The majority of the epigenomic reports in hepatocellular carcinoma have focused on identifying novel differentially methylated drivers or passengers of the oncogenic process. Few reports have considered the technologies in place for clinical translation of newly identified biomarkers. The aim of this study was to identify epigenomic technologies that need only a small number of samples to discriminate HCC from non-HCC tissue, a basic requirement for biomarker development trials. To assess that potential, we used quantitative Methylation Specific PCR, oligonucleotide tiling arrays, and Methylation BeadChip assays. Concurrent global DNA hypomethylation, gene-specific hypermethylation, and chromatin alterations were observed as a hallmark of HCC. A global loss of promoter methylation was observed in HCC with the Illumina BeadChip assays and the Nimblegen oligonucleotide arrays. HCC samples had lower median methylation peak scores and a reduced number of significant promoter-wide methylated probes. Promoter hypermethylation of
Promoter-wide alterations of DNA methylation have been described at all stages that encompass hepatocarcinogenesis, precancerous lesions, and tumor initiation to unresectable HCC [
The earlier methylation studies of HCC used the candidate gene approach and first generation methylation microarrays, which study less than 7K CpG islands [
The contribution of DNA methylation to the development of HCC is not yet elucidated. A methylation study in HCC is also challenging as there are several well-known risk factors for HCC, such as alcohol-induced cirrhosis and chronic viral hepatitis B or C infection [
We selected two existing methylation platforms to separately distinguish between HCC and non-HCC liver tissue in a small number of samples: an oligonucleotide methylation tiling array (MeDIP-chip, Nimblegen’s 385K Promoter, and CpG Island methylation array) and the Infinium Human Methylation 27K BeadChip assay (Illumina). We then generated a list of hypermethylated genes based on both the frequency in which the genes had been identified as methylated in different studies and also in the methylation arrays we used. From this list, we chose three genes for validation in an independent cohort comprised of HCC and adjacent nonpathological samples using quantitative Methylation-Specific PCR (qMSP). The focus of the study was to identify whether methylation platforms stratifying a small sample size together with publicly available genomic and epigenomic databases could be deployed in biomarker development trials. The methylation platforms can be used as stand-alone tools or as complementary platforms to other transomic tools, depending on the scientific question.
Deidentified frozen primary HCC, adjacent nontumor (cirrhotic and noncirrhotic), and normal liver (noncirrhotic tissue obtained from autopsies) tissue samples were obtained from the Johns Hopkins University School of Medicine and the Human Cooperative Tissue Network. The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the Johns Hopkins Institutional Review Board. All patients had not undergone therapy prior to sample collection. The samples were frozen in liquid nitrogen and stored at −80°C.
Tissue samples were digested with 1% SDS and 50
Bisulfite-treated DNA from 3 HCC samples and 3 adjacent normal liver samples was hybridized to the Human Methylation 27K BeadChip, which quantitatively interrogates 27,578 CpG loci covering more than 14,000 genes at single-nucleotide resolution. The Infinium Methylation assay detects cytosine methylation at CpG islands based on highly multiplexed genotyping of bisulfite-converted genomic DNA (gDNA). The assay interrogates these chemically differentiated loci using two site-specific probes, one designed for the methylated locus (M bead type) and another for the unmethylated locus (U bead type). Single-base extension of the probes incorporates a labeled ddNTP, which is subsequently stained with a fluorescence reagent. The level of methylation for the interrogated locus can be determined by calculating the ratio of the fluorescent signals from the methylated versus unmethylated sites.
DNA (500 ng) from 3 liver tissue samples (1 HCC and 2 noncirrhotic normal liver samples) enriched with MeDIP were hybridized to Nimblegen Promoter plus CpG Island 385K oligonucleotide tiling arrays. A single array design covers 28,226 CpG islands and promoter regions for 17,000 RefSeq genes. The promoter region covered is 1 kb long: 800 bp upstream from the TSS and 200 bp downstream from the TSS. Small CpG islands are extended at both ends for a total additional coverage of 700 bp for more reliable detection. DNA methylation positive control regions, such as the HoxA gene cluster, H19/IGF2 cluster, KCNQ1 cluster, and IGF2R gene, are also included on the array.
The beta values of all probes on the Illumina Infinium arrays were subjected to log10 transformation in order to generate a dendrogram and corresponding heat map based on unsupervised hierarchical clustering with Spotfire (Somerville, MA). The clustering was performed with the unweighted average method using correlation as the similarity measure and ordering by average values.
The Microarray Core at Johns Hopkins School of Medicine performed the bioinformatics analysis of the Infinium array data using Illumina’s proprietary BeadStudio software package to provide average methylation Beta values for each probe. Nimblegen performed the bioinformatics analysis for the 385K CpG Island Plus Promoter Array. Nimblegen uses the ACME algorithm to identify hypermethylated genes that have a statistically significant methylation peak score above 2 [
Candidate gene selection for promoter methylation analysis was performed utilizing existing databases of known methylation events in cancer [
DNA from 27 HCC and 22 adjacent normal tissue samples (cirrhotic, noncirrhotic, and cryptogenic) was bisulfite treated and analyzed with qMSP. Fluorogenic PCR reactions were carried out in a reaction volume of 20
Amplification reactions were carried out in 384-well plates in a 7900 HT Fast Real-Time PCR System (Applied Biosystems) and were analyzed by SDS 2.2.1 Sequence Detector System (Applied Biosystems). Thermal cycling was initiated with a first denaturation step at 95°C for 3 minutes, followed by 40 cycles of 95°C for 15 seconds, and 58°C for 1 minute. Each plate included patient DNA samples, positive (Bisulfite-converted Universal Methylated Human DNA Standard, Zymo Research) and negative (normal leukocyte DNA or DNA from a known unmethylated cell line) controls, and multiple water blanks. Serial dilutions (60 ng, 6 ng, 0.6 ng, 0.06 ng, and 0.006 ng) of Bisulfite-converted Universal Methylated Human DNA Standard were used to construct a standard curve for each gene.
qMSP values were adjusted for DNA input by expressing results as ratios between 2 absolute measurements. The relative level of methylated DNA for each gene in each sample was determined as a ratio of qMSP for the amplified gene to
Patient characteristics are summarized in Table
Hepatocellular carcinoma risk factors per participant.
ID | Age | Sex | Race | Type | Etiology | Histology | Size | AFP | B4GALT1 | RASSF1A | SSBP2 |
---|---|---|---|---|---|---|---|---|---|---|---|
16 | 53 | M | W | Normal | Cryptogenic | Cirrhosis | · | · | U | U | U |
22 | 58 | M | W | Normal | HCV | Cirrhosis | · | · | U | U | U |
30 | 65 | M | W | Normal | HCV | Cirrhosis | · | · | U | U | U |
31 | 67 | M | W | Normal | HCV | Cirrhosis | · | · | U | M | M |
21 | 58 | M | W | Normal | Other | Cirrhosis | · | · | U | U | U |
15 | 50 | F | B | Normal | Cryptogenic | Noncirrhotic | · | · | U | U | U |
2 | 19 | F | B | Normal | HCV | Noncirrhotic | · | · | U | U | M |
29 | 62 | F | W | Normal | HCV | Noncirrhotic | · | · | U | U | U |
9 | 40 | F | B | Normal | Other | Noncirrhotic | · | · | U | U | U |
4 | 26 | F | W | Normal | Other | Noncirrhotic | · | · | U | U | M |
13 | 45 | F | W | Normal | Other | Noncirrhotic | · | · | U | U | U |
35 | 81 | F | W | Normal | Other | Noncirrhotic | · | · | U | M | U |
18 | 54 | M | B | Normal | Cryptogenic | Noncirrhotic | · | · | U | U | U |
10 | 42 | M | W | Normal | Other | Noncirrhotic | · | · | U | M | U |
20 | 55 | M | W | Normal | Other | Noncirrhotic | · | · | U | M | M |
36 | unk | unk | unk | Normal | unk | unk | · | · | U | U | M |
37 | unk | unk | unk | Normal | unk | unk | · | · | U | U | U |
38 | unk | unk | unk | Normal | unk | unk | · | · | U | U | M |
39 | 53 | M | M | Normal | ETOH | Noncirrhotic | · | · | U | U | U |
40 | 19 | F | F | Normal | HCV | Noncirrhotic | · | · | U | U | U |
41 | 57 | M | M | Normal | HCV | Cirrhosis | · | · | U | U | U |
42 | 60 | M | M | Normal | HCV | Cirrhosis | · | · | U | U | U |
5 | 28 | F | W | Tumor | Other | HCC | 2.5 | 1 | M | U | U |
28 | 62 | F | W | Tumor | HCV | HCC | 2.8 | 17 | U | M | M |
17 | 53 | M | W | Tumor | Cryptogenic | HCC | 4 | 10 | U | M | U |
7 | 40 | F | B | Tumor | Other | HCC | 3.7 | 1 | U | U | M |
25 | 59 | F | AS | Tumor | HBV | HCC | 4 | 20 | M | M | M |
3 | 20 | F | W | Tumor | HCV | HCC | 4 | unk | M | U | U |
27 | 60 | M | B | Tumor | HCV/ETOH | HCC | 4 | 20 | M | M | M |
11 | 45 | M | W | Tumor | HCV | HCC | 5.0 | unk | M | M | M |
8 | 40 | M | W | Tumor | HCV | HCC | 5.5 | 11110 | U | U | U |
32 | 67 | M | W | Tumor | HCV | HCC | 6 | 2 | M | M | M |
33 | 73 | M | W | Tumor | Other | HCC | 6 | 2 | U | M | M |
6 | 37 | M | B | Tumor | Other | HCC | 7.0 | 54071 | U | M | U |
23 | 58 | M | W | Tumor | Other | HCC | 7.2 | 4659 | M | U | M |
14 | 50 | F | B | Tumor | Cryptogenic | HCC | 9 | 1594 | U | M | U |
26 | 60 | M | W | Tumor | Cryptogenic | HCC | 12 | 146 | M | M | M |
19 | 55 | M | W | Tumor | Other | HCC | 17 | 5 | U | M | M |
12 | 45 | F | B | Tumor | HVB/HCV | HCC | unk | unk | U | U | U |
24 | 58 | M | W | Tumor | HCV | HCC | unk | unk | M | M | M |
1 | 19 | F | B | Tumor | HCV | HCC | 25 | 19764 | U | M | M |
34 | 74 | F | W | Tumor | Cryptogenic | HCC | 4.5 | unk | M | M | U |
43 | 42 | M | M | Tumor | Other | HCC | 15 | NA | M | M | U |
44 | 26 | F | M | Tumor | Other | HCC | 8 | NA | M | M | U |
45 | 12 | F | M | Tumor | Other | HCC | NA | NA | M | M | U |
46 | NA | NA | NA | Tumor | NA | HCC | NA | NA | U | M | U |
47 | NA | NA | NA | Tumor | NA | HCC | NA | NA | U | M | M |
48 | NA | NA | NA | Tumor | NA | HCC | NA | NA | U | U | M |
49 | 45 | F | F | Tumor | Cryptogenic | HCC | 1.5 | NA | M | M | M |
M: Methylated; U: Unmethylated; unk: unknown.
We used scatterplots to compare differential DNA methylation values between HCC and normal liver tissue samples hybridized to the 385K Nimblegen tiling array after DNA enrichment with MeDIP (MeDIP-chip). Figure
Scatterplots and histograms for a representative set of one tumor sample and two normal samples hybridized to oligonucleotide methylation tiling arrays. The methylation score is on the
We used unsupervised clustering of the Illumina BeadChip array results to create a heat map based on correlation, which clearly separates the three HCC samples from the adjacent normal liver sample (Figure
Heat map of the promoter-wide methylation data obtained by hybridizing to the Infinium array three hepatocellular carcinoma (HCC) samples and three nontumor liver samples from patients with no known liver disease. A dendrogram (tree graph) of the average beta values for three HCC samples and three nontumor samples was created with Spotfire (Somerville, MA). Unsupervised hierarchical clustering was performed with the unweighted average method using correlation as the similarity measure and ordering by average values. The color red was selected to represent high scores and the color green to represent low scores.
Our search of publicly available methylation databases found a combined total of 549 methylated genes when searching for hepatocellular carcinoma (389) and hepatoma (160), 451 of which were unique genes. After crossing that list with the list of frequently methylated genes we identified using methylation arrays, we chose three genes for validation, one gene that was already found to be hypermethylated in HCC by several groups (
Quantitative MSP results of hepatocellular carcinoma samples and adjacent normal liver samples that were bisulfite treated to examine the promoter methylation status of
ROC curves were used to determine the sensitivity and specificity of the three genes individually and combined in a biomarker panel (Figure
Specificity, sensitivity, and area under the curve results for RASSF1A, B4GALT1, and SSBP2 in HCC, individually, and in a combined panel of the three genes.
RASSF1A | B4GALT1 | SSBP2 | Combined | |
---|---|---|---|---|
Specificity | 100% | 100% | 100% | 100% |
Sensitivity | 52% | 52% | 38% | 68% |
AUC | 0.73 | 0.75 | 0.58 | 0.82 |
ROC curves for a panel of the three genes
When the methylation status of these three genes was included in a logistic regression model together with gender, age, and etiology, the sensitivity was 87%, the specificity 100%, and the AUC was 0.91 (Figure
HCC is the most common primary malignancy of the liver in adults, the fifth most common solid tumor, and the third most common cause of cancer death worldwide [
Epigenetic lesions in DNA without mutations in the coding regions have been shown to be common phenomena in the pathogenesis of a wide range of cancers, especially the methylation-mediated silencing of tumor suppressor genes such as VHL, p16INK4a, E-cadherin, hMLH1, BRCA1, and LKB1 [
Differential methylation has been identified from the early precancerous stages, in association with inflammation and/or persistent infection with HBV or HCV seen in chronic hepatitis or liver cirrhosis to HCC lesions [
By using a study principle that combines promoter-wide and gene-specific methylation platforms that interrogate the promoter region, we were able to distinguish HCC from non-HCC tissue. Our group and others have previously shown that analytical platforms, which quantified global DNA methylation in repetitive regions of the genome, could also distinguish between HCC and non-HCC tissue [
The primary goal of our study was to test whether a small sample size is sufficient to provide information on methylation-related studies by using Illumina BeadChip assays and the Nimblegen oligonucleotide arrays. Our discovery set, although including a limited number of samples, was able to identify genes differentially methylated in HCC when compared to normal samples. Among them, there were genes previously reported as also genes with a known role in HCC and other cancer types. To further validate our findings and the power of a genome-wide analysis based on a small sample size, we generated a list of hypermethylated genes in HCC. We ranked the list based on both the frequency in which the genes had been identified in different studies and also in the methylation arrays we used.
As our knowledge of the HCC epigenome increases, new therapeutic and clinical management strategies may be developed and new serum-based screening or needle biopsy-based diagnostic tools may become available for subgroups at risk for HCC. The pace of DNA methylation translational research is expected to increase exponentially due to the rapid advancement of high-throughput promoter-wide technologies, such as microarray and next-generation sequencing, as well as the advent of user-friendly commercial kits for methylation enrichment [
The aim of this paper was not to provide robust conclusions about specific biomarkers but rather to demonstrate that discrimination between HCC and non-HCC liver tissue using currently available technologies that quantify promoter-wide and gene-specific DNA methylation alterations is feasible. The usefulness of the markers showcased in this paper still needs to be determined in follow-up studies. However, the technological platforms we have used in this project can have an immediate impact on clinical and biomarker development studies.
Promoter-wide microarray technologies may be used to identify methylation patterns that distinguish between HCC and non-HCC tissue. These technologies are well suited for personalized diagnostics and clinical management. As utilization costs of microarrays decrease, population-based studies may also consider using custom-made microarrays to examine large numbers of participants in prevention and early detection studies. Furthermore, we have also shown how qMSP analyses can be used for fast, accurate, and cost-effective high-throughput validation of methylation frequencies in a large number of samples. There is a potential to test the arising genes’ lists to detect biomarkers for early HCC detection in bodily fluids such as plasma, serum, or urine and provide a noninvasive method to clinicians to stratify patients of higher risk for HCC. As the field of translational epigenomics moves forward, clinical tests using these technologies will be warranted to determine their usefulness and reliability in novel screening and clinical management approaches for HCC.
The authors declare that there is no conflict of interests regarding the publication of this paper.
This research was supported in part by the following Grant Awards: NCI U01CA084986 and K01CA164092. The authors want to thank Regina Santella for providing normal liver and hepatocellular carcinoma samples and Rafael Irizarry for his help and supervision in bioinformatics and biostatistics.