Identifying Actionable Microbial Biomarkers for Cancer Therapy Using Big Data Approaches

Nearly 100 trillion of microbes cohabit the human body. They modulate digestive, nerve, and immune systems to help maintain health. However, their dysbiosis were also linked to many diseases, e.g., obesity, autism, allergies, and cancers. Recently, gut microbiota was found predictive of immune checkpoint inhibitor’s efficacy, which achieved durable response in many advanced cancers, however, with a low response rate. It is thus significant if one can modulate the gut flora to improve ICI response, which was successful in pre-clinical models. To truly realize this potential, we are working on integrating big multiomics data sources and developing machine learning approaches for discovering and validating actionable microbial biomarkers for cancer therapies. Our approaches combine state-of-art genome technologies and data sciences and include: large-scale data mining of microbial and immune features in public data space; building a cancer-microbiome data hub; and developing machine learning models to identify markers predictive of immune infiltration and patient outcomes, among others. We expect our effort to provide a comprehensive set of cross-platform and open-source bioinformatics and statistical tools for cancer metagenomics and cancer genomics data analysis and to initiate a new paradigm for future cancer genomics by the integrative analysis of human and microbiome data.

Summary of Results

Dysbiosis, an imbalance of an organ’s microbiota, is associated with colorectal cancer pathogenesis. Diverse microbial organisms are associated with the cells found in tumor biopsies. Characterizing this mucosa-associated microbiome through genome sequencing has advantages compared to culture-based profiling. However, there are notable technical and analytical challenges in characterizing universal features of tumor microbiomes. Colorectal tumors demonstrate a significant extent of microbiome variation among individuals from different geographic and ethnic origins. To address these issues, we identified a consensus microbiome for colorectal cancer through analyzing 924 tumors from eight independent RNA-Seq data sets. A standardized meta-transcriptomic analysis pipeline was established with quality control metrics. Microbiome profiles across different CRC cohorts were compared and recurrently altered microbial shifts specific to CRC were determined. We identified cancer-specific set of 114 microbial species associated with tumors that were found among all investigated studies. Firmicutes, Bacteroidetes, Proteobacteria, and Actinobacteria were among the four most abundant phyla for CRC microbiome. Member species of Clostridia were depleted and Fusobacterium nucleatum was one of the most enriched bacterial species in CRC. Associations between the consensus species and specific immune cell types were noted indicating the potential role of these microbes in altering the immune tumor microenvironment.

Data Resources

To promulgate our findings to the wider biomedical research community, we developed a data resource website that releases our genomic results. This interface provides access to the consensus microbiome features derived from over 900 colorectal cancers. Our results are freely available as a web data resource for other researchers to explore:

https://crc-microbiome.stanford.edu