Systematic evaluation of TCGA tumor microbiota reveals context-dependent reliability

Microbial profiles from The Cancer Genome Atlas (TCGA) are widely used to study the tumor microbiota, a key component of the cancer ecosystem, yet their reliability remains unclear. Here, we systematically benchmarked two leading TCGA microbial profiles (TMPs) to define their consistency, accuracy, and reliability in host-microbe association studies across 24 cancer types, with a primary focus on the bacterial component. We found that while the TMPs showed substantial agreement in microbial composition, their accuracy in detecting known oncomicrobes was variable, ranging from excellent for human papillomavirus (HPV) to poor for Helicobacter pylori. The concordance of downstream host-microbe associations was moderate for gene expression but nearly absent for methylation and protein data. Our permutation-based framework revealed that while most individual associations were statistically reliable, those involving cell type composition and patient survival were statistically spurious. To empower future research with these insights, we introduced Multi-Omics and Microbiome Associations in Cancer 2 (MOMAC2), an interactive web portal that stratifies all associations by confidence level. We demonstrated its utility by using high-confidence associations to confirm HPV-driven methylation-gene expression axes and guide a novel experimental investigation. Co-culture with Streptococcus anginosus not only validated its predicted gene expression changes in oral cancer cells but also revealed a significant promotion of cancer cell proliferation and migration. Our study provides a rigorous framework for interpreting TCGA's tumor microbiome and highlights that these data require careful, multi-layered validation to yield robust biological insights.

Bacteria living inside tumors can influence how cancer grows and responds to treatment, but the field has been hampered by controversy over the reliability of the data. Our study provides a much-needed road map for researchers. We rigorously tested the massive Cancer Genome Atlas data set and developed a statistical framework to separate true biological signals from random noise. We discovered that many widely reported links are statistically unreliable and likely false leads. Importantly, our framework successfully pinpoints trustworthy signals. We used it to identify a specific bacterium, Streptococcus anginosus, and proved in the lab that it makes oral cancer cells grow faster and spread. Our publicly available Multi-Omics and Microbiome Associations in Cancer 2 (MOMAC2) web portal now allows scientists to use these reliability-graded findings to accelerate robust cancer microbiome research.