Hepatocellular carcinoma displays distinct DNA methylation signatures with potential as clinical predictors

Abstract

Background: Hepatocellular carcinoma (HCC) is characterized by late detection and fast progression, and it is believed that epigenetic disruption may be the cause of its molecular and clinicopathological heterogeneity. A better understanding of the global deregulation of methylation states and how they correlate with disease progression will aid in the design of strategies for earlier detection and better therapeutic decisions.

Methods and findings: We characterized the changes in promoter methylation in a series of 30 HCC tumors and their respective surrounding tissue and identified methylation signatures associated with major risk factors and clinical correlates. A wide panel of cancer-related gene promoters was analyzed using Illumina bead array technology, and CpG sites were then selected according to their ability to classify clinicopathological parameters. An independent series of HCC tumors and matched surrounding tissue was used for validation of the signatures. We were able to develop and validate a signature of methylation in HCC. This signature distinguished HCC from surrounding tissue and from other tumor types, and was independent of risk factors. However, aberrant methylation of an independent subset of promoters was associated with tumor progression and etiological risk factors (HBV or HCV infection and alcohol consumption). Interestingly, distinct methylation of an independent panel of gene promoters was strongly correlated with survival after cancer therapy.

Conclusion: Our study shows that HCC tumors exhibit specific DNA methylation signatures associated with major risk factors and tumor progression stage, with potential clinical applications in diagnosis and prognosis.

Figure 1. Unsupervised analysis of CpG methylation bead arrays in HCC.

A. Clustering analysis of 76 HCC samples included in the bead array assay (HCC tumor and surrounding tissue). For the upper part of the cluster, names are given manually according to the enrichment of specific clusters. 1505 CpG sites are included. Yellow indicates hypomethylated, and red hypermethylated CpG sites. B. Representative logarithmic plot of two replicates included in the array, showing proper consistency of methylation (r2 value is included on the plot). C. Average promoter methylation of all 1505 CpG sites, in HCCs and surrounding tissues. D. Clustering analysis after grouping the samples by ethological factors. E. Average methylation for all 1505 CpG sites from the same ethological groups shown in (d). Significant differences (P<0.05) between tumor and surrounding tissue are represented with an asterisk (*).

Figure 2. Signature and predictor of HCC by methylation profiling.

A. Differential methylation analysis was performed with the class comparison tool of BRBArrayTools software, as described in Materials and Methods. The heat map represents those CpG sites distinguishing HCC from surrounding tissue (n = 87) with a P value<0.001. The full list of CpG sites is presented as Table S2. Yellow indicates hypomethylated, and red hypermethylated CpG sites. B. Representation of the misclassification error as a function of the number of genes, as assessed with the PAM prediction analysis. The upper panel shows the correlation for the grouped samples; the lower panel shows the independent correlation for tumor and surrounding samples. Sensitivity and specificity of the predictor is included in the Figure. C. A heat map with the 20 CpG sites included in the HCC predictor was obtained for an independent series of HCC samples and HCC surrounding tissues, after unsupervised hierarchical clustering analysis.

Figure 3. Methylation profile according to risk factor and tumor progression.

Class comparison analyses were performed, as described in Figure 2. A. The heat map represents 27 CpG sites distinguishing the different HCC samples according to their TNM classification, with a P value<0.05. B. The heat map represents 17 CpG sites distinguishing the different HCC samples according to their ethological exposure, with a P value<0.01. HBV or HCV infection, EtOH  =  ethanol consumption, and Unknown  =  unknown risk factor.

Figure 4. Survival risk predictor in HCC.

A. Survival analysis using BRB-ArrayTools. A survival signature was developed using fitted Cox proportional-hazards model and leave-one-out crossvalidation, considering the time of biopsy as the starting point. Survival curves show a significant difference between two groups of HCC patients. B. A 58 CpG sites predictor (selected from the analysis shown in A.) was correlated with survival after treatment. Only the first 10 CpG sites (with the lowest P value) are shown. C. Pathway analysis for the 58 CpG sites included in the survival predictor showing the 5 significantly enriched pathways. D. Quantitative RT-PCR was performed for some of the genes with the highest ability to predict survival in HCC (MYLK, FLT1, CDKN1C and TAp73, in a subset of samples with high (H) and low (L) risk.