Discovering Microbial Features for Oral Cancer Diagnosis and Prediction

Professor Huang, Shi   (Principal Investigator (PI))

Professor Khoo Ui Soon   (Co-Investigator)

Dr Liu Pei   (Co-Investigator)

Professor Su Yuxiong   (Co-Investigator)

Dr Koohi-Moghadam Hongzhe   (Co-Investigator)

12

2022-05-10

150000

Discovering Microbial Features for Oral Cancer Diagnosis and Prediction

Diagnosis, FFPE, Microbiome, Oral cancer, Prediction

Molecular BiologyCancer

202107185075

Seed Fund for Basic Research for New Staff

2021

Completed

Introduction Oral cancer (OC) is the 18th most common malignancy worldwide and accounts for many head and neck cancers in contemporary clinical practice. Over 90% of OCs are oral squamous cell carcinoma (OSCC). Oral carcinogenesis may be associated with a lengthy pre-pathologic phase, which features the occurrence of diseases with increased risk of malignancy, known as oral potentially malignant disorders (OPMD). Appropriate recognition and management of OPMDs are essential to ensure early recognition of malignancy, delivery of effective treatment with reduced morbidity, and, ultimately, to improve long-term prognosis and survival for OC patients. Malignant transformation potential (MTP) of OPMDs, unfortunately, varies substantially between 0.13 and 85%. The accurate prediction of transformation risk for such lesions on an individual basis remains challenging(1). In addition to genetics, microbiome also contribute to the development of OC(2, 3), and many cancer types (such as breast cancer, lung cancer, etc.)(4-6). Indeed, recent studies have demonstrated that OC shows substantial microbial contributions from oral microbiome (7, 8). Although it is still unclear if microbiome dysbiosis can cause OC, recent studies showed that bacterial compositions within tumor tissues are different from those normal healthy tissues from controls(9). This promoted us to think about analysing microbiome in formalin-fixed, paraffin-embedded (FFPE) tissue samples, the gold standard of preserving tumor biopsy specimens, for predicting the MTP of OPMDs (5). Yet, tissue-associated microbiome has not been well considered in the OC prediction studies due to the lack of (1) well-designed longitudinal cohorts; and (2) cost-effective and high-throughput sequencing methods. Regarding cohort design, previous studies all focus on post-diagnosis cross-sectional data that have very small sample size. Such post-diagnosis cross-sectional data only provide the feasibility of a diagnostic model for OC based on tissue-associated microbiome, rather than an early detection of OC. We still lack a large longitudinal cohort study to validate the tissue-associated microbiome-based early detection of OC. Regarding sequencing methods, conventional microbiome profiling based on marker-gene sequencing or whole-metagenome sequencing (WMS) all have their strength and weakness (Table 1). None of them offers the perfect solution for tissue microbiome profiling. Indeed, it is either impossible or prohibitively expensive for those conventional methods to handle DNA samples that are in a trace amount, heavily degraded, or dominated by host DNA, e.g., in human tissue or blood(11, 12). Here, we emphasize that the microbial load in human tissue is extremely low. In fact, 95.15%~99.92% reads in WMS were found to be originated from human(13). We know that a sequencing depth of 1~2 million reads is typically encouraged in WMS(14). If 99% reads are from human, then the sequencing depth should be 100~200 million reads, which is way too expensive for WMS. This explains why in a previous study(13) the authors performed a very deep sequencing (~400 million reads/sample) for only 5 samples. We illustrated that a reduced metagenome sequencing -- type IIB Restriction site-Associated DNA sequencing for microbiome: 2bRAD-M (i.e., STREAM here; Figure 1) can quantify bacteria, archaea, and fungi from extremely low-biomass samples (e.g., 1pg total DNA, 50bp highly degraded DNA, or 99% host contaminated DNA; Figure 2) in a cost-efficient way(10). We further detected cervical cancer markers in the FFPE microbiome and constructed a species-resolved classifier that discriminated the healthy tissue, pre-invasive cancer, and invasive cancer with high accuracy (AUROC: 0.96, Figure 3). Other than cervical tissue samples, this method needs validations in more types of FFPE samples. Our central hypothesis is that the presence of microorganisms or microbial DNA in the FFPE samples is related to the risk of OC and can be used as a biomarker for diagnosis or even early detection of OC. Objectives 1. Adopt a new sequencing pipeline to profile oral FFPE tissue microbiomes. We plan to apply a 2bRAD sequencing pipeline on analyzing oral mucosal FFPE tissue microbiome, while it was previously validated in the FFPE samples associated with cervical cancer (Figure 3). We will rename this method as STREAM (Species/strain-resolved Type-IIB REstriction site-Associated Metagenomic sequencing). 2. Develop a machine-learning model to predict OC based on FFPE microbiome profiles. We will formalize the classification problem as a supervised classification problem using Random Forests. The strict 10-fold cross-validation approaches will be applied to examine if the OC-associated microbiome confounds with other host factors. Then we will apply it to the species-level taxonomic profiles of FFPE samples from this cohort to (1) distinguish OC from OPMD; (2) predict the OC development. (3) To improve the predictive model of OC, we will include microbial features from FFPE microbiome assays with EHR features(1) in the modeling. Key references: 1. Adeoye J, et al. Deep Learning Predicts the Malignant-Transformation-Free Survival of Oral Potentially Malignant Disorders. Cancers (Basel). 2021;13(23). 2. Irfan M, et al. The Oral Microbiome and Cancer. Front Immunol. 2020;11:591088. 3. Teles FRF, et al. Association or Causation? Exploring the Oral Microbiome and Cancer Links. J Dent Res. 2020;99(13):1411-24. 4. Nejman D, et al. The human tumor microbiome is composed of tumor type-specific intracellular bacteria. Science. 2020;368(6494):973-80. 5. Poore GD, et al. Microbiome analyses of blood and tissues suggest cancer diagnostic approach. Nature. 2020;579(7800):567-74. 6. Peneder P, et al. Multimodal analysis of cell-free DNA whole-genome sequencing for pediatric cancers with low mutational burden. Nat Commun. 2021;12(1):3230. 7. Zhao H, et al. Variations in oral microbiota associated with oral cancer. Sci Rep. 2017;7(1):11773. 8. Zhang L, et al. The Oral Microbiota May Have Influence on Oral Cancer. Front Cell Infect Microbiol. 2019;9:476. 9. Gopinath D, et al. Differences in the bacteriome of swab, saliva, and tissue biopsies in oral cancer. Sci Rep. 2021;11(1):1181. 10. Sun Z, et al. Species-resolved sequencing of low-biomass or degraded microbiomes using 2bRAD-M. Genome Biol. 2022;23(1):36. 11. Quince C, et al. Shotgun metagenomics, from sampling to analysis. Nature Biotechnology. 2017;35(12):1211-. 12. Knight R, et al. Best practices for analysing microbiomes. Nat Rev Microbiol. 2018;16(7):410-22. 13. Huang YF, et al. Analysis of microbial sequences in plasma cell-free DNA for early-onset breast cancer patients and healthy females. BMC Med Genomics. 2018;11(Suppl 1):16. 14. Hillmann B, et al. Evaluating the Information Content of Shallow Shotgun Metagenomics. mSystems. 2018;3(6).