Distinctive Metabolism-Associated Gene Clusters That Are Also Prognostic in Intrahepatic Cholangiocarcinoma and Hepatocellular Carcinoma

Abstract

Objective: To offer new prognostic evaluations by exploring potentially distinctive genetic features of hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC).

Methods: There were 12 samples for gene expression profiling processes in this study. These included three HCC lesion samples and their matched adjacent nontumor liver tissues obtained from patients with HCC, as well as three ICC samples and their controls collected similarly. In addition to the expression matrix generated on our own, profiles of other cohorts from The Cancer Genome Atlas (TCGA) program and the Gene Expression Omnibus (GEO) were also employed in later bioinformatical analyses. Differential analyses, functional analyses, protein interaction network analyses, and gene set variation analyses were used to identify key genes. To establish the prognostic models, univariate/multivariate Cox analyses and subsequent stepwise regression were applied, with the Akaike information criterion evaluating the goodness of fitness.

Results: The top three pathways enriched in HCC were all metabolism-related; they were fatty acid degradation, retinol metabolism, and arachidonic acid metabolism. In ICC, on the other hand, additional pathways related to fat digestion and absorption and cholesterol metabolism were identified. Consistent characteristics of such a metabolic landscape were observed across different cohorts. A prognostic risk score model for calculating HCC risk was constructed, consisting of ADH4, ADH6, CYP2C9, CYP4F2, and RDH16. This signature predicts the 3-year survival with an AUC area of 0.708 (95%CI = 0.644 to 0.772). For calculating the risk of ICC, a prognostic risk score model was built upon the expression levels of CYP26A1, NAT2, and UGT2B10. This signature predicts the 3-year survival with an AUC area of 0.806 (95% CI = 0.664 to 0.947).

Conclusion: HCC and ICC share commonly abrupted pathways associated with the metabolism of fatty acids, retinol, arachidonic acids, and drugs, indicating similarities in their pathogenesis as primary liver cancers. On the flip side, these two types of cancer possess distinctive promising biomarkers for predicting overall survival or potential targeted therapies.

Figure 1

Differentially expressed genes between HCC and normal (a), between ICC and normal (b), and between HCC and ICC (c). (d) DEGs that are differentially expressed between two histological types and significantly regulated in each type, respectively, compared to normal tissue.

Figure 2

Significantly enriched pathways based on KEGG and GO databases: (a) functional pathways exhibited by genes that are differentially expressed in HCC compared to its control and differentially expressed compared to ICC samples (left: summary of enriched pathways in KEGG annotation; middle: top 20 enriched pathways; right: Gene Ontology classification); (b) functional pathways exhibited by genes that are differentially expressed in ICC compared to its control and differentially expressed compared to HCC samples (left: summary of enriched pathways in KEGG annotation; middle: top 20 enriched pathways; right: Gene Ontology classification).

Figure 3

Principal component analysis indicated the distinctive natures of ICC and HCC versus the nontumor site in terms of metabolism-related genes.

Figure 4

Differential analyses conducted in TCGA-CHOL and TCGA-LIHC in comparison with nontumor samples (a) revealed 217 commonly differentiated metabolism-related genes (b) significantly enriched to fatty acid metabolism, retinol metabolism, and drug metabolism (c).

Figure 5

Predicted protein interaction analysis revealed 9 enriched major clusters of proteins.

Figure 6

Performance (ROC) of using enrichment scores of key clusters to classify HCC from ICC: (a) individual ROC curves when using enrichment scores of each cluster separately; (b) combinative ROC curve when using enrichment scores of clusters 1 and 2 together; (c) comparisons of actual enrichment levels of clusters 1 and 2 in two histological groups.

Figure 7

The efficacy of the risk model of HCC in predicting survival in TCGA-LIHC. The distribution of risk score and survival status of all samples are shown as (a, b); the heat map of expression of the genes included in the model is shown in (c); (d) Kaplan-Meier plot showing the overall survival based on the relatively high and low risk divided by the optimal cut-off point; (e) time-dependent ROC curve analysis of survival prediction by the prognostic model.

Figure 8

The efficacy of the risk model of ICC in predicting survival in TCGA-LIHC. The distribution of risk score and survival status of all samples are shown as (a, b); the heat map of expression of the genes included in the model is shown in (c); (d) Kaplan-Meier plot showing the overall survival based on the relatively high and low risk divided by the optimal cut-off point; (e) time-dependent ROC curve analysis of survival prediction by the prognostic model.

Figure 9

Kaplan-Meier plots showing the overall survival based on the relatively high and low risk divided by the optimal cut-off point. Both scoring models were replied to GEO cohorts of HCC (a) and ICC (b). (c) Using the expression level of CYP26A1 to predict overall survival in this ICC cohort; (d) using the expression level of NAT2 to predict overall survival in this ICC cohort.