Characteristics of Gut Microbiota in Patients with Severe Pneumonia and its Potential Clinical Relevance

Introduction

Severe pneumonia (SP) is a critical pulmonary infection caused by various pathogens, often accompanied by complications such as acute respiratory failure and multiple organ dysfunction, which consequently lead to high mortality and disability risks (Torres et al. 2021). In recent years, researchers have been consistently investigating the diverse factors influencing lung infections in clinical practice, with particular focus on risk factors associated with malnutrition in pneumonia patients. Findings have indicated that factors such as obesity and advanced age exert a significant influence on the susceptibility to severe pneumonia (Silverio et al. 2021). Elucidating the specific pathways through which these factors influence the pathogenesis and progression of pneumonia remains a question worthy of in-depth exploration. Previous investigations have demonstrated that host-related factors, such as obesity and age, can modulate the human microbiota, disrupt host immune homeostasis, and thereby contribute to disease development (Honarpisheh et al. 2022). The human body harbors a microecosystem characterized by a diverse composition, in which the microbiota, comprising bacteria, viruses, and fungi, inhabits various anatomic sites throughout the body, notably the lungs and gut (Hashimoto 2023).

The lung microbiota has been found to be associated with a substantial number of pulmonary diseases. Tian et al. (2024) recognized that respiratory tract microbiota-based interventions hold promise for revolutionizing asthma management. Budden and Romani (2025) through an analysis of the regulatory roles of respiratory tract microbiota in pneumonia, lung cancer, and chronic obstructive pulmonary disease (COPD), further concluded that microbiota-targeted therapeutic approaches are feasible. These studies suggest that the lung microbiome may modulate pulmonary inflammation; however, the specific mechanisms underlying this remain to be elucidated. Additionally, whether this microbiota is associated with the gut microbiota remains inconclusive.

Compared to the lung microbiome, the gut microbiota has long been the focus of extensive research due to its high relative abundance and species diversity, and has been more thoroughly investigated to elucidate its mechanistic roles. Accumulating evidence demonstrates that gut microbiota dysbiosis is closely associated with the onset and progression of diverse pathological conditions, including common gastrointestinal disorders such as inflammatory bowel disease (Qiu et al. 2022) and colorectal cancer (Wong and Yu 2023), as well as systemic diseases such as obesity (Gomes et al. 2018), diabetes (Iatcu et al. 2021), and rheumatoid arthritis (Zhao et al. 2022).

Studies have demonstrated that the use of composite probiotic formulations in the treatment of pneumonia patients can effectively improve prognosis and accelerate recovery (Gutiérrez-Castrellón et al. 2022). With the identification of the impact of gut health on pulmonary immunity, the concept of the “gut-lung axis” has garnered significant attention. Intestinal microbiota exert both local effects within the gastrointestinal tract and distal influences on distant organs, such as the lungs. This bidirectional interaction, termed the “gut-lung axis”, serves as the foundational mechanism underlying the use of the microbiome in the prevention or treatment of respiratory diseases (Bulanda and Wypych 2022). In recent years, dysbiosis of the gut microbiota has been considered to play a key role in the occurrence, progression, and prognosis of pneumonia (Nagata et al. 2023). These findings provide important insights for further exploration of the mechanisms of the gut-lung axis. Patients with pulmonary diseases are frequently associated with dysbiosis, which is characterized by the over proliferation of pathogenic bacteria and a reduction in beneficial bacteria. This dys-biotic state may exacerbate pulmonary inflammatory responses and potentially influence clinical outcomes (Li et al. 2024). In severe pulmonary infections, alterations in the gut microbiota are recognized as a critical factor contributing to immune dysfunction, excessive inflammatory activation, and a poor prognosis (Yang and Cong 2021). The gut and lung engage in bidirectional interactions via multiple mechanisms, including circulating inflammatory cells and mediators, the intestinal barrier, oxidative stress, immune responses, and other related processes (Wang et al. 2023).

The current research is increasingly focusing on exploring the relationship between the gut microbiota and pneumonia. However, studies on the gut microbiota in patients with severe pneumonia remain relatively limited. This study aims to analyze gut microbiota characteristics in patients with severe pneumonia using a machine learning algorithm and to discuss potential correlations between gut microbiota and clinical prognosis in this population.

Experimental

Results

Analysis of bacterial characteristics in severe pneumonia patients

To compare gut microbiota between patients with severe pneumonia and healthy controls, this study collected fecal samples from both groups and analyzed microbial community diversity using 16S rRNA sequencing. Statistical analysis showed no significant differences in age or gender between the two groups (Table SII). Fig. 1A–1F presents the α-diversity analysis of patients with severe pneumonia, where the Richness, Chao 1, and Observed Species indices were significantly higher than those in the healthy control group (p = 0.017). These indices suggest potential alterations in the gut microbiota of patients with severe pneumonia. Subsequent principal component analysis (PCA) and principal coordinate analysis (PCoA) revealed partial differences in the clustered distribution patterns of microbial communities between the disease and control groups, further indicating potential discrepancies in microbial community distribution between the two groups (Fig. 1G–1H).

Fig. 1.

To further analyze the composition of intestinal microbiota in patients with SP, we selected the top 10 microorganisms ranked by relative abundance in both patients with SP and healthy individuals for comparison. We found that there were significant differences in microbial composition between the two groups of fecal samples at the level of phylum, class, order, family, genus, and species (Fig. 2A–2F). At the phylum level, compared to the healthy population, severe pneumonia patients exhibited a reduced relative abundance of Firmicutes and Bacteroidota, and an increased relative abundance of Proteobacteria and Actinobacteriota in their gut microbiota. At the genus level, the healthy population exhibited a higher abundance of Bacteroides, Faecalibacterium, Phascolarctobacterium, and Megamonas in their gut microbiota. At the species level, severe pneumonia patients had a gut microbiota with higher abundances of Enterobacteriaceae and Bacteroidaceae.

Fig. 2.

Furthermore, LEfSe analysis revealed significant differences in microbiota abundance at various taxonomic levels between the two groups (Fig. 2G–2H). These results indicate that the gut microbiota of patients with SP appears to differ from that of healthy individuals, providing important insights into the characteristic microbiota of SP patients.

Machine learning methods were used to screen for core microbial markers

To further identify intestinal microbiota biomarkers specific to patients with severe pneumonia, we employed 16S rRNA sequencing to analyze fecal samples and applied multiple machine learning algorithms to identify microbial biomarkers. The regularization path plot (Fig. 3A) and cross-validation error plot (Fig. 3B) of the LASSO algorithm demonstrate that as the λ value increases, both the model’s feature coefficients and the cross-validation tend to converge to zero. In the RF algorithm, as the number of decision trees increases, the model error exhibits a rapid decline followed by stabilization (Fig. 3C). Additionally, we present the importance rankings of the top 15 microbial biomarkers (Fig. 3D). Employing the 5-fold cross-validation methodology, the SVM algorithm demonstrates a decreasing cross-validation error rate alongside increasing accuracy, with the latter stabilizing as the number of microbial biomarkers increases (Fig. 3E–3F). Ultimately, the LASSO algorithm identified 16 genera, the RF algorithm selected 23 genera, and the SVM algorithm screened 101 genera.

Fig. 3.

Subsequently, we selected the genera that consistently appeared across the microbial communities generated by all three algorithms, ultimately identifyingnine microbial genera as characteristic markers for SP patients (Fig. 4A). These nine characteristic microbial markers were Finegoldia, Monoglobus, Mogibacterium, Peptoniphilus, Campylobacter, Parasutterella, Roseburia, Porphyromonas, and Faecalibacterium. The correlation analysis revealed that these nine characteristic microbial markers were not correlated with the blood test parameters (Table SIII and SIV). Comparing the microbial genus abundance between the two groups, we observed that, compared to the healthy group, the SP group had a reduced relative abundance of Roseburia, Monoglobus, Faecalibacterium, and Parasutterella. In contrast, the relative abundance of Mogibacterium, Campylobacter, Finegoldia, Peptoniphilus, and Porphyromonas was increased (Fig. 4C–4K). We constructed a diagnostic prediction model based on the nine identified microbial genera and generated its corresponding ROC curve (Fig. 4B). The area under the curve (AUC) was approximately 0.969 (95% CI: 0.946-0.992). The model demonstrated 95.24% sensitivity and 91.49% specificity in diagnosing severe pneumonia. This indicates that the microbiota-based diagnostic prediction model performs well in diagnosing SP patients. Moreover, as characteristic gut microbial genera specific to SP patients, these microbial communities are highly likely to play critical roles in disease progression.

Fig. 4.

Discussion

Previous studies have demonstrated that interactions between the host and gut microbiota can influence the occurrence and progression of diseases. Specifically, the host’s physiological status modulates the homeostasis of the gut microbiota, while alterations in gut microbial communities can indirectly affect the host’s disease state through immune responses (Marsland et al. 2015). In this study, we collected clinical data from patients with severe pneumonia to characterize the host’s physiological status and further applied machine learning algorithms to analyze the compositional features of the host’s gut microbiota. We aimed to seek evidence of the interaction between the intestine and the lungs, and to explore the mechanism of the lung-intestine axis in severe pneumonia.

Our research has revealed that the diversity of the intestinal microbiota in patients with SP has changed. In the alpha diversity analysis, we found that the SP group exhibited significantly higher values for Richness, Chao1, and Obs indices than the normal group. This is a rather unique finding. This is because the body is in an inflammatory state, leading to specific fluctuations in the intestinal microbiota, increased species diversity (especially harmful bacteria), and thus a higher alpha diversity trend compared to the normal group. The β-diversity analysis also revealed differences in species distribution between the two groups.

Using a machine learning algorithm, we screened nine distinct gut microbiota compositions at the genus level for characteristic microbial markers that could identify patients with SP. The general trend in the relative abundance of these nine characteristic microorganisms showed an increase in the relative abundance of harmful bacteria, such as Finegoldia, Mogibacterium, Peptoniphilus, Campylobacter, and Porphyromonas, while the abundance of beneficial bacteria, such as Monoglobus, Parasutterella, Roseburia, and Faecalibacterium, decreased. Previous studies have shown that gut microbiota can play a role in resisting the development of disease-related inflammation in the body (Pickard et al. 2017). For example, Faecalibacterium is a genus associated with overall health, and its low abundance is linked to the onset and progression of inflammation (Martín et al. 2023). Porphyromonas has been found to be associated with several systemic diseases, including cardiovascular disease, Alzheimer’s disease, and rheumatoid arthritis. Ju et al. isolated Parasutterella from the core intestinal microbiome of mice, and a series of characterization experiments demonstrated its significant role in metabolic processes, offering valuable insights into host health metabolism (Ju et al. 2019). Campylobacter, one of the four leading causes of gastroenteritis worldwide, plays a significant role in promoting inflammation (Costa and Iraola 2019). Gou et al. (2023) also found, through animal experiments, that intestinal pathogens, including Campylobacter and Peptoniphilus, could be depleted after drug intervention, thereby reducing intestinal inflammation and protecting the intestinal barrier. Therefore, we speculate that the microbial marker identified in this study using machine learning algorithms will help alleviate inflammation in severe pneumonia.

Previous studies have focused on various complications of pneumonia, including cardiac involvement. Xu et al. (2020) found that elevated CRP levels, advanced age, underlying comorbidities, and the severity of COVID-19 were the main risk factors for cardiac involvement in patients with COVID-19. Tsuchihashi et al. (2021) analyzed the relationship between clinical characteristics of novel coronavirus infection and severe outcomes and found that advanced age, male sex, and underlying disease were associated with severe outcomes. Similarly, our study collected clinical data for logistic regression and found that age and the total length of hospital stay were significant predictors of recovery and discharge in patients with severe pneumonia.

Among them, age is a significant risk factor for recovery and discharge in patients with severe pneumonia, indicating that older patients are less likely to recover and be discharged. Previous studies have shown that the elderly population has notable differences compared to younger individuals in immune system function, metabolic capacity, and gut microbiota stability (Yu et al. 2024). As individuals age, immune function gradually declines, and the abundance of probiotics in the gut microbiome decreases (Vemuri et al. 2018), which may reduce the body’s ability to resist pneumonia. In this study, the total length of hospital stay was identified as a protective factor for patient improvement and discharge from severe pneumonia. This may be because a more extended hospital stay provides patients with more opportunities to receive comprehensive treatment, including nutritional supplementation, medication adjustments, physical therapy, psychological support, and health education. These interventions help maintain normal metabolism and promote the body’s recovery process, including the restoration of gut probiotics. This enables the timely regulation of gut microbiota imbalances, contributing to the patient’s recovery and meeting the discharge criteria. Furthermore, the target population of this study comprises patients with severe pneumonia, who require an extended treatment and recovery period. A longer hospital stay may reflect timely intervention and monitoring, offering patients more opportunities for recovery and increasing their chances of meeting the discharge criteria, despite the severity of their condition.

This study investigated the correlation between gut microbiota characteristics and severe pneumonia patients, though several limitations remain. First, the research was conducted at a single center with all samples collected from the same location. All machine learning analyses were performed on a single dataset, which significantly limits the study’s generalizability and carries the risk of overfitting. Second, mild cases were excluded because the project primarily focused on critical care patients, leading to insufficient analysis of microbial profiles in these cases. Third, the lack of clinical test data from healthy controls prevented us from identifying differences in clinical factors between patients and healthy individuals.

Although grouping based on “Whether improved and discharged” could provide some evidence for the research, further improvement in the health control group testing is still necessary in subsequent studies.

When establishing a diagnostic prediction model, other risk factors, such as genetic and clinical indicators, were not included in the analysis, and the disease prediction based solely on the nine core microbial markers selected still has some errors. Additionally, changes in gut microbiota are influenced by multiple factors, including antibiotic use, dietary habits, and gastrointestinal disorders.

Future studies should incorporate samples from more centers to refine machine learning results. If possible, Blood samples from healthy populations should be collected for testing, and multi-center cohort studies should further explore causal relationships between gut microbiota and blood test indicators. Furthermore, precise microecological regulation remains a critical challenge for improving patient outcomes. Future research could explore probiotics, prebiotics, or fecal microbiota transplantation to restore gut balance and enhance immune function, thereby improving clinical outcomes in patients with severe pneumonia.

Conclusion

In summary, we identified a set of gut microbial markers at the genus level that could effectively identify patients with severe pneumonia by using multiple machine learning algorithms. Compared with the normal population, the intestinal microbiota of patients with severe pneumonia has changed. By integrating gut microbiota analysis with patients’ clinical profiles, we explored the potential role of the microbiota in the onset of the disease. Our findings will contribute to future research on the gut-lung axis and provide valuable insights for clinical diagnosis and treatment.