Gut microbiome-driven modulation of the tumor immune microenvironment optimizes dual checkpoint blockade in advanced non-small-cell lung cancer.

Introduction
Lung cancer remains the leading cause of cancer-related deaths worldwide.1 Immune checkpoint inhibitors (ICIs) have revolutionized the treatment of advanced lung cancer, improving survival outcomes with an acceptable safety profile. ICIs target various immune regulatory pathways such as programmed cell death protein 1 (PD-1),2,3 programmed death-ligand 1 (PD-L1),4 and cytotoxic T-lymphocyte-associated protein 4 (CTLA-4).5 They are now considered as key components of modern lung cancer therapy.
The combination of immuno-oncology has demonstrated favorable efficacy as a dual immune checkpoint blockade in several clinical trials.5,6 Notably, these regimens have also shown efficacy in patients with low or negative PD-L1 expression, thus further broadening their therapeutic utility. However, the lack of robust predictive biomarkers and criteria for integrating chemotherapy into combination regimens remains critical challenges when optimizing patient selection.
Recent studies have highlighted the pivotal role of the gut microbiome in modulating antitumor immunity and shaping the efficacy of ICIs. Among these, certain commensal bacteria such as Ruminococcaceae, Akkermansia muciniphila, and Bifidobacterium have been associated with improved ICI responses when treating various cancer types, including lung cancer.7, 8, 9, 10 The gut microbiome is known to influence the recruitment and activation of various immune cell populations that are critical for antitumor immunity, including CD8+ T cells and regulatory T cells (Tregs).8,11 However, dysbiosis caused by the use of antibiotics or proton pump inhibitors (PPIs) may disrupt immune homeostasis and ultimately lower antitumor immunity, thereby potentially reducing the efficacy of ICIs.12, 13, 14, 15 Despite these findings, the specific pathways through which gut microbiota changes influence the tumor immune microenvironment and impact ICI efficacy remain poorly understood.
To address these challenges, we conducted a prospective study on 50 patients with advanced non-small-cell lung cancer (NSCLC) who were scheduled to receive ipilimumab + nivolumab (I–N) therapy. Gut microbiome analyses were carried out to assess bacterial composition and diversity, and multiplexed immunofluorescence staining was used to evaluate immune cell populations, including CD8+ T cells and Tregs, in the tumor microenvironment.
By integrating these datasets into this cohort study, we aimed to elucidate the relationship between the state of the gut microbiota and crosstalk between localized immune responses, thereby clarifying their impact on the therapeutic efficacy of immunotherapy. We also investigated potential biomarkers to predict the treatment outcomes of I–N therapy and examined the added benefit of chemotherapy to combination therapy. This study aimed to contribute to the development of personalized therapeutic strategies to improve treatment outcomes for patients with advanced NSCLC, by elucidating the novel role of the gut microbiota in the process.

Materials and methods

Patients
We prospectively enrolled 50 patients with advanced or recurrent NSCLC who were scheduled to be treated with I–N, with or without chemotherapy, at eight Japanese institutions between April 2021 and September 2023. The study included patients who met the following criteria: had histologically and cytologically confirmed unresectable advanced or recurrent NSCLC and had not been previously treated with chemotherapy or immunotherapy (not including molecular targeted therapy). Patients with active multiple primary cancers were excluded. All patients provided written informed consent before being enrolled and were followed up from the start of treatment until October 2024. The study’s protocol was approved by the Ethics Committee of the University Hospital, Kyoto Prefectural University of Medicine, Kyoto, Japan (approval number: ERB-C-1950), and the study was conducted in accordance with the tenets of the Declaration of Helsinki. This protocol was registered at the University Medical Hospital Information Network Clinical Trials Registry (UMIN 000044811).

Sequencing of the 16S rRNA gene
Fecal samples were prospectively collected using Mykinso kits containing guanidine thiocyanate solution (TechnoSuruga Laboratory, Shizuoka, Japan) and were stored at 4°C. DNA was extracted using the GENE PREP STAR PI-1200A system (Kurabo Industries, Osaka, Japan), following the manufacturer’s protocol. The V1-V2 region of the 16S ribosomal RNA (rRNA) gene was amplified using the KAPA HiFi Hot Start Ready Mix (Roche, Basel, Switzerland) with the 16S_27Fmod and 16S_338R primers. Indexed libraries were prepared using the Nextera XT kit (Illumina, San Diego, CA), quantified via quantitative PCR with the KAPA SYBR FAST qPCR kit (Roche), and sequenced on an Illumina MiSeq platform using the MiSeq Reagent Kit v2 (Illumina; 500 cycles, 250 bp paired-end).

Microbiome analysis
Processing of the sequencing data, including chimera check, amplicon sequence variant (ASV) definition, and taxonomy assignment, was carried out using Quantitative Insights Into Microbial Ecology 2 version 2024.5.16 Singletons were excluded from this study. The taxonomy assignment of the resulting ASV was completed using the Sklearn classifier algorithm against the SILVA_138 database.
The Chao1 (an estimation of ASV richness) and Shannon (an estimation of ASV evenness) phylogenetic diversity indexes were calculated for microbiota variations within single samples (i.e. α-diversity) and were statistically analyzed using the Wilcoxon rank-sum test. The variation in microbial communities across the samples (i.e. β-diversity) was estimated using the Bray–Curtis metric,17 which calculates the distances between the samples and visualizes them via Principal Coordinates Analysis. This was then statistically examined using permutational multivariate analysis of variance. The final figures were generated using R v4.4.2 (The R Foundation for Statistical Computing, Vienna, Austria).

Predictive functional profiling of gut microbial communities
Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt) v2.1.4 software18 was used to characterize the metagenomic-based function of the microbiome in each identified cluster. PICRUSt was used to obtain relative Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway abundance information from metagenomic data. The predicted data were collapsed into hierarchical categories so that the relative abundances of the gut metabolic functions could be calculated. We calculated the nearest sequenced taxon index and excluded nine ASVs with the nearest sequenced taxon indexes of >2.

Linear discriminant analysis effect size
Linear discriminant analysis (LDA) effect size,19 a biomarker discovery method, was used to determine the gut bacteria and their metabolic pathways that best characterized each study group. The LDA score was used to indicate the effect size of each pathway. Taxa with LDA scores of >2.5 (P < 0.05) for the microbiota and KEGG metabolic pathway were considered significant. These analyses were carried out in R, using the phyloseq and MicrobiotaProcess packages.20,21

Multiplex immunofluorescence staining
Tumor samples collected from surgical resection or endoscopic biopsy were fixed in formalin and paraffin-embedded (FFPE) before being sectioned and prepared into microscope slides. Four micron-thick FFPE tissue sections were then deparaffinized and rehydrated by serial passage through xylene and graded ethanol for analysis. Multiplex fluorescent immunohistochemistry was carried out using the Opal 6-Plex Detection Kit (Akoya Biosciences: Marlborough, MA), following the manufacturer’s protocol. All slides were stained with anti-human CD8 (clone C8/144b; Abcam: Cambridge, UK), anti-human FOXP3 (clone 236A/E7; DAKO), anti-human PD-1 (clone EPR4877; Abcam), anti-human CTLA-4 (clone CAL49; Abcam), anti-human CD3 (clone SP162; Abcam), and anti-human cytokeratin (clone AE1/AE3, Abcam) antibodies. The slides were then cover-slipped using ProLong Diamond Antifade Mountant with DAPI (Invitrogen: Carlsbad, CA).
Multiplex fluorescence-labeled images of randomly selected fields were captured using the PhenoImager Fusion version 2.1.0 automated imaging system (Akoya Biosciences). Image analysis was carried out using Phenochart version 2.0.0 (Akoya Biosciences) and InForm software version 3.0 (Akoya Biosciences) for cell segmentation, based on DAPI counterstaining and phenotype definition. All cells that stained positive for CD8, FOXP3, PD-1, CTLA-4, CD3, and cytokeratin were counted in 5-10 high-power microscopic fields, before the mean values were calculated and recorded. The subsequent analysis of these data was conducted using HALO version 3.6.4134.362 (Indica Labs: Albuquerque, NM). All slides were annotated to identify tumor regions, explicitly excluding artifacts such as air bubbles. Cells of each immune phenotype were counted in three to five high-power fields, and the average cell density was calculated automatically. Two researchers (AF and YK) evaluated the stained slides independently.

Statistical analysis
Progression-free survival (PFS), overall survival (OS), and objective response rate (ORR) were defined according to RECIST v1.1. Patients without events were censored at the date of their final follow-up. The cut-off values for Chao1, Shannon index, and tumor-infiltrating lymphocyte (TIL) density were set at their respective medians. All statistical tests were two-sided, and P values <0.05 were considered statistically significant. The relationship between TILs and gut microbiome diversity was examined using Spearman’s correlation coefficient. Comparisons of Chao1, Shannon index, and TIL density between two groups were carried out using the Wilcoxon rank-sum test. ORR and other categorical variables were compared between groups using the chi-square test or Fisher’s exact test, as appropriate. PFS and OS were first estimated using unweighted Kaplan–Meier curves, and differences between groups were compared using the log-rank test.
To address potential treatment selection bias, we used inverse probability of treatment weighting (IPTW). Propensity scores for receiving I–N with chemotherapy (I–N + C) were estimated using a logistic regression model including baseline covariates considered a priori to influence treatment allocation and outcomes: number of metastatic sites, PD-L1 status, brain metastases, post-operative recurrence, and baseline use of systemic corticosteroids at ICI initiation. Stabilized IPTWs were derived from these scores to construct a weighted pseudo-population. Covariate balance before and after weighting was evaluated using absolute standardized differences (<0.1 considered adequate). Hazard ratios (HRs) for PFS comparing I–N with I–N + C were estimated using Cox models with robust variance estimators. Prespecified subgroup analyses according to baseline gut microbiota diversity (high versus low by median Shannon index, with Chao1 richness used in sensitivity analyses) were also carried out using IPTW-weighted Cox models.
Cox proportional hazards models were used to estimate HRs and 95% confidence intervals (CIs). Antibiotic usage, sex, age (≥75 years), PD-L1 status, histology, and baseline systemic corticosteroid use were included as covariates in conventional multivariable Cox models. All time-to-event and regression analyses were carried out using complete-case data; patients with missing values for a given variable were excluded from the corresponding models. All statistical analyses were carried out using EZR statistical software version 1.40,22 which is a graphical user interface for R (The R Foundation for Statistical Computing).

Results

Patient characteristics
Of the 50 eligible patients who were initially identified, 2 were excluded: 1 did not receive I–N after enrollment, and 1 had poor-quality fecal samples that could not be analyzed. Therefore, the final cohort comprised 48 patients (Supplementary Figure S1, available at https://doi.org/10.1016/j.esmoop.2026.106077). The median follow-up period was 16.2 months. The baseline characteristics of the overall cohort are presented in Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2026.106077. The median age was 71 years, and 45 (94%) had Eastern Cooperative Oncology Group (ECOG) performance status (PS) of ≤1. Twelve patients (25%) received antibiotics 1 month before dual checkpoint blockade initiation and 17 (35%) received PPIs at the start of treatment. A total of 21 patients were treated with I–N, whereas 27 were treated with I–N + C. Compared with the I–N + C group, the I–N group had a significantly higher proportion of patients with post-operative recurrence (43% versus 4%, P < 0.001) and a tendency toward more patients with ECOG PS ≥2 (14% versus 0%, P = 0.07).

Treatment outcomes of ipilimumab plus nivolumab with and without chemotherapy
The overall ORR was 44% (95% CI 30% to 59%) and the disease control rate was 63% (95% CI 47% to 76%; Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2026.106077). The median PFS and OS were 5.8 months (95% CI 3.6-11.3 months) and 24.8 months [95% CI 11.3 months-not estimable (NE)], respectively (Supplementary Figure S2A and B, available at https://doi.org/10.1016/j.esmoop.2026.106077). The ORR was significantly higher in the I–N + C group than in the I–N group (59% versus 24%, P = 0.02), and the disease control rate showed a similar trend (74% versus 48%, P = 0.08; Supplementary Table S1, available at https://doi.org/10.1016/j.esmoop.2026.106077). The median PFS and OS values, however, were comparable between the I–N + C and I–N groups [9.9 (95% CI 3.0-14.1) versus 4.3 (95% CI 2.8-9.5) months, HR 0.70, 95% CI 0.37-1.32, P = 0.27; 30.2 (95% CI 7.7-NE) versus 18.3 (95% CI 5.6-NE) months, HR 0.92, 95% CI 0.41-2.06, P = 0.80, respectively; Supplementary Figure S2C and D, available at https://doi.org/10.1016/j.esmoop.2026.106077]. When stratified by tumor PD-L1 expression (0% versus 1%-49%), broadly similar patterns were observed within each subgroup (Supplementary Figure S3A-D, available at https://doi.org/10.1016/j.esmoop.2026.106077).

Impact of gut microbiome diversity and composition on the efficacy of immuno-oncology treatment
Patients with PFS of ≥6 months were classified as the responder (R) group, whereas those with PFS of <6 months comprised the non-responder (NR) group. In the overall population, the R group tended to have higher α-diversity than the NR group (Figure 1A and B), whereas β-diversity was comparable between the two groups (Figure 1C).
Gut microbiota diversity was comparable between the I–N and I–N + C groups (Supplementary Figure S4A and B, available at https://doi.org/10.1016/j.esmoop.2026.106077). Within the I–N group, the R subgroup exhibited significantly higher α-diversity than the NR subgroup (Figure 1D and E), with β-diversity also differing significantly (Figure 1F). Within the I–N + C group, α- and β-diversity did not differ between the R and NR subgroups (Figure 1G-I).
Median PFS and OS were comparable between the high and low α-diversity groups in the overall population (Figure 2A and B; Supplementary Figure S5A and B, available at https://doi.org/10.1016/j.esmoop.2026.106077). In the I–N group, patients with higher α-diversity had significantly longer median PFS and tended to have longer OS than those with low α-diversity (Figure 2C and D; Supplementary Figure S5C and D, available at https://doi.org/10.1016/j.esmoop.2026.106077). Conversely, within the I–N + C group, median PFS and OS were comparable between the high and low α-diversity groups (Figure 2E and F; Supplementary Figure S5E and F, available at https://doi.org/10.1016/j.esmoop.2026.106077). Similar patterns were observed when stratified by tumor PD-L1 expression (Supplementary Figure S6A-H, available at https://doi.org/10.1016/j.esmoop.2026.106077).

IPTW analysis
We compared PFS between I–N and I–N + C within subgroups defined by baseline gut microbiota diversity. Patients were divided into high- and low-diversity groups according to the median Shannon index. I–N + C was associated with longer PFS in the low-diversity group, whereas no clear advantage was observed in the high-diversity group (Supplementary Figure S7A and B, available at https://doi.org/10.1016/j.esmoop.2026.106077). Similar patterns were observed when stratified by Chao1 richness (Supplementary Figure S7C and D, available at https://doi.org/10.1016/j.esmoop.2026.106077).
After IPTW, all baseline covariates had absolute standardized differences <0.1, indicating good balance (Supplementary Table S2, available at https://doi.org/10.1016/j.esmoop.2026.106077). In the IPTW-weighted pseudo-population, there was no clear difference in PFS between I–N and I–N + C (HR 0.94, 95% CI 0.42-2.09, P = 0.87; Supplementary Figure S8A, available at https://doi.org/10.1016/j.esmoop.2026.106077). Among patients with low Shannon diversity, I–N + C was associated with longer PFS (IPTW-adjusted HR 0.18, 95% CI 0.05-0.67, P = 0.01), whereas in the high-diversity group, I–N + C tended to be associated with shorter PFS (IPTW-adjusted HR 2.76, 95% CI 0.999-7.61, P = 0.05; Supplementary Figure S8B and C, available at https://doi.org/10.1016/j.esmoop.2026.106077). Similar trends were observed using Chao1 richness (Supplementary Figure S8D and E, available at https://doi.org/10.1016/j.esmoop.2026.106077).

Gut microbiome signatures associated with response to I–N but not I–N + C
In the I–N group, short-chain fatty acid (SCFA)-producing bacteria such as Lachnospiraceae, Coprococcus, Eubacterium, and Agathobacter23 were enriched in the R subgroup (Figure 3A), whereas the NR group showed enrichments in Escherichia–Shigella, Ruminococcus gnavus group, and Clostridium innocuum group. KEGG pathway revealed enrichment of pathways related to peptidoglycan biosynthesis, RNA degradation, and NOD-like receptor signaling in the R subgroup, while pathways associated with pathogenic Escherichia coli infection, glucose metabolism, and sulfur and selenium metabolism were enriched in the NR subgroup (Figure 3B). In the R subgroup of the I–N + C group, few enriched bacteria were observed (Supplementary Figure S9A, available at https://doi.org/10.1016/j.esmoop.2026.106077), and KEGG analysis showed no clear differences in metabolic pathways.

Association of gut microbiome characteristics with TILs assessed by multiple immunostaining
Among 48 patients, 28 provided appropriate FFPE specimens for multiplexed immunostaining (Supplementary Figure S1, available at https://doi.org/10.1016/j.esmoop.2026.106077). Baseline characteristics and representative images are shown in Supplementary Table S3 and Supplementary Figure S10A and B, available at https://doi.org/10.1016/j.esmoop.2026.106077, respectively. No differences in TIL density were observed between I–N and I–N + C groups (Supplementary Figure S11A-F, available at https://doi.org/10.1016/j.esmoop.2026.106077). CD8+ TIL density correlated significantly with α-diversity, particularly PD-1+CD8+ TIL density (Figure 4A-F; Supplementary Figure S12A-F, available at https://doi.org/10.1016/j.esmoop.2026.106077). High PD-1+CD8+ expression was associated with enrichment of SCFA-producing bacteria such as Eubacterium and Clostridia UCG-01423,24 resembling those in the I–N R subgroup (Figure 4G), and KEGG analysis revealed enrichment of immune-activating metabolic pathways (Figure 4H). SCFA-producing bacteria positively correlated with TIL densities, while inflammation-associated bacteria negatively correlated (Supplementary Figure S13, available at https://doi.org/10.1016/j.esmoop.2026.106077).
Even within these small datasets, gut microbiota diversity showed a significant correlation with PFS in I–N therapy (Supplementary Figure S14A-D, available at https://doi.org/10.1016/j.esmoop.2026.106077). In joint Cox models including Shannon α-diversity and PD-1+CD8+ TIL density, α-diversity remained associated with PFS, whereas PD-1+CD8+ TIL density was not independently associated (Supplementary Table S4, available at https://doi.org/10.1016/j.esmoop.2026.106077). Similarly, TIL density did not correlate with PFS in the overall cohort or in I–N and I–N + C groups (Supplementary Figures S15-S17, available at https://doi.org/10.1016/j.esmoop.2026.106077).

Impact of concomitant medications on dual checkpoint blockade treatment efficacy
Antibiotic use within 1 month before treatment was associated with shorter median PFS and OS in patients treated with I–N, with and without chemotherapy [PFS 9.7 (95% CI 4.3-13.6) versus 2.8 (95% CI 1.5-4.5) months, P = 0.001; OS 30.2 (95% CI 13.6-NE) versus 6.4 (95% CI 2.5-18.3) months; P = 0.005; Figure 5A and B]. In contrast, PPI use was not associated with a difference in median PFS but was associated with shorter OS [not reached (95% CI 13.9-NE) versus 11.3 (95% CI 3.7-30.2) months, P = 0.02; Supplementary Figure S18A and B, available at https://doi.org/10.1016/j.esmoop.2026.106077]. Multivariate analysis indicated that antibiotic use (HR 2.46, 95% CI 1.07-5.67, P = 0.03) was independently associated with shorter PFS (Supplementary Table S5, available at https://doi.org/10.1016/j.esmoop.2026.106077).
Stratified by chemotherapy, antibiotic use was associated with shorter PFS in both I–N and I–N + C groups [5.3 (95% CI 3.3-9.7) versus 2.8 (95% CI 1.5-NE) months, P = 0.06; 13.0 (95% CI 6.2-24.9) versus 2.9 (95% CI 0.8-13.8) months, P = 0.03, respectively; Supplementary Figure S19A and B, available at https://doi.org/10.1016/j.esmoop.2026.106077], whereas PPI use showed no clear differences [4.9 (95% CI 2.1-9.5) versus 4.0 (95% CI 1.4-11.3), P = 0.67; 11.6 (95% CI 3.0-25.1) versus 4.1 (95% CI 2.6-13.8), P = 0.12; Supplementary Figure S19C and D, available at https://doi.org/10.1016/j.esmoop.2026.106077].
We further characterized antibiotic exposure within 30 days before immunotherapy initiation among the 12 exposed patients (Supplementary Table S6, available at https://doi.org/10.1016/j.esmoop.2026.106077). β-Lactam agents predominated (11/12), and antibiotics were most commonly prescribed prophylactically after biopsy (8/12), and most courses were short (≤3 days in 7/12). The last antibiotic dose occurred 15-30 days before ICI initiation in 8/12 patients. In exploratory univariable analyses, longer antibiotic duration and shorter intervals between the last antibiotic dose and ICI initiation were each associated with numerically shorter PFS (Supplementary Table S7, available at https://doi.org/10.1016/j.esmoop.2026.106077).

Correlation between gut microbiome and concomitant medication use
Antibiotic use was not associated with lower α-diversity as assessed by the Chao1 and Shannon indexes (P = 0.51 and P = 0.47, respectively; Figure 5C and D), and β-diversity did not differ between groups (P = 0.22; Figure 5E). Patients who did not receive antibiotics had feces enriched in Bifidobacterium and Agathobacter, key taxa involved in maintaining gut microbial balance and immunological modulation (Figure 5F). PPI use was also not associated with α-diversity by Shannon and Chao1 indexes or β-diversity (Supplementary Figure S18C-E, available at https://doi.org/10.1016/j.esmoop.2026.106077). Patients receiving PPIs had feces enriched in oral commensal bacteria, including Streptococcus, Actinomyces, Fusobacterium, and Prevotella (Supplementary Figure S18F, available at https://doi.org/10.1016/j.esmoop.2026.106077).
These observations suggest that antibiotics and PPI-induced changes in gut microbiota may influence tumor immunity and reduce dual checkpoint blockade efficacy.

Discussion
Pharmacological therapies, including immunotherapy, have shown favorable outcomes for treating advanced NSCLC. Among them, I–N, with or without chemotherapy, has demonstrated efficacy even in patients with low PD-L1 expression, emphasizing the need for novel biomarkers. Given that the addition of chemotherapy to treatment regimens often raises concerns regarding adverse events, biomarkers to assess its necessity are also of interest. In this study, patients with high gut microbiome diversity and SCFA-producing bacteria showed greater CD8+ TIL infiltration and better responses to I–N, whereas those with dysbiosis benefited from the addition of chemotherapy to their regimens. To our knowledge, this is the first study to identify the gut microbiome as a biomarker for predicting I–N efficacy and guiding chemotherapy decisions in NSCLC, while also offering critical insights into the gut–immune microenvironment interaction that profoundly impacts the therapeutic efficacy of pharmacological treatments.
SCFA-producing bacteria are essential for creating a favorable immune microenvironment that supports the recruitment and activation of TILs.25 SCFAs, particularly butyrate, enhance CD8+ T-cell functionality by improving energy metabolism and promoting interferon (IFN)-γ-mediated cytotoxic responses, while also facilitating their differentiation into long-lived memory cells for sustained antitumor immunity.26, 27, 28, 29 Additionally, butyrate up-regulates NOD2 expression by enhancing histone acetylation in the Nod2 promoter region.30 Through the recognition of bacterial peptidoglycan, this NOD-like receptor signaling activates the production of inflammatory cytokines and chemokines, leading to dendritic cell maturation and the subsequent activation of CD8+ T cells.31, 32, 33 The presence of Lachnospiraceae has been linked to enhanced immune responses in the tumor microenvironment, potentially through the modulation of T-cell metabolism and activity.34 These findings suggest that gut microbiota composition not only predicts the efficacy of dual checkpoint blockade therapy but also highlights the potential of prebiotic-driven SCFA-producing bacteria activation to enhance immunotherapy outcomes.
In this study, gut microbiota diversity demonstrated the strongest correlation with the infiltration of PD-1+CD8+ T cells. Previous studies have shown that PD-1+CD8+ T cells are predictive of the clinical efficacy of PD-1 inhibitors, highlighting the critical influence of gut microbiota diversity on the therapeutic effects of these agents.35 Additionally, it has been reported that combining CTLA-4 inhibitors with PD-1 inhibitors reduces the immunosuppressive function of Tregs and promotes selective depletion of intratumoral Tregs.36,37 Therefore, gut microbiota diversity may exhibit a stronger correlation with therapeutic efficacy in dual checkpoint blockade therapy by modulating Treg function through CTLA-4 inhibition.
Patients with low gut microbiota diversity in our cohort showed a higher abundance of pro-inflammatory and potentially immunosuppressive bacterial taxa such as Escherichia–Shigella and R. gnavus. These bacterial species are associated with intestinal inflammation and disruptions in gut homeostasis that may negatively influence the immune microenvironment and impair antitumor immune responses.38, 39, 40 Furthermore, inflammation resulting from dysbiosis exacerbates T-cell exhaustion, reducing the number of CD8+IFN-γ+ T cells in the tumor microenvironment and ultimately weakening antitumor immunity.41
Microbiota diversity was associated with the efficacy of I–N in this study, but this correlation disappeared when chemotherapy was added to the treatment regimen. This may be attributed not only to the direct cytotoxic effects of chemotherapy but also to the release of tumor antigens, which improves the immune microenvironment. Chemotherapy-induced immunogenic cell death is thought to overcome microbiota-related immunosuppression by releasing tumor antigens and danger-associated molecular patterns such as calreticulin, ATP, and HMGB1—all of which can activate dendritic cells and potentially enhance CD8+ T-cell responses, even in patients with low microbiota diversity.42, 43, 44
Consistent with previous findings, we identified that prior antibiotic use was associated with shorter PFS and OS in this study.13,14,45 This effect may be attributable to dysbiosis characterized by the loss of beneficial commensal bacterial genera such as Bifidobacterium and Agathobacter.7,46,47 Alternatively, it may reflect the influence of systemic conditions or inflammation requiring antibiotic use, both of which can negatively impact ICI efficacy. Similarly, PPI use was associated with an enrichment of oral commensal bacteria, including Streptococcus, Prevotella, and Actinomyces. These oral taxa can promote the production of inflammatory cytokines such as interleukin (IL)-6 and IL-1β, potentially leading to an immunosuppressive tumor microenvironment.14,48,49 Previous gut metagenomic and metatranscriptomic NSCLC cohorts have linked oral streptococci, increased Actinomycetota activity, and broader microbial functional shifts to poorer ICI outcomes, although taxa-level signatures differ across studies.50,51 Based on these findings, careful management of antibiotic and PPI use before the initiation of ICI therapy may be important for minimizing their potential adverse effects on treatment efficacy.
This study was subject to several important limitations. Firstly, although 48 patients were initially included, many analyses relied on smaller subsets of the cohort, which reduced statistical power and limited the robustness of subgroup and taxa-level comparisons. Because the number of events was limited, we were only able to include a restricted set of covariates in the multivariable Cox models to avoid overfitting. As a result, we could not adjust for other clinically relevant factors such as comorbidities, markers of infection or systemic inflammation, perioperative status, or detailed concomitant medications. The risk of residual confounding, particularly in the analyses involving antibiotic and PPI exposure, remains substantial. As an observational study, our analysis can identify associations but cannot establish causality. Dynamic changes in gut microbiota during treatment were not tracked, and potential confounders such as diet were not systematically controlled. Moreover, the cohort’s limited geographic and ethnic diversity may restrict the generalizability of our findings. In addition, these exploratory taxa- and pathway-level findings from LEfSe/PICRUSt2 were not adjusted for multiple comparisons, which increases the risk of false-positive discoveries. Consequently, our results should be considered hypothesis-generating, and confirmation in larger, more diverse cohorts with richer clinical annotation and longitudinal microbiome profiling is warranted.
This study indicates that gut microbiota diversity, particularly the abundance of SCFA-producing bacteria, is associated with improved efficacy of I-N therapy and greater TIL activation. Chemotherapy may complement I-N therapy in patients with lower gut microbiota diversity. These findings establish the gut microbiota as a potential biomarker for optimizing ICI-based therapies and highlight the synergy between immunotherapy and chemotherapy in addressing microbiota-driven resistance.