DNA methylation-based prognosis and epidrivers in hepatocellular carcinoma

Abstract

Epigenetic deregulation has emerged as a driver in human malignancies. There is no clear understanding of the epigenetic alterations in hepatocellular carcinoma (HCC) and of the potential role of DNA methylation markers as prognostic biomarkers. Analysis of tumor tissue from 304 patients with HCC treated with surgical resection allowed us to generate a methylation-based prognostic signature using a training-validation scheme. Methylome profiling was done with the Illumina HumanMethylation450 array (Illumina, Inc., San Diego, CA), which covers 96% of known cytosine-phosphate-guanine (CpG) islands and 485,000 CpG, and transcriptome profiling was performed with Affymetrix Human Genome U219 Plate (Affymetrix, Inc., Santa Clara, CA) and miRNA Chip 2.0. Random survival forests enabled us to generate a methylation signature based on 36 methylation probes. We computed a risk score of mortality for each individual that accurately discriminated patient survival both in the training (221 patients; 47% hepatitis C-related HCC) and validation sets (n = 83; 47% alcohol-related HCC). This signature correlated with known predictors of poor outcome and retained independent prognostic capacity of survival along with multinodularity and platelet count. The subset of patients identified by this signature was enriched in the molecular subclass of proliferation with progenitor cell features. The study confirmed a high prevalence of genes known to be deregulated by aberrant methylation in HCC (e.g., Ras association [RalGDS/AF-6] domain family member 1, insulin-like growth factor 2, and adenomatous polyposis coli) and other solid tumors (e.g., NOTCH3) and describes potential candidate epidrivers (e.g., septin 9 and ephrin B2).

Conclusions: A validated signature of 36 DNA methylation markers accurately predicts poor survival in patients with HCC. Patients with this methylation profile harbor messenger RNA-based signatures indicating tumors with progenitor cell features.

Fig. 1.

Distribution of DNA-methylation–derived MI in the training set (Heptromic) as per density plot (A). Box plots (with overimposed individual values) for different clinical and pathological variables known to be prognostic in HCC and the mortality index (B-I).

Fig. 2.

Predicted HR for survival based on DNA-methylation–derived MI values (A). Kaplan-Meier’s plots for outcome analysis in Heptromic (n = 221, training; B-C) and French data sets (n = 83, validation; D).

Fig. 3.

Box plots (with overimposed individual values) of MI according to S and EpCAM mRNA-based molecular subclasses (A). Patients in the S2 and EpCAM classes (reported initially as identifying tumors with a progenitor cell origin) have significantly higher levels of DNA-methylation–derived MI based on predictions from the methylation signatures. (B) Bottom panel represents enrichment of the 205 samples with methylation and mRNA data based on the levels of risk from the MI score. Top 20% patients with the highest DNA-methylation–derived MI score are significantly enriched in S2 and EpCAM classes.

Fig. 4.

Genome-wide DNA methylation profiling of HCC patients. Genomic distribution of the differentially methylated probes between normal liver and HCC samples according to: (1) functional genomic distribution (promoter, body, 3’ UTR and intergenic) and (2) CpG content and neighborhood context (island, shore, shelf, and open sea) (A). Unsupervised hierarchical clustering of 10 normal liver (yellow), 9 cirrhotic liver (orange), and 221 HCC samples (blue) using the top 100 differentially methylated probes according to F score (B).

Fig. 5.

(Left panels) Box plots represent demethylation status of new candidate epidrivers generated from the training cohort (Heptromic). Right panels show their values on the French cohort (validation). P values computed comparing HCC versus normal liver. Abbreviations: CDKL2, cyclin-dependent kinase-like 2; DRD4, dopamine receptor D4; FAM196A, family with sequence similarity 196, member A; FOXE4, forkhead box protein E3; TBX15, T-box 15.