Altered microbiota of the lower respiratory tract and its association with COVID-19 severity analysed by metagenomics and metatranscriptomics

Abstract

The interaction between gut and oropharyngeal microbiota plays a significant role in the viral infections like SARS-CoV-2, but role of the lower respiratory tract microbiota remains unclear. Our study utilized metatranscriptomics and metagenomics to analyze the microbial composition of bronchoalveolar lavage fluid and sputum samples from 116 COVID-19 patients, categorized into mild, severe, and critical groups. Our analysis revealed significant differences in viral genotypes across disease stages. As disease severity increased, the Chao index also rose. The mild group was predominantly dominated by Firmicutes, while the severe group showed an increase in Bacteroidetes. The critical group was characterized by a higher abundance of Actinobacteria and Proteobacteria. Notably, the abundance of Streptococcus and Rothia decreased as the disease progressed. Additionally, the Shannon index correlated with mortality risk, while the Chao index was associated with ICU admission, mechanical ventilation, and patient survival. These findings highlight the strong link between microbial composition and COVID-19 severity, providing valuable insights for assessing disease progression.

Mechanistic modeling of enhancer-RNA sheds light on the role of transcription factors and their interactions in the context of estrogen-induced transcriptional program in breast cancer.

Introduction

The outbreak of SARS-CoV-2 in 2019, which led to the global COVID-19 pandemic, has prompted extensive research on the virus’s impact on the respiratory system^1,2. Studies show that in healthy individuals, the lungs support a small but diverse community of beneficial bacteria³, which play a crucial role in maintaining respiratory health^4,5. However, in COVID-19 patients, the microbial composition in the lower respiratory tract often shifts to resemble that of the upper respiratory tract, indicating a disruption in the normal microbial balance^6,7.

The upper respiratory tract consists of the nostrils, nasal cavity, sinuses, nasopharynx, mouth, oropharynx, and larynx⁸. Research shows that the mucosal surfaces of these areas host distinct microbial communities, which play a key role in interacting with and regulating the host’s immune system⁹. Changes in the composition of oropharyngeal microbes have been linked to the severity of COVID-19, as they influence the body’s inflammatory response^10–12. Understanding microbial dynamics in the lower respiratory tract is critical for improving outcomes in COVID-19-related lung infections.

Infections with Influenza A and SARS-CoV-2 primarily alter the abundance and composition of local bacterial communities¹³. Studies show that patients infected with these viruses experience an increase in opportunistic pathogens and symbiotic bacteria, including Pseudomonas, Enterobacteriaceae, and Acinetobacter, within their lung microbiota^14–18. Viral infections in the upper respiratory tract can also disrupt the microbiota in both the upper and lower respiratory tracts, increasing susceptibility to secondary bacterial infections^18,19. This disruption contributes to higher morbidity and mortality rates^20,21. Therefore, analyzing microbial composition and function is crucial for understanding the mechanisms behind these changes and the interactions between SARS-CoV-2, microbial communities, and the host immune system.

In our study, we analyzed 116 samples from COVID-19 patients, including 70 bronchoalveolar lavage fluid (BALF) and 46 sputum samples, using both metatranscriptomic and metagenomic sequencing. We observed distinct shifts in microbiome composition across disease severity, with significant differences at both the genus and species levels. In critically ill patients, Actinobacteria and Proteobacteria were predominant, while the relative abundance of Streptococcus and Rothia declined as severity increased. The Chao index varied significantly among severity groups and was associated with ICU admission, mechanical ventilation, and survival status. In contrast, the Shannon index was closely linked to mortality risk. These findings highlight a strong relationship between microbial diversity and COVID-19 severity, offering valuable insights to inform future diagnostic, preventive, and therapeutic strategies.

Materials and methods

Patients and clinical samples

Between November 1, 2022, and January 29, 2023, 188 COVID-19 patients with confirmed pneumonia were recruited at West China Hospital of Sichuan University, following the inclusion/exclusion criteria and diagnostic protocols outlined in the 7th edition of the National Health Commission guidelines²². Eligible participants were adults aged 18 or older. Exclusions included 22 cases with substandard sample quality, 12 with duplicate samples, and 2 with non-lower respiratory tract samples. Patients under 18 and samples without RNA sequencing were also excluded. Ultimately, 116 samples were included in the study, consisting of 46 sputum samples and 70 alveolar lavage fluid samples (see Table 1 and Supplementary Data 1).

Table 1.

Clinical characteristics of the study cohorts

	Samples from all patients (n = 116)	BALF samples (n = 70)	Sputum sample (n = 46)
Age
< 60, n (%)	34 (29.3%)	20 (28.6%)	14 (30.4%)
≥ 60, n (%)	82 (70.7%)	50 (71.4%)	32 (69.6%)
Sex
Female, n (%)	29 (25.0%)	18 (25.7%)	11 (23.9%)
Male, n (%)	87 (75.0%)	52 (74.3%)	35 (76.1%)
COVID-19
Mild	27 (23.3%)	10 (14.3%)	17 (37.0%)
Severe	24 (20.7%)	5 (7.1%)	19 (41.3%)
Critical	65 (56.0%)	55 (78.6%)	10 (21.7%)
ICU admission	65 (56.0%)	54 (77.1%)	11 (24.0%)
Mechanical ventilation	70 (60.3%)	59 (84.3%)	11 (24.0%)
Tracheal intubation	62 (53.4%)	54 (77.1%)	8 (17.4%)
Mortality	46 (39.7%)	36 (51.4%)	10 (21.7%)
History of infection	8 (6.9%)	7 (10.0%)	1 (2.2%)
History of treatment
Antibiotic treatment	74 (63.4%)	56 (80.0%)	18 (39.1%)
Antifungal treatment	44 (37.9%)	36 (51.4%)	8 (17.4%)
Antitubercular treatment	6 (5.2%)	6 (8.6%)	0 (0.0%)
Sequencing type
Metatranscriptome sequencing	116 (100%)	70 (100%)	46 (100%)
Metagenomic sequencing	116 (100%)	70 (100%)	46 (100%)

Participants in this study were hospitalized at West China Hospital with laboratory-confirmed COVID-19. Clinical severity was classified into three groups: mild, severe, and critical, based on national guidelines. Mild cases were characterized by non-severe symptoms and normal chest imaging. Severe cases included fever, respiratory compromise, and radiological pneumonia. Critical cases met at least one of the following criteria: (1) respiratory rate ≥30 breaths/min, (2) resting SpO2 ≤ 93%, (3) PaO₂/FiO₂ ratio ≤300 mmHg, (4) need for mechanical ventilation, (5) circulatory shock, or (6) multi-organ dysfunction requiring intensive care. All patients provided written informed consent prior to sample collection. The study was conducted in accordance with the ethical standards of West China Hospital, Sichuan University.

Processing of samples and DNA extraction

Sputum and alveolar lavage fluid samples, ranging from 1.5 to 3 mL, were collected following standard procedures. A 450 μL portion of each sample was mixed with 11.5 μL of 1.0% saponin (final concentration: 0.025%) and thoroughly mixed for 15 s. The mixture was then equilibrated at 25 °C for 5 min according to standard protocols. Next, 75 μL of the mixture was extracted and used as the reaction solution, which was mixed again for 15 s and incubated at 37 °C for 10 min in a water bath. After centrifugation at 18,000 g for 5 min, 450 μL of the supernatant was carefully pipetted, leaving 70-80 μL of liquid at the bottom. The remaining solution was mixed with 800 μL of PBS, vortexed, and centrifuged again at 18,000 g for 5 min. Following this, 800 μL of the supernatant was aspirated, and 370 μL of TE buffer was added to the lysate, followed by vortexing to solubilize the nucleic acids. Next, Lyticase (7.2 μL, RT410-TA, TIANGEN) was used for cell wall digestion via enzymatic lysis. Mechanical disruption was performed with 250 μL of 0.5 mm glass beads, followed by thorough agitation. Finally, 300 μL of the lysate was isolated for further analysis.

Construction of DNA library

DNA was extracted from samples of 116 patients using the TIANMicrobe Magnetic Bead Pathogenic Microbial DNA Extraction Kit (NG550-01), following the manufacturer’s instructions. The extracted DNA was then processed through enzymatic digestion, terminal repair, ligation, and PCR to construct sequencing libraries (BGI). Fragment sizes, approximately 300 bp, were confirmed using the Agilent 2100 Bioanalyzer, and library concentrations were measured with the Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific). The library concentration of each sample was adjusted according to the required sequencing depth. The libraries were then pooled and subjected to a circular reaction to form a circular DNA structure. DNA nanoball (DNB) synthesis was achieved through rolling circle amplification (RCA), and the DNBs were loaded onto a microfluidic chip for sequencing on the BGIseq platform. Post-sequencing, we deduplicated the samples using barcodes and ensured data quality by verifying sequencing depth and conducting quality control.

Construction and sequencing of RNA library

Microbial RNA was purified from 116 clinical specimens of SARS-CoV-2-infected individuals using the QIAamp Viral RNA Mini Kit (Qiagen), following the manufacturer’s guidelines. The RNA was then fragmented, reverse-transcribed, end-repaired, and ligated to adapters. The final library was amplified by PCR. Library quality was assessed with the Agilent 2100 system, and those meeting the standards were sequenced on the MGISEQ-2000 platform using high-throughput methods.

Bioinformatics analyses and statistical analyses

In this study, sequencing data were preprocessed using fastp software to remove low-quality bases and trim adapter sequences²³. The filtered sequences were then aligned to the GRCh38 reference genome using hisat2 (version 2.2.1)²⁴, with any human genome sequences excluded. The remaining sequences were analyzed with Kraken2 software for species identification, utilizing a RefSeq database constructed following the Kraken2 protocol²⁵. Relative abundance data were subsequently imported into R (version 4.2.3) via RStudio for statistical analysis and visualization using the ggplot2 package.

After removing host sequences, BWA-MEM v0.7.18²⁶ was used to align the fastq files to the SARS-CoV-2 reference genome (NC_045512.2), and PCR duplicates were removed using samtools (v1.21)²⁷. Genetic variants were called with Lofreq v2.1.5, and SNPs were annotated and their functional impacts, including missense and nonsense mutations, were predicted using SnpEff 5.2e^28,29. COVID-19 lineage analysis was performed with Pangolin v4.3.1. Phylogenetic analysis was conducted using mafft v7.525, IQtree v2.3.6, and iTol v7^30–32.

α-diversity was assessed in mothur³³, based on metric data distribution, while beta diversity (Bray-Curtis dissimilarity) was calculated with the “vegan” package in R. The Wilcoxon rank-sum test was applied to assess significant differences between two groups, and the Kruskal-Wallis test was used for comparisons involving three groups. Principal Coordinate Analysis (PCoA) based on Bray-Curtis distance metrics was performed in R to generate bi-dimensional plots³⁴.

The permutational analysis of variance (PERMANOVA) test was performed using the adonis function in the vegan package. Visualizations were generated with the Python packages Seaborn, Matplotlib, and Pandas, along with the R package ggplot2. To identify microbial biomarkers distinguishing different groups, we conducted Linear Discriminant Analysis (LDA) Effect Size (LEfSe) analysis³⁵. LDA score greater than 3 was considered indicative of significant differences in microbial abundance. Additionally, a heatmap of metagenomic biomarkers (at both the genus and species levels) was created using ggplot2 in R.

To assess the association between clinical factors and lower respiratory tract microbial diversity, we performed multivariate linear regression analyses. To compare bacterial biomarkers and clinical outcomes between groups, we first used the Wilcoxon rank-sum test to generate initial p-values. These were subsequently adjusted using the Benjamini-Hochberg method to control for the false discovery rate. The relative abundance between non-treated and treated groups was represented by the log₂ fold change. Taxa were considered upregulated when FDR values were below 0.05 and Log2FoldChange exceeded 1, and downregulated when FDR values were below 0.05 and Log2FoldChange was below -1. A volcano plot was generated using GraphPad Prism 8.0, based on the FDR values and Log2FoldChange.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Results

Clinical features of COVID-19 patients

This study collected 116 samples from COVID-19 patients, including 70 bronchoalveolar lavage fluid (BALF) samples and 46 sputum samples (Fig. 1, Table 1 and Supplementary Data 1). The metagenomic sequencing (mNGS) generated an average of 1.92 × 10⁸ sequence reads, with 1.21 × 10⁸ metatranscriptome reads (Supplementary Data 1, 2 and Supplementary Data 2-2). COVID-19 virus detection was successful in 111 of the 116 cases, resulting in an overall detection rate of 95.69% (Supplementary Table 1). Among the BALF samples, 66 of 70 tested positive, yielding a detection rate of 94.29%, while 45 of 46 sputum samples were positive, with a detection rate of 97.83%. The clinical characteristics of the patients are summarized in Fig. 1, with 75% (87/116) being male. The patients were categorized as mild (23.28%, 27/116), severe (20.69%, 24/116), and critical (56.03%, 65/116).

Fig. 1. Overview of the mNGS datasets from BALF and sputum samples of COVID-19 patients.

Summary of sample types and patient information used in this study.

Genotyping and variant detection in COVID-19 patients

Genotyping analysis was performed on 86 complete SARS-CoV-2 genomes from the 116 COVID-19 positive patients. The results showed variation in viral genotype distribution across patients with different clinical severities (Fig. 2A; Supplementary Data 3). In mild cases, eight viral branches were identified, including B (the first discovered), B.1.1.529 (Omicron), BA.5.2.48, BA.5.2.49, DY.1, DY.2, DY.4, and BF.7.14.5 (Omicron variants). In severe cases, the B.1.1.161 (early COVID-19 variant) branch was also detected. Critically ill patients exhibited an expanded viral genotype profile, with the addition of B.1.1.161, BF.7.14.4, and BF.7.14 (Omicron variants), bringing the total number of branches to 10 (Fig. 2A; Supplementary Data 4).

Fig. 2. Genotypic classification, SNP variation, and genetic mutations of COVID-19 patients.

A Distribution of SARS-CoV-2 lineages across mild, severe, and critical patients. The X-axis represents the sample count, and the Y-axis displays the different SARS-CoV-2 lineages. B Phylogenetic tree depicting 86 pathogens. C Heatmap showing 11 gene mutations for each sample. The X-axis corresponds to the samples, and the Y-axis represents the SARS-CoV-2 genes.

While SARS-CoV-2 genotypes were largely similar across patients with different disease severities, variations emerged as the disease progressed, suggesting a potential role of viral evolution in disease progression. To investigate this further, we constructed phylogenetic trees for each strain and analyzed single nucleotide polymorphisms (SNPs). SNPs were detected in all 86 cases (Fig. 2B; Supplementary Data 5). Additionally, mutation distribution was examined across 11 genes, revealing prominent mutations in the spike protein (S) and ORF1ab genes. Although mutation counts varied among patients, no significant differences were observed between mild, severe, and critically ill cases (Fig. 2C).

Species-level microbial diversity differences in lower respiratory tract samples from mild, severe, and critical COVID-19 patients

In this study, we analyzed 70 BALF and 46 sputum samples from 116 COVID-19 patients to evaluate respiratory microbiome diversity across mild, severe, and critical cases. Significant differences in microbial composition were observed at the species level (Fig. 3A). The Chao index increased with disease severity, indicating greater species richness in critically ill patients (Fig. 3B). However, no significant differences were found in the Shannon and Simpson indices at the species level (Fig. 3C, D).

Fig. 3. Species-level microbial diversity characteristics of the lower respiratory tract in COVID-19 patients with varying disease severity.

A–D Comparison of metatranscriptome microbiome α-diversity at the species level among mild (light beige), severe (peach), and critical (dark brown) patient groups based on metatranscriptomic data. A, Observed species count (feature number). B Chao index. C Shannon index. D Simpson index. E Comparison of microbiome β diversity at the species level based on Bray-Curtis dissimilarity index, covering mild, severe, and critical patient groups. F Principal Coordinate Analysis (PCoA) based on Bray-Curtis dissimilarity index. Mild patients are represented by light beige points, severe patients by peach points, and critical patients by dark brown points. R and P values correspond to the results of the ANOSIM test. Statistical significance is indicated as follows: ns (P > 0.05), *(P < 0.05), **(P < 0.01), ***(P < 0.001), and ****(P < 0.0001).

We calculated the Bray-Curtis distance between species in mild, severe, and critically ill patients. Critically ill patients exhibited a significantly greater Bray-Curtis distance compared to mild and severe cases, with no significant difference observed between the mild and severe groups (Fig. 3E). PCoA analysis further revealed that the microbiota abundance in critically ill patients was significantly higher than in the other groups (Fig. 3F). Overall, these findings suggest that the respiratory microbiome in critically ill COVID-19 patients shows greater species-level variability and dispersion than in mild and severe cases.

Next, we analyzed species-level diversity in sputum samples and found no significant differences in alpha diversity among mild, severe, and critically ill patients (Supplementary Fig. 1A–D). However, β-diversity analysis revealed significant differences between mild and critically ill patients, as well as between severe and critically ill patients, indicating notable variations in overall diversity (Supplementary Fig. 1E). PCoA further highlighted distinct microbial differences among these groups (Supplementary Fig. 1F). In BALF samples, alpha diversity at the species level also showed no significant differences across patient groups (Supplementary Fig. 1G–J). However, β-diversity analysis identified significant differences between mild and critically ill patients, reflecting substantial shifts in overall microbial diversity (Supplementary Fig. 1K). Unlike sputum samples, the PCoA plot for BALF did not indicate significant differences in species abundance (Supplementary Fig. 1L).

In addition, we analyzed species-level metagenomic diversity in 70 BALF and 46 sputum samples (Supplementary Fig. 2A). The Chao index increased significantly with disease severity, while the Shannon index decreased, indicating reduced microbial evenness (Supplementary Fig. 2B, C). The Simpson index showed no significant differences (Supplementary Fig. 2D). β-diversity analysis revealed significant differences between mild and critically ill patients, as well as between severe and critically ill patients, highlighting notable variations in overall microbial composition (Supplementary Fig. 2E). PCoA further illustrated distinct microbial differences among these groups (Supplementary Fig. 2F).

Genus-level microbial diversity differences in lower respiratory tract from mild, severe, and critical COVID-19 patients

We further performed a comprehensive analysis of microbiome differences at the genus level among mild, severe, and critical ill COVID-19 patients. The Chao index, used to measure α-diversity, revealed significant differences in species composition across the three groups, with higher values associated with increased disease severity (Fig. 4A, B). In contrast, the Shannon and Simpson indices showed no significant differences (Fig. 4C, D). We also calculated the Bray-Curtis distance at the genus level and found a significant increase in distance between critically ill patients and both mild and severe cases, while no significant difference was observed between mild and severe groups (Fig. 4E). PCoA analysis further demonstrated a significant difference in the absolute abundance of microbiomes between critically ill and mild patients (Fig. 4F). Overall, these findings suggest that the respiratory microbiome of critically ill COVID-19 patients is more diverse and dispersed at the genus level compared to mild and severe cases.

Fig. 4. Genus-level microbial diversity characteristics of the lower respiratory tract in COVID-19 patients with varying disease severity.

A–D Comparison of metatranscriptome microbiome α-diversity at the genus level among mild (light beige), severe (peach), and critical (dark brown) patient groups based on metatranscriptomic data. A Observed genus count (feature number). B Chao index. C Shannon index. D Simpson index. E Comparison of microbiome β diversity at the genus level based on Bray-Curtis dissimilarity index, covering mild, severe, and critical patient groups. F Principal Coordinate Analysis (PCoA) based on Bray-Curtis dissimilarity index. Mild patients are represented by light beige points, severe patients by peach points, and critical patients by dark brown points. R and P values correspond to the results of the ANOSIM test. Statistical significance is indicated as follows: ns (P > 0.05), *(P < 0.05), **(P < 0.01), ***(P < 0.001), and ****(P < 0.0001).

Additionally, we further examined α-diversity in sputum samples at the genera level and found no significant differences among mild, severe, and critically ill patients (Supplementary Fig. 3A–D). However, β-diversity analysis revealed a significant difference between severe and critically ill patients, although overall diversity remained relatively stable (Supplementary Fig. 3E, F). Similarly, in BALF samples, α-diversity at the genera level showed no significant differences across the groups (Supplementary Fig. 3G–J). In contrast, β-diversity analysis of BALF samples indicated a significant difference between mild and critically ill patients, with overall diversity showing notable changes (Supplementary Fig. 3K, L).

We also investigated the metagenomics diversity of 70 BALF samples and 46 sputum samples at the genera level (Supplementary Fig. 4A). Notably, the observed species and the Chao index increased significantly with disease severity, while the Shannon index and Simpson indices did not demonstrate significant differences at the species level (Supplementary Fig. 4B–D). However, β-diversity analysis revealed significant differences between mild and critically ill patients, as well as between severe and critically ill patients, with overall diversity showing notable variations (Supplementary Fig. 4E). The PCoA plot further highlighted distinct differences among these groups (Supplementary Fig. 4F).

Microbial composition in lower respiratory tract from mild, severe, and critical COVID-19 patients

To gain a clearer understanding of microbial distribution across varying severities of COVID-19, we analyzed BALF and sputum samples from the lower respiratory tract. Our results revealed significant shifts in microbial composition among mild, severe, and critically ill patients (Fig. 5 and Supplementary Fig. 5). The detail information at phylum level was shown in Supplementary Table 2 and Fig. 5A. In mild cases, Firmicutes were the dominant microbiota, but their prevalence dramatically decreased to 4.57% in critically ill patients (Fig. 5A). In severe cases, Bacteroidetes were most prominent, accounting for 39.16% of the microbiota, compared to 12.89% in mild cases and 2.91% in critically ill cases (Fig. 5A). Furthermore, among critically ill patients, Actinobacteria and Proteobacteria were the most dominant, making up 59.35% and 31.63% of the microbiota, respectively. In contrast, these groups accounted for 36.8% and 6.83% in mild cases and 9.62% and 5.68% in severe cases.

Fig. 5. Phylum- and genus-level compositional characteristics of the microbiota in COVID-19 patients with varying disease severity.

A Average composition and relative abundance of phylogeny microbial communities in the mild, severe and critical patients with COVID-19. B Average composition and relative abundance of microbial communities in genus patients with mild, severe and critical illness.

Of note, as shown in Fig. 5B and Supplementary Table 3, the proportions of the genera Streptococcus and Rothia decreased progressively with worsening disease severity, with Rothia becoming undetectable in critically ill patients. On the other hand, the genera Corynebacterium and Pantoea were observed only in severe and critically ill patients, comprising 38.50% and 8.13% of the microbiota in critically ill patients, respectively. Additionally, the genera Streptomyces was found exclusively in critically ill patients (Fig. 5B).

To investigate the differences in bacteria, fungi, and viruses across mild, severe, and critical patients, we also analyzed the top 50 genera with the highest relative abundance for each microbe. The Kruskal-Wallis test was used for statistical analysis, with a significance level of p-value < 0.05. Finally, 46 bacterial, 31 fungal, and 10 viral genera showed significant differences. Hierarchical cluster analysis based on their relative abundances generated heatmaps for these differentially abundant genera (Supplementary Fig. 5A, B).

Microbial biomarkers in lower respiratory tract from mild, severe, and critical COVID-19 patients

We conducted a detailed analysis of the microbial community structure associated with the mild, severe, and critical groups using LEfSe, an algorithm designed to identify high-abundance biomarkers and estimate the effect size of each taxon differing among the groups (Fig. 6A, B; Supplementary Table 4 and Supplementary Table 5).

Fig. 6. Microbial community structure characteristics of the lower respiratory tract in COVID-19 patients with varying disease severity.

A, B Microbial biomarkers were identified using the Linear Discriminant Analysis Effect Size (LEfSe) algorithm. The histogram of unique biomarkers with an LDA score > 3.5 is presented, where the length of the bars represents the effect size of significantly different species. A Biomarkers at the genus level. B Biomarkers at the species level. R and P values correspond to the results of the ANOSIM test. Statistical significance is indicated as follows: ns (P > 0.05), *(P < 0.05), **(P < 0.01), ***(P < 0.001), and ****(P < 0.0001).

A total of 30 distinct genera were identified across the groups. In the mild group, 10 potential markers were identified, including Tannerella, Thermobaculum, Granulicatella, Moraxella, Parvimonas, Abiotrophia, Rothia, Porphyromonas, Actinomyces and Streptococcus. The severe group revealed 8 potential markers, such as Mogibacterium, Veillonella, Alloprevotella, Haemophilus, Candida, Leptotrichia, Capnocytophaga and Prevotella. For the critical group, 12 potential markers were identified, including Bacillus, Mycobacterium, Vibrio, Escherichia, Cutibacterium, Pedobacter, Acinetobacter, Klebsiella, Herbaspirillum, Streptomyces, Pantoea and Corynebacterium (Supplementary Table 4). A total of 42 distinct species were identified across the groups. In the mild group, 18 potential markers were identified, including Actinomyces graevenitzii, Streptococcus pneumoniae, Herbaspirillum autotrophicum, Rothia mucilaginosa, and Streptococcus mitis. The severe group revealed 15 potential markers, such as Candida albicans, Streptococcus oralis, Capnocytophaga granulosa, Haemophilus influenzae, and Porphyromonas endodontalis. For the critical group, 9 potential markers were identified, including Corynebacterium striatum, Pantoea endophytica, Pedobacter nutrimenti, Klebsiella pneumoniae, Cutibacterium acnes, Acinetobacter baumanni, Escherichia coli, Shewanella chilikensis, and Human alphaherpesvirus 1 (Supplementary Table 5).

We also drew the heatmap depicting the relative abundance of bacteria biomarkers identitied in LEfSe across mild, severe, and critically ill patients. The results showed that Shewanella chilikensis, Pedobacter nutrimenti, Pantoea endophytica, Klebsiella pneumoniae, Escherichia coli, Cutibacterium acnes, Corynebacterium striatum, Acinetobacter baumannii, and Herbaspirillum autotrophicum were more prevalent in critically ill patients, with other strains being less abundant (Supplementary Fig. 6A). The significant differences in Candida albicans among the groups also showed in Supplementary Fig. 6B. Specifically, there were marked differences in Candida albicans between mild and severe patients, and between severe and critically ill patients, but not between mild and critically ill patients.

Human herpesvirus 1 (HHV-1)³⁶, also known as herpes simplex virus type 1 (HSV-1), is a common human virus that causes infections in various parts of the body, most commonly the oral region. Further statistical (Supplementary Fig. 6C) analysis indicated that HHV-1 differed significantly between mild and severe patients, as well as between mild and critically ill patients, with no significant difference between severe and critically ill patients.

Differential microbial diversity in lower respiratory tract samples across clinical groups

To further explore the relationship between the lower respiratory microbiota and COVID-19 severity, we examined the association between the Shannon index, Chao index and clinical outcomes. The Chao index was significantly associated with ICU admission, use of mechanical ventilation, and patient survival. However, it showed no clear relationship with intubation, underlying infection diseases, or treatments including antibiotics, antifungals, and anti-tuberculosis drugs, nor with patient age (Fig. 7A–I). In contrast, the Shannon index was not significantly associated with ICU admission, mechanical ventilation, intubation, comorbidities, or any of the treatments or age. Notably, however, it was significantly correlated with patient mortality (Fig. 8A–I).

Fig. 7. Correlation analysis of the Chao index with clinical outcomes.

A ICU Admission. B Mechanical Ventilation. C Endotracheal Intubation. D Mortality Rate. E Underlying Diseases. F Antibiotic Treatment. G Antifungal Treatment. H Antitubercular Treatment. I Age Effect. R and P values correspond to the results of the ANOSIM test. Statistical significance is indicated as follows: ns (P > 0.05), *(P < 0.05), **(P < 0.01), ***(P < 0.001), and ****(P < 0.0001).

Fig. 8. Correlation analysis of the Shannon index with clinical outcomes.

A ICU Admission Rate. B Mechanical Ventilation Rate. C Endotracheal Intubation Rate. D Mortality Rate. E, Underlying Diseases. F Antibiotic Treatment. G Antifungal Treatment. H Antitubercular Treatment. I Age Effect. R and P values correspond to the results of the ANOSIM test. Statistical significance is indicated as follows: ns (P > 0.05), *(P < 0.05), **(P < 0.01), ***(P < 0.001), and ****(P < 0.0001).

To study the relative abundance differences between groups for different clinical outcomes and bacteria biomarkers from lefse, A volcano plot was shown in Fig. 9 and Supplementary Table 6. First, compared to non-ICU patients, four bacterial species were significantly upregulated in ICU patients, including Prevotella denticola, Tannerella, Capnocytophaga granulosa, and Capnocytophaga sputigena, while Lactobacillus casei, Acinetobacter, and Bifidobacterium were significantly downregulated. These findings provide further support for the potential link between the lower respiratory microbiome and the severity of COVID-19 in patients.

Fig. 9. Volcano plot of intergroup relative abundance differences of lefse biomarkers under different clinical outcomes.

A–D, Microbes with significant abundance differences are represented in orange (upregulated microbes) or cyan (downregulated microbes); microbes with no significant abundance differences are marked in gray. A ICU Admission. B Mortality. C Mechanical Ventilation. D Endotracheal Intubation. R and P values correspond to the results of the ANOSIM test. Statistical significance is indicated as follows: ns (P > 0.05), *(P < 0.05), **(P < 0.01), ***(P < 0.001), and ****(P < 0.0001).

Comparing mortality to non-mortality, we found a significant increase in Acinetobacter abundance (Fig. 9A and Supplementary Table 7). Conversely, 19 bacterial species showed a notable decrease, including Haemophilus influenzae, Lactobacillus casei, Prevotella denticola, Porphyromonas endodontalis, Prevotella salivae, Porphyromonas KLE 1280, Prevotella pallens, Alloprevotella tannerae, Granulicatella elegans, Prevotella veroralis, Porphyromonas sp. 279, Rothia mucilaginosa, Streptococcus oralis, Actinomyces graevenitzii, Capnocytophaga granulosa, Streptococcus pneumoniae, Capnocytophaga sputigena, Prevotella nigrescens, and Streptococcus mitis (Fig. 9B).

Next, we compared the microbiome patients requiring mechanical ventilation to those not needing it. Nine bacterial species showed significantly increased abundance, including Lactobacillus casei, Acinetobacter HMSC24B09, Prevotella denticola, Tannerella sp. HOT-286, Porphyromonas KLE 1280, Capnocytophaga sputigena, Porphyromonas sp. 279, Abiotrophia defectiva, and Capnocytophaga granulosa. In contrast, Granulicatella elegans showed a significant decrease (Fig. 9C and Supplementary Table 8).

In addition, when comparing patients who required intubation with those who did not, seven bacterial species had significantly increased abundance, including Prevotella denticola, Tannerella sp. HOT-286, Abiotrophia sp. HMSC24B09, Capnocytophaga sputigena, Capnocytophaga granulosa, Abiotrophia defectiva, and Capnocytophaga gingivalis, while Lactobacillus casei showed a significant decrease (Fig. 9D and Supplementary Table 9). To further assess the relationship between clinical factors and lower respiratory tract microbiota diversity, we constructed multivariate linear regression models to evaluate the impact of ICU admission, mechanical ventilation, intubation, mortality, and age on diversity indices. The analysis revealed a statistical association between mortality and both the Chao and Shannon indices, although the explanatory power was low (R² = 0.1221) (Supplementary Data 6). Based on this, we further analyzed the association between eight clinical variables and diversity indices in a subset of 74 samples and found no significant correlations (Supplementary Table 10). Taken together, these microbiome shifts likely reflect the progression of disease and varying treatment needs, suggesting a link between microbial composition and clinical outcomes.

Discussion

In this study, we conducted metatranscriptomic and metagenomic sequencing on BALF and sputum samples from COVID-19 patients to comprehensively examine the microbiome across different sample types. Our findings provide the first insight into the microbial community structure in the lower airways of COVID-19 patients across disease severity gradients (mild, severe, critical). Notably, we observed significant differences in both alpha and beta diversity among patients with varying disease severity. These results suggest that SARS-CoV-2 infection disrupts both the oropharyngeal microbiome and the microbiome of the lower respiratory tract, leading to microbial imbalances and potentially increasing the risk of secondary pulmonary infections.

The upper respiratory tract, being directly exposed to the environment, is highly susceptible to respiratory viral infections^37,38. Previous studies have demonstrated that severe infections can result in the dominance of specific microbial populations, causing a substantial reduction in microbial diversity, particularly in areas like the oral cavity³⁹. Reported also have documented comparable results in the upper airway microbial communities of SARS-CoV-2-infected individuals, highlighting dynamic shifts in microbial composition over time⁴⁰. Despite growing interest in COVID-19 pathogenesis, the lung microbial community dynamics in lower airway microbiota have yet to be comprehensively investigated using metagenomic approaches. Our study reveals significant differences in microbial diversity at both the species and genera levels among patients with mild, severe, and critical COVID-19. A significant increase in Chao diversity was observed in the lower respiratory tract microbiota of critically ill SARS-CoV-2-infected individuals, with disease severity demonstrating a dose-dependent relationship with the index. Bray-Curtis dissimilarity analysis revealed marked compositional disparities in the respiratory microbiota between critically ill patients and severe cases, while no statistical differences were detected between mild and severe groups. Our findings indicate that the airway microbiota of critically ill SARS-CoV-2-infected individuals exhibit higher alpha diversity compared to non-critical cases. Emerging metagenomic data further demonstrate that viral infection induces respiratory and gut microbial dysbiosis, characterized by decreased commensal populations, increased pathogenic taxa, and reduced overall biodiversity^41,42. A small study on mild COVID-19 cases showed a shift from early dysbiosis to increased diversity during hospitalization^40,43. Similarly, our research found a general decline in lower respiratory microbiome diversity across patients as disease severity increased.

From an immunomicrobiology perspective, the human microbiota exerts profound influence over host immune responses, with microbiota-targeted interventions showing promise for enhancing COVID-19 therapeutic efficacy and post-infection recovery^44,45. In our study, we observed that in mild COVID-19 cases, the Firmicutes phylum is dominant. However, its proportion drops significantly to 4.57% in critically ill patients. Conversely, the Bacteroidetes phylum is prevalent in severe cases, accounting for 39.16%. Among critically ill patients, Actinobacteria and Proteobacteria become the dominant groups, making up 59.35% and 31.63%, respectively. In comparison, these phyla represent 36.8% and 6.83% in mild cases, and 9.62% and 5.68% in severe cases. As disease severity increases, the proportions of the Streptococcus and Rothia genera decrease. Using Lefse analysis, we identified bacterial biomarkers that are associated with varying levels of disease severity. Several species—such as Klebsiella pneumoniae, Escherichia coli, Propionibacterium acnes, Corynebacterium striatum, Acinetobacter baumannii, and Herbaspirillum autotrophicum—showed strong correlations with COVID-19 severity, a finding supported by previous studies^46–48. Statistically significant differences in Candida albicans abundance were identified between mild and severe cases (p < 0.05), as well as between severe and critically ill cases, but no significant difference in mild and critically ill cases. This suggests that Candida albicans may influence COVID-19 severity, though its exact role requires further investigation. Given its potential as both a predictor of disease severity and a therapeutic target, addressing antibiotic resistance and maintaining microbial balance during clinical care could improve patient outcomes^49,50.

This study has several limitations. It was based on a retrospective cohort and lacked prospective validation. Differences in disease stages at admission, despite stratified analyses of ICU admission, mechanical ventilation, intubation, mortality, and age, may have introduced time-related biases affecting the accuracy of the results. Although multivariable regression was used to adjust for confounders, the complexity of treatment interventions likely limited the interpretability and generalizability of the findings, as well as the robustness of causal inferences. In addition, more than half of the patients were critically ill and received antibiotic treatment. Microbial diversity indices were compared between patients treated with a single antibiotic and those treated with multiple antibiotics, but no significant differences in the Chao index were observed. This does not contradict existing evidence that antibiotics impact the microbiota and is likely due to residual confounding.

Moreover, the widespread use of antibacterial and antifungal therapies may have altered the composition of the lower respiratory tract microbiota. A significant difference in Candida albicans abundance was detected across groups with varying disease severity, although the underlying mechanisms remain unclear and warrant further investigation. At the technical level, early use of low-depth metagenomic sequencing limited genome recovery, and species identification relied heavily on database alignment, potentially affecting the accuracy of functional enrichment analyses. To address this, strain-level variation and phylogenetic analyses were performed. In light of these limitations, future studies should be based on large, prospectively designed cohorts, apply high-depth sequencing, and incorporate multivariable analytical approaches to better characterize the changes in the lower respiratory tract microbiota associated with COVID-19.

Extensive research has examined the microbiota changes in the oral cavity, pharynx, and both the upper and lower respiratory tracts of COVID-19 patients^39,51–53. These studies have provided detailed insights into how the respiratory microbiome shifts with varying infection severity⁵². These microbiome alterations hold promise as potential biomarkers for predicting COVID-19 severity, offering valuable guidance for clinical diagnosis and treatment. Our study examines the microbial characteristics of the lower respiratory tract by analyzing BALF and sputum samples from patients with varying severities of SARS-CoV-2 infection. We observed significant differences in the Chao index at both the genus and species levels between mild, moderate, and critically ill patients. Notably, Candida albicans levels varied between mild and moderate patients, as well as between severe and critically ill patients, suggesting a potential role in disease progression. These findings highlight possible pathogens associated with mixed infections in the lungs, which may guide targeted antibiotic treatment for secondary bacterial infections. We also identified key strains, such as Veillonella dispar, which require further investigation to understand their contribution to SARS-CoV-2 pathogenicity and severe disease progression. Overall, our study underscores the strong link between microbiome composition and COVID-19 severity, presenting it as a promising biomarker to improve diagnosis, prevention, and treatment strategies.

Statistics and reproducibility

Statistical analyses were conducted using GraphPad Prism 8 (GraphPad Software, California, USA). An unpaired t-test was used for group comparisons. R and P values reported in the results are derived from the ANOSIM test. Statistical significance is denoted as follows: ns (P > 0.05), *(P < 0.05), **(P < 0.01), ***(P < 0.001), and ****(P < 0.0001).

Author contributions

All authors contributed to the study and approved the final manuscript. Denian Wang: Conceptualization, Data curation, Formal analysis, Methodology, Funding acquisition, Writing-Original draft preparation. Yishang Duan: Data curation, Formal analysis, Investigation, Writing-Original draft preparation. Lijuan He: Formal analysis, Investigation, Visualization. Ke Yuan: Formal analysis, Investigation, Visualization. Juan Jiang: Formal analysis, Investigation, Visualization. Jinghong Xian: Formal analysis, Investigation, Visualization. Rui Zhang: Formal analysis, Investigation, Visualization. Huohuo Zhang: Formal analysis, Investigation, Visualization. Jing Wang: Formal analysis, Investigation, Visualization. Ming Huang: Formal analysis, Investigation. Nan Li: Investigation, Validation. Min Huang: Investigation. Chenggong Hu: Data curation, Investigation. Sifeng Lu: Data curation, Formal analysis, Investigation. Taibing Deng: Data curation, Investigation, Zhongwei Zhang: Data curation, Investigation. Bojiang Chen: Conceptualization, Data curation, Supervision, Writing-Reviewing and Editing. Weimin Li: Conceptualization, Data curation, Funding acquisition, Supervision, Writing-Reviewing and Editing.

Peer review

Peer review information

Communications Biology thanks Michaela Hyblova and Anna Gorska for their contribution to the peer review of this work. Primary Handling Editors: Tobias Goris and Rosie Bunton-Stasyshyn.

Data availability

This study was conducted in full compliance with the Regulations of the People’s Republic of China on the Administration of Human Genetic Resources and all relevant ethical standards. The raw sequencing data generated in this study, including microbial DNA and RNA sequencing data with human sequences removed, have been deposited in the Genome Sequence Archive (GSA: CRA025488) at the National Genomics Data Center, part of the China National Center for Bioinformation, Beijing Institute of Genomics, Chinese Academy of Sciences. The data are publicly available at https://ngdc.cncb.ac.cn/gsa. In addition, sample-specific clinical information and experimental groupings related to microbial DNA and RNA datasets are provided in Supplementary Data 1. Sequencing details, including raw and clean reads, are available in Supplementary Data 2. Genotyping data from 116 COVID-19 patients are presented in Supplementary Data 3. Source data for Table 1 and Figs. 3 to 9 are included in Supplementary Data 1. Source data for Fig. 2A are available in Supplementary Data 3, for Fig. 2B in Supplementary Data 4, and for Fig. 2C in Supplementary Data 5. For additional information or data access requests, please contact Weimin Li (weimi003@scu.edu.cn) or Denian Wang (wangdenian623@wchscu.edu.cn).

Competing interests

The authors declare no competing interests.

Ethical approval

Written informed consent was obtained from all patients before sample collection. The study followed the ethical guidelines of West China Hospital, Sichuan University, and received approval from the institutional ethics committee (Approval No. 2023-30).

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-025-08234-1.