Transcriptome profiling of patient-derived tumor xenografts suggests novel extracellular matrix-related signatures for gastric cancer prognosis prediction

Background: A major obstacle to the development of personalized therapies for gastric cancer (GC) is the prevalent heterogeneity at the intra-tumor, intra-patient, and inter-patient levels. Although the pathological stage and histological subtype diagnosis can approximately predict prognosis, GC heterogeneity is rarely considered. The extracellular matrix (ECM), a major component of the tumor microenvironment (TME), extensively interacts with tumor and immune cells, providing a possible proxy to investigate GC heterogeneity. However, ECM consists of numerous protein components, and there are no suitable models to screen ECM-related genes contributing to tumor growth and prognosis. We constructed patient-derived tumor xenograft (PDTX) models to obtain robust ECM-related transcriptomic signatures to improve GC prognosis prediction and therapy design.

Methods: One hundred twenty two primary GC tumor tissues were collected to construct PDTX models. The tumorigenesis rate and its relationship with GC prognosis were investigated. Transcriptome profiling was performed for PDTX-originating tumors, and least absolute shrinkage and selection operator (LASSO) Cox regression analysis was applied to extract prognostic ECM signatures and establish PDTX tumorigenicity-related gene (PTG) scores. The predictive ability of the PTG score was validated using two independent cohorts. Finally, we combined PTG score, age, and pathological stage information to establish a robust nomogram for GC prognosis prediction.

Results: We found that PDTX tumorigenicity indicated a poor prognosis in patients with GC, even at the same pathological stage. Transcriptome profiling of PDTX-originating GC tissues and corresponding normal controls identified 383 differentially expressed genes, with enrichment of ECM-related genes. A robust prognosis prediction model using the PTG score showed robust performance in two validation cohorts. A high PTG score was associated with elevated M2 polarized macrophage and cancer-associated fibroblast infiltration. Finally, combining the PTG score with age and TNM stage resulted in a more effective prognostic model than age or TNM stage alone.

Conclusions: We found that ECM-related signatures may contribute to PDTX tumorigenesis and indicate a poor prognosis in GC. A feasible survival prediction model was built based on the PTG score, which was associated with immune cell infiltration. Together with patient ages and pathological TNM stages, PTG score could be a new approach for GC prognosis prediction.