A Large-Scale Single-Cell Atlas Reveals the Peripheral Immune Panorama of Bacterial Pneumonia

Abstract

Rationale

Bacterial pneumonia poses a substantial global health burden; yet, the immunological mechanisms driving disease pathogenesis and resolution are incompletely understood.

Methods

We generated a large-scale single-cell transcriptomic atlas of peripheral blood immune cells from 100 individuals: 39 with severe bacterial pneumonia, 31 with mild disease, and 30 healthy control subjects.

Objectives

Integrating single-cell RNA sequencing with clinical and molecular data revealed profound remodeling of the peripheral immune landscape across disease severities.

Measurements and Main Results

Severe pneumonia was characterized by lymphopenia and monocytosis, accompanied by distinct shifts in T cell, B cell, and myeloid cell subset composition. Classical monocytes emerged as central orchestrators of the cytokine storm observed in severe cases, displaying elevated expression of proinflammatory genes (e.g., S100A8/9/12) and enhanced TLR4-MYD88 signaling. Exhaustion of innate-like CD8⁺ T cells, marked by upregulation of canonical inhibitory receptors, was a hallmark of severe disease. In contrast, mild pneumonia exhibited robust CD8⁺ T effector and helper memory cell activation, together with effective humoral immunity, evidenced by plasma cell expansion and coordinated T follicular helper cell–B cell interactions. B cells in mild cases showed enhanced antigen recognition, BCR signaling, and costimulatory gene expression, whereas those in severe cases displayed signs of dysfunction. Myeloid cell alterations in severe pneumonia included increased monocytic myeloid-derived suppressor cells and nonclassical monocytes, contributing to immunosuppression and complement overactivation, respectively.

Conclusions

This high-resolution atlas of peripheral immune responses in bacterial pneumonia identifies key cellular and molecular drivers of disease severity, providing potential therapeutic targets for immunomodulation and improved outcomes.

At a Glance Commentary

Scientific Knowledge on the Subject

The host immune response is a critical determinant of clinical outcome in bacterial pneumonia; yet, its complexity is not fully understood. Although phenomena such as lymphopenia and cytokine storms are recognized hallmarks of severe disease, the specific peripheral immune cell subsets driving these responses remain poorly characterized. Furthermore, the molecular features that distinguish a protective from a pathogenic immune response have not been defined at a high-resolution, single-cell level. Herein, a comprehensive atlas linking cellular dynamics to disease severity has been lacking.

What This Study Adds to the Field

This study provides the first large-scale, single-cell transcriptomic atlas of the peripheral immune response in bacterial pneumonia. We identify classical monocytes as the central orchestrators of the cytokine storm in severe disease, primarily through the S100A8/A9/A12-TLR4-MYD88 signaling axis. Severe disease is further characterized by profound exhaustion of innate-like CD8⁺ T cells and the expansion of immunosuppressive monocytic myeloid-derived suppressor cells. In contrast, mild disease is marked by a robust effector T cell response and effective humoral immunity underscored by coordinated T follicular helper–B cell interactions. These findings highlight distinct cellular and molecular drivers of disease severity, providing novel therapeutic targets for immunomodulation in bacterial pneumonia.

Bacterial pneumonia, an acute inflammatory disease of the lung caused by bacterial infection, is broadly classified as community-acquired pneumonia (CAP) or hospital-acquired pneumonia (1). Both significantly contribute to global morbidity and mortality (2). Lower respiratory tract infections, including bacterial pneumonia, caused ∼2.56 million deaths in 2017 (fifth leading global cause), with disproportionately higher mortality in sub-Saharan Africa and South/Southeast Asia according to the 2017 Global Burden of Disease, Injuries, and Risk Factors Study (3). In developed countries, adult CAP incidence ranges from 0.2% to 1.1%, with a 2–14% mortality rate (4). Bacterial pneumonia–associated child mortality and adult hospitalization rates continue to rise in lower- and middle-income countries. Notably, gram-negative bacteria accounted for 72.6% of infections in elderly Chinese patients with CAP (5). Therefore, bacterial pneumonia remains a major public health concern, posing substantial health and economic burdens globally.

Bacterial pneumonia presents with a spectrum of clinical manifestations, from mild, self-limiting respiratory symptoms to severe respiratory failure and death (6). Disease severity is influenced by pathogen virulence, patient age, immune status, and prior antibiotic exposure. Mild cases typically present with fever, cough, and sputum production (7). Severe cases involve multisystem complications, driven by complex host–pathogen interactions determining infection and tissue damage extent (8). The immune response can be broadly categorized as hyperinflammatory or immunosuppressed. Upon alveolar pathogen invasion, the host initiates an inflammatory cascade, releasing cytokines and chemokines (e.g., tumor necrosis factor [TNF-α], IL-1, IL-6, and CXCL8) that recruit and activate neutrophils, monocytes, and the complement system (9–11). However, excessive inflammation can trigger a cytokine storm, causing lung injury and potentially fatal outcomes (12). Immunosuppression, prevalent in the elderly because of immunosenescence and comorbidities, impairs pathogen clearance, leading to prolonged illness or clinical deterioration (13). Thus, a detailed understanding of the human immune response in bacterial pneumonia is crucial for developing effective therapies; yet, this area remains underinvestigated.

Understanding the immune dysregulation driving bacterial pneumonia pathogenesis is crucial for improving patient outcomes and reducing morbidity and mortality in severe disease (14). Single-cell RNA sequencing (scRNA-seq) offers unparalleled resolution for dissecting complex host–pathogen interactions and immune responses at the cellular level (15). Although peripheral blood immune cell studies have provided insights into bacterial respiratory infections such as tuberculosis (Mycobacterium tuberculosis) (16), brucellosis (Brucella spp.) (17), and atypical pneumonia (Mycoplasma pneumoniae) (10), comprehensive characterization of peripheral immune cell dynamics in bacterial pneumonia remains elusive. Hence, applying scRNA-seq to peripheral immune cells from patients with bacterial pneumonia can reveal disease-specific alterations in cellular composition, transcriptional programs, and immune response dynamics, revealing key inflammatory pathways and potential therapeutic targets. Moreover, integrating scRNA-seq data with clinical parameters (e.g., disease severity) and other molecular data (e.g., cytokine profiles) provides a more holistic understanding of host–pathogen interactions in pneumonia, potentially enabling the discovery of novel diagnostic biomarkers and the development of personalized therapies tailored to individual patient immune profiles.

Here, we performed scRNA-seq on peripheral blood immune cells from a large-scale cohort of 100 participants: 39 with severe bacterial pneumonia (SBP), 31 with mild bacterial pneumonia (MBP), and 30 healthy control subjects (HCs). Integrating scRNA-seq with clinical and molecular data, we characterized high-resolution transcriptomic changes across disease severities, highlighting the relationship between disease stage and host immune response. We also identified key variations in the disease characteristics of bacterial pneumonia, providing a valuable resource for investigating inflammatory features. These findings provide insights into the pathogenic mechanisms and protective immune responses underlying bacterial pneumonia, with implications for disease control and treatment.

Methods

Ethical Approval

This study was conducted in accordance with the Declaration of Helsinki and approved by the ethics committee of the Chinese People’s Liberation Army General Hospital (309202305011312). All participants or their legal guardians provided written informed consent.

Study Design and Participants

This study enrolled 70 individuals with bacterial pneumonia and 30 HCs (see Table E1 in the online supplement). Two HCs were recruited specifically for this study, with the remaining 28 derived from a previous study (18). Participants were recruited between October 2023 and June 2024 from seven Beijing hospitals. Bacterial infections were confirmed by PCR, high-throughput sequencing, or culture. Pneumonia diagnoses were established on the basis of characteristic clinical features (cough, fever, sputum production, pleuritic chest pain) and chest radiography findings (19), because physical examination findings (e.g., rales, bronchial breath sounds) are less sensitive and specific for diagnosis (20). Particularly, to minimize confounding by illness duration when comparing severity groups, specific enrollment timing criteria were applied. Both groups were sampled within 24–72 hours of hospital admission after clinical confirmation of disease severity. Patients with MBP were sampled after being confirmed to have stable, non–ICU-requiring disease. Patients with SBP were sampled within the same 24–72-hour window from hospital admission, a time frame that typically coincided with their transfer to the ICU, thereby capturing a comparable phase of early, established severe illness. Furthermore, to enhance group comparability and control for potential confounders, patient groups were also matched on key baseline characteristics (Table E1).

Severe pneumonia was defined using the 2007 Infectious Diseases Society of America/American Thoracic Society criteria for severe CAP (19). These criteria require fulfillment of at least one major criterion or three or more minor criteria: 1) major criteria include septic shock requiring vasopressors or respiratory failure requiring mechanical ventilation, and 2) minor criteria include respiratory rate ≥30 breaths/min, Pa_O2/Fi_O2 ratio ≤250, multilobar infiltrates, confusion/disorientation, uremia (blood urea nitrogen ≥20 mg/dl), leukopenia (white blood cell count <4,000 cells/μl), thrombocytopenia (platelet count <100,000/μl), hypothermia (core temperature <36°C), or hypotension requiring aggressive fluid resuscitation.

Patients were excluded if they had any of the following: 1) age younger than 18 years, 2) viral pneumonia, 3) autoimmune disease, 4) immunosuppression (e.g., corticosteroid or chemotherapy treatment, organ transplant, hematologic malignancy, or HIV infection with a CD4⁺ T cell count below 200 cells/μl), or 5) specified comorbidities (e.g., asthma, chronic obstructive pulmonary disease, cystic fibrosis, or bronchiectasis).

scRNA-seq and Data Analysis

Peripheral blood mononuclear cells (PBMCs) were isolated from fresh blood samples collected within 2 hours of diagnosis from patients with MBP or SBP. Cell viability was assessed using a Countstar cell viability kit with viability exceeding 90% (21). scRNA-seq libraries were prepared using the Chromium Single Cell 5′ Kit version 2 (10x Genomics; PN-1000263) according to the manufacturer’s instructions and sequenced on an Illumina NovaSeq 6000 platform (2 × 150-bp paired-end reads).

scRNA-seq data were analyzed as previously described (16, 22). Briefly, a filtered gene expression matrix for all 100 samples was generated using kallisto/bustools (kb version 0.24.4) and concatenated using anndata (ad version 0.7.6) (22). Low-quality cells and doublets were removed using Scanpy (sc version 1.9.2). Data were normalized to 10,000 reads per cell, and highly variable genes (HVGs) were identified (22). Batch effects were corrected with Harmony after sample-specific HVG selection (23). A consensus list of 1,500 HVGs, excluding ribosomal, mitochondrial, and immunoglobulin genes, was generated on the basis of recovery rates across all samples. Principal component analysis (PCA; 20 components) was performed on the selected HVGs. The resulting PCA matrix was batch corrected using Harmony via Scanpy’s external.pp.harmony_integrate function, with sample and dataset variables as covariates (theta = 2.5 and 1.5, respectively) (23, 24). Cell clustering was performed on the batch-corrected matrix using the Louvain algorithm implemented in Scanpy, and marker genes were identified using the rank_genes_groups function (25).

Cell Clustering and Annotations

Cells were clustered using the Louvain algorithm (sc.tl.louvain function) in two consecutive rounds. An initial round (resolution = 2.0) identified 11 major cell types according to canonical marker genes (CellMarker 2.0 database: http://117.50.127.228/CellMarker/) (Figure E3): plasma cells, B cells, CD4⁺ T cells, CD8⁺ T cells, mucosal-associated invariant T (MAIT) cells, γδ T cells, natural killer (NK) cells, dendritic cells (DCs), double-negative T (DNT) cells, monocytes, and megakaryocytes. Subclusters were manually annotated on the basis of canonical marker genes (CellMarker 2.0 database) and further validated using differentially expressed genes (DEGs) identified with the sc.tl.rank_genes_groups function.

Cell trajectories were inferred using partition-based graph abstraction (PAGA) implemented in Scanpy (version 1.5.1) (26). PAGA revealed continuous transitions between cell types. DEGs were identified using Scanpy’s sc.tl.rank_genes_groups function (use_raw=True) on the basis of cluster membership or disease condition (Wilcoxon rank-sum test with an adjusted P value <0.01 and a fold change threshold >1.5). Ligand–receptor interaction analysis was performed using CellphoneDB (https://github.com/ventolab/CellphoneDB). Significant interactions were identified on the basis of the tool’s permutation testing framework using the recommended P value threshold of <0.05.

Identifying Changes in Immune Cell Proportion

The relative abundances of immune cell types and subtypes were compared across disease conditions using the Kruskal-Wallis test. For each cell population, three pairwise post hoc comparisons (SBP vs. HCs, MBP vs. HCs, SBP vs. MBP) were performed, with the Bonferroni correction applied to control the family-wise error rate across these comparisons. The influence of disease stage and potential interactions on cell-type proportions was assessed using multivariate ANOVA (16). Disease-specific enrichment of cell populations was quantified by calculating the ratio of observed to expected cell counts (R_O/E) (16).

Determining Cell State Scores

Cell type–specific activation states and physiological activities were compared across disease phases using gene set scoring. Gene sets representing proinflammatory cytokines, inflammatory responses, naive and exhausted states, cytotoxic activity, and regulatory effector functions were compiled from published studies (Table E3) (12, 21, 27). Scores were calculated in Scanpy (sc.tl.score_ genes function) as the normalized average expression of genes within each set. These scores were initially compared across the three groups using the Kruskal-Wallis test. Subsequently, for each score, three pairwise post hoc comparisons (SBP vs. HCs, MBP vs. HCs, SBP vs. MBP) were performed, with the Bonferroni correction applied to control the family-wise error rate across these comparisons.

Plasma Cytokine Quantification and Flow Cytometry

Plasma cytokine concentrations were quantified using the T-helper type 1 cell (Th1)/Th2 34-plex human ProcartaPlex kit (Thermo Fisher Scientific) according to the manufacturer’s instructions (28). Flow cytometry was performed as previously described (27).

Statistical Analysis

Statistical analyses and data visualization were performed using Python and R. Statistical significance is indicated in all figures as follows: ns, not significant (P > 0.05); *P ⩽ 0.05; **P ⩽ 0.01; ***P ⩽ 0.001; ****P ⩽ 0.0001.

Code Availability

Customized data analysis scripts used in this study are available upon reasonable request.

Results

Pan-Peripheral Immune Cell Landscape in Bacterial Pneumonia

To delineate the immune response landscape in bacterial pneumonia, we conducted scRNA-seq on PBMCs from individuals with varying disease severity. Using the 10x Genomics platform, we sequenced 855,314 cells from 100 samples, including 31 MBP, 39 SBP, and 30 HCs. Predominant causative agents included common bacterial pneumonia pathogens such as Acinetobacter baumannii, Pseudomonas aeruginosa, and Staphylococcus aureus (Figure E1). Detailed clinical and laboratory data for the enrolled individuals are presented in Table E1 and Figure E1. After computational doublet removal, 826,638 cells passed quality control (>200 genes detected, <10% mitochondrial reads per cells) (Figure E2). These high-quality cells averaged 4,625 unique molecular identifiers representing 1,765 genes (Figure E2). The dataset comprised 182,148 cells (22.06%) from HCs, 278,471 cells (33.73%) from patients with MBP, and 365,019 cells (44.21%) from patients with SBP (Figure E2). After normalization, high-quality cells were integrated into a single dataset for PCA (see Methods).

Unsupervised graph-based clustering and uniform manifold approximation and projection (UMAP) revealed 11 major cell types based on canonical marker gene expression: B cells, plasma cells, CD4⁺ T cells, CD8⁺ T cells, MAIT cells, γδ T cells, NK cells, DCs, DNT cells, monocytes, and megakaryocytes (Figures E3A and E3B). DNT cells expressed low concentrations of the canonical T lymphocyte markers (e.g., CD3D, CD3E) and lacked expression of markers characteristic of other peripheral blood immune cells, confirming their identity (Figures E3A and E3B). UMAP visualization exhibited T and NK cells segregated from B lymphocytes, whereas myeloid cells showed transcriptional profiles from distinct lymphocytes (Figure E3A). UMAP colored by disease severity revealed distinct clustering, particularly between patients with SBP and HCs, indicating bacterial infection–induced transcriptional changes and unique immune profiles for different disease states (Figures E3B and 1). Within these 11 major clusters, 47 distinct cell subtypes were identified using canonical markers, highly expressed genes, and DEGs (Figure 1B and Table E2). This comprehensive PBMC atlas provides a valuable resource for the precise annotation and in-depth analysis of these cell types at multiple resolutions.

Figure 1.

An overview of the study design and results for peripheral blood mononuclear cell single-cell transcriptomic study. (A) Diagram outlining the overall study design, which included 100 individuals, including 70 patients (31 patients with mild disease and 39 patients with severe disease) and 30 healthy donors. (B) The clustering result (left row) of 47 cell subtypes (right row) from 100 samples. Each point represents a single cell, colored according to cell type. (C) Uniform manifold approximation and projection of the healthy donors and those with mild and severe disease. (D) Disease preference of major cell clusters as estimated using the ratio of observed to expected cell counts.

Cell clusters exhibited distinct enrichment patterns across sample sources and disease severities (Figures 1D, E3D, and E3E). To further explore the unique immunological signatures associated with different disease states, we examined the individual immune cell composition (Figures 1D, E3D, and E3E). At the major cell-type level, patients with SBP exhibited substantial depletion of most lymphocyte populations and enrichment of monocytes compared with HCs and patients with MBP (Figure E3D). This finding, consistent with R_O/E analysis (a ratio of observed to randomly expected cell number) (Figure E3E) (29), suggests that lymphopenia and monocytosis may be key immunological features of SBP. These scRNA-seq findings are mirrored in clinically measured blood cell counts (Figure E1D), supporting the reliability of our analysis (Table E1). At the subtype level, R_O/E analysis revealed distinct cell subtype preferences across patient groups (Figure 1D). Plasma cells, plasmablasts, and most monocyte subtypes were notably enriched in bacterial pneumonia, increasingly so with disease severity, whereas most T-lymphocyte subsets were depleted in severe cases (Figure 1D). In addition, the large cohort size enabled us to analyze the association between disease severity and alterations in immune cell composition within PBMCs (Figure E3F). ANOVA revealed that most lymphocyte subsets and intermediate monocyte subtypes were significantly associated with SBP, implicating these populations in SBP pathogenesis. Collectively, these findings indicate that distinct immunological profiles characterize different severities of bacterial pneumonia.

Monocytes Contribute to the Inflammatory Storm in Severe Disease

A robust host inflammatory response is crucial for effective resolution of bacterial pneumonia (30). To dissect cellular sources of cytokines, we developed cytokine and inflammation scores for each immune cell type using established gene sets (Table E3). These interconnected scores enabled assessment of the relative contribution of each peripheral immune cell subset to the overall inflammatory response. A Kruskal-Wallis test (Bonferroni corrected) was performed for identifying hyperinflammatory cell subtypes by comparing the scores of each subtype against all other subtypes. Patients with bacterial pneumonia exhibited significantly elevated cytokine and inflammatory gene expression, indicative of active inflammation (Figures 2A and E4A). Notably, patients with severe disease exhibited significantly higher inflammatory and cytokine profiles than HCs and patients with MBP, indicative of a peripheral cytokine storm (Figures 2A and E4A). Consistently, C-reactive protein concentrations were significantly elevated in patients with SBP (Figure E1D), confirming heightened inflammation.

Figure 2.

Monocytes are the primary contributors to the production of proinflammatory cytokines in patients with severe bacterial pneumonia. (A) Uniform manifold approximation and projections of peripheral blood mononuclear cells. Colored on the basis of 11 major cell types (top left), 7 hyperinflammatory cell subtypes (top right), cytokine score (middle), and inflammatory score (bottom). (B) Pie charts depicting the relative contribution of each inflammatory cell subtype to the cytokine and inflammatory scores in patients with severe bacterial pneumonia. (C) Heatmap depicting the expression of cytokines within each hyperinflammatory cell subtype identified. (D) Bar chart depicting the relative contribution of the top 10 cytokines in patients with severe disease. (E) Boxplots of S100A8/A9/A12 expression based on single-cell RNA-sequencing profiling for healthy control subjects, patients with mild disease, and patients with severe disease. Significance was evaluated using the Kruskal-Wallis test with Bonferroni correction (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ^nsP > 0.05). (F) Heatmap plots of the sum of significant interaction among the seven hyperinflammatory cell subtypes. (G) Dot plot illustrating the strength and significance of ligand–receptor interactions between major inflammatory subtype (Mono_01_Classical_CD24) and other inflammatory subtypes. Dot size corresponds to the −log₁₀(P value), and dot color indicates the mean expression level of the interacting pair. Displayed interactions were filtered for nominal significance (P < 0.05).

Thirteen cell subtypes, including four T lymphocyte, two NK cell, and seven monocyte subtypes, exhibited significantly elevated cytokine and inflammation scores (Figure E4B), implicating them as potential sources of the cytokine storm. Seven monocyte subtypes exhibited significantly elevated cytokine and inflammatory scores in patients with SBP compared with HCs and patients with MBP (Figure E5). These monocyte subtypes were thus identified as likely drivers of the cytokine storm in severe cases. Moreover, these seven inflammatory monocyte subsets were also expanded in patients with SBP (Figure E4C), further implicating them in the exacerbated inflammatory response.

The classical monocyte subset (Mono_01_Classical_CD14) accounted for approximately 70% of the inflammatory and cytokine scores (Figure 2B), identifying it as a principal driver of the SBP cytokine storm. We subsequently analyzed the inflammatory signatures of these identified inflammatory cell subsets, observing distinct proinflammatory cytokine gene expression patterns within each subset (Figure 2C), such as genes encoding S100A8/9/12, CXCL1/2/8, etc. (Figure 2C). Patients with SBP exhibited elevated expression of multiple inflammatory cytokines, notably S100A8/9/12, TNFSF12/13, IL15, and CXCL1 (Figure E6A), indicating a multifactorial peripheral cytokine storm. Elevated cell type–specific proinflammatory cytokine expression within inflammatory monocyte subsets underscores their pivotal role in the pathogenesis of cytokine storms observed in patients with SBP (Figure E6B).

Three S100 proteins (S100A8/A9/A12), predominantly expressed by classical monocytes (Figure E6C), were critical cytokine storm drivers, comprising >99% of cytokine scores in patients with SBP (Figure 2D). S100A8/A9/A12 expression was elevated in patients with SBP (Figure 2E), further supporting our hypothesis. Besides S100 proteins, TNFSF13 and CXCL8 were also significantly elevated in severe cases (Figure E6D) and made up part of the top five proinflammatory cytokines. This suggests that TNFSF13 and CXCL8 may synergize with S100A8/A9 and S100A12 to fuel the inflammatory storm. The plasma concentrations of S100A8/A9 heterodimer, S100A12 molecule, TNFSF13, and CXCL8 are consistent with scRNA-seq analysis, further validating the accuracy of our scRNA-seq results (Figures 2E and E6D). S100A8/A9/A12 primarily signal via Toll-like receptor 4 (TLR4) (31), which recruits MYD88, activating nuclear factor-κB and mitogen-activated protein kinase pathways and amplifying inflammation (31). TLR4 and MYD88 were expressed predominantly in inflammatory monocytes from patients with SBP (Figure E6E). Therefore, we propose that inflammatory monocytes secrete S100A8/A9/A12, thereby activating their own TLR4-MYD88 pathway in a positive feedback loop, exacerbating inflammation and tissue injury. The data identify hyperinflammatory monocyte clusters and the S100A8/A9/A12-TLR4-MYD88 signaling axis as promising therapeutic targets for alleviating SBP immunopathogenesis.

The SBP cytokine storm likely arises from intricate intercellular communication among hyperinflammatory cell populations, mediated by diverse cytokines (16). To investigate this interplay, we analyzed ligand–receptor interaction profiles among seven distinct hyperinflammatory cell subsets identified in patients with SBP (Figure 2F). Our analysis uncovered notable ligand– receptor interactions within inflammatory cell subsets, with enhanced cross-talk involving inflammatory classical monocytes (Figure 2F). These classical monocytes exhibited elevated expression of multiple receptors, including CXCR1, CXCR2, IL1R1, IL10RA, IL15RA, and TNFRSF1A, indicating their responsiveness to a broad spectrum of cytokines from other cell populations (Figure 2G). Importantly, communication between inflammatory classical monocytes and other highly inflammatory cell clusters was mediated predominantly by chemokines (such as CXCL1 and CXCL8) and their cognate receptors (CXCR1 and CXCR2), alongside inflammatory cytokines (e.g., TNF, IL1B) and their respective receptors (TNFRSF1A/B and IL1R1) (Figure 2G). For instance, CXCL8-CXCR1/2 signaling is crucial for monocyte communication and contributes significantly to the co-orchestration of inflammatory responses during infection (32).Together, these results elucidate molecular mechanisms governing the interplay among hyperinflammatory monocytes in severe disease.

Exhaustion of Innate-like CD8⁺ T Cells Is a Hallmark of Severe Illness

Using canonical markers, highly expressed genes, and DEGs, we identified nine distinct CD8⁺ T cell subsets (Figures 3A and E7A and Table E2). These comprised seven conventional CD8⁺ T cell and two innate-like T cell populations, including MAIT and γδ T cells. Because MAIT and γδ T cells express CD8A and CD8B, they were included in the CD8 T cell analysis (Figures E3B and E7A). Broadly, the conventional CD8⁺ T cell populations could be categorized into one naive (CD8T_01_Naive_CCR7_LEF1), one central memory (CD8T_02_cMemory_CCR7_GPR183), one effector memory (CD8T_04_eMemory_GPR183_GZMH), and four effector T subpopulations (Figures 3A and E7A and Table E2). One effector CD8⁺ T cell cluster exhibited high surface expression of CD8A, CD8B, and CD56, together with elevated TYROBP (encoding DAP12) transcription, and therefore was designated as NKT-like cells (Figure E7A). The remaining three effector CD8⁺ T cell clusters were characterized by the distinct expression of cytotoxic genes, such as GZMK (Figure E7A).

Figure 3.

Immunological features of CD8⁺ T cell subsets. (A) The clustering result (left row) of 9 CD8⁺ T cell types (right row) from 100 samples. Each point represents a single cell, colored according to cell type. (B) Box plots showing the exhausted scores in innate-like and effector CD8⁺ T cell subsets across disease conditions. Significance was evaluated using the Kruskal-Wallis test with Bonferroni correction (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ^nsP > 0.05). (C) Box plots showing the cell exhaustion–related markers in innate-like CD8⁺ T cells across disease conditions. Significance was evaluated using the Kruskal-Wallis test with Bonferroni correction (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ^nsP > 0.05). (D) Uniform manifold approximation and projections illustrating IFN-I response and unhelped signature scores for CD8⁺ T cells across disease conditions. (E) Box plots showing the cytotoxic score in effector CD8⁺ T cells across disease conditions. Significance was evaluated using the Kruskal-Wallis test with Bonferroni correction (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ^nsP > 0.05). (F) Dot plots showing the cytotoxicity-related genes in effector CD8⁺ T cell subsets across disease conditions. (G) Bar plots showing the enrichment Gene Ontology terms in effector CD8⁺ T cell subsets from patients with mild and severe bacterial pneumonia.

The CD8⁺ T cell compartment composition differed markedly between HCs and patients with bacterial pneumonia (Figure 1D). Naive and central memory CD8⁺ T cells, spatially segregated in UMAP analysis (Figure 3A), exhibited reduced frequencies in patients (Figure 1D). Conversely, all remaining CD8⁺ T cell subsets increased in patients with MBP, indicating that effector CD8⁺ T cell activation and expansion are key to the early immune response. However, in SBP, the frequencies of two innate-like T cell populations were significantly diminished (Figure 1D), implying that a compromised or dysregulated innate-like T cell response may contribute to the poorer clinical outcomes. These findings indicate that bacterial pneumonia induces complex peripheral CD8⁺ T cell alterations. Analysis of previously defined exhaustion signature genes revealed varying degrees of exhaustion across CD8⁺ T cell subsets (Figure E7B and Table E3). Effector and innate-like CD8⁺ T cells displayed higher exhaustion scores, suggesting a greater degree of functional exhaustion in these populations (Figure E7B). Specifically, innate-like CD8⁺ T cells (NKT, MAIT, and γδ T cells) exhibited increased exhaustion in SBP versus MBP (Figure 3B). This suggests that peripheral innate-like CD8⁺ T cell exhaustion may be a hallmark of SBP. We validated this in an independent SBP cohort (n = 11) by flow cytometry (Figures E8–E12). Mechanistically, innate-like CD8⁺ T cells from severe cases showed increased expression of inhibitory receptors PD-1 (PDCD1), LAG3, and CTLA4 (Figures 3C and E7C), suggesting that these inhibitory molecules contribute to the exhausted phenotype in severe cases. Correspondingly, the key transcription factor (TF) (e.g., PRDM1) was significantly upregulated in exhausted innate-like CD8⁺ T cells from patients with SBP compared with HCs and patients with MBP (Figure E7D). Increased PRDM1 expression correlated with elevated expression of inhibitory receptors and reduced multifunctionality in exhausted CD8⁺ T cells (22).

To investigate the drivers of innate-like CD8⁺ T cell exhaustion in SBP, we examined type I IFN signaling, previously linked to CD8⁺ T cell exhaustion (33). Exhausted innate-like CD8⁺ T cells exhibited significant enrichment of type I IFN signaling–associated genes, indicating a potential link between persistent type I IFN signaling and their exhaustion (Figures 3D and E7E). Moreover, given that compromised CD4⁺ T cell help is also associated with CD8⁺ T cell exhaustion (33), we therefore investigated transcriptional signatures of reduced CD4⁺ T cell help in these cells. As anticipated, exhausted innate-like CD8⁺ T cells upregulated transcripts characteristic of CD8⁺ T cells lacking sufficient CD4⁺ T cell assistance (Figures 3D and E7E). Thus, insufficient CD4⁺ T cell help and persistent type I IFN signaling may contribute to innate-like CD8⁺ T cell exhaustion.

CD8⁺ T cell cytotoxicity is crucial for bacterial clearance and infection control (34), so we assessed the cytotoxic capacity of CD8⁺ T cells in patients with varying disease severities. Effector memory, innate-like, and effector CD8⁺ T cells exhibited a higher cytotoxic capacity than naive and central memory cells (Figure E7F). Among these, three effector CD8⁺ T cell subsets (CD8T_05_Effector_GZMA_GZMM, CD8T_06_Effector_GZMK_Low, CD8T_ 07_Effector_GZMK_High) displayed the highest cytotoxic potential, suggesting their primary role in pathogen clearance and infection control. However, effector CD8⁺ T cell cytotoxic capacity was reduced in severe versus mild disease (Figure 3E). This implies that although robust effector CD8⁺ T cell cytotoxicity is critical for controlling infection in mild cases, its impairment in severe cases may contribute to uncontrolled bacterial growth and worse clinical outcomes. Consistent with this, effector CD8⁺ T cells in patients in mild cases expressed high levels of multiple cytotoxic molecules (Figures 3F and E7G).

We next delved deeper into the transcriptional profiles of effector CD8⁺ T cells by analyzing DEGs and conducting Gene Ontology (GO) enrichment. Compared with HCs, effector CD8⁺ T cells from patients with MBP and SBP had 1,814 and 1,700 upregulated DEGs, respectively (Figure E7H and Table E4). Of these, 1,579 DEGs were shared between the two patient groups (Figure E7H). This substantial transcriptional reprogramming effector CD8⁺ T cells underscores the impact of bacterial infection on their functional state. In mild cases, effector CD8⁺ T cells displayed a robust cytotoxic phenotype, with enriched pathways for T cell activation, signaling, and antigen processing/presentation (Figure 3G). In addition, enriched pathways for cellular regulation, homeostasis, and dynamic cell structure modifications also suggested a well-coordinated response conducive to bacterial clearance (Figure 3G). In contrast, effector CD8⁺ T cells from patients with severe disease exhibited a contrasting profile, with enrichment for stress response, transcriptional/post-translational regulation, and cytoskeletal reorganization pathways (Figure 3G) rather than a dominant cytotoxic signature. This suggests a shift from cytotoxic activity, potentially reflecting a dysregulated or exhausted state caused by the heightened inflammatory environment and persistent antigenic stimulation characteristic of severe disease. Enrichment of DNA damage response, IFN-γ signaling, and protein degradation pathways further supported this notion of cellular stress and altered functionality. These findings highlight divergent effector CD8⁺ T cell functional landscapes in MBP versus SBP, underscoring the dynamic, context-dependent CD8⁺ T cell response to bacterial infection.

Dysregulated CD4⁺ T Cell Subsets in SBP

Analysis of 189,985 CD4⁺ T cells enabled a detailed examination of subpopulation structures and disease-associated alterations in bacterial pneumonia. The CD4⁺ T cell compartment broadly comprised four major populations (35): naive T cells (CCR7⁺), helper memory T cells (ICOS⁺, CXCR3⁺, CCR4⁺, or CCR10⁺), cytotoxic CD4⁺ T cells (GZMK⁺), and regulatory T (Treg) cells (FOXP3⁺) (Figures 4A and E14A and Table E2). Naive CD4⁺ T cells were further divided into three clusters: CD4T_01_ Naive_CCR7_LEF1, CD4T_02_Naive_ CCR7_TCF7, and CD4T_03_Naive_CCR7_ IFITM3 (Figure 3A). All three clusters highly expressed naive markers (e.g., CCR7, TCF7, and LEF1) but exhibited distinct expression profiles of these markers (Figure E14A). The CD4T_03_Naive_CCR7_IFITM3 cluster displayed elevated expression of IFN-related genes (e.g., IFITM3), suggesting a naive IFN phenotype previously reported (35). Patients with SBP were associated with a reduced frequency of naive CD4⁺ T cells compared with HCs and patients with MBP (Figure 1D), suggesting impaired homeostasis and a potential disease severity correlation.

Figure 4.

Immunological features of CD4⁺ T cell subsets. (A) The clustering result (left row) of 11 CD4⁺T cell types (right row) from 100 samples. (B) Box plots showing the percentage of T regulatory (Treg) cells and CD4_05_Treg across disease conditions. Significance was evaluated using the Kruskal-Wallis test with Bonferroni correction (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ^nsP > 0.05). (C) Heatmap showing the expression of selected immunomodulatory molecules in CD4T_11_Treg_FOXP3_IL2RA subsets across disease conditions. (D) Partition-based graph abstraction analysis of CD4⁺ T pseudotime: The associated cell type and the corresponding status are listed. (E) Dot plot of the interactions among CD4T_11_Treg_FOXP3_IL2RA with exhausted innate CD8⁺ T cells in patients with severe bacterial pneumonia. P values are indicated by the circle sizes, as shown in the scale on the right. (F) Dot plots showing the selected genes in helper memory T cells across disease conditions. (G) Box plots showing the IL1RA and CD69 cores in helper memory T cells across disease conditions. Significance was evaluated using the Kruskal-Wallis test with Bonferroni correction (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ^nsP > 0.05). (H) Bar plots showing the enrichment Gene Ontology terms in helper memory T cells in patients with mild and severe disease.

We identified two distinct Treg populations, including CD4T_05_Treg_ CCR7_FOXP3 and CD4T_11_Treg_FOXP3_ IL2RA (Figure 4A). DEG analysis, performed to dissect phenotypic and functional heterogeneity in bacterial pneumonia, revealed divergent transcriptional programs of Treg subsets. CD4T_05_Treg_CCR7_ FOXP3 showed elevated expression of CCR7, SELL (encoding L-selectin), and LEF1, suggestive of a resting/naive Treg phenotype (Figure E14B and Table E2). In contrast, CD4T_11_Treg_FOXP3_IL2RA expressed higher concentrations of IL2RA (CD25), HLA-DRB1, TIGIT, and CTLA4, indicative of an activated or effector Treg phenotype (Figure E14B and Table E2). Patients with SBP were associated with a notable reduction in the total frequency of Tregs compared with HCs and patients with MBP (Figures 1D and 4B), suggesting impaired immune regulation in severe infection. Furthermore, differential abundance testing revealed a selective decrease in the CD4T_05_Treg_CCR7_FOXP3 population in severe cases (Figure 4B), indicating a potential vulnerability of this subset in severe pneumonia.

The activated CD4T_11_Treg_FOXP3_ IL2RA subset in patients with SBP showed marked upregulation of immunomodulatory molecules, including IL10, IL1R2, TGFB1, LAG3, PDCD1 (PD-1), HAVCR2 (TIM-3), ICOS, TNFRSF18, and IL2RA (Figure 4C), suggesting an enhanced activation state of this subset in severe cases. Trajectory inference analysis revealed that CD4T_11_Treg_ FOXP3_IL2RA does not appear to originate from CD4T_05_Treg_CCR7_FOXP3 cells but follows a distinct differentiation path, potentially driven by the inflammatory environment of SBP (Figure 4D). In severe pneumonia, activated CD4T_11_Treg_ FOXP3_IL2RA Tregs expressed higher concentrations of TFs associated with core Treg identity (FOXP3, STAT3, STAT5B) and effector responses (IRF4, BATF, MAF) (Figure E14C). Concurrently, increased expression of HIF1A, PRKAA1, and MTOR indicated metabolic adaptation to the hypoxic, nutrient-depleted microenvironment of severe disease. This metabolic shift, together with altered expression of PRDM1 and TBX21, may contribute to Treg dysregulation, potentially impairing inflammation resolution and bacterial clearance.

Receptor-ligand analysis revealed increased interactions between CD4T_11_ Treg_FOXP3_IL2RA and exhausted innate-like CD8⁺ T cells (NKT, MAIT, and γδ T cells) in SBP (Figure E14D). Specifically, signaling increased through inhibitory pathways, such as PDCD1 (PD-1) with PDCD1LG2 (PD-L2), TGFB1 with its receptors TGFB2/3, and HLA-A with KIR3DL1 (Figure 4E). Concurrently, inflammatory interactions (e.g., CSF1-CELSR3, CCL20-CCR6, and IL7-IL2RG) also increased between these two cell populations (Figure 4D). For example, PDCD1–PDCD1LG2 interactions can suppress CD8⁺ T cell effector function, whereas CCL20-CCR6 can promote Treg and other immune cells’ recruitment to inflammatory sites (36). These aberrant interactions may contribute to sustained inflammation and impaired bacterial clearance, ultimately exacerbating disease severity in severe pneumonia.

Helper memory CD4⁺ T cell subsets also exhibited substantial alterations during bacterial pneumonia (Figure 1D). We identified five distinct populations: T follicular helper (Tfh) cells (CD4T_04_Tfh_ICOS_ SLAMF1), central memory T cells (CD4T_ 06_cMemory_CCR7_CD27), Th2 cells (CD4T_07_Th2_CCR4_GATA3), Th22 cells (CD4T_08_Th22_CCR4_CCR10), and Th1 cells (CD4T_10_Th1_CXCR3_ISG20) (Figure 4A). Although Tfh cells maintained a relatively stable frequency across the three groups, the remaining helper memory subsets increased in MBP but markedly decline in SBP (Figure 1D). This response pattern suggests an initial adaptive immune response involving these subsets, followed by their depletion or dysregulation in severe infection.

DGE analysis revealed that helper memory T cells in mild pneumonia exhibited signatures of enhanced activation and bacterial responsiveness (Figure E14E and Table E5). Upregulated genes included those for T cell activation and migration (e.g., CCR2, CD226, STAT1, CDRT4, SARM1, HSPA1A, BOD1, SMARCA5) (37), suggesting robust early responses facilitating recruitment and antigen presentation (Figure 4F). Increased TNFSF10, GBP5, and PYCARD expression indicated enhanced bacterial killing and inflammatory cytokine production. Elevated JCHAIN expression suggested a contribution to mucosal humoral immunity (38). In severe pneumonia, these cells displayed a transcriptional shift toward dysfunction, with upregulation of inhibitory receptors (e.g., LAG3, PDCD1, TIGIT), tissue damage–associated genes (e.g., S100A8, S100A9, ADAM10) and hypoxic response genes (HIF1A) (Figures 4F and 4G) (39). Although elevated IL1RA and CD69 expression might reflect compensatory activation, their coexpression with exhaustion markers suggests dysregulated responses (Figures 4F and 4G). GO term enrichment analysis supported these findings (Figure 4H), revealing, in mild pneumonia, enrichment of pathways associated with T cell migration, cytokine production, and JAK-STAT signaling, suggestive of a coordinated and effective immune response (Figure 4H). In severe pneumonia, however, enrichment was observed for pathways related to inflammation, IFN-γ production, p38MAPK signaling, and protein phosphorylation, reflecting a dysregulated immune response (Figure 4H). This contrast highlights the critical role of helper memory T cell functionality in disease severity.

Dysregulated B Cell Subsets in SBP

Reclustering of B and plasma cells revealed seven distinct populations, annotated using canonical marker expression (Figures 5A and E17A) and characterized by unique transcriptional profiles (Table E2). These included naive, germinal center, intermediate memory, and memory B cells (Figure 5A). We also observed a population of MZB1⁺ plasmablasts with high expression of the proliferation marker MKI67, consistent with previous studies in other infectious diseases (10, 22), as well as IgA⁺ and IgG⁺ plasma cells (Figure 5A). In patients with bacterial pneumonia, R_O/E analysis revealed a marked expansion of plasmablasts and plasma cells, increasing with disease severity (Figure 1D). However, reanalysis after removing an outlier (sample SBP18) from the patient group with SBP revealed no significant difference in plasmablasts and plasma cell expansion between patients with MBP and SBP (Figure 5B). Plasma cells highly expressed immunoglobulin heavy chain constant region genes IGHA1, IGHA2, IGHG1, IGHG2, and IGHM (Figure 5C), suggesting an antigen-specific antibody-secretory function. Expression of IGHA2 and IGHM was elevated in mild disease, suggesting a potential role for IgA2 and IgM antibodies in early disease resolution or containment (Figure 5D). Further investigation is required to elucidate the functional implications of these isotype-specific responses and their contribution to protective immunity.

Figure 5.

Immunological features of B cell subsets. (A) The clustering result (left row) of 7 B cell types (right row) from 100 samples. (B) Box plots showing the percentage of B_05_Plasma_IGHA1, B_06_Plasmablast_MKI67, and B_07_Plasma_IGHA1_IGHG1across disease conditions. Significance was evaluated using the Kruskal-Wallis test with Bonferroni correction (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ^nsP > 0.05). (C) Classes of heavy chains for plasma cells and plasmablasts. (D) Heatmap plots showing the expression of selected genes in plasma cells across disease conditions. (E) Correlation between plasma with T follicular helper cells in patients with mild disease. (F) Dot plots showing the selected genes in plasmablasts and plasma cells across disease conditions. (G) Dot plots showing the selected genes in B cells across disease conditions. (H) Venn diagram illustrating the number of upregulated genes in B cells.

Although circulating Tfh cell numbers were reduced in patients with bacterial pneumonia (Figure 1D), a strong positive correlation between Tfh and plasma cell counts was evident in mild disease (Figure 5E) but absent in severe disease (Figure E17B). This suggests that coordinated T and B cell responses may contribute to effective humoral immunity during mild disease, a process potentially disrupted in severe disease. This observation aligns with previous studies, demonstrating Tfh cell necessity for optimal antibody responses and high-affinity neutralizing antibody generation during infection (22, 40). DGE analysis revealed 4,685 upregulated genes in plasmablasts and plasma cells from patients with MPB compared with HCs, whereas 3,607 upregulated genes were identified in these cell populations from patients with SBP (Figure E17C). This difference in upregulated gene numbers further supports distinct B cell immunological landscapes in MBP versus SBP. Of these DGEs, 3,470 were upregulated in both mild and severe pneumonia cohorts, whereas 1,215 were uniquely upregulated in mild cases (Figure E17C). GO analysis of this mild disease–specific gene set revealed significant enrichment of pathways related to mRNA processing, splicing, and translation, as well as endoplasmic reticulum protein folding, complex assembly, and secretory pathways (Table E6 and Figure E17D). These findings suggest a heightened capacity for antibody production and quality control during early infection in contrast to a potential exhaustion or dysregulation of these cellular processes in severe cases. Consistently, key genes and TFs involved in antibody trafficking, secretion, folding, function, stability, targeting, degradation, and expression regulation were upregulated in MBP (Figure 5F). Overall, these results highlight the dynamic B cell response during bacterial pneumonia, indicating effective humoral immunity in mild disease that may become impaired with increasing severity.

B cell activation and differentiation into antibody-secreting plasma cells are crucial for resolving bacterial infections. To determine B cell roles in bacterial pneumonia severity, we examined B cell heterogeneity and activation status in mild and severe disease. In mild cases, all B cell subsets exhibited increased expression of genes associated with antigen recognition, B cell receptor (BCR) signaling, and costimulation (e.g., CD79A, CD79B, BTK, and BLK) (Figures 5G and E17E). Further DGE analysis revealed specific upregulation of 135 genes in mild cases, some of which are directly related to B cell activation and function, including TRAF3IP2-AS1, GRAP2, and SPIB (Figure 5H and Table E7). For example, SPIB encodes a TF essential for B cell development and antibody responses (41) (Figure 5H). Other upregulated genes include ORAI3 (mediating calcium signaling critical for B cell activation), LEF1 (a Wnt pathway downstream target associated with B cell development and proliferation) (42), and TLR9 (a bacterial DNA receptor triggering B cell activation) (43) (Figure 5H). Consistently, GO enrichment analysis of mild disease-specific DEGs confirmed upregulation of pathways related to antigen presentation, positive regulation of B cell activation, and T cell receptor signaling (Figure E17F–E17H). These results suggest that robust B cell activation and development in mild cases may contribute to the observed plasma cell expansion and efficient antibody production, potentially facilitating more effective bacterial clearance.

Unlike the robust B cell activation in mild cases, severe cases exhibited a transcriptional signature indicative of B cell dysfunction (Table E6). B cells from severe cases showed reduced expression of genes essential for antigen recognition, BCR signaling, and costimulation, including CD79A, CD79B, BTK, and BLK, among others (Figures 5G and E17E). DGE analysis revealed a distinct transcriptional profile in B cells from patients with SBP, characterized by the upregulation of genes such as DONSON, SAMD4A, HIVEP1, NCOR2, etc. (Table E7). For instance, NCOR2 encodes a transcriptional corepressor that modulates gene expression by interacting with nuclear hormone receptors and other TFs, and its increased expression could suppress key genes in B cell activation and antibody production (44). Similarly, CXCR4, a chemokine receptor involved in B cell migration and lymphoid tissue positioning, was upregulated, which might disrupt appropriate B cell trafficking, hindering interactions with other immune cells and antigen presentation (45). This B cell dysregulation was further corroborated by GO enrichment analysis (Figure E17F and E17G), which revealed potential impairment in B cell responses during severe cases, potentially contributing to the more severe clinical manifestations. These data highlight the need for further investigation into the mechanisms underlying B cell dysfunction in severe cases and for exploring therapeutic strategies to restore effective humoral immunity.

Dysregulated Myeloid Cells in SBP

Reclustering based on canonical cell markers and highly expressed genes identified 13 myeloid cell subsets, including monocytes, DCs, and megakaryocytes (Figures 6A and E18A and Table E2). Monocytes comprised nine subsets: three classical (Mono_01_ Classical_CD14, Mono_02_Classical_CD83_ HLA-DPB1, and Mono_03_Classical_ CD83_CCL5), two intermediate (Mono_04_ Intermediate_CD14_CD16 and Mono_05_ Intermediate_CD83_HLA-DPB1), one nonclassical (Mono_06_Non-classical_CD16_ C1QA), two myeloid-derived suppressor cells (MDSCs; Mono_07_MDSC_CD24_ARHGAP26 and Mono_08_MDSC_S100A8_ S100A9), and one proliferative (Mono_09_ Proliferation_MKI67) subset (Figures 6A and E18A). The Mono_09_Proliferation_MKI67 subset, previously identified in other infectious diseases by flow cytometric analysis of blood (46), transcriptionally resembled CD14⁺ monocytes and was characterized by proliferative markers (e.g., MKI67 and TYMS), confirming its identity as a proliferating monocyte population. DCs were subdivided into three established subsets: myeloid DCs (DC_01_mDC_CD1C_CLEC10A), plasmacytoid DCs (DC_02_pDC_LILRA4_ ITM2C), and type 1 conventional DCs (DC_03_DC1_CLEC9A_CADM1) (Figures 6A and E18A). Among these myeloid cell subsets, all monocyte subsets except nonclassical monocytes were notably elevated in patients with SBP. Conversely, all DC subsets were significantly decreased in these same patients (Figures 6A and E18A). These results suggest that monocyte expansion and concomitant DC population reduction characterize the myeloid response in patients with SBP.

Figure 6.

Immunological features of myeloid cells. (A) The clustering result (left row) of 13 myeloid subtypes (right row) from 100 samples. Each point represents a single cell, colored according to myeloid subtype. (B) Bar plots showing the expression of S100A8/A9 and HLA-DRA/B5/B1 in monocytic myeloid-derived suppressor cells across disease groups. (C) Heatmap plots of the selected genes in monocytic myeloid-derived suppressor cells across disease groups. (D) Heatmap plots of selected genes (C1QA/B/C) in nonclassical monocyte subset (Mono_06_Non-classical_CD16_C1QA) across disease conditions. (E) Partition-based graph abstraction analysis of monocyte pseudotime in patients with severe (left) and mild (right) disease: The associated cell type and the corresponding status are listed. (F) Venn diagram illustrating the number of upregulated genes in classical monocytes. (G) Bar plots showing cytokine and inflammatory scores in classical monocytes across disease groups. (H) Dot plots showing the selected genes in classical monocytes across disease conditions.

Within the monocyte subsets, two monocytic MDSC (mMDSC) subtypes were identified (Figure 6A), representing a heterogeneous population of immature myeloid cells known to expand during inflammation and suppress T cell responses (22). In peripheral blood, classical monocytes are characterized by a CD14⁺HLA-DR⁺ phenotype, whereas mMDSCs exhibit a CD14⁺ HLA-DR⁻/^low phenotype (Figure E18B). mMDSCs are further characterized by increased calprotectin expression and immunosuppressive activity. Elevated calprotectin and HLA-DR downregulation in Mono_07_MDSC_CD24_ARHGAP26 and Mono_08_MDSC_S100A8_S100A9 from patients with SBP indicated that these clusters closely resemble canonical MDSCs (Figure 6B). In severe disease, mMDSC displayed elevated expression of multiple immunomodulatory molecules, including arginase 1 (ARG1), indoleamine 2,3-dioxygenase 1 (IDO1), programmed cell death 1 ligand 1 (PD-L1; encoded by CD274), inducible nitric oxide synthase (iNOS; encoded by NOS2), NADPH oxidase 1 (NOX1), IL-10, transforming growth factor-β1 (TGF-β1) and TGF-β3, and ectonucleoside triphosphate diphosphohydrolase 1 (ENTPD1) (Figure 6C). This transcriptional signature suggests that mMDSCs are equipped for multifaceted T cell suppression. For example, ARG1 can deplete L-arginine, essential for T cell proliferation and function, whereas iNOS can generate nitric oxide (NO) that inhibits T cell proliferation, induces apoptosis, and disrupts cytokine production. Although ARG1 and iNOS compete for their common substrate L-arginine, potentially influencing the relative activity of each pathway, the concurrent upregulation of both ARG1 and NOS2 implies a potent capacity of these mMDSCs to create an immunosuppressive microenvironment through both L-arginine consumption and NO generation. PD-L1 binding to PD-1 on T cells inhibits their activation and promotes exhaustion, and IDO1-mediated tryptophan catabolism, depleting this essential amino acid, leads to T cell anergy and apoptosis (47, 48). Ligand–receptor interaction analysis validated PD-L1 (CD274) and TGF-β1/3 interactions on mMDSCs with PD-1 and TGF-βR2/R3 on T cells, respectively (Figure E18). Other ligand–receptor interactions, such as CD274-CD80, LAMC1-ITGA2, and VSTM1-ADGRG3, among others, may also contribute to T cell suppression. These findings implicate mMDSCs as critical immunosuppression mediators in SBP, potentially impairing pathogen clearance.

A nonclassical monocyte subset (Mono_06_Non-classical_CD16_C1QA), characterized by a CD16⁺CD14⁻ phenotype and high C1QA/B/C expression, was also identified (Figure 6A). Further analysis revealed that monocytes were the primary source of C1 complement components in peripheral blood (Figure E19). Increased C1QA/B/C expression was observed in patients with SBP (Figure 6D), indicating complement overactivation. This activation can induce several pathological changes, including enhanced inflammation, tissue damage, impaired lung function, systemic complications, and impaired bacterial clearance. For instance, C1 activation triggers a cascade generating anaphylatoxins C3a and C5a, thereby increasing vascular permeability. This process results in edema and inflammatory cell infiltration into the lung tissue. Moreover, C3a and C5a are potent chemoattractants for neutrophils and macrophages, amplifying the inflammatory response (49). These recruited cells release proteases and reactive oxygen species, thereby contributing to lung tissue damage. These results suggest that nonclassical monocytes, via C1q production, may contribute to complement-mediated inflammation and lung injury in severe cases.

PAGA trajectory inference of monocyte subsets revealed distinct developmental pathways in bacterial pneumonia, differing markedly by disease severity (Figure 6E). In severe pneumonia, classical monocytes apparently sourced both intermediate monocytes and mMDSCs, aligning with inflammatory stimuli driving mMDSC differentiation from precursors (50). A separate trajectory led to nonclassical monocytes, suggesting divergent development despite a shared myeloid progenitor. Both mMDSC and nonclassical monocyte trajectories were prominent in severe disease, corroborated by increased cell numbers and stronger network connections. Conversely, the mild cases’ more interconnected PAGA graph indicated enhanced interplay among monocyte subsets (Figure 6E). This suggests a monocyte differentiation shift toward immunosuppressive and complement-modulating phenotypes (mMDSCs and nonclassical monocytes) in severe pneumonia, whereas milder inflammation permits greater monocyte plasticity and interconnectivity.

Because classical monocytes are the predominant myeloid population (Figure E19B) and drive inflammatory responses in bacterial pneumonia (Figure 2), we investigated their transcriptional heterogeneity across severities. DEG analysis identified 2,140 and 2,370 upregulated genes in severe and mild pneumonia, respectively (Figure 6F and Table E7). Although 1,895 genes were common to both, 245 and 230 were uniquely upregulated in severe and mild cases, respectively (Figure 6F), indicating shared and distinct immune responses across severities. In severe pneumonia, classical monocytes exhibit a transcriptional profile indicative of enhanced inflammation (e.g., IL1R1, complement components [e.g., C1QA and C2], and S1PR4), metabolic reprogramming (CD38, altered lipid metabolism genes [e.g., PLIN2, STARD13, ABHD2, and ACOX2]) (Figures 6F–6H). Concurrently, markers of oxidative stress (MAOA, ALOX5AP) and tissue remodeling (THBS1) suggested ongoing lung injury (Figures 6F–6H). These pathophysiological adaptations in classical monocyte function appear to be orchestrated by a transcriptional network involving KLF9, ETFB, and ETS family members, highlighting complex response regulation in severe pneumonia (Figures 6F–6H). In contrast, classical monocytes in mild cases display a transcriptional signature of early innate immune activation (TLR1, TLR8), modulated inflammatory signaling (IL17RA, SOCS3), and altered cellular processes (Figures 6F–6H). Enhanced vesicular trafficking/sorting (e.g., SORT1, SNX family, SEC22A, RTN1, RAB11FIP1, LYST) and cytoskeletal regulation (e.g., ROCK2, AHNAK, MYO5A, DIAPH1, PIP5K1B) likely facilitate pathogen handling and immune cell migration (Figures 6F–6H). A complex interplay of kinases, phosphatases, and TFs (e.g., JAK1, RUNX2, EGR family members, STAT3, and CTCF) underpinned these functional changes, reflecting coordinated monocyte transcriptional responses in mild pneumonia. GO enrichment analysis further revealed potential classical monocyte impairment in severe cases, possibly worsening clinical manifestations (Figure E19C).

Discussion

This study presents the first large-scale single-cell transcriptomic analysis of peripheral immune responses in bacterial pneumonia, providing insights into both disease pathogenesis and protective mechanisms. Our findings systematically reveal profound peripheral immune remodeling, where disease severity significantly shapes immune cell composition and transcriptional profiles. Observed lymphopenia (Figure 1), particularly marked in severe pneumonia, suggests compromised adaptive immunity, potentially worsening disease progression (51). The mechanisms for this widespread T cell depletion (e.g., increased cell death, decreased proliferation, or circulatory egress) require further investigation. In contrast, expanded monocyte compartments, especially classical monocytes, in severe pneumonia likely reflect their key roles in initiating and amplifying inflammatory responses (Figure 2). Although essential for bacterial clearance, this monocyte-driven inflammation can also exacerbate lung injury via cytokine storms, underscoring the critical protective–pathogenic immune balance. Plasma cell and plasmablast enrichment, predominantly in mild pneumonia, indicates a robust humoral response associated with effective pathogen control. Increasing monocyte expansion with concomitant lymphocytes as disease severity worsens highlights the complex innate–adaptive interplay in shaping disease outcome.

Our analysis revealed a central role for monocytes, particularly classical monocytes, in driving the cytokine storm characteristic of severe disease, a hyperinflammatory state likely contributing to adverse outcomes. We identified several monocyte subsets with heightened inflammatory signatures, with classical monocytes being substantial contributors to overall cytokine production (Figure 2), implicating them as primary orchestrators of the cytokine storm, potentially via mediators such as S100A8/9/12, IL-15, and CXCL1/8. Increased plasma S100A8/9/12 further supports their clinical relevance and biomarker potential. The S100A8/9/12-TLR4-MYD88 signaling axis in these monocytes suggests a potential positive feedback loop amplifying inflammation and contributing to lung injury. Particularly, although IL-15 is known to promote T cell proliferation and effector function (52, 53), its increased expression by inflammatory monocytes in the context of SBP, alongside numerous other proinflammatory and potentially immunosuppressive signals, suggests a complex role. The elevation of IL-15 here may reflect an attempted compensatory response or contribute to the broader inflammatory milieu, the net effect of which on T cell functionality in SBP requires consideration of the overall cytokine and cellular interaction landscape. In addition to IL-15, the pronounced peripheral elevation of CXCL8, a potent chemokine with proinflammatory cytokine properties, contributes to the systemic inflammatory milieu and strongly signals for neutrophil mobilization, activation, and recruitment (32), complementing local lung neutrophilia (14). Although neutrophils were not quantified in our PBMC cohort, interactions between systemically primed neutrophils and the dysregulated T, B, and monocyte populations could exacerbate immunopathology and impair pathogen clearance, a common feature of severe bacterial infections. Intricate ligand–receptor interactions among inflammatory monocyte subsets, mediated by molecules such as CXCL8, TNF, and IL-1β, underscore the complex intercellular communication networks driving this hyperinflammation. Targeting these pathways, including S100 proteins and their downstream signaling cascades, may therefore provide therapeutic strategies for SBP.

Beyond monocyte-driven inflammation, SBP also induces profound T cell dysregulation, potentially contributing to disease severity. Innate-like CD8⁺ T cells (MAIT, NKT, and γδ T cells) exhibited marked exhaustion in severe cases (Figure 3), characterized by increased expression of inhibitory receptors (PD-1, LAG-3, CTLA-4) and the TF PRDM1. This exhaustion highlights functional impairment of these crucial early responders, further evidenced by upregulated IFN signaling–associated genes and signatures of reduced CD4⁺ T cell help. Although conventional CD8⁺ T cells initially expanded in MBP, they displayed reduced cytotoxic capacity in SBP, potentially leading to uncontrolled bacterial growth. Transcriptionally, effector CD8⁺ T cells in severe cases were enriched for stress response and regulatory pathways over cytotoxic functions, suggesting a shift from pathogen control toward a dysfunctional state. This shift, potentially driven by antigen exposure and/or an altered inflammatory milieu, may contribute to the uncontrolled bacterial burden observed in severe pneumonia. Within the CD4⁺ T compartment, Tregs, particularly the resting/naive subset, were significantly reduced in patients with SBP (Figure 4). Although activated Tregs in severe cases expressed higher immunomodulatory molecule levels, they also displayed transcriptional signatures of metabolic adaptation and potential functional dysregulation, implying a compromised Treg response that could exacerbate the dysregulated immune environment and contribute to pathogenesis. Furthermore, increased inhibitory interactions between these activated Tregs and exhausted innate-like CD8⁺ T cells fostered an immunosuppressive environment. Finally, although helper memory CD4⁺ T cell subsets initially expanded in mild pneumonia, their frequencies declined in severe cases, showing transcriptional signatures of dysfunction. This multifaceted dysregulation across innate-like and conventional T cell subsets underscores the impact of SBP on adaptive immunity and highlights potential therapeutic targets.

Our results reveal a complex interplay of B cell activation and dysfunction in bacterial pneumonia, significantly impacting humoral immunity and disease outcome. Although plasmablast and plasma cell expansion occurred in pneumonia, it did not differ significantly between mild and severe cases (Figure 5), suggesting that plasma cell numbers alone are not reliable indicators of disease severity. Instead, humoral response quality appears to be crucial. In mild pneumonia, Tfh and plasma cell numbers were positively correlated, suggesting coordinated Tfh–B cell interaction, crucial for generating high-affinity, antigen-specific antibodies and promoting effective pathogen clearance. This coordinated response was absent in severe disease, potentially indicating disrupted Tfh-mediated B cell help and impaired humoral immunity. Particularly, elevated IGHA2 and IGHM expression in plasma cells from mild cases suggests specific roles for IgA2 and IgM in early disease resolution via mucosal immunity and complement activation, respectively; thus further investigation into their isotype-specific responses is warranted. In addition, increased antibody production and quality control in plasmablasts and plasma cells from mild cases, evidenced by upregulated genes related to mRNA processing, endoplasmic reticulum protein folding, and secretion, further highlight the effectiveness of the humoral response during early infection. This robust early response contrasts sharply with the B cell dysfunction in severe pneumonia. B cells from severe cases exhibited reduced expression of genes essential for antigen recognition, BCR signaling, and costimulation, suggesting an impaired ability to mount an effective antibody response. This dysfunction was corroborated by differential gene expression analysis, revealing upregulation of genes associated with transcriptional repression (e.g., NCOR2) and altered B cell trafficking (e.g., CXCR4) in severe cases. These transcriptional changes may contribute to the impaired antibody production and ultimately hinder pathogen clearance. Combined B cell dysfunction and disrupted Tfh–B cell interaction likely creates a permissive environment for bacterial persistence and dissemination, exacerbating disease progression. Therefore, therapeutic strategies aimed at restoring B cell function and promoting Tfh–B cell cooperation may improve outcomes in SBP.

Myeloid cell dynamics critically shape the immune response to bacterial pneumonia (54), with distinct alterations in severe disease (Figure 6). Severe pneumonia featured significant expansion of monocyte subsets (classical, intermediate, and MDSCs) alongside a marked reduction in all DC subsets. This myeloid compositional shift suggests a bias toward innate immune activation and potentially impaired antigen presentation, hindering the development of effective adaptive immunity. We identified two mMDSC subtypes exhibiting elevated expression of immunomodulatory molecules (ARG1, IDO1, PD-L1, iNOS) in severe disease, implicating mMDSCs in driving immunosuppression. Particularly, mMDSC-mediated immunosuppression likely contributed to the observed T cell dysfunction and impaired bacterial clearance (Figure 6). Furthermore, elevated C1QA/B/C expression by nonclassical monocytes in severe cases points to complement overactivation, a potential driver of inflammation and lung injury. Trajectory inference revealed distinct monocyte differentiation pathways in severe versus mild disease. In severe pneumonia, a prominent trajectory toward mMDSCs and nonclassical monocytes suggests preferential differentiation toward immunosuppressive and complement-modulating phenotypes, potentially exacerbating immune dysregulation. This contrasts with the more interconnected monocyte differentiation in mild pneumonia, indicating greater plasticity and potentially more effective immune responses. Classical monocytes also exhibited distinct transcriptional programs associated with disease severity (Figure 6). In severe cases, they upregulated genes associated with inflammation, metabolic reprogramming, oxidative stress, and tissue remodeling, reflecting a dual role in both pathogen control and concomitant lung damage. In contrast, classical monocytes in mild cases displayed transcriptional signatures of early innate immune activation, modulated inflammatory signaling, and enhanced cellular processes (e.g., vesicular trafficking, cytoskeletal regulation), potentially aiding pathogen clearance and immune cell migration. Collectively, these findings underscore the dynamic and context-dependent nature of myeloid cell responses in bacterial pneumonia. Targeting specific myeloid subsets and pathways, such as mMDSC-mediated immunosuppression or complement overactivation, may therefore provide promising therapeutic strategies for modulating the immune landscape and improve clinical outcomes in SBP.

Our present study of PBMCs provides a systemic counterpoint to recent characterizations of the local lung immune microenvironment in bacterial pneumonia using BAL fluid (BALF) from a similar patient cohort (14). Although the BALF study detailed the lung-specific inflammatory storm and the dysregulation of neutrophils and macrophages at the site of infection, the present study delineates the peripheral immune panorama, with key novel findings on the systemic inflammatory signatures and pronounced dysregulation of circulating T cell, B cell, and monocyte compartments. For instance, although both studies identify an inflammatory storm driven by similar cytokines (e.g., S100A8/A9 and CXCL8), our PBMC analysis pinpoints classical monocytes as major peripheral contributors and elucidates the S100A8/A9/A12-TLR4-MYD88 axis within them (Figure 2), suggesting coordinated systemic and local inflammatory dysregulation. Although T cell exhaustion is evident in both compartments, this study reveals a distinct and profound exhaustion signature in peripheral innate-like CD8⁺ T cells (MAIT, γδ T, and NKT-like cells) as a hallmark of severe cases in blood (Figure 3), differing in emphasis and specific subset involvement from the conventional T cell exhaustion patterns in BALF. In addition, the present study provides detailed insights into peripheral B cell dynamics (Figure 5), including plasmablast expansion and antibody isotype profiles, which are critical systemic responses. Thus, by examining PBMCs, this study extends our understanding beyond the local infection site, revealing unique aspects of systemic immune dysfunction and highlighting how interconnected peripheral and lung immune responses exhibit distinct characteristics and cellular contributions during bacterial pneumonia. Furthermore, comparing our findings with the well-characterized immune response in viral pneumonia (e.g., COVID-19) helps delineate bacteria-specific pathogenic mechanisms. Although SBP shares features with severe COVID-19, including profound lymphopenia and T cell exhaustion (18, 33), our data point to distinct upstream triggers and cellular responses. A key distinction is the prominent S100A8/A9/A12-TLR4-MyD88 inflammatory axis we identified in classical monocytes, characteristic of a response to bacterial components, which contrasts with the dominant type I IFN signature often reported in severe viral pneumonia (55). Moreover, the profound exhaustion signature we observed specifically within the peripheral innate-like CD8⁺ T cell compartment (MAIT, γδ T, and NKT-like cells) provides a more granular view of immunosuppression that appears particularly pronounced in this bacterial pneumonia cohort.

Limitations

This study, although providing a comprehensive overview of peripheral immune responses in bacterial pneumonia, has several limitations. First, focusing solely on PBMCs may not fully capture the immunological landscape within the lung microenvironment. Analysis of local samples (e.g., BALF) would provide a more complete understanding. Second, this study did not include a viral pneumonia cohort for direct comparison, which would be valuable for delineating bacteria-specific versus shared respiratory infection immune signatures (56). Third, despite our cohort size, the inherent heterogeneity of bacterial pneumonia (e.g., pathogens, comorbidities, prior treatments) may have influenced our findings. Furthermore, given the prevalence of gram-negative pathogens in our cohort, our findings regarding the S100A8/A9/A12-TLR4-MyD88 axis are most directly applicable to these infections. Gram-positive pneumonia, in contrast, typically engages alternative pattern recognition pathways. For example, gram-positive pathogens are recognized via TLR2, and many, such as Streptococcus pneumoniae, produce pore-forming toxins such as pneumolysin that can directly activate the NLRP3 inflammasome, leading to pyroptosis and a distinct inflammatory cascade. Although the initial triggers differ, it remains to be determined whether these distinct upstream pathways converge on shared downstream inflammatory responses and cellular dysregulation. Finally, our study primarily provides a descriptive analysis of transcriptional results at single-cell resolution. Functional studies are crucial to establish causality and investigate the therapeutic potential of identified pathways and cellular subsets.

Conclusions

In conclusion, our high-resolution single-cell atlas of the peripheral immune response in bacterial pneumonia reveals cellular and molecular signatures of disease severity. Monocytes are key drivers of the cytokine storm characteristic of severe disease, accompanied by substantial T and B cell dysregulation, including T cell exhaustion, impaired cytotoxicity, and B cell dysfunction. We further characterized dynamic myeloid alterations, highlighting mMDSCs and complement overactivation contributions to the immunosuppressive environment in severe pneumonia. These findings offer crucial insights into innate–adaptive immune interplay in bacterial pneumonia severity and establish a framework for future mechanistic studies and therapeutic development.