Background: Cholangiocarcinoma (CCA) is a rare malignant carcinoma characterized by high mortality, challenging diagnosis, and poor prognosis. A powerful prediction biomarker is urgently needed for the early diagnosis and individualized treatment of CCA patients.
Methods: A systematic bioinformatics analysis was conducted based on mRNA expression data and clinical information from The Cancer Genome Atlas (TCGA), Gene Expression Omnibus (GEO) and National Genomics Data Center (NGDC) datasets. Differentially expressed genes (DEGs) between tumor tissues and adjacent counterpart controls were identified in the TCGA and GSE107943 datasets. A nine-gene prediction model was constructed, and its effects on CCA prognosis were analyzed using univariate, multivariate and LASSO Cox proportional hazards regression models, Kaplan-Meier plotter, CIBERSORT and OncoPredict in the discovery and validation cohorts. Additionally, the expression profiles of the target genes were determined via qRT-PCR and DEG analyses in an independent cohort.
Results: A nine-gene signature (HELLS, HOXC6, MFSD2A, OTX1, PTGES, PYGB, SMOC1, TEX30 and ZBTB12) displayed excellent predictive performance for the overall survival of CCA. According to the prognostic signature, CCA patients were classified into high-risk and low-risk groups, with obvious differences in overall survival probabilities. The low-risk group had a significantly better prognosis than the high-risk group in both the discovery cohort (n = 66) and replication cohort (n = 255) (P < 0.0001, P < 0.0001). Additionally, five cancer drugs (Erlotinib, ML323, AGI-6780, Gallibiscoquinazole and AZD3759) presented clearly specific sensitivities for high-risk and low-risk group patients. Moreover, according to tumor microenvironment analyses, high-risk group patients had a higher level of M0 macrophage infiltration than low-risk group patients (P = 0.025, P = 0.0048). In replicating the expression patterns of the nine genes, eight of the nine genes (except TEX30) were found to have significant expression profiles between 15 tumor tissues and adjacent counterpart controls; moreover, the qRT-PCR results validated the abnormal expression pattern of the target genes in CCA.
Conclusions: Collectively, we established an effective prognostic model for different populations of CCA patients based on nine DEGs. These findings may provide potential benefits for the development of new prognostic biomarkers and therapeutic targets for CCA.