Smoking-associated modulation of gut microbiota shapes response to immune checkpoint inhibitors in non-small cell lung cancer Authors: Po-Yu Chien, Wen-Chien Cheng, Chin-Chuan Hung, Chih-Yen Tu, Te-Chun Hsia, Der-Yang Cho, Po-Ren Hsueh, Zi-Lun Lai, Yi-Cheng Shen, Yu-Chao Lin Published: 2026 ABSTRACT AbstractBackgroundSmoking status has been associated with differences in immune checkpoint inhibitor (ICI) efficacy in non-small cell lung cancer (NSCLC), though the underlying mechanisms remain unclear. This prospective cohort study evaluated whether smoking-related changes in the gut microbiota are linked to ICI response.MethodsBaseline fecal samples from 225 patients with NSCLC were analyzed using full-length 16S rRNA sequencing. Microbial composition, predicted functional features, and clinical variables were examined in relation to treatment response and progression-free survival (PFS). Associations with treatment response were assessed using multivariable ordinal logistic regression. PFS was evaluated using Kaplan–Meier analysis and the log-rank test.ResultsSmoking was associated with selected taxon-level microbial differences despite no significant differences in alpha or beta diversity. Compared with never-smokers, smokers exhibited reduced abundance of Bifidobacterium longum and enrichment of Gram-negative taxa. Functional prediction indicated increased potential for lipopolysaccharide, Kdo₂-lipid A, and lipid IVA biosynthesis in smokers. Predicted lpxM abundance, encoding a lipid A myristoyltransferase, was positively correlated with the Gram-negative bacterial fraction. Moreover, increased abundance of lpxM-linked Gammaproteobacteria was associated with improved ICI response in the NSCLC cohort. Conversely, lower B. longum abundance was associated with favorable ICI response and prolonged PFS.ConclusionsThese results indicate that smoking-related gut microbial alterations, particularly reduced B. longum abundance, are associated with enhanced ICI efficacy in NSCLC, supporting a role for the gut microbiota in smoking-associated differences in immunotherapy outcomes.Keywords: Non-small cell lung cancer, Bifidobacterium longum, Immune checkpoint inhibitors, Microbiome Introduction Non-small cell lung cancer (NSCLC) accounts for approximately 85% of all lung cancer cases and remains a leading cause of cancer-related morbidity and mortality worldwide [1]. Immune checkpoint inhibitors (ICIs) have reshaped the therapeutic landscape of NSCLC by restoring antitumor immune surveillance, resulting in durable clinical benefit for a subset of patients [2]. Nevertheless, responses to ICIs are highly variable, and their efficacy is constrained by primary or acquired resistance as well as immune-related toxicities. Cigarette smoking is a well-established etiological factor in NSCLC, driving tumor initiation and progression through the accumulation of genetic alterations [3]. Paradoxically, smoking history has been repeatedly associated with improved clinical outcomes in patients receiving ICIs. Smokers often exhibit higher tumor mutational burden and increased neoantigen load, features linked to enhanced immunogenicity. Notably, emerging evidence suggests that smoking may partially mitigate immunotherapy resistance in EGFR-mutant NSCLC, a subgroup traditionally refractory to ICIs [4, 5]. Compared with never-smokers, smokers harboring EGFR mutations display more pronounced immune activation signatures, implying that smoking-induced alterations in the tumor–immune microenvironment may sensitize otherwise resistant tumors to immunotherapy [6]. Beyond its direct oncogenic effects, cigarette smoking exerts a profound influence on the gut microbiota. Exposure to cigarette smoke components—including nicotine, polycyclic aromatic hydrocarbons, and volatile organic compounds—can disrupt microbial homeostasis by altering mucosal integrity, luminal pH, and antimicrobial pressures [7]. Clinical studies have reported smoking-associated shifts in microbial composition, characterized by reduced abundance of beneficial taxa such as Bifidobacterium and Faecalibacterium, alongside broader changes in dominant bacterial phyla [8]. Accumulating evidence underscores the central role of the gut microbiota in modulating systemic immunity and determining responsiveness to ICIs [9]. Specific microbial taxa and metabolites have been shown to enhance antigen presentation, T cell activation, and effector function, highlighting the relevance of the gut–lung axis in cancer immunotherapy [10]. However, the mechanistic link connecting smoking-induced microbiota alterations to immunotherapy outcomes has not been systematically investigated. In particular, whether smoking reshapes gut microbial composition and function in a manner that contributes to enhanced ICI efficacy in NSCLC remains unknown. We therefore hypothesized that cigarette smoking modulates the gut microbiota and its metabolic capacity, thereby influencing host immune responses to immune checkpoint blockade. The aim of this study was to delineate the relationship among smoking status, gut microbial features, and clinical outcomes following ICI therapy in NSCLC. Elucidating smoking-associated microbial signatures may provide mechanistic insight into differential immunotherapy responses and inform the development of microbiota-targeted strategies to improve ICI efficacy. Materials and methods Study design and participants This prospective cohort study enrolled patients with non-small cell lung cancer (NSCLC) treated at China Medical University Hospital, Taichung, Taiwan (R.O.C.). A total of 271 patients were initially screened, and 225 were ultimately included in the final analysis after excluding those without baseline stool samples or with incomplete clinical data. Clinical information, including smoking status, EGFR mutation subtype (Del-19, L858R, wild-type, and rare mutations), tumor stage, and treatment history, was collected from medical records. The study protocol was approved by the Research Ethics Committee of China Medical University Hospital (IRB number: CMUH111-REC1-211) and was performed in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants prior to enrollment and sample collection. Sample collection and DNA extraction Baseline stool samples were collected prior to treatment initiation. Microbial genomic DNA was isolated using the QIAamp PowerFecal Pro DNA Kit on the QIAcube HT automated extraction system to minimize contamination. The quantity of extracted genomic DNA (gDNA) was measured using a Qubit 3.0 Fluorometer with the Qubit dsDNA HS Assay Kit (Invitrogen, Life Technologies). A minimum yield of 500 ng gDNA per sample was required for downstream processing. All DNA extracts were stored at -80 °C until library preparation. 16S rRNA full-length amplicon production Full-length 16S rRNA genes were amplified using PacBio barcoded primers 27F (5’-AGR GTT YGA TYM TGG CTC AG-3’) and 1492R (5’-RGY TAC CTT GTT ACG ACT T-3’). PCR reactions (25 μL total volume) contained KAPA HiFi HotStart 2X ReadyMix PCR reagent (Kapa Biosystems, Woburn, MA, USA), 0.375 μM of each primer, and 500–1000 ng of gDNA template. The cycling program consisted of an initial denaturation at 95 °C for 3 min, followed by 25 cycles of 95 °C for 30 s, 57 °C for 30 s, and 72 °C for 1 min, with a final 72 °C extension for 1 min. PCR products were verified by 0.8% agarose gel electrophoresis and quantified by Qubit fluorometry. Amplicons were purified using AMPure PB magnetic beads (PacBio, Menlo Park, CA, USA) at a 1:1 ratio to remove impurities. Library construction and sequencing Approximately 500 ng of purified PCR products was used for library preparation with the SMRTbell Express Template Prep Kit 3.0 (PacBio, Menlo Park, CA, USA). Following DNA end-repair, adapters were ligated according to the manufacturer’s protocol. Library quality and fragment size distribution were evaluated using the Agilent 4200 TapeStation system with the D5000 ScreenTape assay (Agilent Technologies, Santa Clara, CA, USA). Sequencing was performed on the PacBio Sequel IIe system using SMRT Cell 8 M v3 chemistry. Primary base calling and filtering were conducted on the Sequel IIe instrument, while secondary data processing was carried out using SMRT Link 9.0 software. Sequencing bioinformatic analysis The raw PacBio HiFi reads were processed through a dedicated pipeline optimized for full-length 16S data. Initial quality control was performed using seqkit [11], followed by cutadapt to trim primer sequences [12]. High-fidelity amplicon sequence variants (ASVs) were generated using DADA2 implemented within QIIME2 [13, 14], which also included chimera detection and removal. Taxonomic classification was carried out with VSEARCH against the SILVA v138 database, resulting in ASV tables [15, 16]. Diversity analyses included calculation of Shannon index and observed OTUs for alpha diversity, compared across groups by the Kruskal–Wallis test. Beta diversity was assessed using Bray–Curtis dissimilarities, with group differences tested by PERMANOVA. Differential abundance testing was performed using LEfSe, applying an LDA score cutoff of ≥ 2 [17]. Functional prediction of microbial communities was conducted using PICRUSt2 [18], while pathway enrichment and statistical comparisons were visualized using STAMP [19]. Statistical analysis Microbial differences between smokers and non-smokers were assessed using White’s non-parametric t-test as implemented in STAMP software for exploratory two-group comparisons of microbial relative abundance. Comparisons across EGFR subtypes were tested by ANOVA. Multiple testing correction was performed using the Benjamini–Hochberg FDR method. In ICI-treated patients, clinical response was modeled as an ordinal outcome with three levels (progressive disease [PD] < stable disease [SD] < partial response [PR]). We fitted cumulative‐logit proportional-odds models using the clm function (R, ordinal package). Covariates included age, sex, body mass index (BMI), smoking status, PD-L1 expression. Available baseline dietary habit variables were incorporated into the multivariable models to reduce potential confounding by diet. These variables included vegetable intake, fruit intake, probiotic drink intake, and fried food intake. Odds ratios (ORs) and 95% CIs were obtained by exponentiating model coefficients. In this framework, exp(β) represents the common ratio of cumulative odds of achieving a better response category across cut points. Progression-free survival (PFS) analysis was restricted to stage IV patients treated with pembrolizumab and evaluated using Kaplan–Meier curves and log-rank test. All analyses were performed in R (version 4.4.1) with tidyverse, dplyr, ggplot2, effectsize, ordinal, survival, and survminer. GraphPad Prism (version 5) was used for visualization. A two-sided p-value < 0.05 was considered significant. Results Patient enrollment and baseline characteristics A total of 271 patients with non-small cell lung cancer (NSCLC) were prospectively enrolled (Fig. 1). After exclusion of 46 patients due to missing fecal samples or incomplete clinical data, 225 patients were included in the final microbiota analysis. Baseline fecal samples were collected prior to treatment and subjected to 16S rRNA gene sequencing. Patients were stratified by smoking status into nonsmokers (n = 121) and smokers (n = 104). Study design and patient inclusion flowchart. Prospective cohort study of patients with non-small cell lung cancer (NSCLC) Baseline characteristics of the study cohort are summarized in Table 1. The median age was 67 years (range 28–93), and 56.0% of patients were male. Most patients presented with stage IV disease (68.4%). Adenocarcinoma was the predominant histologic subtype (80.9%), followed by squamous cell carcinoma (14.2%) and adenosquamous carcinoma (2.2%). The mean body mass index (BMI) was 23.46 (range, 14.89–36.59). EGFR mutations were identified in 75 patients (33.3%), most commonly L858R (57.3%) and exon 19 deletion (Del-19; 32.0%), with rare variants detected in 10.7% of cases. Wild-type EGFR was observed in 92 patients (40.9%). Table 1Baseline clinical characteristics of the study cohortPatient characteristicsN = 225Age, median (range)67 (28–93)Sex, n (%)Female99 (44.0%)Male126 (56.0%)History of smoking, n (%)Nonsmokers121 (53.8%)Smokers104 (46.2%)Current71/104 (68.3%)Former33/104 (31.7%)Histology, n (%)Adenocarcinoma182 (80.9%)Squamous32 (14.2%)Adenosquamous5 (2.2%)Other6 (2.7%)Stage, n (%)I34 (15.1%)II12 (5.3%)III25 (11.1%)IV154 (68.4%)BMI, mean (range)23.46 (14.89–36.59)EGFR mutation, n (%)aPositive75 (33.3%)Del-1924/75 (32.0%)L858R43/75 (57.3%)Rare mutationsb8/75 (10.7%)Wild type92 (40.9%)PD-L1 expression, n (%)cTPS ≥ 50% (high expression)36 (32.4%)TPS 1–49% (low expression)42 (37.8%)TPS < 1% (no expression)33 (29.7%)ICI used, n (%)dPembrolizumab45/52 (86.5%)Atezolizumab3/52 (5.8%)Nivolumab2/52 (3.8%)Durvalumab1/52 (1.9%)Ipilimumab + pembro + nivo1/52 (1.9%)Data are presented as n (%) unless otherwise indicatedpembro, pembrolizumab; nivo, nivolumab; TPS, tumor proportion scoreaEGFR testing available for n = 167; subtype percentages are calculated among EGFR-positive casesbRare EGFR mutations included G719X, L861Q, Exon 20 insertion, Exon 20, Exon 21 + 20, Exon 21 Leu861Gln, and Del-19 + L858RcPD-L1 testing available for n = 111. TPS categories are reported as a proportion of tested casesdICI-treated subgroup n = 52. Drug usage proportions are calculated within ICI-treated patients Baseline clinical characteristics of the study cohort Data are presented as n (%) unless otherwise indicated pembro, pembrolizumab; nivo, nivolumab; TPS, tumor proportion score aEGFR testing available for n = 167; subtype percentages are calculated among EGFR-positive cases bRare EGFR mutations included G719X, L861Q, Exon 20 insertion, Exon 20, Exon 21 + 20, Exon 21 Leu861Gln, and Del-19 + L858R cPD-L1 testing available for n = 111. TPS categories are reported as a proportion of tested cases dICI-treated subgroup n = 52. Drug usage proportions are calculated within ICI-treated patients PD-L1 tumor proportion score (TPS) was ≥ 50% in 36 patients (32.4%), 1–49% in 42 patients (37.8%), and < 1% in 33 patients (29.7%). Among the 52 patients who received immune checkpoint inhibitors (ICIs), pembrolizumab was the most frequently administered agent (86.5%), followed by atezolizumab (5.8%), nivolumab (3.8%), durvalumab (1.9%), and combination therapy with ipilimumab, pembrolizumab, and nivolumab (1.9%). Gut microbiota composition according to smoking status Global gut microbial profiles were first compared between nonsmokers and smokers (Fig. 2). Across all samples, 26,879 amplicon sequence variants (ASVs) were identified, of which 5,627 were shared between groups, while 11,367 and 9,885 ASVs were unique to nonsmokers and smokers, respectively (Fig. 2A). Overall comparison of gut microbiota between nonsmokers and smokers. (A) Venn diagram showing shared and unique amplicon sequence variants (ASVs) between groups; N denotes the number of ASVs. (B) Relative abundance of dominant bacterial genera. (C-D) Alpha diversity indices (Shannon entropy and observed OTUs) showing no significant difference between groups. (E) Principal coordinate analysis (PCoA) based on Bray–Curtis dissimilarity illustrating similar global community structures (PERMANOVA, p = 0.175). (F) LEfSe cladogram highlighting bacterial taxa differentially enriched between groups. Green nodes indicate taxa enriched in nonsmokers; red nodes indicate taxa enriched in smokers. ASV, amplicon sequence variant; PCoA, principal coordinate analysis; LEfSe, linear discriminant analysis effect size At the genus level, dominant taxa—including Bacteroides, Prevotella, Sutterella, Faecalibacterium, Escherichia–Shigella, Parabacteroides, Blautia, Megamonas, Bifidobacterium, and Akkermansia—did not differ significantly in relative abundance between groups (Fig. 2B). Nonetheless, smokers exhibited a consistent trend toward higher relative abundance of genera commonly associated with dysbiosis or pro-inflammatory states (e.g., Prevotella, Sutterella, and Escherichia–Shigella), accompanied by reduced abundance of putatively beneficial genera, including Bifidobacterium, Faecalibacterium, and Akkermansia. Alpha diversity, assessed by Shannon entropy and observed OTUs, did not differ significantly between nonsmokers and smokers (p = 0.755 and p = 0.901, respectively; Fig. 2C–D). Similarly, principal coordinate analysis based on Bray–Curtis dissimilarity revealed no clear separation between groups (p = 0.175; Fig. 2E), indicating that smoking did not markedly alter overall microbial richness or global community structure. Exploratory LEfSe analysis further identified selected taxon-level differences between groups, including enrichment of taxa within Clostridia and Peptostreptococcales in nonsmokers and higher abundance of Coriobacteriales Incertae Sedis and Streptococcaceae in smokers (Fig. 2F). Together, these findings suggest selected taxon-level differences rather than broad restructuring of the overall gut microbial community. Smoking-associated microbiota features and predicted functional pathways Given prior evidence that smoking may attenuate the immune-desert phenotype associated with EGFR mutations and that Del-19 tumors typically exhibit inferior responses to ICIs compared with L858R [4, 20], we next sought microbial taxa exhibiting opposing regulation patterns between Del-19 status and smoking exposure (Fig. 3A–B). The intersection of taxa enriched in Del-19 tumors but depleted in smokers yielded five taxa lacking definitive annotation (Fig. 3A). In comparison, Bifidobacterium longum emerged as a notable candidate, displaying increased abundance in Del-19 tumors but reduced abundance in smokers (Fig. 3B–C). Identification of smoking-associated microbiota and pathways potentially linked to ICI responsiveness. (A) Overlap analysis identifying bacterial taxa showing opposite regulation between the Del-19 and smoker groups; N denotes the number of species. (B) B. longum emerged as the key intersecting taxon; N denotes the number of species. (C) Box plot showing significantly reduced B. longum abundance in smokers compared with nonsmokers. (D) Predicted microbial functional pathways differing between groups, with emphasis on LPS biosynthesis-related pathways potentially associated with immune activation. (E) Correlation between B. longum abundance and Gram-negative bacterial fraction using Spearman’s correlation analysis. (F) Correlation between Gram-negative bacterial fraction and lpxM (hexa-acylated LPS-associated enzyme) abundance using Spearman’s correlation analysis. (G) Relative abundance of Gammaproteobacteria across ordered ICI response categories (PD → SD → PR); monotonic trend assessed by the Jonckheere-Terpstra test. LPS, lipopolysaccharide; TLR4, Toll-like receptor 4 To explore potential functional consequences associated with reduced B. longum abundance, predicted metagenomic pathway analysis was performed (Fig. 3D). Pathways enriched in smokers were predominantly related to lipopolysaccharide (LPS) biosynthesis, fatty acid metabolism, and bacterial secretion systems. Notably, the superpathway of LPS biosynthesis (PWY-1269), Kdo₂-lipid A biosynthesis (LPSSYN-PWY), and lipid IVA biosynthesis (LIPA-PWY) were among the most prominently enriched pathways. As LPS production is largely attributed to Gram-negative bacteria, we next examined whether reduced B. longum abundance was associated with a shift toward Gram-negative-dominant microbial communities. Across the cohort, B. longum relative abundance was inversely correlated with the proportion of Gram-negative taxa (Fig. 3E), suggesting that lower B. longum levels coincide with increased Gram-negative representation and potentially greater LPS exposure. Because LPS immunogenicity is determined by its acylation pattern [21], we further assessed the relationship between Gram-negative enrichment and predicted abundance of lpxM, an enzyme responsible for the final acylation step generating hexa-acylated lipid A. A positive correlation was observed between Gram-negative bacterial fraction and lpxM abundance (Fig. 3F). As lpxM is predominantly encoded by members of the class Gammaproteobacteria, we next evaluated whether Gammaproteobacteria abundance was associated with ICI response. Using a Jonckheere–Terpstra trend test, higher Gammaproteobacteria abundance was significantly associated with more favorable ICI responses (Fig. 3G). Collectively, these findings suggest an association between smoking-associated microbiota remodeling and predicted enrichment of LPS biosynthesis-related pathways. Increased abundance of lpxM-linked Gram-negative features may represent a microbial signature associated with ICI responsiveness. Association of Bifidobacterium longum abundance with ICI response and progression-free survival To assess the clinical relevance of B. longum, we examined its association with ICI response among NSCLC patients (Table 2). In this subset, the median age was 67 years, with a predominance of male patients (75.0%) and smokers (73.1%). Adenocarcinoma (55.8%) and squamous cell carcinoma (36.5%) were the most common histologic subtypes, and pembrolizumab was the most frequently administered ICI (86.5%). Table 2Baseline clinical characteristics of the ICI-treated cohortPatient CharacteristicsN = 52Age, median (range)67 (43–86)Sex, n (%)Female13 (25.0%)Male39 (75.0%)History of smoking, n (%)Nonsmokers14 (26.9%)Smokers38 (73.1%)Current29/38 (76.3%)Former9/38 (23.7%)Histology, n (%)Adenocarcinoma29 (55.8%)Squamous19 (36.5%)Adenosquamous1 (1.9%)Other3 (5.8%)Stage, n (%)I1 (1.9%)II0 (0.0%)III10 (19.2%)IV41 (78.9%)BMI, mean (range)23.91 (16.9–32.8)PD-L1 expression, n (%)aTPS ≥ 50% (high expression)19 (47.5%)TPS 1–49% (low expression)16 (40.0%)TPS < 1% (no expression)5 (12.5%)Data are presented as n (%) unless otherwise indicatedaPD-L1 testing available for n = 40. TPS categories are reported as a proportion of tested cases Baseline clinical characteristics of the ICI-treated cohort Data are presented as n (%) unless otherwise indicated aPD-L1 testing available for n = 40. TPS categories are reported as a proportion of tested cases After adjustment for potential confounders, including age, BMI, sex, PD-L1 expression, smoking history, and dietary habits, higher B. longum relative abundance was independently associated with poorer response to ICI therapy (Fig. 4A; Table 3). These results indicate that reduced B. longum abundance is associated with enhanced ICI responsiveness, potentially reflecting a microbiota configuration more permissive to immune activation. Association of B. longum abundance with ICI response and progression-free survival in NSCLC patients. (A) Multivariable ordinal logistic regression evaluating the association between B. longum abundance and clinical response to immune checkpoint inhibitors. Clinical response was defined according to RECIST v1.1 as an ordinal outcome from progressive disease (PD) to stable disease (SD) to partial response (PR), with odds ratios greater than 1 indicating higher odds of a better response. The model was adjusted for age, BMI, sex, PD-L1 expression, smoking history, and dietary habits. (B) Kaplan–Meier analysis of progression-free survival in pembrolizumab-treated stage IV NSCLC patients stratified by the median value of B. longum abundance (high, n = 16, progression events = 13; low, n = 19, progression events = 9). Numbers at risk are shown below the plot. Differences between groups were assessed using the log-rank test Table 3Multivariable ordinal logistic regression analysis evaluating the association of B. longum and clinical covariates with ICI responsePredictorOR (95% CI)p-valueB. longum (log1p)0.44 (0.22 – 0.89)0.022Age (years)1.18 (1.03 – 1.36)0.021Sex (female vs. male)1.09 (0.12 – 10.22)0.943BMI (per 1 unit)0.58 (0.38 – 0.89)0.012PD-L10.02 (0.0003 – 1.72)0.087Smoking13.53 (0.47 – 392.42)0.129Vegetable intake3.77 (0.26 – 54.48)0.330Fruit intake76.53 (1.43 – 4105.77)0.033Probiotic drink intake0.14 (0.01 – 1.82)0.131Fried food intake0.10 (0.01 – 1.80)0.119 Multivariable ordinal logistic regression analysis evaluating the association of B. longum and clinical covariates with ICI response Progression-free survival (PFS) analysis was subsequently restricted to patients receiving pembrolizumab, the standard first-line therapy for stage IV NSCLC [22]. Patients with lower B. longum abundance exhibited significantly prolonged PFS compared with those with higher abundance (log-rank p = 0.044; Fig. 4B). Together, these findings support B. longum as a potential negative predictive biomarker for ICI benefit in NSCLC, particularly in pembrolizumab-treated patients. Discussion In this prospective cohort of patients with NSCLC, we provide integrated evidence linking smoking-associated alterations in the gut microbiota to clinical outcomes following immune checkpoint inhibition. While previous studies have reported greater relative benefit from ICIs among smokers, often attributed to higher tumor mutational burden and increased neoantigen availability [3, 23], our findings extend this paradigm by identifying a microbiota-mediated component that may contribute to this effect. Specifically, smoking was associated with a relative depletion of Bifidobacterium longum and enrichment of Gram-negative taxa, accompanied by increased predicted abundance of lpxM, a lipid A myristoyltransferase critical for generating immunostimulatory hexa-acylated LPS. Clinically, lower B. longum abundance correlated with improved ICI response and prolonged progression-free survival, particularly among pembrolizumab-treated patients with stage IV disease. Together, these data suggest that the gut microbiota may serve as a functional intermediary linking smoking exposure to systemic immune modulation and ICI efficacy in NSCLC. Cigarette smoking is known to disrupt intestinal homeostasis through multiple mechanisms, including epithelial barrier injury, oxidative stress, immune perturbation, and exposure to combustion-derived chemicals [7, 8]. These factors collectively favor microbial dysbiosis, often characterized by expansion of inflammation-associated taxa and loss of health-associated commensals [24]. Although we did not observe significant differences in overall microbial diversity between smokers and nonsmokers, smokers exhibited a compositional shift toward taxa such as Prevotella, Sutterella, and Escherichia–Shigella, alongside reduced abundance of Bifidobacterium, Faecalibacterium, and Akkermansia. Among these, B. longum showed the most pronounced reduction in smokers and demonstrated a robust inverse relationship with Gram-negative bacterial representation. Notably, Gram-negative taxa within the class Gammaproteobacteria, which commonly encode lpxM, were positively associated with favorable ICI responses [25]. This pattern is consistent with a model in which smoking-associated microbiota remodeling may be linked to altered LPS-related immune signaling. Such changes may contribute to a host immune milieu more permissive to checkpoint blockade. Importantly, smoking status was not associated with improved ICI responsiveness in our cohort, possibly reflecting the heterogeneous and potentially opposing effects of smoking in NSCLC. In contrast, smoking-associated microbial features may represent more proximal biological correlates of ICI responsiveness than smoking status alone. The gut microbiota has emerged as a key regulator of antitumor immunity and ICI efficacy, influencing dendritic cell maturation, antigen presentation, and T cell priming [26, 27]. Several taxa, including Akkermansia muciniphila, have been shown to potentiate PD-1 blockade by reinforcing intestinal barrier function and promoting IL-12-dependent CD8⁺ T cell activation [9]. However, the role of B. longum in ICI responsiveness remains context-dependent and incompletely resolved. Jin et al. reported enrichment of B. longum among responders in a single-cohort study of advanced NSCLC treated with PD-1 inhibitors [28]. Komatsu et al. also reported higher abundance of Bifidobacteriaceae in responders [29]. Conversely, Routy et al. observed reduced B. longum abundance in NSCLC patients with prolonged PFS in a multicenter analysis [30]. Dora et al. similarly reported that Actinobacteria-related signatures were associated with unfavorable immunotherapy outcomes [31]. These discrepant findings may reflect differences in taxonomic resolution, cohort characteristics, treatment regimens, and response definition. In particular, Jin et al. grouped PR and SD together as responders, whereas our study modeled RECIST response as an ordered outcome. Moreover, the analysis by Komatsu et al. was performed at the family level, whereas our study specifically evaluated B. longum at the species level using full-length 16S rRNA sequencing. By integrating microbial composition, predicted functional pathways, and longitudinal clinical outcomes within a prospectively enrolled NSCLC cohort, our study suggests that smoking-associated depletion of B. longum, together with enrichment of lpxM-linked Gram-negative taxa, may mark a microbiota configuration associated with improved ICI outcomes. Mechanistically, B. longum is a well-established commensal within the Actinobacteria phylum and is generally regarded as a contributor to mucosal homeostasis and immune tolerance [32]. Its reduction in smokers was associated with a shift toward Gram-negative dominance, a finding of particular relevance given the immunomodulatory properties of bacterial LPS. Importantly, LPS immunogenicity is highly dependent on lipid A acylation status, with hexa-acylated forms acting as potent TLR4 agonists [25]. The terminal acylation step mediated by lpxM increases the likelihood of generating hexa-acylated lipid A, thereby amplifying innate immune signaling. Recent experimental studies have directly linked microbiota-derived hexa-acylated LPS to enhanced responsiveness to immune checkpoint blockade, establishing lipid A structure as a functionally relevant determinant of immunotherapy efficacy [21]. In our dataset, Gammaproteobacteria, a major reservoir of lpxM, tracked positively with ICI response, supporting the biological plausibility of this pathway in NSCLC. Body mass index (BMI) has been proposed as a modifier of immunotherapy efficacy, with several studies reporting improved outcomes among overweight or obese patients receiving ICIs [33]. However, findings in advanced NSCLC have been inconsistent, and large-scale analyses suggest that BMI alone is not a robust or reliable predictor of immunotherapy benefit [34, 35]. In our cohort, lower BMI was associated with improved ICI response, further supporting the notion that BMI may inadequately capture the complex host factors influencing immunotherapy efficacy. These observations suggest that more direct assessments of body composition, metabolic state, or systemic inflammatory profiles may provide greater clinical relevance for risk stratification than BMI alone. Intriguingly, we also observed that B. longum abundance tended to be higher in tumors harboring EGFR exon 19 deletions. EGFR-mutant NSCLC, particularly the Del-19 subtype, is characterized by an immune-cold microenvironment with limited T cell infiltration and reduced responsiveness to ICIs [4, 5, 36]. The relative enrichment of B. longum in this subgroup may therefore reflect a less inflammatory gut microbiota configuration that parallels intrinsic tumor immune quiescence. This observation raises the possibility that host–microbiota interactions vary according to oncogenic driver status and may contribute to molecular subtype–specific differences in immunotherapy response. Although PD-L1 TPS remains an important clinical biomarker for ICI treatment selection in NSCLC, it is not a sufficient standalone predictor of ICI outcomes [37]. The lack of a statistically significant association between PD-L1 TPS and ICI responsiveness in our results may reflect the limited sample size and exploratory nature of this analysis. In addition, ICI response is influenced by multiple factors beyond PD-L1 expression alone, including tumor immunogenicity and host immune status, and potentially the gut microbiome [37]. Several strengths of this study warrant emphasis. The prospective design, combined with systematic integration of clinical variables, microbial taxonomy, and predicted functional pathways, allowed us to construct a coherent and biologically grounded framework linking smoking-associated microbiota remodeling to ICI efficacy. Importantly, our findings challenge the notion that smoking-associated microbial alterations are uniformly detrimental and instead suggest that specific microbiota configurations induced by smoking may enhance immunotherapy responsiveness. This insight helps reconcile prior inconsistencies regarding the role of B. longum and highlights a translational opportunity: identifying microbiota-modulating strategies that recapitulate beneficial immune effects of smoking without its well-established harms. This study also has limitations. First, although full-length 16S rRNA sequencing improves taxonomic resolution, it cannot capture strain-level heterogeneity or directly measure microbial metabolites. Second, the functional analyses were based on PICRUSt2 predictions derived from 16S rRNA sequencing data. Therefore, these results represent predicted microbial functional potential rather than direct evidence of microbial gene expression, metabolite production, or host immune pathway activation. Third, although available dietary habit variables were incorporated into the multivariable analyses, these variables did not capture detailed nutrient composition, total caloric intake, longitudinal dietary changes, or short-term dietary variation before stool collection. Therefore, residual confounding by diet cannot be excluded. Finally, the ICI-treated cohort represented a subset of the original microbiome cohort. Therefore, analyses linking smoking-associated microbial features with ICI responsiveness and prognosis should be interpreted as an internal exploratory analysis. In summary, our study demonstrates that smoking-associated gut microbiota remodeling, characterized by depletion of B. longum and enrichment of lpxM-high Gammaproteobacteria, is associated with improved ICI response and prolonged progression-free survival in NSCLC (Fig. 5). These findings identify a previously unrecognized microbial axis linking smoking exposure to immunotherapy efficacy and suggest that targeted modulation of specific bacterial taxa or lipid A biosynthetic pathways may represent a novel strategy to optimize ICI outcomes. Proposed model linking smoking, gut microbiota remodeling, and immunotherapy efficacy in NSCLC Acknowledgements We thank Yi-Yao Hsu (Department of Laboratory Medicine, China Medical University Hospital, Taichung, Taiwan, ROC) for his extensive assistance with preliminary gut microbiome sequencing analyses and for training our team in bioinformatics methods. Author contributions Po-Yu Chien: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing—Original Draft, Writing—Review & Editing. Wen-Chien Cheng: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing—Original Draft, Writing—Review & Editing. Chin-Chuan Hung: Validation, Writing—Review & Editing, Supervision. Chih-Yen Tu: Project administration, Resources. Te-Chun Hsia: Project administration, Resources. Der-Yang Cho: Project administration. Po-Ren Hsueh: Project administration. Zi-Lun Lai: Software. Yi-Cheng Shen: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing—Review & Editing, Funding acquisition. Yu-Chao Lin: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing—Review & Editing, Supervision. Funding This study was supported by China Medical University Hospital [C1110812016-2]. The funders had no role in study design, data collection, analysis, decision to publish, or preparation of the manuscript. Data availability The datasets presented in this study are available in NCBI Sequence Read Archive with the accession number PRJNA1388452. The data that support the findings of this study are available from the corresponding author upon reasonable request. Declarations Conflict of interest The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Po-Yu Chien and Wen-Chien Cheng contributed equally to this work. Contributor Information Yi-Cheng Shen, Email: greywolf0127@gmail.com. Yu-Chao Lin, Email: 010001@tool.caaumed.org.tw. References Associated Data Data Availability Statement The datasets presented in this study are available in NCBI Sequence Read Archive with the accession number PRJNA1388452. The data that support the findings of this study are available from the corresponding author upon reasonable request.