Oral microbiota as a novel predictive biomarker of response to Immune Checkpoint Inhibitors in Non-Small Cell Lung Cancer patients

Background: Immunotherapy has emerged as the new frontier of anticancer treatment, yet robust biomarkers able to predict who will benefit from immunecheckpoint inhibitors (ICIs) are still lacking. Growing evidence shows that the salivary microbiota substantially overlaps with that of the lower airways, suggesting it might mirror the tumour micro-environment in thoracic malignancies. We therefore explored whether baseline oral microbial composition can act as a non-invasive predictor of ICI response in advanced non-small cell lung cancer (NSCLC). Methods: We prospectively enrolled 71 stage IV NSCLC patients scheduled for single-agent ICI (pembrolizumab, nivolumab, cemiplimab or atezolizumab) between February 2019 and May 2023. Pre-treatment saliva was subjected to 16S rRNA V1V3 sequencing. Diversity metrics were generated with MicrobiomeAnalyst; differential taxa were identified by LEfSe and validated with Mann-Whitney tests. Taxa were dichotomised by ROC-defined cut-offs and correlated with progressionfree survival (PFS) and overall survival (OS) using multivariable Cox models that included age, sex, ECOG status, histology, smoking and PD-L1. Median follow-up was 18 months. Results: After the exclusion of one low-depth library, 70 patients were evaluable (median age 70 years; 71% male; 77% adenocarcinoma; 56% with PD-L1 ≥50%). Durable clinical benefit (disease control ≥12 months) was achieved by 18 individuals (27%), whereas 52 (7%) progressed within the first year. Alpha-diversity measured with the Shannon index was comparable between the two outcome groups (all p >0.20), yet Bray-Curtis beta-diversity separated them at phylum level (PERMANOVA pseudo-F = 2.31, p = 0.028), a difference driven almost entirely by Actinobacteria. LEfSe confirmed a coherent Actinobacteria signal spanning phylum, class, order, family and genus (Actinomyces), each over-represented in nonbenefitting patients (LDA >3.5, FDR ≤0.049), while no other phylum retained significance after multiple-testing correction. Receiver-operating-characteristic analysis yielded an optimal cut-off of 16% relative abundance for phylum-level 1 Actinobacteria (AUC 0.79, 95% CI 0.68–0.89; sensitivity 78%, specificity 73%); analogous thresholds were 16% for class Actinobacteria, 13% for order Actinomycetales and 10% for both family Actinomycetaceae and genus Actinomyces. Patients exceeding the 16% phylum cut-off experienced markedly worse outcomes: median progression-free survival was 3.0 months (95% CI 2.8–4.2) versus 8.1 months (95% CI 6.4–12.5), and overall survival 6.2 months (95% CI 4.5–9.5) versus 15.4 months (95% CI 11.2–22.7); both log-rank p <0.001. In multivariable Cox models adjusting for age, sex, ECOG status, histology, smoking and PD-L1, high Actinobacteria remained independently adverse for progression (HR 1.92, 95% CI 1.18–3.12, p = 0.009) and death (HR 2.08, 95% CI 1.26–3.45, p = 0.004. When treated as a continuous variable, each 5-percentage-point rise in Actinobacteria conveyed a 12% higher risk of progression (HR 1.12, 95% CI 1.05–1.20) and a 14% higher risk of death (HR 1.14, 95% CI 1.06–1.22), confirming a dose-response relationship. Taken together, these findings identify Actinobacteria - particularly the Actinomyces lineage - as a robust, dose-dependent and clinically independent predictor of unfavourable ICI outcome. Discussion: Our findings strengthen the idea that the oral microbiota is not a passive reflection of health status but an active participant in systemic antitumour immunity. The association between Actinobacteria abundance and resistance to immunecheckpoint inhibition emerges consistently across successive taxonomic ranks - from phylum down to the genus Actinomyces - suggesting a lineage-specific biological effect rather than a random fluctuation of individual taxa. Moreover, survival curves display a clear dose-response pattern: progression-free survival and overall survival decrease stepwise as the proportion of Actinobacteria rises beyond approximately 10%, 13% and 16%, implying that quantitative increases in this lineage progressively erode the efficacy of PD-L1 blockade. Mechanistic evidence from colorectal transcriptomic studies supports a causal role. Actinomyces species appear to activate TLR2/NF-κB signaling, potentially promoting the release of interleukin-8 and interleukin-10, the recruitment of myeloid-derived suppressor cells, and a reduction in CD8⁺ T cell infiltration, features consistent with microenvironment. an immune checkpoint inhibitor (ICI)-refractory tumor environment. Conclusions: In summary, this prospective study demonstrates that a high baseline abundance of the Actinobacteria lineage - particularly the genus Actinomyces - in the oral microbiota independently predicts primary resistance and inferior survival in stage IV NSCLC patients receiving PD-(L)1 blockade. These data position salivary microbial profiling as a rapid, non-invasive tool that could refine patient selection beyond PD-L1 and tumour-mutational burden, and they highlight the oral cavity as a modifiable micro-environment whose manipulation may enhance immunotherapy efficacy. Future multicentre cohorts, longitudinal sampling and mechanistic models are now warranted to validate the signature, unravel causal pathways and test microbiota-directed interventions aimed at converting “microbiota-high-risk” patients into durable responders.