揭示肺癌免疫治疗反应的肿瘤微生物和免疫生物标志物

Background: Although immune checkpoint inhibitors (ICIs) have shown promise, fewer than half of advanced Non-Small Cell Lung Cancer (NSCLC) patients respond, and biomarkers such as PD-L1 and tumor mutational burden poorly predict response. The intratumoral microbiota directly interacts with cancer and immune cells, yet its relationship with immune subsets in ICI-treated NSCLC is unclear. We hypothesize that integrating tumor microbiota with immune features could improve prediction of response and patient stratification. Methods: We analyzed total RNA-Seq data from 120 stage IV NSCLC patients whose tumor samples were collected prior to ICI therapy. Patients were classified as 30 responders or 82 non-responders,and progression-free survival was evaluated. A dual RNA-Seq approach profiled both host and microbiota: human reads were removed (Bowtie2), non-human reads classified (Kraken2). Microbial alpha-diversity (Chao1, Shannon), beta-diversity (Bray-Curtis). Host immune profiling included differential expression (DESeq2) and immune infiltration (CIBERSORT), with group differences tested by t-test. Results: We detected 273 microbial species. alpha- and beta-diversity did not differ between responders and non-responders. Differences between responders and non-responders were mainly driven by Gram-negative taxa. Streptomyces was the only Gram-positive genus enriched in responders. Immune profiling showed that responders had higher expression of antigen-presentation genes, greater CD8⁺ T-cell cytotoxicity, and upregulation of PD-1 and CTLA-4, whereas non-responders had impaired antigen presentation, reduced effector T cell activity, and a neutrophil-dominant suppressive axis with increased S100A8 and ARG1; GSVA indicated enrichment of neutrophil-mediated Gram-negative killing pathways. Conclusions: As microbial signals in low-biomass lung tumor RNA-seq are susceptible to contamination, we removed taxa commonly associated with environmental sources. Notably, Streptomyces-a genus producing several anticancer natural products-remained more frequent in responders, suggesting a potential link to favorable outcomes. In addition, our results suggest that responders exhibit antigen presentation and activated effector T cells, yet also display high expression of PD-1 and CTLA-4, placing them in an inflamed but suppressed state that becomes responsive upon checkpoint inhibition. Conversely, non-responders are dominated by innate immune programs with suppressed adaptive immunity, and therefore fail to mount an effective antitumor response even when treated with immune checkpoint inhibitors. Machine learning approaches integrating immune-microbiome crosstalk are ongoing. If validated, these intratumoral microbial and immunological predictors could optimize patient selection for immunotherapy and support precision treatment strategies in NSCLC.