Abstract
BACKGROUND: Existing research suggested that the intratumoral microbiome participates in the progression and metastasis of lung cancer. However, its role in distinguishing benign from early-stage malignant pulmonary tumors remains poorly characterized. METHODS: In a single-center retrospective cohort of surgically resected nodules (2021–2023), specimens underwent five-region (V2, V3, V5, V6, and V8) 16S rRNA sequencing. After 2:1 propensity-score matching and sequencing QC, 98 FFPE specimens were analyzed (66 malignant; 32 benign). Diversity and community structure were compared using Chao1/Shannon indices, Bray - Curtis PCoA, and PERMANOVA (999 permutations). An eight-genus logistic-regression classifier was internally validated by 1,000-iteration bootstrap and five-fold stratified cross-validation. RESULTS: Malignant lesions exhibited greater microbial richness and evenness than benign lesions (Chao1, p = 0.007; Shannon, p = 0.029) and a distinct overall community structure (PERMANOVA R(2) = 0.020, p = 0.021). Lipopolysaccharide immunostaining indicated a higher bacterial signal in malignant tissue (median H-score 3.3 vs 1.5; p < 0.001). We developed an eight-genus diagnostic model, with five genera (Myroides, Knoellia, Pelomonas, Peptostreptococcus, Porphyrobacter) enriched in the malignant group. The model demonstrated stable discrimination on internal validation, achieving a bootstrap median AUC of 0.769 (95% CI 0.574–0.903) with sensitivity 0.864 and a five-fold cross-validation AUC of 0.783 (95% CI 0.661–0.905); when trained on the full dataset, the AUC was 0.829 (95% CI 0.744–0.913). Among malignant cases, higher model-derived risk scores showed a non-significant trend toward lymph node metastasis (p = 0.075), and DFS did not differ across risk strata (5/66 events; p = 0.474). PICRUSt2-based functional prediction suggested enrichment of pathways related to focal adhesion, calcium signaling, O-glycan biosynthesis, and immune-associated processes in the malignant group. CONCLUSIONS: The intratumoral microbiome is distinctly altered in early-stage malignant pulmonary tumor. A microbiota-based diagnostic model demonstrates good accuracy for discriminating malignant from benign pulmonary lesions, highlighting its potential as a novel diagnostic biomarker. Prospective multicenter validation and assessment in minimally invasive specimens are warranted. GRAPHICAL ABSTRACT: [Image: see text] SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12967-026-08078-1.