Salivary microbial signature highlighting actinomyces as a predictor of immune-checkpoint inhibitor monotherapy response in advanced non–small cell lung cancer

Abstract Background Immune checkpoint inhibitors (ICIs) have improved survival in advanced non-small cell lung cancer (NSCLC), yet reliable biomarkers beyond programmed death-ligand 1 (PD-L1) expression remain limited. Increasing evidence links the gut microbiome to ICI activity, but the predictive value of the salivary microbiome is poorly defined. Methods We prospectively analyzed baseline saliva from 71 stage IV NSCLC patients treated with anti–PD-1/PD-L1 (ICI) monotherapy. After quality control, 70 samples underwent 16 S rRNA gene sequencing of the V1–V3 region. Microbial diversity, differential abundance (LEfSe, Mann-Whitney/Kruskal-Wallis with false discovery rate correction) and survival associations (Kaplan-Meier; Cox proportional-hazards with LASSO-based variable selection and 1000-fold bootstrap validation) were examined. In this cohort, an exploratory genus-level cut-off was derived by receiver operating characteristic (ROC) analysis. Results α-diversity and β-diversity did not differ between responders (progression-free survival (PFS) ≥ 12 months; n = 18) and non-responders ( n = 52). Differential‑abundance profiling revealed a graded enrichment of the phylum Actinobacteria across all lower ranks, class Actinobacteria, order Actinomycetales, family Actinomycetaceae and genus Actinomyces ,in non‑responders (LEfSe LDA > 3.5; p = 0.001 for each level; FDR ≤ 0.049). ROC analysis suggested an Actinomyces relative abundance of 11% (AUC = 0.768; sensitivity 0.94; specificity 0.44) as a data-driven threshold, classifying patients into low (≤ 11%, n = 46) and high (> 11%, n = 24) groups. High abundance was associated with shorter PFS (median 3 vs. 4 months; HR = 2.16, 95% CI 1.21–3.88, p = 0.009) and overall survival (OS) (median 5 vs. 9 months; HR = 2.61, 95% CI 1.48–4.61, p < 0.001) after multivariable adjustment for ECOG status, treatment line, corticosteroid and opioid use, smoking, histology and metastatic sites. Bootstrap validation supported model stability, with median bootstrap HRs of 2.56 (PFS) and 2.63 (OS), with narrow percentile CIs (PFS 1.57–4.49; OS 1.40–6.34) overlapping the original estimates. Conclusions In this exploratory cohort, salivary microbiome signature characterized by high Actinomyces abundance was independently associated with poorer ICI outcomes in NSCLC. Saliva profiling is non-invasive and, if validated in larger and independent cohorts, may complement tumour PD-L1 and clinical factors to refine patient stratification.