Abstract
BACKGROUND: Esophageal cancer (EC) has a high incidence and is highly invasive. It is meaningful to employ invasion-related genes (IRGs) to predict patients' prognosis. METHODS: We launched a weighted correlation network analysis to screen for EC tumor differentially expressed IRGs (module genes) from the TCGA-ESCA dataset. By executing univariate-LASSO-multivariate Cox regression analyses, we stepwise selected module genes to obtain prognostic feature genes and create a model. We validated the model using the external GEO dataset GSE53624. We assessed the model's independent prognosis prediction ability. A nomogram was plotted and further validated later. GSEA was undertaken on high-risk groups (Group H). We compared immunity and tumor mutations in Group H and the low-risk group (Group L) and made small molecular drug predictions on prognostic genes. RESULTS: A risk prognostic model consisting of 10 genes (ARMCX2, RGS16, APLN, TRIM28, AKAP4, ZC3H12B, MAD1L1, TWIST1, TMTC2, and TADA2B) was created. ZC3H12B was significantly linked with TADA 2B, AKAP4, and TWIST1 (P < 0.01). The model exhibited a good predictive performance, functioning as an independent prognostic factor. The predictive accuracy of the nomogram was relatively high. Pathways that were significantly enriched in Group H included base excision repair, cysteine and methionine metabolism, and porphyrin and chlorophyll metabolism (P < 0.05). Compared with Group L, Group H had higher expression of relevant immune genes and a higher degree of tumor mutation (P < 0.05). ZC3H12B was significantly linked with immune cells (macrophages and iDCs), showing a high degree of mutation. The IC(50) values of Lomustine, Dexrazoxane, Batracylin, and Buthioninesulphoximine were significantly positively linked with the expression of ZC3H12B (P < 0.01). CONCLUSION: The 10-gene prognostic model can independently predict patients' prognosis. The great correlation between ZC3H12B and multiple feature genes and immune cells may be tightly linked to EC progression.