0.043). Genera depleted in the AP group also displayed correlative patterns. Acinetobacter abundance was nominally positively correlated with CD3⁺% (r=0.35, P=0.003), whereas Stenotrophomonas showed a nominal positive correlation with MONO (r=0.26, P=0.031) (Figure 4A). At the species level, Staphylococcus haemolyticus, nominally depleted in AP patients, had nominal positive correlations with blood pH (r=0.30, P=0.012) and CD19⁺% (r=0.25, P=0.040), and nominal negative correlations with NK% (r=−0.30, P=0.016) and blood PaCO₂ (r=−0.29, P=0.016) (Figure 4B). Nevertheless, none of these correlations retained statistical significance after FDR correction for multiple testing (all FDR‑adjusted P>0.05).

Figure 4.

Associations between clinical indicators and bacteria relative abundances in the overall cohort before FDR correction. (A) Heatmap illustrating nominal correlation coefficients between the top 20 most abundant genera and clinical indicators. (B) Heatmap illustrating nominal correlation coefficients between the top 20 most abundant species and clinical indicators. Statistical significance was evaluated using Spearman correlation analysis with the R package “cor.test”. Asterisks (*) indicate P < 0.05 prior to FDR correction. * p-value < 0.05; ** p-value < 0.01; *** p-value < 0.001. None of these correlations retained statistical significance after FDR correction for multiple testing (all FDR‑adjusted P>0.05).

Microbiome Characteristics Between Survival and Non-Survival AP Patients

To further explore the relationship between microbial features and patient outcomes, a detailed analysis of clinical parameters and microbiota was conducted between survival and non-survival within the AP group.

Patients in the non-survival group were significantly older than those in the survival group (P=0.03), suggesting that advanced age may be a risk factor for mortality in AP patients (Table 2). APACHE II scores were significantly higher in non-survivors than in survivors (P=0.014). Furthermore, the non-survival group exhibited significantly lower percentages of multiple lymphocyte subsets, including CD3⁺% (P=0.044), CD4⁺% (P=0.027), CD8⁺% (P=0.002), and NK% (P=0.003), while the CD19⁺% was markedly elevated (P<0.0001). These findings indicate that advanced age, APACHE II scores, and altered lymphocyte subset profiles were associated with poor prognosis in AP patients.

Table 2.

Characteristics and Clinic Parameters of the Non-Survival and Survival Group in AP Patients

Characteristics	Non-survival (n=26)	Survival (n=23)	P-value
Age	80.65±9.04	74.48±8.63	0.03*
Gender, n (% male)	21(80.77)	18(78.26)	1
Comorbidities
Hypertension, n (%)	17(65.38)	13(56.52)	0.57
Diabetes, n (%)	4(15.38)	7(30.43)	0.30
Coronary heart disease, n (%)	6(23.08)	3(13.04)	0.47
Cerebral infarction, n (%)	8(30.77)	8(34.78)	1.00
Parkinson’s disease, n (%)	3(11.54)	1(4.35)	0.61
Atrial fibrillation, n (%)	4(15.38)	2(8.70)	0.67
Chronic obstructive pulmonary disease, n (%)	1(3.85)	5(21.74)	0.09
APACHE II score	25.62±7.04	20.70 ±6.38	0.014*
Laboratory findings
WBC (×109/L)	11.0(6.0,16.5)	8.4(6.2,11.8)	0.155
Neutrophils (×109/L)	10.1(4.7,14.6)	7.8(5.3,10.3)	0.176
Monocyte (×109/L)	0.5(0.2,0.9)	0.4(0.1,0.6)	0.078
Lymphocytes (×109/L)	0.7(0.4,1.1)	0.4(0.3,0.7)	0.118
CRP (mg/L)	43.6(17.3,86.1)	53.3(38.1,130.4)	0.452
PCT (mg/L)	2.2(0.5,6.3)	1.0(0.5,2.5)	0.249
pH	7.3(7.2,7.4)	7.3(7.3,7.4)	0.609
PaO2 (kpa)	10.9(9.2,17.6)	15.8(12.0,19.5)	0.029*
PaCO2 (kpa)	5.8(4.8,8.1)	5.5(4.8,8.3)	0.928
Lactate (mmol/L)	2.4(1.5,3.5)	1.8(1.3,2.9)	0.189
CD3+%	56.4(53.8,61.3)	59.0(54.2,63.0)	0.044*
CD4+%	31.0(28.8,33.2)	33.1(31.1,37.1)	0.027*
CD8+%	22.4(13.3,22.4)	24.0(23.4,27.3)	0.002**
CD19+%	21.8(16.6,22.2)	11.7(10.7,14.4)	<0.001***
NK%	20.7(14.5,21.0)	28.4(20.6,30.2)	0.003**

Notes: Data are presented as n (%) for categorical variables and as median (IQR) or mean ± SD for continuous variables, as appropriate. P values were calculated using the chi-square test or Fisher’s exact test for categorical variables and the Mann–Whitney U-test for continuous variables. *p < 0.05; **p < 0.01; ***p < 0.001.

No significant differences were observed in alpha or beta diversity between non-survivors and survivors in AP group (Figure S3). Gemella exhibited higher relative abundance compared with survivors, although the difference did not reach statistical significance after Benjamini‑Hochberg false discovery rate (FDR) correction (P=0.014, FDR-adjusted P=0.27; Figure 5A). Because Gemella abundance was zero-inflated, Gemella was analyzed both as relative abundance and as detection status (detected vs undetected; detected defined as relative abundance > 0). Within the AP cohort (n = 49), Gemella was detected in 8 of 26 non-survivors (30.8%) and in 1 of 23 survivors (4.3%); the detection rate was significantly higher in non-survivors (Fisher’s exact test, P=0.026; Table S1). Spearman correlation analysis further suggested that Gemella abundance was nominally negatively correlated with CD3+% (r=−0.34, P=0.017, FDR‑adjusted P=0.68) and nominally positively correlated with CD19+% (r=0.30, P=0.038, FDR‑adjusted P=0.68) (Figure 5B). However, neither difference remained statistically significant after Benjamini-Hochberg FDR correction.

Figure 5.

Exploratory analyses of microbial relative abundances and their correlations with clinical indicators in the AP group before FDR correction. (A) Nominally differential relative abundance of top 20 most abundant genus between Survival and Non-survival subgroup in the AP group before FDR correction. (B) Heatmap of Spearman correlation between top 20 most abundant genera and clinical indicators in the AP group before FDR correction. Asterisks (*) indicate P < 0.05 prior to FDR correction. * p-value < 0.05; ** p-value < 0.01. None of these correlations retained statistical significance after FDR correction for multiple testing (all FDR‑adjusted P>0.05).

Because Gemella was the only genus among the top 20 genera to show a nominal difference between non-survivors and survivors based on the unadjusted analysis, a exploratory analysis was performed to assess the association between Gemella detection and in-hospital mortality. Because Gemella was detected in only nine AP patients, including eight non-survivors and one survivor, Firth’s penalized logistic regression was used to reduce small-sample and sparse-data bias (Table 3). In the unadjusted Firth penalized logistic regression model, Gemella detection was associated with higher odds of in-hospital mortality (OR 6.892, 95% CI 1.360–69.062; P=0.018). After adjustment for APACHE II score in the primary model, the magnitude of the association was attenuated and was no longer statistically significant (adjusted OR 4.708, 95% CI 0.859–48.395; P=0.076). In the alternative model adjusted for age, Gemella detection remained positively associated with mortality (adjusted OR 7.407, 95% CI 1.374–77.621; P=0.018). Conventional maximum-likelihood logistic regression yielded estimates in the same direction (Table S2). The unadjusted OR was 9.778 (95% CI 1.116–85.648; P=0.039), the APACHE II-adjusted OR was 6.735 (95% CI 0.727–62.362; P=0.093), and the age-adjusted OR was 10.919 (95% CI 1.165–102.346; P=0.036). In an influence analysis excluding the single Gemella-positive survivor, the Firth estimate increased to an OR of 13.563, but its confidence interval was extremely wide (95% CI 1.406–1820.642; P=0.020) (Table S2). This marked change indicated that the magnitude and precision of the association were sensitive to an individual observation. Overall, although the direction of association was positive across the fitted models, the estimates were imprecise and depended on the selected adjustment variable and the sparse distribution of Gemella detection. These findings warrant validation in a larger-sample cohort.

Table 3.

Firth Penalized Logistic Regression Analyses of the Association Between Gemella Detection and in-Hospital Mortality Among Patients with Aspiration Pneumonia (n = 49)

Model	Predictor	OR (95% CI)	P value
Firth: unadjusted	Gemella detected	6.892 (1.360–69.062)	0.018*
Firth: adjusted for APACHE II	Gemella detected	4.708 (0.859–48.395)	0.076
Firth: adjusted for age	Gemella detected	7.407 (1.374–77.621)	0.018*

Notes: Outcome was in-hospital mortality (coded as 1 = death and 0 = survival). Gemella detection was defined as abundance > 0 (coded as 1 = detected and 0 = undetected). *p < 0.05.

Abbreviations: OR, odds ratio; CI, confidence interval; APACHE II, Acute Physiology and Chronic Health Evaluation II.

Discussion

In this study, mNGS was employed to systematically analyze the composition and diversity of the lower respiratory tract microbiome in BALF samples from patients with AP and CAP. Furthermore, the associations between differentially abundant microbiota and patients’ immune phenotypes as well as clinical outcomes were investigated.

We first compared the diversity and composition of the lower respiratory tract microbiota between the two patient groups. It was found that patients with AP exhibited significantly higher alpha diversity (Shannon index and Simpson index) compared to those with CAP. This observation appears to contradict the prevailing view in previous studies that respiratory microbial diversity is generally reduced in AP.^18,22 The discrepancy observed in the present study may be attributed to the following reasons: First, AP is often caused by a single, substantial introduction of oropharyngeal or gastroesophageal reflux contents, which inherently carry a large number of anaerobic commensal bacteria.²³ This explanation is supported by the nominal enrichment of oral-origin microbiota, such as Streptococcus and Rothia, detected in the AP group in this study. In contrast, CAP is typically acquired via microaspiration or aerosol transmission, with a pathogen spectrum concentrated in a limited number of facultative anaerobes such as Streptococcus pneumoniae and Haemophilus influenzae, resulting in inherently lower initial diversity.²⁴ Secondly, differences in the timing of antibiotic exposure may have amplified the observed disparities. Patients with CAP typically seek medical attention early, and all CAP patients included in this study were admitted to the ICU and had already been subjected to broad-spectrum antibiotic therapy prior to sample collection. In contrast, AP patients often experience recurrent microaspiration during hospitalization, leading to disease progression and a relatively delayed initiation of antibiotic treatment.²⁵ Currently, there is no unified clinical diagnostic criterion for aspiration pneumonia,²⁶ and variations in sampling methods and detection techniques across previous studies may also influence the results of the Shannon and Simpson indices.²⁷

Although no significant separation in beta diversity was observed between the AP group and the CAP group (P>0.05) in this study, suggesting that the overall microbial structure in both groups still belongs to a “pneumonia-type” community. However, exploratory analysis of the relative abundance of differentially abundant taxa between the AP and CAP groups may help characterize microbiota-level differences between these two cohorts. First, Streptococcus and Rothia, which were nominally enriched in the lower respiratory tract of AP patients, are both core commensals of the oropharyngeal microbiota. This finding aligns with the pathological model of “overt aspiration,” wherein oropharyngeal microorganisms are directly disseminated into the lower airways. In a study by Akata et al involving 16S rRNA sequencing of BALF from 77 pneumonia patients, the detection rate of oral streptococci in BALF was significantly higher in patients with high aspiration risk than in those without aspiration risk (31.0% vs 14.7%, P=0.009).²⁸ Another study analyzing lower respiratory tract specimens from 17 patients with aspiration pneumonia also identified a high relative abundance of Streptococcus and Rothia genera by both culture and 16S rRNA sequencing,²⁹ supporting the pathway linking “oropharyngeal-aspiration” to enrichment of Streptococcus and Rothia. Notably, detection of oropharyngeal flora in AP patients may signal aspiration events; nevertheless, potential contamination arising during bronchoscopy must also be ruled out. In addition, the CAP cohort in this study consisted of critically ill ICU patients, in whom prior antibiotic exposure and the ICU microbial ecology may shift the respiratory microbiome toward healthcare-associated taxa.³⁰ Consistent with this hypothesis, Acinetobacter was significantly enriched in the CAP group. Thus, the predominance of Acinetobacter and other healthcare-associated taxa within the CAP group may therefore reflect ICU-related exposure or antibiotic-driven selection, rather than features typical of community-acquired pneumonia.

Spearman correlation analysis was further performed to explore relationships between dominant bacterial microbiome in different groups and peripheral immune markers. This study examined the correlations between differentially abundant bacterial taxa and significantly altered peripheral immune markers. The relative abundance of Streptococcus was nominally negatively correlated with peripheral blood CD3⁺% and nominally positively correlated with NK%. In line with these observations, nominal enrichment of Streptococcus in the AP group was accompanied by a nominal decrease in CD3⁺% and an increase in NK%. This phenomenon may suggest the potential rapid proliferation of orally derived commensal bacteria in the lower respiratory tract following substantial aspiration. Previous studies have hypothesized that bacterial cell wall components such as lipoteichoic acid (LTA) and peptidogly may participate in TLR2/4-related immune signaling.^31–33 However, whether such mechanisms contributed to the immune-cell patterns observed in the present study cannot be determined from our cross-sectional data. In the CAP group, the relative abundance of the enriched genus Acinetobacter was nominally positively correlated with CD3⁺% levels, yet displayed no significant correlation with NK% levels. Prior clinical data demonstrated that patients within the CAP group presented higher CD3⁺% and lower NK% levels. This nominal positive correlation between Acinetobacter abundance and CD3⁺ percentage may reflect differences in host immune status or ICU‑related microbial exposure; however, this finding is exploratory and requires further validation. This may also partly explain the relatively modest variation in NK% levels observed among CAP patients with Acinetobacter enrichment. At the species level, correlation analysis showed that Veillonella parvula, a commensal oral bacterium nominally enriched in the AP group, exhibited poor correlation with the tested immune markers. In contrast, Staphylococcus haemolyticus, which was nominally enriched in the CAP group, exhibited a nominal negative correlation with NK%. However, this association did not remain statistically significant following FDR correction. Staphylococcus haemolyticus is generally regarded as a mucosal colonizer of the respiratory tract. While several studies have reported rising incidence of S. haemolyticus as a causative agent of CAP,³⁴ its detection in our cohort may reflect ICU‑related colonization or healthcare‑associated acquisition rather than genuine community‑acquired infection. Unlike other coagulase-negative staphylococci, Staphylococcus haemolyticus has been shown to evade host immune clearance (such as NK cells) through biofilm formation. Additionally, by altering surface protein expression, it reduces the likelihood of phagocytosis by neutrophils and macrophages.³⁵ The biological significance of Staphylococcus haemolyticus detection in this cohort remains uncertain and may reflect colonization or healthcare-associated exposure rather than causative infection.

To further investigate the correlation between the lower respiratory tract microbiota and prognosis in patients with AP, we specifically compared clinical indicators and lower respiratory tract microbiota changes between the non-survival group and survival group of AP patients. The study revealed no significant differences in community diversity (alpha diversity and beta diversity) between the non-survival and survival subgroups of AP patients, a finding that diverges from previous studies. A 2024 cohort study on pediatric acute lower respiratory tract infections reported no significant difference in alpha diversity (P>0.05) but significant separation in beta diversity between survival and non-survival groups.³⁶ In contrast, another cohort study involving adults with severe pneumonia observed higher beta diversity in non-survivors.³⁷ These discrepancies may be attributed to differences in the diseases studied and cohort sizes. The lack of significant differences in microbial diversity between the non-survival and survival groups in the present study may be due to the early timing of sample collection, during which changes in microbial diversity might not yet have become apparent. The non-survival group exhibited a significantly lower T-cell immune level, while the proportion of CD19⁺ B cells involved in humoral immunity was markedly elevated, suggesting that differences in immune response pathways may lead to divergent clinical outcomes in AP patients.

Exploratory Spearman correlation analysis also revealed nominal associations between Gemella and peripheral immune markers, showing a nominal negative correlation with CD3+% and a nominal positive correlation with CD19+% (P<0.05). Unfortunately, the correlations were not statistically significant after FDR correction. Additionally, Gemella was more frequently detected among non-survivors than among survivors, and the unadjusted Firth regression indicated a positive association with in-hospital mortality. However, adjustment for APACHE II score attenuated the association, which was no longer statistically significant. Although the age-adjusted model retained statistical significance and conventional logistic regression produced estimates in the same direction, all models yielded wide confidence intervals. The influence analysis further demonstrated the limitations imposed by the sparse data. Exclusion of the single Gemella-positive survivor increased the estimated OR but produced an extremely wide confidence interval, reflecting complete or near-complete separation rather than increased certainty. Therefore, this result does not strengthen the evidence for an association; instead, it indicates that the magnitude of the estimate was strongly influenced by an individual observation. Taken together, Gemella detection may reflect aspiration burden, transfer of oral microbiota, illness severity, host immune status, or other unmeasured factors. However, the present analysis cannot determine whether Gemella has a direct role in adverse outcomes. Given the small number of Gemella-positive patients, residual confounding, model dependence, and lack of external validation, Gemella should be regarded only as a hypothesis-generating microbial feature and not as an independent prognostic factor or clinically useful biomarker. Larger prospective multicenter studies are required for validation. Gemella is a genus of Gram-positive facultative anaerobic cocci commonly found in the normal microbiome of the human oral cavity, upper respiratory tract, and gastrointestinal tract.³⁸ As opportunistic pathogens, Gemella may act as the etiological agents of lung abscess. Although studies on Gemella in aspiration pneumonia are limited, recent research on the respiratory microbiota of hospitalized COVID-19 patients has identified Gemella as a potential biomarker for distinguishing disease severity.³⁹ Another study used pleural fluid mNGS to clarify the microbial spectrum of aspiration pneumonia, among which oropharyngeal flora were the most common, including Gemella.⁴⁰ These observations are hypothesis-generating but do not establish causality or independent prognostic value for Gemella.⁴¹

This study has several limitations. First, it was a single-center retrospective exploratory study with a relatively small sample size, which limited the statistical power, especially for survival and multivariable analyses. The number of Gemella-positive cases was small. Therefore, its association with in-hospital mortality should be considered exploratory. Second, during bronchoscopy, the introduction of oral microorganisms into BALF samples is difficult to completely avoid. However, BALF collection was performed according to a standardized protocol in all patients to minimize potential bias caused by procedural differences. Finally, the observed associations do not establish causality and require validation in larger prospective multicenter studies.

Conclusion

In this single-center retrospective exploratory study of critically ill patients, AP was associated with higher Shannon and Simpson diversity in BALF, whereas overall microbial community composition did not differ significantly from that of CAP. Unadjusted analyses suggested higher representation of selected oral-associated taxa in AP, but most taxon-level differences were not retained after Benjamini–Hochberg FDR correction; only Acinetobacter remained more abundant in the CAP group. Associations between bacterial taxa and peripheral immune variables were nominal and did not remain statistically significant after FDR correction. Gemella was detected more frequently among AP non-survivors, but its association with mortality was no longer statistically significant after adjustment for APACHE II score and was sensitive to the sparse exposure-outcome distribution and individual observations. Therefore, the Gemella finding should be regarded as hypothesis-generating rather than evidence of an independent prognostic factor or clinically useful biomarker. Larger prospective multicenter studies are required for validation.

Funding Statement

This study was supported by the Science and Technology Commission of Shanghai Municipality (2024ZDXK0012) and Center of Emergency and Critical Medicine, Jinshan Hospital of Fudan University, Research Center for Chemical Injury, Emergency and Critical Medicine of Fudan University, Key Laboratory of Chemical Injury, Emergency and Critical Medicine of Shanghai Municipal Health Commission (82272243, 22Y11900800, JSZK2023A01).

Abbreviations

AP, Aspiration Pneumonia; ARDS, Acute Respiratory Distress Syndrome; CAP, Community-Acquired Pneumonia; CD3⁺%, CD3 T-Lymphocyte Percentage; CD4⁺%, CD3CD4 T-Helper Cell Percentage; CD8⁺%, CD3CD8 Cytotoxic T Cell Percentage; CD19⁺%, CD19 B-Lymphocyte Percentage; COPD, Chronic Obstructive Pulmonary Disease; CRP, C-Reactive Protein; ICU, Intensive Care Unit; mNGS, Metagenomic Next-Generation Sequencing; NK%, Natural Killer Cell Percentage; PaCO₂, Partial Pressure of Carbon Dioxide; PaO₂, Partial Pressure of Oxygen; pH, Potential of Hydrogen; PCT, Procalcitonin; VAP, Ventilator-Associated Pneumonia; WBC, White Blood Cell Count; OR, odds ratio; CI, confidence interval; APACHE II, Acute Physiology and Chronic Health Evaluation II.

Data Sharing Statement

The datasets presented in this study can be found in online repositories. The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA033525). The dataset is publicly available at: https://ngdc.cncb.ac.cn/gsa/s/J0dy437q.

Ethics Approval and Consent to Participate

The study was performed in accordance with the Declaration of Helsinki and was approved by the ethics committee of the Jinshan Hospital of Fudan University (approval ID: JIEC 2024-S99). Written informed consent was obtained from all participants or their legally authorized representatives. All data were de-identified and anonymously processed.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

CZ and MS are employees of Dinfectome Inc. The authors declare no other competing interests in this work.