Multi-omics Identification and Route-Specific Characterization of Metastasis-specific EMT Genes and Their Microenvironmental Interactions

Background Gastric cancer (GC) constitute a significant cause of cancer-related mortality worldwide, with metastatic patterns including hematogenous, peritoneal, and ovarian routes. Although GC gene expression patterns have been extensively researched, the metastasis-specific gene expression landscape remains largely unexplored. Methods We undertook a whole transcriptome sequencing analysis of 66 paired primary and metastatic (hematogenous, peritoneal, or ovarian) GC tumors from 14 patients. Public databases including The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) was used for validation. Single-cell RNA sequencing (scRNA-seq) of four ascites from serosa positive GC patients and five primary tumors by layer (superficial and deep) were analyzed. Results Through differential expression analysis between paired primary and metastatic tumors by routes identified 122 unique metastasis-specific epithelial-mesenchymal transition (msEMT) genes. These genes demonstrated varying expression patterns depending on the metastatic route, suggesting route-specific molecular mechanisms in GC metastasis. High expression of msEMT genes in primary tumors was associated with more frequent CDH1 mutations, the genomically stable subtype, and poor prognosis in TCGA GC cohort. This association was further corroborated by poor prognosis and high predictive performance for peritoneal/ovarian recurrence in two independent cohorts (GSE66229; n=300, GSE84437; n=433). scRNA-seq analysis of five primary tumors (GSE167297) and four independent ascites samples from GC patients revealed that msEMT genes were predominantly expressed in diverse fibroblast sub-populations, rather than cancer cells. Conclusions This study illuminates the route-specific mechanisms and underlines the significance of msEMT genes and cancer-associated fibroblasts in peritoneal metastasis of GC. ### Competing Interest Statement The authors have declared no competing interest. * GC : gastric cancer TCGA : the cancer genome atlas MSI : microsatellite instability EBV : Epstein-Barr virus CIN : chromosomal instability GS : genomic stable msEMT : metastasis-specific epithelial-mesenchymal transition scRNA : single-cell RNA FFPE : formalin-fixed paraffin-embedded TNM : Tumor Node Metastasis WTS : whole transcriptome sequencing NMF : Non-negative Matrix Factorization HR : hazard ratio CI : confidence interval EN : Elastic Net regression GBM : Gradient Boosting Machine RF : Random Forests SVM : Support Vector Machine PCANN : Probabilistic Classifier Artificial Neural Networks DEA : Differential Expression Analysis CAF : Cancer Associated Fibroblast myo-CAF : myofibroblastic CAF i-CAF : inflammatory CAF v-CAF : vascular CAF