This study consisted of two cohorts, including a retrospective cohort and a prospective cohort. For the retrospective cohort, we collected real-world medical records from Taipei Veterans General Hospital, a tertiary medical center in Taiwan. Patients with stage IV NSCLC who harbored wild-type EGFR and ALK treated with either chemotherapy alone, immunotherapy alone, or both were included. The immunotherapy included PD-1/PD-L1 inhibitors, and the chemotherapy consisted of platinum-doublet chemotherapy and single-agent chemotherapy according to the clinical practice guidelines (45). The following patients were excluded from the analysis: (1) those with known EGFR or ALK mutations who had received first-line targeted therapy; (2) those diagnosed with small-cell carcinoma; and (3) those who underwent double immunotherapy. The period of patient enrollment spanned from January 1, 2018, to December 31, 2020, while the data collection and follow-up time for the monitoring of the infection episodes was from January 1, 2018, to March 31, 2022. The study was approved by the Institutional Review Board of Taipei Veterans General Hospital (2022-08-010BC), and the requirement for informed consent was waived.

For the prospective cohort, patients with NSCLC were recruited from Taipei Veterans General Hospital from 2020 to 2022. Patients with stage IV NSCLC undergoing systemic treatment, such as immunotherapy (IO), chemotherapy (C/T), or both were included. Patients were categorized into three groups based on their treatment modalities: the C/T alone group, the IO alone group, and the IO+C/T group. The C/T alone group received chemotherapy only, the IO alone group received immunotherapy only, and the IO+C/T group received a combination of immunotherapy and chemotherapy. The following patients were excluded from this study: (1) patients who had received steroids at a dosage exceeding prednisolone 10 mg/day within the preceding 14 days and (2) patients who had undergone major surgery or radiotherapy within the preceding 30 days; (3) patients with known EGFR or ALK mutation, received first-line targeted therapy; (4) patients diagnosed with small-cell carcinoma; (5) patients who underwent double immunotherapy. Medical records were reviewed to collect crucial clinical characteristics, including age, sex, tumor node metastasis (TNM) staging, systemic treatment regimen, and the timing of infection. The blood specimens were collected for the immunophenotyping analysis. The study was approved by the Institutional Review Board of Taipei Veterans General Hospital (2020-05-006B), and all participants provided informed consent prior to participation in this study.

Whole-blood specimens were collected through phlebotomy from the patients on the day before starting systemic treatment (Day 0) and on one week after receiving first dose of medication (Day 8). Peripheral blood mononuclear cells (PBMCs) were isolated by Ficoll density gradient centrifugation within 6 hours from the time of blood collection. The isolated PBMCs were cryopreserved until subsequent testing and analysis.

The immunophenotyping of various lymphocyte subpopulations and monocytes were performed by mass cytometry (CyTOF). The experiment started with a careful design of the antibody/probe panel. This was followed by sample analyzing by CyTOF 2 mass cytometer (Fluidigm, San Francisco, CA, USA), uploading the flow cytometry standard (FCS) file to the online FCS file-processing platforms, and finally data analysis ( Supplementary Figure S1 ). The CyTOF staining process involved several sequential steps. Initially, the cells were treated with a rhodium viability staining reagent (Fluidigm) for 10 min before commencing the staining procedure. Subsequently, Cell-ID Intercalator-103Rh (Fluidigm, Catalog #201103A) was added to the cells, with incubation for 10 minutes before the introduction of CyTOF staining antibodies. The dilution of antibodies started at the concentration recommended by the manufacturer (1 μl antibody per 100 μl cell suspension containing 3 × 10⁶ cells). According to the manufacturer’s protocol for staining the antibody, 50 μL of the 2X antibody cocktail was added to each tube and the total staining volume is 100 μL (50 μL of cell suspension + 50 μL 2X antibody cocktail). Then, the cells were incubated with intracellular antibodies for 30 minutes. Following incubation, the cells were washed once with CyTOF staining buffer (a solution containing cell staining buffer, calcium/magnesium-free PBS, 0.2% BSA, and 0.05% sodium azide; Fluidigm; Catalog #201068, San Francisco, CA, USA). To fix the cells, 1.6% formaldehyde buffer (16% formaldehyde [weight by volume], methanol-free dilution; Thermo Scientific™, Catalog #28908, Waltham, MA, USA) was used for 10 minutes. After fixation, the cells were washed twice with CyTOF staining buffer. Finally, the cells were diluted in distilled water following the last washing step and then injected into the mass cytometer. These cells were subsequently analyzed using a CyTOF 2 mass cytometer (Fluidigm) in accordance with the manufacturer’s guidelines. Data analysis was performed using OIMQ data analysis software (OMIQ, Inc., Santa Clara, CA, USA). Lastly, the gated CD45⁺ cell population was clustered based on all labeled phenotypic markers through the spanning-tree progression analysis of density-normalized events (SPADE) method (OMIQ, Inc. Santa Clara, CA, USA).

The immunophenotyping of the following lymphocyte and monocyte subpopulations was conducted through CyTOF (Fluidigm) under various conditions: T cells (CD45⁺CD3⁺) labeled with antibodies of Catalog #3141009C and #3170007C; cytotoxic T cells (CD45⁺CD3⁺CD8⁺) labeled with antibodies of Catalog #3141009C, #3170007C, and #3146001C; helper T cells (CD45⁺CD3⁺CD4⁺) labeled with antibodies of Catalog #3141009C, #3170007C and #3145001B; natural killer (NK) cells (CD45⁺CD16⁺) labeled with antibodies of Catalog #3141009C and #3209002C; exhausted T cells (CD45⁺CD3⁺CD8⁺PD1⁺ or CD45⁺CD3⁺CD8⁺TIM3⁺ or CD45⁺CD3⁺CD8⁺LAG3⁺) labeled with antibodies of Catalog #3141009C, #3170007C, #3146001C, #3155009C, #3153008C, and #3165037C; B cells (CD45⁺CD19⁺) labeled with antibodies of Catalog #3141009C, and #3158032C; and monocytes (CD45⁺HLADR⁺CD14⁺) labeled with antibodies of Catalog #3141009C, #3174001C, and #3160006B (all from Fluidigm) (46–50). To manage the multidimensional data obtained from CyTOF, we used specialized tools in the OIMQ data analysis software (OMIQ, Inc.), including the optimized t-Distributed Stochastic Neighbor Embedding (opt-SNE) plot and histogram, which facilitated dimensionality reduction and offered a comprehensive visualization of events within a 2-dimensional map derived from the multidimensional data (51). Immune signatures in PBMCs from healthy controls (n = 3) were analyzed through mass cytometry (CyTOF), and the opt-SNE plots and histograms revealed the obvious expression of CyTOF markers, including CD45, CD3, CD8, CD4, CD16, PD-1, TIM-3, and LAG-3, in the PBMCs ( Supplementary Figure S2 ).

The blood specimens collected from the various treatment groups were centrifuged at 4°C and 1500 rpm. The resulting plasma was collected and cryopreserved until further testing. To determine the concentrations of various proteins in the plasma, we employed the Quantikine enzyme-linked immunosorbent assay (ELISA) Kit (R&D Systems, Minneapolis, MN, USA; Human interleukin [IL]-2 ELISA Kit, Catalog #D2050; Human IL-6 ELISA Kit, Catalog #D6050; Human IL-10 ELISA Kit, Catalog #D1000B; Human tumor necrosis factor-alpha [TNF-α] ELISA Kit, Catalog #DTA00D; Human interferon-gamma [IFN-γ] ELISA Kit, Catalog #DIF50C). According to the manufacturers’ protocol, the processed specimens were incubated at room temperature for 2 hours, after which they were subjected to four washes with the given wash buffer. The recommended volume of conjugate antibody was added to each well. The washing step was then repeated, after which 200 μL of substrate solution was added to each well, and the processed specimens were incubated at room temperature for 30 minutes. Then, 50 μL of stop solution was added to each well. The experimental plates were read on a Spectramax iD3 plate reader (Molecular Devices, San Jose, CA, USA) at 450 nm within 30 minutes. Positive and negative controls were included in the analysis. The change of cytokine levels was demonstrated by using Log₂ fold change in the concentration of different cytokines in Day 8 (D8) compared with that in Day 0 (D0) (52).

Clinical data, including age, sex, smoking history, performance status, comorbidities, and the lines of cancer treatment, were collected and analyzed. Radiotherapies, both in the thoracic and extrathoracic areas, were recorded. Neutropenia, defined as an absolute neutrophil count (ANC) of <500/mm³, was recorded. IrAEs were recorded based on the review of medical records. The diagnosis of infections or pneumonia in the patients with lung cancer was made by clinical physicians in adherence to established clinical practice guidelines (53–55). A comprehensive evaluation following practice guidelines was conducted to differentiate between infectious pneumonia and pneumonitis related to immunotherapy or radiation. The evaluation involved assessing symptoms, performing chest CT scans, bronchoscopy with bronchoalveolar lavage (BAL), evaluating pulmonary function, conducting lung biopsy and determining the appropriate therapy (56–58). In the retrospective cohort, steroid exposure was defined as the daily administration of prednisone or other equivalent steroids at a dose of 10 mg or higher for at least 10 days. Infection episodes requiring the administration of either oral or intravenous antibiotics were identified through a review of medical records. The types of infection included (1) pneumonia, (2) urinary tract infection, (3) bacteremia, (4) skin and soft tissue infection, and (5) others (including intraabdominal infection, colitis, neutropenic fever, and occult infection). The infection episodes were counted, and any instances that necessitated hospitalization or admission to the intensive care unit (ICU) were also documented. Treatment response evaluation was performed according to the Response Evaluation Criteria in Solid Tumors (RECIST) group criteria (version 1.1). Progression-free survival (PFS) was calculated from the date of treatment commencement to the earliest identified sign of disease progression, as determined by the RECIST criteria, or the date of death from any cause. If disease progression had not occurred at the last follow-up visit, PFS was considered censored at that time point. Furthermore, overall survival (OS) was computed from the date of treatment commencement until the date of death or last follow-up visit. Throughout the follow-up period, any incidence of pneumonia/irAEs was also recorded.

Categorical variables were assessed using Pearson’s chi-squared test or Fisher’s exact test. Continuous variables were compared using Student’s t test and the Mann–Whitney U test. To assess the risk factors for infection and pneumonia, logistic regression models were used. Survival analysis was conducted by using the Kaplan–Meier method with a log-rank test. The Cox proportional-hazards regression model was applied for univariate and multivariate survival analyses. Variables that exhibited a significance level of p < 0.1 in the univariate analysis were included in the multivariate analysis. Statistical significance was set at p < 0.05 (two-sided). SPSS software (version 21.0, SPSS Inc., Chicago, IL, USA) was used for all the analyses.

In the retrospective cohort, a total of 283 patients with stage IV NSCLC were included. These patients were categorized into three groups according to their treatment regimen: C/T group (n = 139), IO group (n = 63), and C/T+IO group (n = 81). The median age of the cohort was 63 years (range, 33–96 years). Among all patients, 45.9% were male, and 37.7% were ever-smokers. The majority of the patients exhibited an Eastern Cooperative Oncology Group performance status (ECOG PS) of 0–1, and 24.3% of the patients had received more than two lines of treatment. Among the three groups, the median age was significantly higher in the IO alone group. A greater proportion of patients were male and had smoking history, ECOG PS 0–1, and radiotherapy in the IO alone group. The patients who received the C/T+IO treatment exhibited higher rates of neutropenia and previous steroid exposure. A greater proportion of patients who received C/T alone had more than two lines of treatment ( Table 1 ). No difference in the frequency of comorbidities such as diabetes mellitus (DM), chronic obstructive pulmonary disease (COPD), and chronic kidney disease (CKD) was noted among the three groups. All the baseline characteristics are listed in Table 1 .

Of the 283 patients, 170 (40.9%) had one or more episodes of infection ( Supplementary Table S1 ). No significant differences in the incidence of infection were found among the three groups. Pneumonia was the most prevalent type of infection, accounting for 36% of infections and exhibiting a relatively consistent incidence across the three groups ( Supplementary Table S1 ). In addition to pneumonia, the other types of infection episodes were urinary tract infection (1.0%), bacteremia (1.4%), and soft tissue infection (4.8%), with no significant differences in incidence among the three groups. Notably, a higher incidence of other infection types was observed in the C/T+IO group (22.2%) compared with the C/T alone group (7.9%) and IO alone group (14.3%) (p = 0.011; Supplementary Table S1 ). Among patients who developed infections, 36.3% required hospitalization, and 10.3% were indicated for ICU admission ( Table 2 ). However, the incidence of hospitalization and ICU admission did not significantly differ among three groups ( Table 2 ). Only one patient (0.4%) who received IO alone treatment had coronavirus disease 2019 (COVID-19) pneumonia.

The risk factors for infection and pneumonia were analyzed using logistic regression. In the univariate analysis, male sex and CKD exhibited increased odds ratios (ORs) for infection. However, in the multivariate analysis, only CKD was significantly associated with a higher risk of infection (OR: 5.05, 95% confidence interval [CI]: 1.06–24.15, p = 0.043; Supplementary Table S2 ). Immunotherapy was not associated with the risk of infection. In the context of pneumonia incidence, comorbidities, sex, and smoking history exhibited no statistical significance ( Table 3 ). However, immunotherapy was associated with a lower risk of pneumonia after adjustment for other factors (OR: 0.46, 95% CI: 0.25–0.84, p = 0.012; Table 3 ). The IO+C/T treatment was not associated with higher risk of pneumonia compared to the IO alone treatment (OR: 1.17, 95% CI: 0.57–2.40, p = 0.663; Table 3 ). In addition, immunotherapy was not related to the risk of COVID-19 pneumonia in the univariate and multivariate analysis ( Supplementary Table S3 ). Furthermore, a survival analysis was performed for the patients with or without pneumonia. The patients were categorized into the chemotherapy group and immunotherapy group which consisted of IO alone group and C/T+IO group according to their exposure to immunotherapy. The patients without pneumonia exhibited more favorable OS than those with pneumonia in both the aforementioned groups. Among all groups, the patients who received immunotherapy and did not develop pneumonia exhibited the most favorable survival outcome, with a median OS of 47.7 months (95% CI: 25.9–69.5, p < 0.001; Figure 1 ). In C/T alone group, patients with pneumonia had lower OS than those without (median OS: 12.1 vs. 20.4 months, p < 0.001). In IO alone group, patients with pneumonia had lower OS than those without (median OS: 9.4 vs. not reached, p < 0.001). In C/T+IO alone group, patients with pneumonia tended to have lower OS than those without (median OS: 23.7 vs. 47.7 months, p = 0.161) (Data not shown). The patients with pneumonia other than that related to COVID-19 did not exhibit better OS than those with COVID-19 pneumonia (p = 0.325). In summary, the findings suggest that the administration of immunotherapy in patients with lung cancer not only is related to a lower risk of pneumonia but may also confers survival benefits by preventing pneumonia.

Circulating biomarkers and immune signatures that could predict the occurrence of post-treatment pneumonia among the patients receiving various lung cancer therapies were investigated through mass cytometry (CyTOF). A total of 30 patients were enrolled in the prospective cohort, and the patients’ blood specimens were analyzed through CyTOF. The included patients were categorized into three groups based on their treatment regimens: C/T alone group (n = 6), IO alone group (n = 18) and C/T+IO group (n = 6). The patient characteristics were listed in Supplementary Table S4 . There was no patient with COVID-19 pneumonia in the prospective cohort. In the context of pneumonia incidence, the comorbidities, age, sex, smoking history, and treatment modalities showed no statistical significance ( Supplementary Table S5 ).

This study employed opt-SNE plots and overlay histograms to visually depict the changes in lymphocyte number in PBMCs among different treatment groups ( Figure 2 ). The spectral colors overlaid on the opt-SNE plots illustrate the marker expression patterns of the different cell populations. In the prospective cohort, ten of the thirty patients received more than two lines of treatment before enrolment in this study. Therefore, the influence of previous treatment and cancer status may lead to the different marker expression patterns from those in the healthy controls ( Supplementary Figure S2 ). In the C/T alone group and C/T+IO group ( Figures 2A, E ), the expressions of CD45, CD3, CD8, and CD4 were not similar to those in the healthy controls and those in the IO alone group with obvious marker expression ( Supplementary Figure S2 and Figure 2C ). In the IO alone group ( Figure 2C ), marked expressions of CD45, CD3, CD8, and CD4, and CD16 were found but the expressions of PD-1, TIM-3, and LAG-3 were not similar to those in the healthy controls with obvious marker expression. The different proportions of patients who received more than two lines of treatment among the C/T alone, IO alone, and C/T+IO alone group ( Supplementary Table S4 ) may lead to the different intensity of marker expressions among these groups. In addition, part of the cell populations expressing CD45 in the C/T alone and C/T+IO group ( Figures 2A, E ) were clustered separately from the cell populations expressing CD3. There are several potential reasons for their partial separation in opt-SNE plots. First, the intensity and variance of CD45 expression across different cell types after treatment may be higher compared with CD3, leading to their partial separation in dimensionality reduction mappings. Second, CD45 has a large extracellular domain leading to epitope spreading across leukocytes and this high dimensionality in CD45 staining may separate it out from CD3. Third, biological heterogeneity in the cell-activation states, bound ligands etc. may alter CD45, and CD3 expression and lead to their partial separation (51, 59, 60). The C/T alone group exhibited markedly low numbers of PD-1⁺ cells, TIM-3⁺ cells, and LAG-3⁺ cells ( Figures 2A, B ). The overlay histograms further illustrated that the numbers of PD-1⁺ cells, TIM-3⁺ cells, and LAG-3⁺ cells in the C/T alone group increased after treatment and the number of PD-1⁺ cells in the IO alone group decreased after treatment ( Figures 2B, D ). To provide a comprehensive illustration of these findings, this study employed SPADE analysis to cluster and visualize distinct subpopulations of immune cells, including NK cells, CD8⁺ (cytotoxic) T cells, CD4⁺ T cells, B cells, and monocytes, in different treatment groups both before and after treatment ( Figures 3A, C, E ). The average percentages of various immune cell types among the total cells before and after treatment are depicted in Figures 3B, D, F , respectively. Notably, the proportions of NK cells were significantly higher after treatment than before treatment in the IO alone group (p < 0.001) and C/T+IO group (p < 0.01; Figures 3D, F ). The proportion of CD4⁺ T cells was significantly higher after treatment than before treatment in the C/T+IO group (p < 0.001) ( Figure 3F ). However, the proportions of different immune cells did not exhibit significant changes before and after chemotherapy ( Figure 3B ).

An immunophenotyping analysis of different lymphocyte subpopulations was conducted before and after various treatments ( Figure 4 ). The cell counts per microliter of lymphocyte subpopulations for the study population and for each treatment group, namely C/T alone, IO alone, and C/T+IO, are presented in Figures 4A–C, D–F, G–I, J–L , respectively. With stratification by different exhausted T-cell subpopulations, the analysis revealed that PD-1⁺CD8⁺ (cytotoxic) T cells, and PD-1⁺CD4⁺ T cells (both p < 0.01; Figure 4F ) significantly increased in the C/T alone group after treatment. However, in the IO alone group, TIM-3⁺ cytotoxic T cells (p < 0.05; Figure 4G ), PD-1⁺CD8⁺ T cells (p < 0.05; Figure 4I ) decreased after treatment. In both the IO alone and C/T+IO groups, PD-1⁺CD4⁺ T cells decreased after treatment (p < 0.001 and p < 0.01, respectively; Figures 4I, L ). Subsequently, the cell counts per microliter of different lymphocyte subpopulations were assessed in the patients with and without pneumonia after treatment ( Figure 5 ). The lymphocyte subpopulations in the C/T alone group, IO alone group, and C/T+IO group are presented in Figures 5A–C , respectively. In both C/T alone and C/T+IO groups, PD-1⁺CD8⁺ T cells significantly increased in patients with pneumonia after treatment (p < 0.05; Figures 5A, C ). Nonetheless, in the IO alone group, PD-1⁺CD8⁺ T cells significantly decreased in the patients without pneumonia after treatment (p < 0.05; Figure 5B ). Collectively, the findings indicate that the increased cell counts per microliter of PD-1⁺CD8⁺ T cells after treatment might be related to the occurrence of pneumonia following chemotherapy or immunotherapy. Survival analysis was further performed for patients with pneumonia and those without pneumonia in the prospective cohort. Patients were categorized into chemotherapy group and immunotherapy group which consisted of IO alone group and C/T+IO group according to their exposure to immunotherapy. Among all groups, there was no significant difference in OS between patients with pneumonia and those without pneumonia (p = 0.978; Supplementary Figure S3 ).

Cytokine profiles among different treatment groups were evaluated to elucidate the interplay between different treatments and the immune response. The levels of cytokines, including IL-2, IL-6, IL-10, TNF-α, and IFN-γ, were measured in peripheral blood specimens at baseline (Day 0) (D0) and after treatment (Day 8) (D8) through ELISA ( Figure 6 ). Three treatment group showed different levels of cytokines observed at D0, indicating that there might be different immune-related characteristics among groups.

As presented in Figure 6A , TNF-α level significantly increased after IO alone treatment (p < 0.05) but TNF-α level significantly decreased after C/T alone treatment (p < 0.01; Figure 6B ). IL-2 significantly decreased after C/T+IO treatment (p < 0.01; Figure 6C ). The Figure 6D showed the change of cytokine levels by using Log₂ fold change in the concentration of different cytokines in Day 8 (D8) compared with that in Day 0 (D0). TNF-α revealed higher level of Log₂ fold change in the IO group than that in the C/T alone group (p < 0.05; Figure 6D ). These findings suggest that IO alone, C/T alone, and C/T+IO treatment exert varying effects on the responses of TNF-α and IL-2.