A lung microbial signature for postoperative recurrence in stage I-II non-small cell lung cancer.

Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Respir Res . 2026 Mar 5;27:166. doi: 10.1186/s12931-026-03609-2 Search in PMC Search in PubMed View in NLM Catalog Add to search A lung microbial signature for postoperative recurrence in stage I–II non-small cell lung cancer Chunyan Gu Chunyan Gu 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China 2 Department of Laboratory Medicine, Precision Medicine Translational Research Center, Frontiers Science Center for Disease-related Molecular Network, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Chunyan Gu 1, 2, # , Yawen Qi Yawen Qi 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Yawen Qi 1, # , Yufang Xie Yufang Xie 3 Department of Respiratory and Critical Care Medicine, Jiujiang First People’s Hospital, Jiujiang, Jiangxi China Find articles by Yufang Xie 3, # , Lan Yang Lan Yang 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Lan Yang 1, # , Gang Wang Gang Wang 2 Department of Laboratory Medicine, Precision Medicine Translational Research Center, Frontiers Science Center for Disease-related Molecular Network, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Gang Wang 2 , Jing Zhou Jing Zhou 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Jing Zhou 1 , Jing Ren Jing Ren 4 The Integrated Care Management Center, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Jing Ren 4 , Sha Liu Sha Liu 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Sha Liu 1 , Guonian Zhu Guonian Zhu 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Guonian Zhu 1 , Weimin Li Weimin Li 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Weimin Li 1 , Wenchuang Hu Wenchuang Hu 2 Department of Laboratory Medicine, Precision Medicine Translational Research Center, Frontiers Science Center for Disease-related Molecular Network, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Wenchuang Hu 2, ✉ , Yaxin Chen Yaxin Chen 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Yaxin Chen 1, ✉ , Dan Liu Dan Liu 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China Find articles by Dan Liu 1, ✉ Author information Article notes Copyright and License information 1 Department of Respiratory and Critical Care Medicine, Institute of Respiratory Health, Frontiers Science Center for Disease-related Molecular Network, State Key Laboratory of Respiratory Health and Multimorbidity, West China Hospital, Sichuan University, Chengdu, Sichuan China 2 Department of Laboratory Medicine, Precision Medicine Translational Research Center, Frontiers Science Center for Disease-related Molecular Network, West China Hospital, Sichuan University, Chengdu, Sichuan China 3 Department of Respiratory and Critical Care Medicine, Jiujiang First People’s Hospital, Jiujiang, Jiangxi China 4 The Integrated Care Management Center, West China Hospital, Sichuan University, Chengdu, Sichuan China ✉ Corresponding author. # Contributed equally. Received 2025 Aug 29; Accepted 2026 Feb 26; Issue date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13072651  PMID: 41787488 Abstract Background Postoperative recurrence remains a prevalent and lethal threat to survival in early-stage non-small cell lung cancer (NSCLC), affecting approximately 20% of patients within five years. While emerging evidence implicates the lung microbiota and microbiota-related metabolites in carcinogenesis, their prognostic value for predicting NSCLC recurrence remains underexplored. Methods Using preoperative bronchoalveolar lavage fluid from 72 patients with stage I–II NSCLC (17 recurrence and 55 non-recurrence), we performed microbial 2bRAD-M sequencing to profile bacterial, fungal, and archaeal communities, along with untargeted metabolomics. Recurrence-associated microbial signatures were identified through machine learning approaches and network analysis. Results Patients with recurrence exhibited significantly reduced alpha diversity, shorter recurrence-free survival but enhanced microbial interactions compared with recurrence-free controls. We identified two predictive microbial signatures (32-species and 30-genus) that significantly outperformed conventional clinical factors, achieving mean area-under-the-curve of the receiver operating characteristic curve values of 0.92 (species-level) and 0.81 (genus-level), respectively. Notably, four potential recurrence-related drivers, namely the genera Paenibacillus , Cupriavidus , and Pseudomonas , along with the species Pseudomonas aeruginosa , were significantly increased in recurrence patients and associated with poorer survival. An integrated metabolomic analysis revealed that two microbiota-related metabolites, indole-3-lactic acid and pyridoxal, potentially mediated elevated recurrence risk through the tryptophan and vitamin B 6 metabolic pathways. Conclusion These findings elucidate the prognostic role of microbiota and metabolites in the lower respiratory tract of early-stage NSCLC, and highlight their dual potential as biomarkers and intervention targets. Supplementary Information The online version contains supplementary material available at 10.1186/s12931-026-03609-2. Keywords: non-small cell lung cancer, tumor recurrence, lung microbiome, metabolites, bronchoalveolar lavage fluid Introduction Surgical resection remains the primary curative treatment for early-stage NSCLC. Nevertheless, approximately 20% of stage I–II patients develop recurrence within five years postoperatively, predominantly within the first two years, which results in increased mortality due to aggressive tumor behavior and limited treatment response [ 1 , 2 ]. This persistent clinical challenge underscores the imperative to improve long-term survival. While adjuvant therapies, such as chemotherapy [ 3 , 4 ], targeted therapies (e.g., EGFR and ALK inhibitors) [ 5 – 7 ], and immunotherapies (e.g., anti-PD-1/PD-L1 antibodies) [ 8 ], have improved disease-free survival (DFS) for selected patients, their efficacy remains limited, partly due to the lack of robust biomarkers to reliably identify high-risk individuals. Mounting evidence indicates that DNA, RNA, and protein molecules, as well as their associated modifications, derived from clinical samples such as lung tissue, bronchoalveolar lavage fluid (BALF), and blood of cancer patients, hold prognostic significance in predicting recurrence or guiding adjuvant therapies [ 9 – 13 ]. For instance, a 14-gene transcriptional signature ( BAG1 , BRCA1 , CDC6 , CDK2AP1 , ERBB3 , FUT3 , IL11 , LCK , RND3 , SH3BGR , WNT3A , ESD , TBP , and YAP1 ) has been shown to stratify recurrence risk in resected stage I NSCLC, and to identify patients who might benefit from adjuvant chemotherapy [ 10 , 11 ]. Another example is HOXA9 methylation. Hypermethylation of HOXA9 has been found to be associated with worse recurrence-free survival (RFS) in early-stage lung adenocarcinoma (LUAD) [ 12 ]. However, these studies focus predominantly on human-derived molecules, neglecting the prognostic potential of the human microbiome—a pivotal regulator of carcinogenesis, immune microenvironment remodeling, and therapeutic response [ 14 – 16 ]. Although traditionally considered sterile [ 17 ], the lung harbors a microbiome, albeit with lower biomass and diversity than other sites like the gut and oral cavity. To some extent, this characteristic has impeded research on the relationship between lung microbiota and respiratory diseases, including lung cancer [ 18 , 19 ]. Distinct microbial profiles stratify histological subtypes and clinical progression in lung cancer. Compared with adjacent normal lung tissue, tumor tissue in NSCLC patients exhibited reduced microbial diversity, with diversity declining as cancer advances [ 20 ]. In bronchial washing fluid samples, specific genera such as Veillonella , Megasphaera , Actinomyces , and Arthrobacter were differentially abundant between LUAD and lung squamous cell carcinoma (LUSC) patients [ 21 ], suggesting their diagnostic potential. Prognostically, the abundance of the genus Stenotrophomonas in tumors is associated with increased NSCLC recurrence risk [ 20 ], and Legionella enrichment correlates with metastasis [ 22 ]. Additionally, elevated Acidovorax levels occur in smokers and TP53 -mutant LUSC tumors [ 23 ], further positioning the lung microbiota as promising biomarkers for cancer subtype classification and clinical outcome prediction. Mechanistically, microbiota in the lower respiratory tract (LRT) may drive carcinogenesis through epithelial damage, immune modulation [ 24 , 25 ], and activation of oncogenic pathways (e.g., IL-17, PI3K, ERK, and MAPK) [ 26 , 27 ]. However, the precise mechanisms by which microbiota orchestrate oncogenic processes remain largely unknown. Microbial signatures show promising diagnostic and prognostic utility in lung cancer. For example, a model incorporating the genera Veillonella and Megasphaera distinguished lung cancer from benign diseases, with an area-under-the-curve of the receiver operating characteristic curve (ROC-AUC) of 0.89 [ 28 ]. In a cohort of 46 stage II NSCLC patients, higher abundances of orders Pseudomonadales, Actinomycetales, and species Marmoricola aurantiacus in tumor tissue were associated with shorter DFS, while increased orders Clostridiales and Bacteroidales in normal lung tissue correlated with worse survival [ 29 ]. Additionally, a BALF-derived 19-genus signature (e.g., Staphylococcus , Bacillus , Anaerobacillus ) predicted stage I NSCLC recurrence (accuracy = 0.89; ROC-AUC = 0.77, 95% confidence interval (CI): 0.62–0.93) [ 30 ]. Plasma circulating microbial DNA models achieved ROC-AUCs of 0.88 (training) and 0.81 (test) for recurrence prediction in a cohort of 101 lung cancer patients (36 recurrence and 65 non-recurrence) [ 31 ]. Although these microbial signatures demonstrate predictive value, the extremely low microbial biomass in the LRT, combined with the species-rank detection limitations of metagenomics approaches (e.g., 16S rRNA gene amplicon sequencing), results in inadequate profiling of the lung microbiome. Additionally, critical limitations endure, including restricted cohort sizes, lack of external validation across diverse cohorts, and translational barriers to clinical implementation of microbial signatures. Microbiota-derived metabolites are recognized as critical mediators of carcinogenesis, inflammation, immune modulation, and immunotherapy response across cancers. For instance, the bacterial toxin colibactin, produced by certain Escherichia coli strains, induces human DNA damage and increases colorectal cancer risk [ 32 ]. Short-chain fatty acids such as butyrate regulate T cell differentiation, macrophage polarization, and cancer cell proliferation, thereby modulating inflammation and tumor progression [ 33 , 34 ]. The gut microbiota-derived metabolite phenylacetylglutamine suppresses T cell activity and attenuates anti-PD-1 efficacy in cancer therapy [ 35 ]. Despite recognizing metabolites as crucial microenvironment modulators, investigations linking the lung microbiome to its local metabolome are notably lacking, particularly in the context of recurrence. Here, we conducted integrated microbiome-metabolome profiling of BALF from 72 stage I–II NSCLC patients to identify recurrence-associated microbial signatures and their associated metabolic factors. We identified predictive species-level and genus-level signatures that collectively outperformed conventional clinical parameters. Additionally, two microbiota-related metabolites were identified as potential mediators of elevated recurrence risk. Our findings offer novel insights into lung microbial-metabolic interactions in postoperative NSCLC recurrence. Methods Patient enrollment and sample collection The prospective cohort study was conducted at West China Hospital of Sichuan University (Chengdu, China) to investigate the potential association between the lung microbiome and postoperative recurrence in NSCLC patients. This study protocol was approved by the Institutional Review Board of West China Hospital of Sichuan University (approval No. 2024[2611]), and written informed consent was obtained from all participants. Patients underwent surgical resection between March 2016 and December 2017, with postoperative follow-up continuing until April 2023. Patients meeting the following criteria were enrolled in this study: (1) pathologically confirmed stage I–II NSCLC; (2) no neoadjuvant therapy before surgery; (3) available BALF samples; (4) no history of concurrent malignancies; and (5) complete follow-up data and clinical records. Tumor staging was determined according to the 9th edition TNM classification system of the International Association for the Study of Lung Cancer [ 36 ]. The primary prognostic endpoint was RFS, defined as the time from surgical resection to the first NSCLC recurrence or tumor-related death. All BALF samples were obtained as a routine procedure via bronchoscopy before surgery and stored at −80 °C for future analysis. 2bRAD-M library preparation and sequencing DNA was extracted from 400 µL BALF using the TIANamp Micro DNA Kit (Tiangen, China), following the manufacturer’s instructions. 2bRAD-M libraries were prepared according to protocols described in previous studies [ 37 , 38 ]. To monitor potential reagent and environmental contamination, negative controls were included and processed in parallel during both DNA extraction and library preparation. Library pools were sequenced using the Illumina NovaSeq 6000 with the PE150 strategy, generating a median of 7,392,761 paired-end reads per sample (interquartile range: 6,655,496–8,188,989). Microbiota data processing Raw sequencing reads were first processed to identify enzyme reads based on the recognition site of the restriction enzyme BcgI , followed by the removal of reads containing > 8% undetermined bases (N) or with > 20% of bases exhibiting Phred quality scores < 30, yielding high-quality 32-bp fragments. Analysis of negative control samples showed only minimal microbial signals, indicating negligible background contamination. Ultimately, a median of 5,473,638 clean reads per sample (interquartile range: 4,921,514–6,033,656) were available, with details of each sample provided in Supplementary Table 1. Taxonomic profiling was performed using the 2bRAD-M computational pipeline ( https://github.com/shihuang047/2bRAD-M ). For each sample, the relative abundance of each species-level taxon was calculated as: where is the number of reads mapped to all 2bRAD markers of the species in a sample and is the number of all theoretical 2bRAD markers for that species. In this study, bacterial, archaeal, and fungal communities were profiled, with 2,764 species across 736 genera identified for downstream analyses. Microbial diversity analysis Based on taxonomic relative abundance, alpha diversity was assessed using four common metrics in the R package vegan (v2.6-10) [ 39 ], namely Shannon diversity index, Simpson diversity index ( ), species richness, and Pielou’s evenness. The Shannon and Simpson indices quantify overall community diversity, species richness reflects the count of distinct taxonomic units per sample, and Pielou’s evenness measures the uniformity of species distribution. Beta diversity was evaluated using Bray-Curtis dissimilarity and visualized using principal coordinates analysis (PCoA). Differences in microbial community composition were statistically evaluated using analysis of similarities (ANOSIM) and permutational multivariate analysis of variance (PERMANOVA) with 9,999 permutations. To confirm that the observed community differences were not driven by variation in within-group dispersion, the homogeneity of multivariate dispersions was evaluated using the permutation test (9,999 permutations). Differentially abundant genera and species between groups were identified using linear discriminant analysis (LDA) effect size (LEfSe) in the R package microeco (v1.14.0) [ 40 ]. Untargeted metabolomic profiling Methanolic extracts of BALF samples (1 mL) were prepared using solid-phase extraction columns and reconstituted in 300 µL methanol/water (4:1, v/v) containing L-2-chlorophenylalanine (4 µg/mL) for ultra-high-performance liquid chromatography–mass spectrometry (UHPLC–MS) analysis. Untargeted metabolomics was performed using an ACQUITY UPLC I-Class Plus system (Waters Corporation, USA) coupled to a Q Exactive™ Plus Hybrid Quadrupole-Orbitrap™ mass spectrometer (Thermo Fisher Scientific, USA), in both electrospray ionization (ESI) positive and negative modes. Quality control samples, prepared by pooling equal volumes of all sample extracts, were injected after every 10 analytical samples to monitor analytical stability and repeatability. The raw data were processed using the software Progenesis QI (v3.0) for peak picking and quantification of each metabolite. Among the extracted ion features, those with > 50% non-zero measurement values in at least one group were retained, and zero values of these features were imputed using quantile regression imputation of left-censored data (QRILC). The intensity of each peak was then normalized to the total spectral intensity in each sample. Metabolite annotation was performed using four common databases, including Human Metabolome Database (HMDB; v5.0) [ 41 ], LIPID Metabolites and Pathways Strategy (LIPID MAPS) [ 42 ], METLIN (v2019) [ 43 ], and Kyoto Encyclopedia of Genes and Genomes (KEGG; v106.0) [ 44 ]. Variable importance in projection (VIP) scores, derived from orthogonal partial least squares discriminant analysis (OPLS-DA) performed using the R package MetaboAnalystR (v4.2.0) [ 45 ], were used to rank the contribution of each metabolite to group discrimination, with statistical significance assessed using the Wilcoxon rank-sum test. To reduce the risk of overfitting, a 7-fold cross-validation strategy and 200-times response permutation testing were applied. Differential metabolites were identified based on the criteria of VIP > 1 and P < 0.05, and subsequently subjected to KEGG pathway enrichment analysis. Microbial signature construction Random forest classifiers [ 46 ] ( scikit-learn v1.5.1) were implemented in Python (v3.12.3) to construct lung microbial signatures for recurrence prediction (recurrence versus non-recurrence ). The data preprocessing pipeline included z-score standardization for numerical variables and one-hot encoding for categorical variables, with independent transformers fitted on each split and applied consistently to both training and test sets. To address class imbalance between recurrence and non-recurrence cases, random oversampling with replacement was applied exclusively during model training. To evaluate the impact of taxonomic resolution on predictive performance, models were independently trained using all features (relative abundances) at the family, genus, and species levels. Additionally, models were trained using clinical variables alone and in combination with microbial features. Hyperparameter optimization was performed using randomized search, evaluating 1,000 parameter combinations across a broad range of options using 20 repetitions of 5-fold cross-validation. Model performance was assessed by ROC-AUC, which quantifies classifier discrimination across all decision thresholds. The model configuration that yielded the highest mean cross-validation ROC-AUC over 100 iterations was selected for further analyses. Associations between the mean ROC-AUC scores and taxonomic resolution (from family to species level) were evaluated using Kendall τ test. Model interpretability was assessed using Shapley additive explanation (SHAP) values, computed via the TreeExplainer method [ 47 ] implemented in the Python package shap (v0.46.0). Global feature importance was quantified by the mean absolute SHAP values over 1,000 iterations to minimize stochastic effects. Directional effects (positive or negative) on predictions were determined by univariate linear regression between SHAP values and feature abundances. Summary statistics of global feature importance are presented in Supplementary Tables 2–4. Given that some features contributed minimally or not at all to prediction, features were ranked by importance, and iterative model refinement was conducted using the top-ranked features until optimal performance was achieved. Microbial interactome analysis Microbial interactions were characterized by performing Spearman rank correlation analyses among all pairwise combinations of bacterial, fungal, and archaeal taxa. For genus-level networks, associations with false discovery rate (FDR)-adjusted P < 0.05 were considered statistically significant and retained for network construction. Interaction modules were identified using the “modularity” function (resolution = 1.0) based on the Louvain community detection algorithm in the software Gephi (v0.10.1). For network visualization, correlations with |Spearman’s rho| > 0.7 were retained and arranged using the Fruchterman–Reingold algorithm implemented in the R package igraph (v2.1.4) [ 48 ]. Characteristics were compared across networks generated with absolute Spearman’s rho cutoffs ranging from 0.4 to 0.8. For species-level networks, except for the application of multidimensional scaling for layout optimization due to increased microbial complexity, all other analytical steps were consistent with those used for the genus-level networks. Key microbes driving interactome shifts between groups were identified using the network module structure shift (NetMoss) algorithm [ 49 ], with statistical significance defined as FDR-adjusted P < 0.05. Phylogenetic tree construction Accession identifiers for 32 recurrence-associated species were obtained from the Genome Taxonomy Database (GTDB), and the corresponding genome assemblies were subsequently downloaded from the NCBI Reference Sequence database using these identifiers. Phylogenetic reconstruction was conducted using GTDB-Tk (v2.4.1) [ 50 ], including three main steps, namely the identification of 120 ubiquitous single-copy bacterial marker genes, multiple sequence alignment, and maximum likelihood tree inference. The phylogenetic tree was annotated and visualized separately using the R packages ape (v5.8-1) [ 51 ] and ggtree (v3.16.0) [ 52 ]. Statistical analysis Kaplan–Meier survival analysis was conducted to assess the impact of identified microbes on RFS, with between-group differences evaluated by stratified log-rank tests. To mitigate non-convergence issues due to sparse microbial abundance data, zero values were replaced with a small pseudo-count (1 × 10 −6 ) prior to centered log-ratio (CLR) transformation. Univariate Cox regression identified significant prognostic clinical variables and microbial features (hazard ratio (HR) ≠ 1 and P < 0.05). To evaluate the independent prognostic value of the identified microbial signatures, we first calculated the cumulative sum of each taxon’s CLR-transformed abundance weighted by its corresponding univariate Cox regression coefficient, and then performed multivariate Cox regression analysis adjusting for clinical confounders. The proportional hazard assumption was tested for all Cox models using the Schoenfeld residual analysis prior to model fitting. These analyses were all completed using the R package survival (v3.8-3) [ 53 ]. Differences in categorical variables (e.g., smoking status, antibiotic use, and sex) were analyzed using Fisher’s exact test, while continuous variables were compared using the Kruskal–Wallis test (with Dunn’s test for multiple comparisons), Student’s t -test, or Wilcoxon rank-sum test, as appropriate. Spearman rank correlation analysis was employed to examine associations between microbial features and either clinical variables or metabolite levels, with P < 0.05 considered statistically significant. To control the FDR for multiple comparisons, P values were adjusted using the Benjamini-Hochberg method where applicable. All statistical analyses were performed using R (v4.5.0) and SPSS (v27.0). Results Clinical characteristics and study design The overall study design is shown in Fig. 1 . A total of 72 participants with stage I–II NSCLC were enrolled, with a median follow-up of 5.8 years (range 0.5–7.1 years). Of these, 70 (97.2%) were pathologically diagnosed with LUAD, and 17 (23.6%) experienced postoperative recurrence. No statistically significant differences were observed between the recurrence (R) and non-recurrence (NR) patients for clinical characteristics previously associated with lung microbial diversity, including age at surgery, antibiotic use, smoking status, or body mass index (Table 1 ). As expected, recurrence patients experienced significantly shorter RFS than non-recurrence patients ( P < 0.0001). Fig. 1. Open in a new tab Study design and analytical framework. A workflow depicting cohort enrollment, sample processing, and multi-omics integration for stage I–II NSCLC patients. BALF, bronchoalveolar lavage fluid; CV, cross-validation; NSCLC, non-small cell lung cancer; NR, non-recurrence; OPLS-DA, orthogonal partial least squares discriminant analysis; R, recurrence; ROC-AUC, area-under-the-curve of the receiver operating characteristic curve; UHPLC–MS, ultra-high-performance liquid chromatography–mass spectrometry; VIP, variable importance in projection Table 1. Baseline clinical characteristics of microbiome-evaluable participants stratified by recurrence status Non-recurrence ( N = 55) Recurrence ( N = 17) P Tumor histology, n (%) LUAD 55 (100.0) 15 (88.2) 0.053 NSCLC 0 2 (11.8) Sex, n (%) Female 37 (67.3) 10 (58.8) 0.568 Male 18 (32.7) 7 (41.2) Age, n (%) Low (<45) 8 (14.5) 3 (17.6) 0.450 Medium [45, 65) 36 (65.5) 13 (76.5) High (≥ 65) 11 (20) 1 (5.9) BMI (kg/m 2 ), n (%) Low (<18.5) 3 (5.5) 0 0.602 Medium [18.5, 25) 38 (69.1) 11 (64.7) High (≥ 25) 14 (25.5) 6 (35.3) Smoking status, n (%) Current 7 (12.7) 4 (23.5) 0.259 Former 3 (5.5) 2 (11.8) Never 45 (81.8) 11 (64.7) Tumor stage, n (%) Tis 2 (3.6) 1 (5.9) 0.062 IA 49 (89.1) 12 (70.6) IB 3 (5.5) 1 (5.9) II 1 (1.8) 3 (17.6) Antibiotic use, n (%) Yes 14 (25.5) 6 (35.3) 0.537 No 41 (74.5) 11 (64.7) Tissue or pleura invasion, n (%) Yes 0 3 (17.6) 0.011 No 55 (100.0) 14 (82.4) Adjunctive therapy after surgery, n (%) Yes 3 (5.5) 4 (23.5) 0.049 No 52 (94.5) 13 (76.5) RFS (days), median (p25; p75) 2,195 (2,109; 2,500) 897 (700; 1,706) <0.0001 Open in a new tab Percentages may total slightly over 100% due to rounding BMI body mass index, LUAD lung adenocarcinoma, NSCLC non-small cell lung cancer, RFS recurrence-free survival Distinct lung microbiota profiles between recurrence and non-recurrence NSCLC patients To characterize the composition of the lung microbiota, we conducted 2bRAD-M sequencing on BALF samples from 72 NSCLC patients. Finally, a total of 2,764 species (736 genera, 202 families) were detected, specifically 2,712 bacterial, 42 fungal, and 10 archaeal species (Supplementary Fig. 1A), consistent with previous reports and slightly outperforming 16S rRNA gene amplicon sequencing in the number of detected microbes [ 54 , 55 ]. Bacteroidota (33.64%), Proteobacteria (27.07%), and Firmicutes (13.30%) collectively accounted for nearly 75% of phylum-level lung microbiota (Supplementary Fig. 1B, C), while Prevotella , Streptococcus , and Neisseria were the most abundant genera (Fig. 2 A, B), in line with earlier findings [ 55 , 56 ]. The three most predominant species were Ralstonia pickettii , Porphyromonas pasteri , and Prevotella intermedia (Fig. 2 C, D). For microbial compositional differences between groups at the genus level, we found that the recurrence group exhibited a 2.44-fold higher abundance of Ralstonia (Wilcoxon rank-sum test; P = 0.039), alongside 23.30% ( P = 0.025) and 52.33% ( P = 0.023) reductions in Prevotella and Neisseria , respectively. At the species level, Ralstonia pickettii increased by 2.45-fold ( P = 0.005), whereas Prevotella pallens and Neisseria subflava decreased by 54.88% ( P = 0.007) and 54.51% ( P = 0.025), respectively. More details are in Supplementary Table 5. Fig. 2. Open in a new tab Lung microbiota divergence between recurrence and non-recurrence patients. A - D Stacked bar plots depicting the microbial composition of BALF samples at the genus level ( A – B ) and species level ( C – D ) across 72 patients (55 NR and 17 R). The top 20 taxa ranked by relative abundances are shown. BALF, bronchoalveolar lavage fluid; NR, non-recurrence; R, recurrence. E - F Box plots showing microbial alpha diversity in R and NR patients, measured by Shannon ( E ) and Simpson ( F ) diversity indices. Between-group differences were assessed using two-sided Student’s t -test (*: P < 0.05, **: P < 0.01). In the box plots, the central line represents the median; the lower and upper edges of the box correspond to the first and third quartiles; and the whiskers extend to 1.5 × the interquartile range. G Box plots depicting the distribution of Bray–Curtis dissimilarities for intragroup (within R/NR) and intergroup (between R and NR) comparisons. P values were obtained using ANOSIM. ANOSIM, analysis of similarities. H PCoA of beta diversity (Bray–Curtis dissimilarity) stratified by recurrence status. The x-axis represents the first principal coordinate (PCo1) and the y-axis indicates the second principal coordinate (PCo2). Ellipses denote 95% of each group’s multivariate normal distribution. Statistical significance was assessed using PERMANOVA (9,999 permutations). PCoA, principal coordinates analysis; PERMANOVA, permutational multivariate analysis of variance Alpha diversity, as measured by the Shannon and Simpson diversity indices, was markedly lower in recurrence patients compared with non-recurrence patients (Shannon diversity index, P = 0.015; Simpson diversity index, P = 0.006) (Fig. 2 E, F), which was associated with shorter RFS (Supplementary Fig. 1D). In contrast, species richness and Pielou’s evenness showed no significant association with recurrence status (Supplementary Fig. 1E, F). Among other clinical factors, current smokers exhibited lower alpha diversity than never-smokers (Simpson diversity index; Kruskal–Wallis test followed by Dunn’s multiple comparisons test, P = 0.034) (Supplementary Fig. 1G), whereas antibiotic exposure showed statistically non-significant effects ( P = 0.282) (Supplementary Fig. 1H, I), suggesting smoking may exert stronger ecological pressure on the lung microbiota than antibiotic use. Beta diversity analysis revealed significantly greater dissimilarity between recurrence and non-recurrence groups than within groups (ANOSIM; R = 0.17, P = 0.009) (Fig. 2 G). PCoA confirmed distinct microbial community separation (Fig. 2 H). This difference in community composition was unlikely to be an artifact of unequal group dispersion, as indicated by the non-significant test for homogeneity of multivariate dispersions ( P = 0.054). Among the clinical variables tested, smoking status showed the strongest association with microbial composition variance (PERMANOVA; R² = 0.07, P = 0.002) (Supplementary Fig. 1J), with full statistics provided in Supplementary Table 6. For the microbial interaction network, a substantially expanded interactome at the genus level was observed in the recurrence group compared with the non-recurrence group, as evidenced by increased involved microbe count and stronger interactions within each network module (Fig. 3 A, B). A similar pattern occurred at the species level, where six out of seven modules were markedly enlarged in the recurrence group (Supplementary Fig. 2A, B). Critically, this trend of enhanced microbial interactions in the recurrence group remained robust when applying cutoffs of |Spearman’s rho| greater than 0.6 across both taxonomic resolutions (Fig. 3 C, Supplementary Fig. 2C). Fig. 3. Open in a new tab Genus-level microbial interactomes in recurrence and non-recurrence groups. A Interaction networks of genus-level microbes in R and NR groups. Each node represents a bacterial (solid circle) or fungal/archaeal (hollow circle) genus, colored according to its corresponding network module. Each edge represents a significant correlation between pairs of genera (FDR-adjusted P < 0.05). A cutoff of 0.7 for the absolute Spearman’s rho was applied for network visualization. NR, non-recurrence; R, recurrence. B Paired pie charts for each network module. Left: module size (number of nodes) and composition, with bacterial genera shown in module-specific colors and fungal/archaeal genera in white. Right: taxonomic composition shows the five most abundant phyla within each module. C The numbers of nodes (dotted lines) and edges (solid lines) in genus-level microbial networks for R and NR groups across varying |Spearman’s rho| cutoffs (0.4 to 0.8; P < 0.05) Identification of the lung microbial signature for recurrence Given significant compositional divergence between the R and NR groups, we established an objective classification model using supervised machine learning. This pipeline integrated two input matrices: (1) eight clinical variables (smoking status, age, sex, antibiotic use, etc.), and (2) microbial features analyzed across family, genus, and species taxonomic resolutions. Predictive performance was evaluated for each matrix both individually and in combination. Clinical variables alone demonstrated limited efficacy in discriminating R from NR, with an ROC-AUC of 0.57 (Fig. 4 A). Despite previously observed associations linking both tissue/pleural invasion (HR = 9.25, 95% CI: 2.51–34.06) and adjunctive therapy after surgery to recurrence (HR = 3.65, 95% CI: 1.18–11.29) (Table 1 ; Supplementary Fig. 3A), these parameters were more effective in identifying poor-prognosis patients than in distinguishing recurrence status, underscoring the constraints of conventional clinical predictors. Fig. 4. Open in a new tab Species-level lung microbial signature outperforms clinical features in recurrence prediction. A ROC-AUCs of optimal recurrence versus non-recurrence classifiers using 20 repetitions of 5-fold cross-validation across taxonomic resolutions, encompassing clinical (yellow), microbiota (blue), and combined (green) feature sets. Data represent mean ROC-AUCs (circles) and SD (error bars) over 100 folds. The regression line (with the 95% confidence interval shaded) shows the association between taxonomic resolution and mean ROC-AUC scores for microbial-only features, with statistical significance assessed by Kendall τ test. ROC-AUC, area-under-the-curve of the receiver operating characteristic curve. SD, standard deviation. B Mean ROC-AUCs of recurrence prediction models using different feature sets, selected based on the ranked feature importance (absolute mean SHAP values) across taxonomic resolutions. Solid lines: microbial features only; dashed lines: microbial-clinical combined features. The vertical dashed line marks the number of species-level features that achieved the highest mean ROC-AUC. Line colors represent taxonomic levels (family, genus, species). SHAP, Shapley additive explanations ( C - D ) Horizontal bar plots showing the contribution of the top 30 genera ( C ) and 32 species ( D ) (1,000 iterations). Color indicates the effect direction: red denotes increased recurrence risk, while blue indicates a protective effect. Data present the percentage of total SHAP contribution for each taxon Microbial signature demonstrated taxonomy-dependent predictive capacity (Fig. 4 B). The genus-level classifier achieved the highest ROC-AUC of 0.81, with 30 genera (|SHAP| ≥ 3.3 × 10 −3 ) selected as recurrence predictors, half of which exhibited positive predictive effects for recurrence (Fig. 4 A-C, Supplementary Fig. 3B). For the species-level classifier, a set of 32 species (|SHAP| ≥ 2.4 × 10 −3 ) outperformed all other taxonomic resolutions, achieving optimal performance (ROC-AUC = 0.92). Only four species showed positive predictive effects, including Capnocytophaga sp000466425 (1.58%), Pseudomonas aeruginosa (0.72%), Prevotella sp003043945 (0.59%), and Gemella morbillorum (0.58%) (Fig. 4 D, Supplementary Fig. 3C). Phylogenetic analysis of these 32 species revealed a broad taxonomic distribution across bacterial lineages, with 18 being Gram-negative and seven, four, and three species belonging to the Prevotella , Neisseria , and Actinomyces genera, respectively (Supplementary Fig. 3D). As expected, a robust positive monotonic association existed between taxonomic resolution and mean ROC-AUCs (Kendall τ = 0.45, P = 5.33 × 10 −24 ), culminating in the species-level abundance achieving superior performance, significantly exceeding genus-level efficacy ( P = 7.20 × 10 −11 ) (Fig. 4 B; Supplementary Table 7). Notably, clinical variables failed to enhance microbiota-based models across different taxonomic resolutions. Predictive value of distinct microbial-rank signatures for recurrence Within the 30-genus signature, the eight genera Cupriavidus , Paenibacillus , Pseudomonas , Ralstonia , Eubacterium , Peptidiphaga , Neisseria , and Rothia were important contributors, collectively accounting for nearly 50% of the model’s feature weight (Supplementary Tables 3 and 8). Of them, five genera, namely Cupriavidus , Paenibacillus , Pseudomonas , Ralstonia , and Eubacterium , were significantly increased in recurrence patients (Fig. 5 A, Supplementary Fig. 4A), with higher relative abundances associated with poorer survival (Fig. 5 B). The remaining three genera were more abundant in non-recurrence patients, with their elevated abundances linked to improved prognosis. Based on the NetMoss algorithm, we identified 591 genera associated with microbial network alterations (FDR-adjusted P < 0.05) between NR and R (Supplementary Table 9). Notably, 22 (73.3%) of the 30 genera overlapped with these network-altered taxa, including six of the eight genera, excluding Eubacterium and Rothia . Univariate Cox regression analysis revealed that two of the eight genera were significantly associated with an increased risk of recurrence (HR > 1 and P < 0.05), namely Cupriavidus and Paenibacillus (Supplementary Fig. 4B). Fig. 5. Open in a new tab Relative abundance and recurrence-free survival analysis of recurrence-associated signature genera. A Box plots displaying the relative abundances of the genera Cupriavidus, Paenibacillus, Pseudomonas, Ralstonia, Eubacterium, Peptidiphaga, Neisseria , and Rothia in R and NR groups. Differences between groups were assessed using the Wilcoxon rank-sum test followed by FDR correction (* for FDR-adjusted P < 0.1 and ** for FDR-adjusted P < 0.05). NR, non-recurrence; R, recurrence. B Kaplan–Meier survival curves for RFS stratified by high versus low relative abundance of each genus. Statistical significance was determined using the log-rank test with FDR correction. RFS, recurrence-free survival. Detailed FDR-adjusted P values are provided in Supplementary Table 10 Within the 32-species signature, four species, namely Capnocytophaga sp000466425, Pseudomonas aeruginosa , Prevotella sp003043945, and Gemella morbillorum , were markedly elevated in recurrence patients (Fig. 6 A, Supplementary Fig. 4C), with higher relative abundances correlating with poorer RFS (Fig. 6 B). In contrast, increased abundances of Peptidiphaga sp000466165, Actinomyces johnsonii , Neisseria subflava , and Rothia dentocariosa were observed in non-recurrence patients and linked to more favorable survival. Additionally, we identified 2,241 species associated with microbial network changes (FDR-adjusted P < 0.05) between NR and R (Supplementary Table 11). Notably, 18 (56.3%) of the 32 species overlapped with network-altered taxa, encompassing all eight aforementioned species except Capnocytophaga sp000466425. Univariate Cox regression analysis identified three of eight species significantly linked to a lower recurrence risk (HR < 1 and P < 0.05), namely Peptidiphaga sp000466165, Actinomyces johnsonii , and Rothia dentocariosa (Supplementary Fig. 4D). Fig. 6. Open in a new tab Relative abundance and recurrence-free survival analysis of recurrence-associated signature species. A Box plots showing the relative abundances of the species Prevotella sp003043945, Capnocytophaga sp000466425, Pseudomonas aeruginosa, Gemella morbillorum, Peptidiphaga sp000466165, Actinomyces johnsonii, Neisseria subflava , and Rothia dentocariosa in R and NR groups. Between-group comparisons were performed using the Wilcoxon rank-sum test followed by FDR correction (*: FDR-adjusted P < 0.1, **: FDR-adjusted P < 0.05). R, recurrence; NR, non-recurrence. B Kaplan–Meier survival curves for RFS, stratified by relative abundances (high versus low) of these species (log-rank test with FDR correction). RFS, recurrence-free survival. The corresponding FDR-adjusted P values are listed in Supplementary Table 10 Subsequent Spearman rank correlation analyses of the 30 genera and 32 species with six clinical parameters revealed significantly lower abundances of the genera Peptidiphaga and Rothia , as well as the species Peptidiphaga sp000466165 in current smokers (Supplementary Fig. 5A, B). No significant associations were observed between these microbial signatures and antibiotic use ( P > 0.05), consistent with our prior finding that alpha diversity remained stable following antibiotic exposure (Supplementary Figs. 1I, 5A, B). Notably, both taxa-rank signatures served as independent risk factors after adjustment for the clinical covariates of tissue/pleural invasion and adjunctive therapy, with the 30-genus signature demonstrating an HR of 1.11 (95% CI: 1.05–1.18) (Supplementary Table 12) and the 32-species signature showing an HR of 1.15 (95% CI: 1.08–1.22) (Supplementary Table 13). Microbiota-related metabolites associated with recurrence Untargeted metabolomic profiling of 72 BALF samples identified 2,400 metabolites. An OPLS-DA model revealed 40 differential metabolites (VIP > 1 and P < 0.05), of which 39 were significantly upregulated in recurrence patients compared with non-recurrence cases, predominantly including lipids and lipid-like compounds (e.g., secoeremopetasitolide A, and oleic acid), amino acid derivatives (e.g., carbamoyl (2R)-2,5-diaminopentanoate, indolelactic acid (indole-3-lactic acid, ILA), and cyclo(L-prolyl-L-valyl)), as well as vitamin-related metabolites (e.g., pyridoxal) (Fig. 7 A, Supplementary Fig. 5C; Supplementary Table 14). KEGG pathway enrichment analysis of these differential metabolites identified three significantly enriched pathways ( P < 0.05), including tryptophan metabolism, vitamin B 6 metabolism, and inositol phosphate metabolism. Fig. 7. Open in a new tab Analysis of recurrence-related metabolites and associations with microbes. A Volcano plot showing changes in metabolite levels between R and NR patients. Red and blue dots represent metabolites that are significantly upregulated or downregulated in R patients (VIP > 1 and P < 0.05), while gray dots indicate non-significant metabolites. Dot size corresponds to VIP scores. NR, non-recurrence; R, recurrence; VIP, variable importance in projection. B – C Heatmaps depicting Spearman correlations (* for P < 0.05, ** for P < 0.01, and *** for P < 0.001) between differential metabolites and 30 recurrence-related genera ( B ) and 32 species ( C ). Dendrograms represent hierarchical clustering of metabolites (left) and microbes (top), based on Euclidean distance with complete linkage. D Sankey diagrams depicting five tumor-associated metabolites and their significantly associated signature microbes ( P < 0.05), as identified in figures B – C . Red and blue strips connecting microbes and metabolites represent positive and negative correlations, respectively Separate correlation analyses between the 30 genera and 32 species and the differential metabolites revealed that a total of 37 metabolites exhibited significant correlations with specific microbes ( P < 0.05). Of these, five metabolites were previously reported as tumor-associated and were upregulated in recurrence patients, namely ILA, pyridoxal, riboprine, cyclo(L-prolyl-L-valyl), and kaempferol 3-(6’’-sinapylglucosyl)-(1→2)-galactoside (Fig. 7 B–D). Specifically, the tryptophan metabolite ILA positively correlated with the genus Eubacterium ( r = 0.25, P = 0.035). In contrast, riboprine showed dual associations: positive with the genus Cutibacterium ( r = 0.23, P = 0.047) and negative with the genus Oribacterium ( r = -0.26, P = 0.026) and the species Pauljensenia sp000308055 ( r = -0.31, P = 0.008). Kaempferol 3-(6’’-sinapylglucosyl)-(1→2)-galactoside exhibited contrasting effects: positive correlations with the genera Paenibacillus ( r = 0.24, P = 0.044) and Cupriavidus ( r = 0.29, P = 0.013), while negative associations with six microbes including Treponema ( r = -0.40, P = 0.0005) and Porphyromonas ( r = -0.39, P = 0.001). Notably, pyridoxal, associated with vitamin B 6 metabolism, demonstrated inverse relationships with the genus Porphyromonas ( r = -0.31, P = 0.009), the species Lautropia dentalis ( r = -0.32, P = 0.006), and three other microbes: Treponema , Prevotella baroniae , and Neisseria mucosa . Similarly, cyclo(L-prolyl-L-valyl) negatively correlated with the species Peptidiphaga sp000466165 ( r = -0.28, P = 0.016). Discussion In this prospective cohort study, we comprehensively characterized the lung microbiome composition in early-stage NSCLC patients, revealing distinct microbial diversity patterns across clinical subgroups. Compared with non-recurrence controls, recurrence patients exhibited markedly reduced alpha diversity alongside enhanced microbial interaction networks, which were collectively associated with shorter RFS. Leveraging these findings, we identified 32-species and 30-genus signatures for recurrence prediction, which demonstrated substantially superior performance relative to conventional clinical factors. Furthermore, an integrated analysis of the lung microbiome and metabolome identified two microbiota-related metabolites significantly linked to recurrence risk, implying that recurrence-associated microbes may modulate tumor progression through these specific metabolic mediators. With the increasing application of microbiome profiling in the diagnosis and prognosis of common cancers, there is growing interest in developing lung microbial signatures for predicting recurrence risk. Notably, using BALF samples consistent with our methodology, Patnaik et al. [ 30 ] identified a 19-genus signature for predicting lung cancer recurrence in 36 stage I NSCLC patients (18 R and 18 NR), achieving an reported ROC-AUC of 0.77. In this study, we used a cohort of 17 R and 55 NR patients to establish a recurrence predictive signature, achieving an ROC-AUC of 0.81 at the genus level. Although our genus-level signature incorporated more microbial features than Patnaik et al.’s study, it demonstrated better predictive performance. Furthermore, we also identified a species-level signature, with an ROC-AUC of 0.92, indicating that species-resolved microbial profiles may provide more precise and reliable recurrence risk stratification. As is well known, opportunistic pathogens are commensal microorganisms that typically colonize human hosts harmlessly but possess latent pathogenic potential [ 57 ]. In our study cohort, four taxa reported as opportunistic pathogens, namely the genera Paenibacillus , Cupriavidus , and Pseudomonas , along with the species Pseudomonas aeruginosa , were found to have elevated abundance in NSCLC recurrence patients, which was linked to increased progression risk. Paenibacillus , a Gram-positive rod-shaped bacterial genus, exhibited increased relative abundance in NSCLC patients with brain metastases [ 58 ]. It invades the body through virulence mechanisms, such as the production of thiol-activated cytolysins [ 59 ], leading to a range of infections, particularly in immunocompromised or immunodeficient individuals, including sepsis, meningitis, and pneumonitis [ 60 ]. Cupriavidus , a glucose-nonfermenting Gram-negative bacterium detected in respiratory secretions and harboring intrinsic antibiotic resistance genes, potentiates pulmonary infections in cancer patients or immunocompromised individuals, with higher abundance linked to a more severe infection risk [ 61 ]. The genus Pseudomonas and its representative species Pseudomonas aeruginosa (a Gram-negative pathogen) exhibit elevated abundance in patients with cystic fibrosis and chronic obstructive pulmonary disease [ 62 ]. Their multiple virulence factors, including lipopolysaccharide, outer membrane proteins, and pyocyanin, collectively impair respiratory epithelial defenses [ 62 ]. Specifically, pyocyanin induces oxidative stress, increases airway barrier permeability, promotes the secretion of pro-inflammatory cytokines, and disrupts macrophage phagocytic function, thereby exacerbating lung damage [ 62 ]. These findings implicate Pseudomonas and Pseudomonas aeruginosa in promoting pulmonary inflammation and tumor recurrence. The pulmonary commensal microbiota plays a significant role in respiratory health, with altered abundance linked to various diseases, including cancer [ 63 ]. Of our identified microbial signature, the genera Neisseria and Porphyromonas were found to be significantly more abundant in non-recurrence patients, which was associated with improved RFS. Neisseria , a Gram-negative bacterial genus commonly found in the oral cavity and LRT of healthy individuals, was found to be decreased in the saliva of lung cancer patients compared with healthy controls [ 64 ]. A higher abundance of Neisseria was associated with improved RFS in NSCLC patients [ 29 ], which aligns with our findings. Porphyromonas is a prevalent genus of anaerobic bacteria in the respiratory tract. It includes the species Porphyromonas gingivalis , a well-characterized oral pathogen associated with periodontitis, and has been closely related to the occurrence and development of multiple tumors [ 65 ], including lung cancer [ 66 ]. However, in our study, the main pulmonary Porphyromonas species was Porphyromonas pasteri , which lacks known virulence genes and is associated with healthy lung conditions [ 67 ]. Interestingly, Porphyromonas has been shown to reduce the risk of Pseudomonas aeruginosa pulmonary infections in patients with cystic fibrosis [ 68 ], suggesting an antagonistic effect against such pathogens. Notably, the genus Peptidiphaga (31.71%) and the species Peptidiphaga sp000466165 (10.34%) demonstrated the highest contribution rates in our prediction model, both associated with decreased recurrence risk. Peptidiphaga , an anaerobic bacterium that selectively metabolizes various amino acids (e.g., arginine, histidine, and proline) [ 69 ], was found at higher abundance in the saliva of healthy controls and in patients successfully treated for aggressive periodontitis [ 70 ], with its abundance positively correlating with anti-inflammatory IL-10 levels in gingival crevicular fluid, suggesting an association of Peptidiphaga with the anti-inflammatory profile and healthy periodontal conditions [ 71 ]. Additionally, in this study, we observed that Peptidiphaga exhibited lower abundance in current smokers, which may suggest that smoking-related lung cancer risk factors promote the LRT microbiota dysbiosis, reducing Peptidiphaga abundance and thereby providing nutrients and space for the proliferation of pathogenic bacteria. Although all studies, including ours, show a close link between lung microbes and disease development, the underlying mechanisms remain unclear and require further research. Lung microbiota-related metabolites are key mediators of local immunological remodeling [ 72 , 73 ]. In this study, two recurrence-related metabolites were identified, namely ILA, a metabolite in the tryptophan metabolic pathway, and pyridoxal, a vitamin B 6 metabolism–related metabolite. Tryptophan is an essential amino acid, and its imbalanced level, coupled with certain dysregulated metabolites, has been linked to various human diseases, with some metabolites known to enhance cancer cell motility and migration [ 74 ]. ILA, an indole derivative mainly derived from the genera Lactobacillus [ 75 ] and Bifidobacterium [ 76 ], reportedly activates aryl hydrocarbon receptor signaling in tumor-associated macrophages, inhibiting CD8⁺TNF-α⁺IFN-γ⁺ T cell infiltration and promoting pancreatic ductal adenocarcinoma progression [ 77 ]. Our findings demonstrated that elevated ILA was associated with Eubacterium enrichment in recurrence patients, suggesting its potential tumor-promoting effect. Pyridoxal and high-dose vitamin B 6 /B 12 supplementation over 10 years were associated with increased lung cancer incidence in male smokers [ 78 ]. Mechanistically, pyridoxal contributes to this risk through inflammation-driven catabolism, evidenced by an increased ratio of 4-pyridoxic acid to the sum of pyridoxal and pyridoxal-5’-phosphate [ 79 ]. In our study, upregulation of pyridoxal was associated with reduced abundance of the genus Porphyromonas , as well as the species Lautropia dentalis and Neisseria mucosa in recurrence patients. Additionally, the metabolites riboprine, cyclo(L-prolyl-L-valyl), and kaempferol 3-(6”-sinapylglucosyl)-(1→2)-galactoside were also significantly upregulated in recurrence patients. Although there are no reports of these metabolites promoting lung cancer, they have been shown to have anticancer or anti-inflammatory properties in other cancer types [ 80 – 82 ]. Our study conducted a comparative analysis of the recurrence-related lung microbiome in patients with early-stage NSCLC. Notably, we utilized BALF samples, which are easier to collect than invasively obtained lung tissue. The 2bRAD-M technique, rather than conventional 16S rRNA gene amplicon sequencing, was employed to simultaneously characterize bacterial, fungal, and archaeal communities. However, several limitations exist. First, the sample size for identification of recurrence-related microbial signatures remained relatively small compared with that in studies on gut microbiota. Expanding the sample size, particularly for recurrence patients, would enhance the precision of identified lung microbial signatures. Second, although we identified recurrence-related microbial signatures with strong predictive performance, validation using external cohorts remains insufficient. Finally, the mechanisms underlying the identified microbiota-related metabolites are unclear and require further experimental validation. Conclusion In summary, we identified two recurrence-related microbial signatures at both the genus and species levels, as well as two microbiota-related metabolites linked to an increased risk of NSCLC recurrence. Our findings highlight the utility of multi-omics integration in elucidating microbiome-metabolome interactions and advance our comprehension of the pulmonary microbiota in NSCLC oncogenesis and progression. Supplementary Information Supplementary Material 1. (21MB, docx) Supplementary Material 2. (409.2KB, xlsx) Acknowledgements Not applicable. Authors’ contributions D.L.—conceptualization, supervision, validation, investigation, methodology, funding acquisition, and project administration. Y.C.—conceptualization, supervision, investigation, methodology, manuscript review and revision, funding acquisition and project administration. W.H.—conceptualization, supervision, validation, manuscript review and revision, and project administration. W.L.—conceptualization, supervision, and project administration. C.G.—investigation, methodology, analysis, visualization, writing of original draft, and manuscript revision. Y.Q.—investigation, interpretation, writing of original draft, and manuscript revision. Y.X.—investigation, sample collection, and data curation. L.Y.—sample collection and data curation. G.W.—manuscript review and revision. J.Z., J.R., S.L., and G.Z.—sample collection. All authors have reviewed and approved the final submitted manuscript. Funding This research was supported by the Noncommunicable Chronic Diseases–National Science and Technology Major Project (No. 2024ZD0529501 / 2024ZD0529500 to D. Liu), the National Natural Science Foundation of China (Nos. 82173182 to D. Liu and 82300011 to Y. Chen), and the Science and Technology Program of Sichuan Province (No. 2023NSFSC1939 to D. Liu). Data availability All raw sequencing data have been deposited in the Genome Sequence Archive for Human (GSA-Human) in National Genomics Data Center (accession number: HRA011114, https://ngdc.cncb.ac.cn/gsa-human/ ). Metabolomic data have been submitted to figshare ( https://figshare.com/s/8bdd4696a3d6b39433f6 ). Declarations Ethics approval and consent to participate The study protocol was conducted in accordance with the ethical principles of the Declaration of Helsinki and was approved by the Institutional Review Board of West China Hospital of Sichuan University (approval No. 2024[2611]). Written informed consent was obtained from all participants before enrollment. Consent for publication Not applicable. Competing interests 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. Chunyan Gu, Yawen Qi, Yufang Xie and Lan Yang contributed equally to this work. Contributor Information Wenchuang Hu, Email: [email protected]. Yaxin Chen, Email: [email protected]. Dan Liu, Email: [email protected]. References 1. Lou F, Sima CS, Rusch VW, Jones DR, Huang J. Differences in patterns of recurrence in early-stage versus locally advanced non-small cell lung cancer. Ann Thorac Surg. 2014;98:1755–61. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Potter AL, Costantino CL, Suliman RA, Haridas CS, Senthil P, Kumar A, Mayne NR, Panda N, Martin LW, Yang C-FJ. Recurrence After Complete Resection for Non-Small Cell Lung Cancer in the National Lung Screening Trial. Ann Thorac Surg. 2023;116:684–92. [ DOI ] [ PubMed ] [ Google Scholar ] 3. Felip E, Altorki N, Zhou C, Csőszi T, Vynnychenko I, Goloborodko O, Luft A, Akopov A, Martinez-Marti A, Kenmotsu H, et al. Adjuvant atezolizumab after adjuvant chemotherapy in resected stage IB–IIIA non-small-cell lung cancer (IMpower010): a randomised, multicentre, open-label, phase 3 trial. Lancet. 2021;398:1344–57. [ DOI ] [ PubMed ] [ Google Scholar ] 4. Pignon J-P, Tribodet H, Scagliotti GV, Douillard J-Y, Shepherd FA, Stephens RJ, Dunant A, Torri V, Rosell R, Seymour L, et al. Lung Adjuvant Cisplatin Evaluation: A Pooled Analysis by the LACE Collaborative Group. J Clin Oncol. 2008;26:3552–9. [ DOI ] [ PubMed ] [ Google Scholar ] 5. Herbst RS, Wu Y-L, John T, Grohe C, Majem M, Wang J, Kato T, Goldman JW, Laktionov K, Kim S-W, et al. Adjuvant Osimertinib for Resected EGFR-Mutated Stage IB-IIIA Non–Small-Cell Lung Cancer: Updated Results From the Phase III Randomized ADAURA Trial. J Clin Oncol. 2023;41:1830–40. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Tsuboi M, Herbst Roy S, John T, Kato T, Majem M, Grohé C, Wang J, Goldman Jonathan W, Lu S, Su W-C, et al. Overall Survival with Osimertinib in Resected EGFR-Mutated NSCLC. N Engl J Med. 2023;389:137–47. [ DOI ] [ PubMed ] [ Google Scholar ] 7. Wu Y-L, Dziadziuszko R, Ahn Jin S, Barlesi F, Nishio M, Lee Dae H, Lee J-S, Zhong W, Horinouchi H, Mao W, et al. Alectinib in Resected ALK-Positive Non–Small-Cell Lung Cancer. N Engl J Med. 2024;390:1265–76. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Felip E, Altorki N, Zhou C, Vallières E, Martínez-Martí A, Rittmeyer A, Chella A, Reck M, Goloborodko O, Huang M, et al. Overall survival with adjuvant atezolizumab after chemotherapy in resected stage II-IIIA non-small-cell lung cancer (IMpower010): a randomised, multicentre, open-label, phase III trial. Ann Oncol. 2023;34:907–19. [ DOI ] [ PubMed ] [ Google Scholar ] 9. Abbosh C, Hodgson D, Doherty GJ, Gale D, Black JRM, Horn L, Reis-Filho JS, Swanton C. Implementing circulating tumor DNA as a prognostic biomarker in resectable non-small cell lung cancer. Trends Cancer. 2024;10:643–54. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Jiang Y, Lin Y, Fu W, He Q, Liang H, Zhong R, et al. The impact of adjuvant EGFR-TKIs and 14-gene molecular assay on stage I non–small cell lung cancer with sensitive EGFR mutations. EClinical Medicine. 2023;64:102205. [ DOI ] [ PMC free article ] [ PubMed ] 11. Kratz JR, He J, Van Den Eeden SK, Zhu Z-H, Gao W, Pham PT, Mulvihill MS, Ziaei F, Zhang H, Su B, et al. A practical molecular assay to predict survival in resected non-squamous, non-small-cell lung cancer: development and international validation studies. Lancet. 2012;379:823–32. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Robles AI, Arai E, Mathé EA, Okayama H, Schetter AJ, Brown D, Petersen D, Bowman ED, Noro R, Welsh JA, et al. An Integrated Prognostic Classifier for Stage I Lung Adenocarcinoma Based on mRNA, microRNA, and DNA Methylation Biomarkers. J Thorac Oncol. 2015;10:1037–48. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Wang C, Li J, Chen J, Wang Z, Zhu G, Song L, et al. Multi-omics analyses reveal biological and clinical insights in recurrent stage I non-small cell lung cancer. Nat Commun. 2025;16:1477. [ DOI ] [ PMC free article ] [ PubMed ] 14. Cullin N, Azevedo Antunes C, Straussman R, Stein-Thoeringer CK, Elinav E. Microbiome and cancer. Cancer Cell. 2021;39:1317–41. [ DOI ] [ PubMed ] [ Google Scholar ] 15. Nejman D, Livyatan I, Fuks G, Gavert N, Zwang Y, Geller LT, Rotter-Maskowitz A, Weiser R, Mallel G, Gigi E, et al. The human tumor microbiome is composed of tumor type–specific intracellular bacteria. Science. 2020;368:973–80. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Sepich-Poore GD, Zitvogel L, Straussman R, Hasty J, Wargo JA, Knight R. The microbiome and human cancer. Science. 2021;371:eabc4552. [ DOI ] [ PMC free article ] [ PubMed ] 17. Hilty M, Burke C, Pedro H, Cardenas P, Bush A, Bossley C, Davies J, Ervine A, Poulter L, Pachter L, et al. Disordered microbial communities in asthmatic airways. PLoS ONE. 2010;5:e8578. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Dickson RP, Huffnagle GB. The lung microbiome: new principles for respiratory bacteriology in health and disease. PLoS Pathog. 2015;11:e1004923. [ DOI ] [ PMC free article ] [ PubMed ] 19. Man WH, de Steenhuijsen Piters WAA, Bogaert D. The microbiota of the respiratory tract: gatekeeper to respiratory health. Nat Rev Microbiol. 2017;15:259–70. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Kim OH, Choi BY, Kim DK, Kim NH, Rho JK, Sul WJ, Lee SW. The microbiome of lung cancer tissue and its association with pathological and clinical parameters. Am J Cancer Res. 2022;12:2350–62. [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Huang D, Su X, Yuan M, Zhang S, He J, Deng Q, Qiu W, Dong H, Cai S. The characterization of lung microbiome in lung cancer patients with different clinicopathology. Am J Cancer Res. 2019;9:2047–63. [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Yu G, Gail MH, Consonni D, Carugno M, Humphrys M, Pesatori AC, et al. Characterizing human lung tissue microbiota and its relationship to epidemiological and clinical features. Genome Biol. 2016;17:163. [ DOI ] [ PMC free article ] [ PubMed ] 23. Greathouse KL, White JR, Vargas AJ, Bliskovsky VV, Beck JA, von Muhlinen N, et al. Interaction between the microbiome and TP53 in human lung cancer. Genome Biol. 2018;19:123. [ DOI ] [ PMC free article ] [ PubMed ] 24. Goto T. Microbiota and lung cancer. Sem Cancer Biol. 2022;86:1–10. [ DOI ] [ PubMed ] [ Google Scholar ] 25. Li B, Wang D, Zhang C, Wang Y, Huang Z, Yang L, et al. Role of respiratory system microbiota in development of lung cancer and clinical application. iMeta. 2024;3:e232. [ DOI ] [ PMC free article ] [ PubMed ] 26. Tsay J-CJ, Wu BG, Sulaiman I, Gershner K, Schluger R, Li Y, Yie T-A, Meyn P, Olsen E, Perez L, et al. Lower Airway Dysbiosis Affects Lung Cancer Progression. Cancer Discov. 2021;11:293–307. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Tsay J-CJ, Wu BG, Badri MH, Clemente JC, Shen N, Meyn P, Li Y, Yie T-A, Lhakhang T, Olsen E, et al. Airway Microbiota Is Associated with Upregulation of the PI3K Pathway in Lung Cancer. Am J Respir Crit Care Med. 2018;198:1188–98. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Lee SH, Sung JY, Yong D, Chun J, Kim SY, Song JH, Chung KS, Kim EY, Jung JY, Kang YA, et al. Characterization of microbiome in bronchoalveolar lavage fluid of patients with lung cancer comparing with benign mass like lesions. Lung Cancer. 2016;102:89–95. [ DOI ] [ PubMed ] [ Google Scholar ] 29. Peters BA, Pass HI, Burk RD, Xue X, Goparaju C, Sollecito CC, et al. The lung microbiome, peripheral gene expression, and recurrence-free survival after resection of stage II non-small cell lung cancer. Genome Med. 2022;14:121. [ DOI ] [ PMC free article ] [ PubMed ] 30. Patnaik SK, Cortes EG, Kannisto ED, Punnanitinont A, Dhillon SS, Liu S, et al. Lower airway bacterial microbiome may influence recurrence after resection of early-stage non–small cell lung cancer. J Thorac Cardiovasc Surg. 2021;161:419–29.e16. [ DOI ] [ PubMed ] [ Google Scholar ] 31. Chen H, Ma Y, Xu J, Wang W, Lu H, Quan C, et al. Circulating microbiome DNA as biomarkers for early diagnosis and recurrence of lung cancer. Cell Rep Med. 2024;5:101499. [ DOI ] [ PMC free article ] [ PubMed ] 32. Nougayrède J-P, Homburg S, Taieb F, Boury M, Brzuszkiewicz E, Gottschalk G, Buchrieser C, Hacker J, Dobrindt U, Oswald E. Escherichia coli Induces DNA Double-Strand Breaks in Eukaryotic Cells. Science. 2006;313:848–51. [ DOI ] [ PubMed ] [ Google Scholar ] 33. Singh N, Gurav A, Sivaprakasam S, Brady E, Padia R, Shi H, Thangaraju M, Prasad Puttur D, Manicassamy S, Munn David H, et al. Activation of Gpr109a, Receptor for Niacin and the Commensal Metabolite Butyrate, Suppresses Colonic Inflammation and Carcinogenesis. Immunity. 2014;40:128–39. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Ma Y, Chen H, Li H, Zheng M, Zuo X, Wang W, et al. Intratumor microbiome-derived butyrate promotes lung cancer metastasis. Cell Rep Med. 2024;5:101488. [ DOI ] [ PMC free article ] [ PubMed ] 35. Zhu X, Hu M, Huang X, Li L, Lin X, Shao X, et al. Interplay between gut microbial communities and metabolites modulates pan-cancer immunotherapy responses. Cell Metabol. 2025;37:806–23.e6. [ DOI ] [ PubMed ] [ Google Scholar ] 36. Rami-Porta R, Nishimura KK, Giroux DJ, Detterbeck F, Cardillo G, Edwards JG, Fong KM, Giuliani M, Huang J, Kernstine KH, et al. The International Association for the Study of Lung Cancer Lung Cancer Staging Project: Proposals for Revision of the TNM Stage Groups in the Forthcoming (Ninth) Edition of the TNM Classification for Lung Cancer. J Thorac Oncol. 2024;19:1007–27. [ DOI ] [ PubMed ] [ Google Scholar ] 37. Sun Z, Huang S, Zhu P, Tzehau L, Zhao H, Lv J, et al. Species-resolved sequencing of low-biomass or degraded microbiomes using 2bRAD-M. Genome Biol. 2022;23:36. [ DOI ] [ PMC free article ] [ PubMed ] 38. Wang S, Meyer E, McKay JK, Matz MV. 2b-RAD: a simple and flexible method for genome-wide genotyping. Nat Methods. 2012;9:808–10. [ DOI ] [ PubMed ] [ Google Scholar ] 39. Oksanen J, Simpson GL, Blanchet FG, Kindt R, Legendre P, Minchin PR, O’Hara RB, Solymos P, Stevens MHH, Szoecs E et al. vegan: Community Ecology Package. 2025. 40. Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS, et al. Metagenomic biomarker discovery and explanation. Genome Biol. 2011;12:R60. [ DOI ] [ PMC free article ] [ PubMed ] 41. Wishart DS, Guo A, Oler E, Wang F, Anjum A, Peters H, Dizon R, Sayeeda Z, Tian S, Lee Brian L, et al. HMDB 5.0: the Human Metabolome Database for 2022. Nucleic Acids Res. 2022;50:D622–31. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Conroy MJ, Andrews RM, Andrews S, Cockayne L, Dennis Edward A, Fahy E, Gaud C, Griffiths William J, Jukes G, Kolchin M, et al. LIPID MAPS: update to databases and tools for the lipidomics community. Nucleic Acids Res. 2024;52:D1677–82. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Smith CA, Maille GO, Want EJ, Qin C, Trauger SA, Brandon TR, Custodio DE, Abagyan R, Siuzdak G. METLIN: A Metabolite Mass Spectral Database. Ther Drug Monit. 2005;27:747–51. [ DOI ] [ PubMed ] [ Google Scholar ] 44. Kanehisa M, Goto S. KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res. 2000;28:27–30. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 45. Pang Z, Chong J, Li S, Xia J. MetaboAnalystR 3.0: Toward an Optimized Workflow for Global Metabolomics. Metabolites. 2020;10:186. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Breiman L. Random Forests. Mach Learn. 2001;45:5–32. [ Google Scholar ] 47. Lundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, Katz R, Himmelfarb J, Bansal N, Lee S-I. From local explanations to global understanding with explainable AI for trees. Nat Mach Intell. 2020;2:56–67. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Csárdi G, Nepusz T, Traag V, Horvát S, Zanini F, Noom D. Müller K: igraph: Network Analysis and Visualization in R. 2025. 49. Xiao L, Zhang F, Zhao F. Large-scale microbiome data integration enables robust biomarker identification. Nat Comput Sci. 2022;2:307–16. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Chaumeil P-A, Mussig AJ, Hugenholtz P, Parks DH. GTDB-Tk v2: memory friendly classification with the genome taxonomy database. Bioinformatics. 2022;38:5315–6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 51. Paradis E, Schliep K. ape 5.0: an environment for modern phylogenetics and evolutionary analyses in R. Bioinformatics. 2019;35:526–8. [ DOI ] [ PubMed ] [ Google Scholar ] 52. Xu S, Li L, Luo X, Chen M, Tang W, Zhan L, et al. Ggtree: a serialized data object for visualization of a phylogenetic tree and annotation data. iMeta. 2022;1:e56. [ DOI ] [ PMC free article ] [ PubMed ] 53. Therneau TM. A Package for Survival Analysis in R. 2024. 54. Zhang Y, Chen X-X, Chen R, Li L, Ju Q, Qiu D, et al. Lower respiratory tract microbiome dysbiosis impairs clinical responses to immune checkpoint blockade in advanced non-small-cell lung cancer. Clin Translational Med. 2025;15:e70170. [ DOI ] [ PMC free article ] [ PubMed ] 55. Zhang J, Wu Y, Liu J, Yang Y, Li H, Wu X, et al. Differential oral microbial input determines two microbiota pneumo-types associated with health status. Adv Sci. 2022;9:2203115. [ DOI ] [ PMC free article ] [ PubMed ] 56. Guo J, Han J, Li F, Ma Q, He J, You F, et al. 16S rRNA sequencing reveals relationships among enrichment of oral microbiota in the lower respiratory tract and pulmonary nodules malignant progression. Microbiol Spectr. 2025;13:e01284–24. [ DOI ] [ PMC free article ] [ PubMed ] 57. Price LB, Hungate BA, Koch BJ, Davis GS, Liu CM. Colonizing opportunistic pathogens (COPs): the beasts in all of us. PLoS Pathog. 2017;13:e1006369. [ DOI ] [ PMC free article ] [ PubMed ] 58. Liu C-G, Lin M-X, Xin Y, Sun M, Cui J, Liu D, et al. Metagenomics and non-targeted metabolomics reveal the role of gut microbiota and its metabolites in brain metastasis of non-small cell lung cancer. Thorac Cancer. 2025;16:e70068. [ DOI ] [ PMC free article ] [ PubMed ] 59. Padhi S, Dash M, Sahu R, Panda P. Urinary Tract Infection due to Paenibacillus alvei in a Chronic Kidney Disease: A Rare Case Report. J Lab Physicians. 2013;5:133–5. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 60. Smith D, Bastug K, Burgoine K, Broach JR, Hammershaimb EA, Hehnly C, Morton SU, Osman M, Schiff SJ, Ericson JE. A Systematic Review of Human Paenibacillus Infections and Comparison of Adult and Pediatric Cases. Pediatr Infect Dis J. 2025;44:455–61. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 61. Kelly SD, Butler J, Green H, Jones AM, Kenna DTD, Pai S, et al. Genomic insights and phenotypic characterization of three multidrug resistant Cupriavidus strains from the cystic fibrosis lung. J Appl Microbiol. 2025;136:lxaf093. [ DOI ] [ PubMed ] 62. Qin S, Xiao W, Zhou C, Pu Q, Deng X, Lan L, et al. Pseudomonas aeruginosa: pathogenesis, virulence factors, antibiotic resistance, interaction with host, technology advances and emerging therapeutics. Signal Transduct Target Therapy. 2022;7:199. [ DOI ] [ PMC free article ] [ PubMed ] 63. Jin C, Lagoudas GK, Zhao C, Bullman S, Bhutkar A, Hu B, et al. Commensal microbiota promote lung cancer development via γδ T cells. Cell. 2019;176:998–1013.e16. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 64. Yan X, Yang M, Liu J, Gao R, Hu J, Li J, Zhang L, Shi Y, Guo H, Cheng J, et al. Discovery and validation of potential bacterial biomarkers for lung cancer. Am J Cancer Res. 2015;5:3111–22. [ PMC free article ] [ PubMed ] [ Google Scholar ] 65. Yuan X, Liu Y, Li G, Lan Z, Ma M, Li H, Kong J, Sun J, Hou G, Hou X, et al. Blockade of Immune-Checkpoint B7-H4 and Lysine Demethylase 5B in Esophageal Squamous Cell Carcinoma Confers Protective Immunity against P. gingivalis Infection. Cancer Immunol Res. 2019;7:1440–56. [ DOI ] [ PubMed ] [ Google Scholar ] 66. Liu Y, Yuan X, Chen K, Zhou F, Yang H, Yang H, et al. Clinical significance and prognostic value of Porphyromonas gingivalis infection in lung cancer. Translational Oncol. 2021;14:100972. [ DOI ] [ PMC free article ] [ PubMed ] 67. Velo-Suarez L, Moalic Y, Guilloux C-A, Ame J, Lamoureux C, Gouriou S, et al. Genomic diversity in Porphyromonas: evidence of Porphyromonas catoniae commensality in lungs. Microb Genomics. 2025;11:001411. [ DOI ] [ PMC free article ] [ PubMed ] 68. Keravec M, Mounier J, Guilloux C-A, Fangous M-S, Mondot S, Vallet S, et al. Porphyromonas, a potential predictive biomarker of Pseudomonas aeruginosa pulmonary infection in cystic fibrosis. BMJ Open Respiratory Res. 2019;6:e000374. [ DOI ] [ PMC free article ] [ PubMed ] 69. Beall CJ, Mokrzan EM, Griffen AL, Leys EJ. Cultivation of Peptidiphaga gingivicola from subgingival plaque: The first representative of a novel genus of Actinomycetaceae. Mol Oral Microbiol. 2018;33:105–10. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 70. Nibali L, Sousa V, Davrandi M, Spratt D, Alyahya Q, Dopico J, Donos N. Differences in the periodontal microbiome of successfully treated and persistent aggressive periodontitis. J Clin Periodontol. 2020;47:980–90. [ DOI ] [ PubMed ] [ Google Scholar ] 71. Khocht A, Orlich M, Paster B, Bellinger D, Lenoir L, Irani C, Fraser G. Cross-sectional comparisons of subgingival microbiome and gingival fluid inflammatory cytokines in periodontally healthy vegetarians versus non-vegetarians. J Periodontal Res. 2021;56:1079–90. [ DOI ] [ PubMed ] [ Google Scholar ] 72. Yan Z, Chen B, Yang Y, Yi X, Wei M, Ecklu-Mensah G, Buschmann MM, Liu H, Gao J, Liang W, et al. Multi-omics analyses of airway host–microbe interactions in chronic obstructive pulmonary disease identify potential therapeutic interventions. Nat Microbiol. 2022;7:1361–75. [ DOI ] [ PubMed ] [ Google Scholar ] 73. Yu W, Wang K, He Y, Shang Y, Hu X, Deng X, et al. The potential role of lung microbiota and lauroylcarnitine in T-cell activation associated with checkpoint inhibitor pneumonitis. eBioMedicine. 2024;106:105267. [ DOI ] [ PMC free article ] [ PubMed ] 74. Platten M, Nollen EAA, Röhrig UF, Fallarino F, Opitz CA. Tryptophan metabolism as a common therapeutic target in cancer, neurodegeneration and beyond. Nat Rev Drug Discovery. 2019;18:379–401. [ DOI ] [ PubMed ] [ Google Scholar ] 75. Wang G, Fan Y, Zhang G, Cai S, Ma Y, Yang L, et al. Microbiota-derived indoles alleviate intestinal inflammation and modulate microbiome by microbial cross-feeding. Microbiome. 2024;12:59. [ DOI ] [ PMC free article ] [ PubMed ] 76. Laursen MF, Sakanaka M, von Burg N, Mörbe U, Andersen D, Moll JM, Pekmez CT, Rivollier A, Michaelsen KF, Mølgaard C, et al. Bifidobacterium species associated with breastfeeding produce aromatic lactic acids in the infant gut. Nat Microbiol. 2021;6:1367–82. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 77. Hezaveh K, Shinde RS, Klötgen A, Halaby MJ, Lamorte S, Ciudad MT, et al. Tryptophan-derived microbial metabolites activate the aryl hydrocarbon receptor in tumor-associated macrophages to suppress anti-tumor immunity. Immunity. 2022;55:324–40.e8. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 78. Brasky TM, White E, Chen C-L. Long-Term, Supplemental, One-Carbon Metabolism–Related Vitamin B Use in Relation to Lung Cancer Risk in the Vitamins and Lifestyle (VITAL) Cohort. J Clin Oncol. 2017;35:3440–8. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 79. Zuo H, Ueland PM, Midttun Ø, Tell GS, Fanidi A, Zheng W, Shu X, Xiang Y, Wu J, Prentice R, et al. Vitamin B6 catabolism and lung cancer risk: results from the Lung Cancer Cohort Consortium (LC3). Ann Oncol. 2019;30:478–85. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 80. Colombo F, Falvella FS, De Cecco L, Tortoreto M, Pratesi G, Ciuffreda P, Ottria R, Santaniello E, Cicatiello L, Weisz A, Dragani TA. Pharmacogenomics and analogues of the antitumour agent N6-isopentenyladenosine. Int J Cancer. 2009;124:2179–85. [ DOI ] [ PubMed ] [ Google Scholar ] 81. Jinendiran S, Teng W, Dahms H-U, Liu W, Ponnusamy VK, Chiu CC-C, et al. Induction of mitochondria-mediated apoptosis and suppression of tumor growth in zebrafish xenograft model by cyclic dipeptides identified from Exiguobacterium acetylicum. Sci Rep. 2020;10:13721. [ DOI ] [ PMC free article ] [ PubMed ] 82. Kaur S, Mendonca P, Soliman KFA. The anticancer effects and therapeutic potential of kaempferol in triple-negative breast cancer. Nutrients. 2024;16:2392. [ DOI ] [ PMC free article ] [ PubMed ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Material 1. (21MB, docx) Supplementary Material 2. (409.2KB, xlsx) Data Availability Statement All raw sequencing data have been deposited in the Genome Sequence Archive for Human (GSA-Human) in National Genomics Data Center (accession number: HRA011114, https://ngdc.cncb.ac.cn/gsa-human/ ). Metabolomic data have been submitted to figshare ( https://figshare.com/s/8bdd4696a3d6b39433f6 ). Articles from Respiratory Research are provided here courtesy of BMC ACTIONS View on publisher site PDF (6.9 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top