Intratumoral Microbiome Predicts Liver Metastasis and Informs Effective Treatment of Gastrointestinal Stromal Tumor

Intratumoral microbiome can promote metastasis, but whether it can accurately predict metastasis remains unexplored. Here, we evaluated the predict efficacy of the intratumor microbiome for liver metastasis (LM) and the treatment efficacy of imatinib against LM of gastrointestinal stromal tumor (GIST). We enrolled two independent cohorts consisted of 108 newly-diagnosed GIST patients who underwent surgical resection and assessed the intratumoral microbiome using 16SrRNA amplicon sequencing. Microbiome-based machine learning classifiers were trained and validated for the postoperative LM outcomes of the patients in our cohorts, and tested them on public datasets. Our microbiome-based classifiers accurately predicted postoperative LM, achieving area under the curves (AUCs) of 0.953 and 0.897 on the discovery and validation cohorts, respectively. A classifier trained on non-treated patients could also predict LM in imatinib-treated patients, suggesting that the intratumoral microbiome determined the treatment efficacy of imatinib. In addition, we also developed classifiers for the risk stratification, which significantly correlated with LM. We further tested the risk stratification in an external cohort, achieving an AUC of 1.0. We identified microbes as potential LM and risk biomarkers. The intratumoral microbiome could accurately predict postoperative LM outcomes of GIST patients. Our microbiome-based machine learning classifiers have potentials in clinical applications.