Abstract 1184: Early-stage lung cancer detection via circulating microbial DNA biomarkers and machine learning classification

Abstract

Abstract Introduction: Human tissues, including tumors, are extensively colonized by taxonomically diverse microbes. Intra-tumoral microbial activity and events of cellular turnover and trafficking contribute to shedding of microbial nucleic acids into the blood stream. Here we characterized microbial signatures (mbDNA) present in primary-tumor tissue and in the blood of patients affected with different cancer types, with particular focus on lung cancer, and we demonstrated the discriminatory power of such microbial signatures for the identification and classification of lung cancer versus other cancer types. We further validated our findings using plasma-derived cell-free microbial DNA (cf-mbDNA) to discriminate between lung cancer and cancer-free control samples. Methods: We examined The Cancer Genome Atlas (TCGA) compendium of treatment-naïve, whole genome and transcriptomic sequencing datasets to extrapolate genetic signatures of microbial origin associated with 33 different tumor types collected from 10,481 patients, which included non-neoplastic tumor-adjacent tissue and blood samples. 7.2% of TCGA sequencing reads were classified as non-human, of which 35.2% could be taxonomically classified using a reference database containing 59,974 total microbial genomes. These taxonomically assigned data sets were then used to train machine learning models (using a 70/30 train/test split for all cancers) to discriminate between and within types and stages of cancer. Results: We demonstrated that mbDNA signatures from whole blood can be used to accurately classify the tissue of origin of 20 unique cancer types, including lung adenocarcinoma and lung squamous cell carcinoma. For lung adenocarcinoma we reported high discrimination between paired tumor tissue and normal-adjacent tissue (Avg. {AUROC,AUPR}={0.85,0.95}) and between primary tumor tissue and all-other cancer types (Avg. {AUROC,AUPR}={0.96,0.69}, n=32 cancer-types). We also demonstrated the high performance of blood-derived mbDNA when discriminating among TCGA cancer types: Avg. {AUROC,AUPR}={0.97,0.80}. Subsequent liquid biopsy results using plasma-derived cf-mbDNA offer compelling evidence that cf-mbDNA signatures can robustly discriminate adenocarcinoma lung-cancer samples from non-cancer controls. Conclusion: mbDNA holds considerable promise as a truly orthogonal means of detecting and classifying lung cancer independently from host genomic alternations. Using only mbDNA signatures we have demonstrated robust discrimination between cancer-free controls and lung cancer samples and have provided early evidence of the applicability of this approach to liquid biopsy. Our present efforts analyzing plasma cf-mbDNA with an expanded sample cohort will serve to fully validate this new class of liquid biopsy biomarkers for lung cancer detection. Citation Format: Gregory D. Sepich-Poore, Serena Fraraccio, Stephen Wandro, Rob Knight, Sandrine Miller-Montgomery, Eddie Adams. Early-stage lung cancer detection via circulating microbial DNA biomarkers and machine learning classification [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1184.