Tumor microenvironment interplay amid microbial community, host gene expression and pathological features elucidates cancer heterogeneity and prognosis risk

Microbial community, host gene expression, and pathological image in the tumor microenvironment (TME) are interrelated, and they can provide valuable insights into tumor heterogeneity. However, the complex and important interconnection patterns among these multimodal omics in pan-cancer remain far from being elucidated. Here, we developed a framework for decoding the association among the microbial community, host gene expression, and pathological image in the TME for a broad spectrum of cancers in multiple independent cohorts. Our results indicated three types of microbe-expression-image triplet (MEI-triplet) interplay patterns in human cancers. MEI1, including esophageal cancer (ESCA) and pancreatic adenocarcinoma (PAAD), was characterized by no significant intratumoral microbe-host interaction. MEI2, which included eight cancers (liver hepatocellular carcinoma [LIHC] is one of these), was characterized by a strong association of tumor microbiota and host gene expression, but changes in the TME were not reflected in pathological images. Meanwhile, MEI3, including colorectal cancer (CRC), stomach adenocarcinoma (STAD), and lung squamous cell carcinoma (LUSC), was characterized by strong associations among tumor microbiota, host gene expression, and pathological images. Further experiments with validation datasets for several representative cancer types also supported these patterns. Out of the three MEI patterns, patients with MEI1 had the worst prognosis risk, followed by those with MEI3 and MEI2. Our findings provide new insights into tumor heterogeneity, which could assist with providing more precise diagnosis and prognosis assessment in clinics.