Carcinogenetic risk estimation based on quantification of DNA methylation levels in liver tissue at the precancerous stage

Authors


Abstract

For appropriate surveillance of patients at the precancerous stage for hepatocellular carcinomas (HCCs), carcinogenetic risk estimation is advantageous. The aim of our study was to establish criteria for such estimation based on DNA methylation profiling. The DNA methylation status of 203 CpG sites on 25 bacterial artificial chromosome (BAC) clones, whose DNA methylation status had been proven to discriminate samples of noncancerous liver tissue obtained from patients with HCC (N) from normal liver tissue (C) samples by BAC array-based methylated CpG island amplification, was evaluated quantitatively using pyrosequencing. The 45 CpG sites whose DNA methylation levels differed significantly between C and N in the learning cohort (n = 22) were identified. The criteria combining DNA methylation status for the 30 regions including the 45 CpG sites were able to diagnose N as being at high risk of carcinogenesis with 100% sensitivity and specificity in the learning cohort and 95.6% sensitivity and 100% specificity in the validation (n = 90) cohort. DNA methylation status for the 30 regions in N samples was significantly correlated with the outcome of patients with HCCs, indicating that clinicopathologically valid DNA methylation alterations have already accumulated at the precancerous stage. The DNA methylation status of the 30 regions did not depend on the presence or absence of hepatitis virus infection, or the status of noncancerous liver tissue (chronic hepatitis or cirrhosis). These criteria may be applicable for carcinogenetic risk estimation using liver biopsy specimens obtained from patients who are followed up because of chronic liver diseases.

Hepatocellular carcinoma (HCC) is a common malignancy worldwide. Hepatitis virus infection is associated with an extremely high risk of HCC development. Although mass vaccination against hepatitis B virus (HBV) has been initiated, HBV-associated liver carcinogenesis will not be stamped out for many years, as the age at presentation of HBV is over 50 years mainly in Asia and Africa.1 The spread of hepatitis C virus (HCV) in Japan that occurred in the 1950s and 1960s has resulted in a rapid increase in the incidence of HCC since 1980s.2 In other countries, including the United States, HCV infection has spread more recently.2 As HCC usually develops in liver already affected by chronic hepatitis or liver cirrhosis associated with hepatitis virus infection, the prognosis of patients with HCC is deemed poor, unless the cancer is diagnosed at an early stage. Therefore, surveillance at the precancerous stage will become a priority. In clinical practice, especially intensive surveillance should be performed on patients at high risk of HCC development, even if the patients are asymptomatic. Thus, risk estimation for HCC development is essential for the management of patients with chronic liver diseases.

Alterations of DNA methylation are among the most consistent epigenetic changes observed during multistage human carcinogenesis.3, 4 Accumulating evidence suggests that alterations of DNA methylation are involved even in the early and precancerous stages.5, 6 With respect to hepatocarcinogenesis, DNA methylation alterations associated with expression and/or splicing abnormalities of DNA methyltransferases are already present in liver tissues exhibiting chronic hepatitis or liver cirrhosis obtained from patients with HCCs.7–11 Differing from alterations of mRNA and protein expression, which can be easily affected by the microenvironment of cancer cells or precursor cells, DNA methylation alterations are stably preserved on DNA double strands by covalent bonds. Therefore, even subtle alterations at the precancerous stage can be detected using highly sensitive methodology. DNA methylation alterations may be optimal indicators for carcinogenetic risk estimation.12, 13

We have already established criteria for estimation of the risk of HCC development using bacterial artificial chromosome (BAC) array-based methylated CpG island amplification (BAMCA),14–19 which can provide an overview of the DNA methylation tendency of individual large regions among all chromosomes;13, 19 25 BAC clones, whose DNA methylation status was able to discriminate noncancerous liver tissue obtained from patients with HCCs in the learning cohort from normal liver tissue obtained from patients without HCCs, were identified.18 However, sensitivity and specificity of such discrimination were not 100% in the validation cohort. Moreover, the CpG sites that are of diagnostic importance are unclear on each of the BAC clones with an average insert size of 170 kbp.20 As the technique of BAMCA requires a large amount of genomic DNA and is somewhat cumbersome, risk estimation using BAMCA may be difficult to apply in a clinical setting.

Here, to identify precisely the CpG sites having the largest diagnostic impact, we quantitatively evaluated the DNA methylation status of 203 CpG sites on these 25 BAC clones using pyrosequencing in tissue specimens. Among the CpG sites, we were able to improve the specificity of carcinogenetic risk estimation by combining those showing the largest diagnostic impact and to apply such risk estimation to a very small amount of genomic DNA with a view to clinical application.

Material and Methods

Patients and tissue samples

As a learning cohort, 10 samples of normal liver tissue (C1–C10) showing no remarkable histological findings were obtained from specimens surgically resected from 10 patients without HCCs who were negative for both HBV surface antigen (HBs-Ag) and anti-HCV antibody (anti-HCV). The patients comprised seven men and three women with a mean (± standard deviation) age of 58.4 ± 9.7 years. Nine patients underwent partial hepatectomy for liver metastases of primary colon cancer, and one patient did so for liver metastases of a gastrointestinal stromal tumor of the stomach at the National Cancer Center Hospital, Tokyo, Japan. A total of 12 samples of noncancerous liver tissue (N1–N12) were obtained from 12 patients who underwent partial hepatectomy for HCCs. These patients comprised nine men and three women with a mean age of 65.3 ± 6.4 years. Among them, six were positive for HBs-Ag and six were positive for anti-HCV. Histological examination of these noncancerous liver tissue samples revealed findings compatible with chronic hepatitis in four and cirrhosis in eight.

As a validation cohort, 45 samples of normal liver tissue (C11–C55) exhibiting no remarkable histological findings were obtained from 45 patients without HCCs who were negative for both HBs-Ag and anti-HCV. The patients comprised 34 men and 11 women with a mean age of 62.2 ± 7.0 years. A total of 39 patients underwent partial hepatectomy for liver metastases from primary colon cancer, three patients did so for liver metastasis from gastric cancer and the remaining three patients did so for liver metastasis from each of gastrointestinal stromal tumor of the stomach, pancreatic cancer and colon carcinoid tumor, respectively. A total of 45 samples of noncancerous liver tissue (N13–N57) were obtained from 45 patients who underwent partial hepatectomy for HCCs. The patients comprised 37 men and eight women with a mean age of 62.3 ± 9.7 years. Of them 13 were positive for HBs-Ag, 29 were positive for anti-HCV, and three were negative for both. Histological examination of these noncancerous liver tissue samples revealed findings compatible with chronic hepatitis in 22 and cirrhosis in 23.

For comparison, 34 samples of primary HCC (T1–T34) were also obtained from specimens surgically resected from the patients who had provided the samples N1–N34. In addition, for comparison, 14 samples of liver tissue (V1–V14) were obtained from 14 patients who were positive for HBs-Ag or anti-HCV, but who had never developed HCCs. The patients comprised six men and eight women with a mean age of 65.1 ± 8.2 years. Of them, 12 patients underwent partial hepatectomy for liver metastases of primary colorectal cancer and two patients did so for liver metastases of gastric cancer.

Our study was approved by the Ethics Committee of the National Cancer Center, Tokyo, Japan. All the patients gave informed consent before their inclusion in our study.

DNA extraction and bisulfite DNA modification

High-molecular-weight DNA from fresh-frozen tissue samples was extracted using phenol–chloroform followed by dialysis. Bisulfite conversion was carried out using 1 μg of genomic DNA and the reagents provided in the EpiTect Bisulfite Kit (QIAGEN GmbH, Hilden, Germany), in accordance with the manufacturer's protocol. This process converts unmethylated cytosine residues to uracil, whereas methylated cytosine residues remain unchanged.21

Pyrosequencing DNA methylation analysis

DNA methylation level was measured by a highly quantitative method using Pyrosequencing™ technology. Polymerase chain reaction (PCR) and sequencing primers were designed based on the converted sequences using Pyrosequencing Assay Design Software ver.1.0 (QIAGEN GmbH). To overcome PCR bias in DNA methylation analysis, we optimized the annealing temperature as described previously.22, 23 Each of the primer sequences and PCR conditions are given in Supporting Information Table 1. The PCR was carried out with 0.6 units of AmpliTaq Gold (Applied Biosystems, Foster City, CA) using 7.5 ng of bisulfite-treated DNA. The biotinylated PCR product was captured on streptavidin-coated beads (Streptavidin Sepharose™ High Performance; GE Healthcare, Uppsala, Sweden). Quantitative sequencing was run on the PyroMark Q24 (QIAGEN GmbH) using the Pyro Gold Reagents (QIAGEN GmbH) in accordance with the manufacturer's protocol. For each assay, the setup included positive controls (Epitect methylated human control DNA; QIAGEN GmbH) and negative controls (Epitect unmethylated human control DNA; QIAGEN GmbH). The PCR products were separated electrophoretically on 3% agarose gel and stained with ethidium bromide to confirm that specific products of the appropriate size and no nonspecific products were obtained on amplification. Representative pyrograms are shown in Figure 1. As outlined in Figure 1, the DNA methylation level (%) at each CpG site is given by the following formula:

equation image
Figure 1.

Pyrosequencing DNA methylation analysis. Examples of pyrograms for a sample of normal liver tissue obtained from a patient without HCC (C6) and a sample of noncancerous liver tissue obtained from a patient with HCC (N9) for exon 1 of the FOXD2 gene (47,677,654, −60, −63 in region 5 in Table 1). Gray columns represent the regions of polymorphic sites after bisulfite modification. x-axis indicates dispensation order (time).

Table 1. Thirty regions that were able to discriminate noncancerous liver tissues (N) from normal liver tissues (C)
inline image

Statistics

Significant differences in DNA methylation levels at each of the CpG sites between groups of samples were analyzed using the Mann-Whitney U test. Survival curves of patient groups with HCCs were calculated by the Kaplan–Meier method, and the differences were compared by log-rank test. Differences at p < 0.05 were considered significant.

Abbreviations

anti-HCV: anti-HCV antibody; BAC: bacterial artificial chromosome; BAMCA: BAC array-based methylated CpG island amplification; HBs-Ag: HBV surface antigen; HBV: hepatitis B virus; HCC: hepatocellular carcinoma; HCV: hepatitis C virus; PCR: polymerase chain reaction

Results

Validation of BAMCA data by pyrosequencing

It has been shown that BAMCA can provide an overview of the DNA methylation tendency of individual large regions among all chromosomes.13, 19 Therefore, using pyrosequencing, we evaluated the DNA methylation levels of all Xma I/Sma I sites, which yielded less than 2,000 bp PCR products that are effective in BAMCA, on representative BAC clones, which had been identified as indicators for carcinogenetic risk estimation in our previous study.18 For example, on clone RP11-17M17, there were 10 Xma I/Sma I sites that were effective in BAMCA (Fig. 2a). The average signal ratio by BAMCA of this BAC clone was significantly lower in samples of noncancerous liver tissue obtained from patients with HCCs than in samples of normal liver tissue and was significantly lower in HCCs than in samples of noncancerous liver tissue obtained from patients with HCCs in our previous study.18 The average DNA methylation levels determined by pyrosequencing of all 10 Xma I/Sma I sites on this BAC clone in 34 samples of noncancerous liver tissue obtained from patients with HCCs were the same as (Xma I/Sma I sites i, ii, vii, viii and ix in Fig. 2a) or significantly lower than (iii, iv, v, vi and x in Fig. 2a) those in 35 samples of normal liver tissue. Moreover, the DNA methylation levels of all Xma I/Sma I sites in 34 HCCs were significantly lower than those in samples of noncancerous liver tissue obtained from patients with HCCs (i–x in Fig. 2a). DNA methylation levels of CpG sites adjacent to the Xma I/Sma I sites that were quantitatively sequenced using the same sequencing primers tended to be close to the DNA methylation levels of the Xma I/Sma I sites themselves in each sample, such as iii and iii′ and iv and iv′ in Figure 2b. Thus, it was confirmed that BAMCA was able to successfully reveal DNA methylation alterations occurring in a coordinated manner on RP11-17M17. In another BAC clone, RP11-799O6, which was also identified as an indicator for carcinogenetic risk estimation, the average signal ratio obtained by BAMCA was significantly higher in samples of noncancerous liver tissue from patients with HCCs than in samples of normal liver tissue in our previous study.18 Although the average DNA methylation levels of seven out of 10 Xma I/Sma I sites, which yielded PCR products of less than 2,000 bp that are effective for BAMCA, by pyrosequencing in samples of noncancerous liver tissue from patients with HCCs were the same as those in samples of normal liver tissue, those of the remaining three Xma I/Sma I sites in samples of noncancerous liver tissue obtained from patients with HCCs were markedly higher than those in samples of normal liver tissue (data not shown). Thus, pyrosequencing data again validated the BAMCA data for BAC clones identified as indicators for carcinogenetic risk estimation.

Figure 2.

Validation of BAMCA data by pyrosequencing. On RP11-17M17 clone, there were 10 Xma I/Sma I sites (i–x) that yielded PCR products of less than 2,000 bp that were effective in BAMCA. The average signal ratio obtained by BAMCA for this BAC clone was significantly lower in samples of noncancerous liver tissue obtained from patients with HCCs (N) than in those of normal liver tissue (C) and was significantly lower in HCCs than in N-samples.18 (a) Scattergrams of DNA methylation levels analyzed by pyrosequencing in C-samples (C1–C35), N-samples (N1–N34) and HCCs (T1–T34) on each Xma I/Sma I site. The average DNA methylation levels obtained by pyrosequencing for all 10 Xma I/Sma I sites on this BAC clone in 34 N-samples were the same as (on i, ii, vii, viii and ix) or significantly lower than (on iii, iv, v, vi and x) those in 35 C-samples. Moreover, DNA methylation levels in 34 HCCs were significantly lower than those in N-samples (on i–x). (b) Pi-charts of DNA methylation levels in C-samples (C1–C10) and N-samples (N1–N12) for each of the CpG sites. CpG sites adjacent to the Xma I/Sma I site (i, iii, iv, vii and ix), which were quantitatively sequenced using the same sequencing primers, are indicated by i′, iii′, iv′, vii′ and ix′, respectively. White indicates unmethylated cytosine and black indicates methylated cytosine. DNA methylation levels of CpG sites adjacent to the Xma I/Sma I sites tend to be close to the DNA methylation levels of the Xma I/Sma I sites themselves, e.g., iii and iii′ and iv and iv′, in each sample.

Criteria for carcinogenetic risk estimation using liver tissue samples based on pyrosequencing

To identify CpG sites having the largest diagnostic impact, DNA methylation levels of 203 CpG sites were measured by pyrosequencing using primer sets encompassing Xma I/Sma I sites, which were effective in BAMCA, on the 25 BAC clones on which we based our previous criteria.18

On 59 CpG sites, the average DNA methylation levels differed significantly between normal liver tissue and noncancerous liver tissue obtained from patients with HCCs in the learning cohort using Mann-Whitney U test (p < 0.001). To establish reproducible criteria, 14 CpG sites whose average DNA methylation levels in both normal liver tissue and noncancerous liver tissue obtained from patients with HCCs were less than 10% were omitted from the list of candidate indicators for carcinogenetic risk estimation, taking the characteristics of Pyrosequencing™ technology into consideration.22 Figure 3a shows scattergrams of the DNA methylation levels in samples of normal liver tissue and noncancerous liver tissue obtained from patients with HCCs on representative CpG sites. Using the cutoff values described in each panel, noncancerous liver tissue obtained from patients with HCCs in the learning cohort was discriminated from normal liver tissue with sufficient sensitivity and specificity (Fig. 3a). On the remaining 45 CpG sites, such discrimination was performed with a sensitivity or specificity of 70% or more than 70%. If several CpG sites were measured using one sequencing primer, one cutoff value was set for the region covered by the sequencing primer using the average DNA methylation levels of the several CpG sites. Then the 30 cutoff values were set for 30 regions including the 45 CpG sites, and their sensitivity and specificity are shown in Table 1. Chromosomal loci and characteristics of the 30 regions (CpG islands or not, exons or introns of specific genes or noncoding regions) are also summarized in Table 1.

Figure 3.

The criteria for carcinogenetic risk estimation based on pyrosequencing. (a) Scattergrams of DNA methylation levels in samples of normal liver tissue (C1–C10) and samples of noncancerous liver tissue obtained from patients with HCCs (N1–N12) in the learning cohort for representative regions. Using the cutoff values (CV, %) described in each panel, N-samples in the learning cohort were discriminated from C-samples with sufficient sensitivity and specificity. (b) Histogram showing the number of regions satisfying the criteria described in Table 1 in samples C1–C10 (clear columns) and N1–N12 (filled columns). On the basis of this histogram, we judged that when the noncancerous liver tissue satisfied the criteria in Table 1 for 15 or more than 15 regions, it was at high risk of carcinogenesis. (c) Validation of the criteria in Table 1 using an additional 90 samples of liver tissue in the validation cohort. All 43 validation samples satisfying the Table 1 criteria for 15 or more regions were N-samples (N13–N36, N38–N41 and N43–N57, filled columns), and 45 of 47 validation samples satisfying the Table 1 criteria for less than 15 regions were C-samples (C11–C55, clear columns). DNA methylation statuses for the 30 regions of N-samples and those of C-samples were completely mutually exclusive in the validation cohort.

A histogram showing the number of regions satisfying the criteria listed in Table 1 for samples C1–C10 and N1–N12 in the learning cohort is shown in Figure 3b. On the basis of Figure 3b, we finally established that when liver tissue satisfied the criteria in Table 1 for 15 or more regions, it was judged to be at high risk of carcinogenesis. Based on this definition both the sensitivity and specificity for diagnosis of noncancerous liver tissue obtained from patients with HCCs in the learning cohort as being at high risk of carcinogenesis were 100%.

To confirm these criteria, an additional 90 samples of liver tissue were analyzed by pyrosequencing as a validation study (Fig. 3c). All of the 43 validation samples satisfying the criteria in Table 1 for 15 or more regions were noncancerous liver tissue obtained from patients with HCCs (N13–N36, N38–N41 and N43–N57), and 45 of the 47 validation samples satisfying the Table 1 criteria for less than 15 regions were normal liver tissue (C11–C55). DNA methylation statuses for the 30 regions of noncancerous liver tissue samples from patients with HCCs and those of normal liver tissue samples were completely mutually exclusive in the validation cohort (Fig. 3c), and our criteria enabled diagnosis of noncancerous liver tissue from patients with HCCs in the validation cohort as being at high risk of carcinogenesis with 95.6% sensitivity and 100% specificity.

Clinicopathological significance of DNA methylation status in the 30 regions

To estimate the clinicopathological significance of DNA methylation status in the 30 regions, 34 samples of noncancerous liver tissue from patients with HCCs (N1–N34) in both the learning and validation cohorts for whom follow-up data had been obtained were divided into two groups according to the number of regions satisfying the criteria (≥23 [the median of the number of regions satisfying the Table 1 criteria] regions vs. <23 regions). The period covered ranged from 11 to 3,936 days (mean: 1,417 days). The cancer-free and overall survival rates for patients with HCCs satisfying the criteria in Table 1 for 23 or more regions in their noncancerous liver tissue were significantly lower than those of patients with HCCs satisfying the Table 1 criteria for less than 23 regions (Fig. 4, p = 0.0023 and p = 0.0015, respectively). These data suggested that clinicopathologically valid DNA methylation alterations associated with patient outcome are already present at the precancerous stage.

Figure 4.

Correlation between DNA methylation status at the precancerous stage and patient outcome. Kaplan–Meier survival curves of patients with HCCs from whom samples N1–N34 were obtained. The cancer-free (a; p = 0.0023) and overall (b; p = 0.0015) survival rates of patients with HCCs satisfying the criteria in Table 1 for 23 (the median of the number of regions satisfying the Table 1 criteria) or more than 23 regions in their samples of noncancerous liver tissue (n = 17, solid lines) were significantly lower than those of patients with HCCs satisfying the criteria in Table 1 for less than 23 regions (n = 17, broken lines).

With respect to all 57 samples of noncancerous liver tissue (N1–N57), the difference in the number of regions satisfying the criteria listed in Table 1 between liver tissue samples showing chronic hepatitis (n = 26, 19.6 ± 3.7) and those showing cirrhosis (n = 31, 22.0 ± 3.9) was marginal (p = 0.0206). For comparison, the DNA methylation levels of 30 regions in 14 additional liver tissue samples (V1–V14) obtained from patients who were infected with HBV or HCV, but who had never developed HCCs, were analyzed by pyrosequencing. The average number of regions satisfying the Table 1 criteria was significantly lower in V1–V14 (12.0 ± 5.0), than in N1–N57 (20.9 ± 4.0, p < 0.0001). These data suggested that our criteria do not simply reflect the presence of hepatitis virus infection, inflammation or fibrosis at the chronic hepatitis and liver cirrhosis stages, but in fact reflect the carcinogenetic risk itself.

Discussion

For appropriate surveillance of patients at the precancerous stage for HCCs, the criteria for carcinogenetic risk estimation should be explored. As considerable numbers of liver tissue samples obtained from patients with HBV or HCV infection indicate a future risk of HCCs, even if HCCs are not yet present, comparison between liver tissue samples obtained from patients with HBV or HCV infection but without HCCs and those obtained from patients with HBV or HCV infection but also showing HCCs, is not an adequate strategy for establishing criteria for carcinogenetic risk estimation. Therefore, in our previous study, we focused on BAC clones whose signal ratios differed significantly between samples of normal liver tissue obtained from patients without HBV or HCV infection and samples of noncancerous liver tissue obtained from patients with HCCs (namely BAC clones on which DNA methylation alterations had occurred at the precancerous stage), and also those on which such DNA methylation alterations had been inherited by HCCs themselves from precancerous conditions. In this way, we successfully established such criteria using BAC array-based methods.18

In our study, the reliability of BAMCA was again confirmed: BAMCA was able to provide an overview of DNA methylation tendency in large regions of chromosomes, and especially was able to detect DNA methylation alterations occurring in a coordinated manner in the entire BAC region. However, the exact CpG sites that are of diagnostic impact are unclear, because several Xma I/Sma I sites that are effective in BAMCA generally exist on each of the BAC clones with an average insert size of 170 kbp.20 Moreover, as BAMCA requires a large amount of genomic DNA, and the technique is somewhat cumbersome, our previous criteria based on BAMCA may not be suitable for clinical uses such as risk estimation based on liver biopsy specimens. Therefore, we employed Pyrosequencing™ technology, which is an excellent tool for quantitative estimation of DNA methylation levels at specific CpG sites.

Although numerous Xma I/Sma I sites are located within CpG islands, one or two Xma I/Sma I sites on each CpG island were analyzed because of difficulties with the design of the PCR and sequencing primers. Then, DNA methylation levels at 203 CpG sites on the 25 BAC clones that comprised our previous criteria based on BAMCA were evaluated quantitatively by pyrosequencing. By combining the 30 regions including 45 specific CpG sites, which were revealed to have a large diagnostic impact, the specificity of the criteria for carcinogenetic risk estimation was successfully improved in comparison with our previous criteria based on BAMCA18: the sensitivity and specificity of the criteria after revision by pyrosequencing were both 100% in the learning cohort and were 95.6% and 100% in the validation cohort, respectively.

Only one region (region 20 in Table 1) among 30 regions that had been used for defining the revised criteria for carcinogenetic risk estimation was located within the promoter region of a specific gene (general receptor for phosphoinositides 1-associated scaffold protein), although DNA methylation alterations in promoter regions are known to be one of most consistent epigenetic changes in human cancers.24 At the risk stage, but not in established cancers, it is feasible that DNA methylation alterations do not expand immediately to the promoter regions of specific genes, such as tumor-related genes. However, 20, 19 and 9 regions that had been used for defining the revised criteria were located within gene bodies, non-CpG islands, and noncoding regions, respectively, which have been overlooked as targets of DNA methylation alterations during multistage human carcinogenesis. Although most of the recently developed detection technologies, such as promoter arrays and CpG island arrays, are sequence-based methods and cannot comprehensively measure the DNA methylation status of gene bodies, non-CpG islands and noncoding regions,25, 26 our findings indicate that meticulous examination of such sequences is also important for establishment of optimal diagnostic indicators.

DNA methylation status in the 30 regions in noncancerous liver tissue at the precancerous stage was significantly correlated with both cancer-free and overall survival rates of patients with HCCs (Fig. 4). Although prognostication before development of HCCs was not a clinically relevant issue, and we never intended to perform such prognostication, we can consider that DNA methylation alterations determining patient outcome had already accumulated at the precancerous stage, based on the data in Figure 4. As DNA methylation status is not randomly altered at the precancerous stage, and DNA methylation profiles in noncancerous liver tissue have been proven to be clinicopathologically valid, it is feasible that such profiles could be optimal indicators for carcinogenetic risk estimation.

The difference in the number of regions satisfying the criteria listed in Table 1 between liver tissue samples showing chronic hepatitis and those showing cirrhosis was marginal, indicating that our criteria were not simply associated with inflammation or fibrosis. In addition, the average number of regions satisfying the Table 1 criteria were significantly lower in liver tissue from patients without HCCs (V1–V14) than in noncancerous liver tissue from patients with HCCs (N1–N34), even though the patients from whom V1–V14 were obtained were infected with HBV or HCV. DNA methylation status in the 30 regions does not depend on hepatitis virus infection but may actually reflect the carcinogenetic risk itself. Therefore, our criteria not only discriminate noncancerous liver tissue from patients with HCCs from normal liver tissues but also may be applicable for classifying liver tissue obtained from patients who are being followed up because of HBV or HCV infections, chronic hepatitis or cirrhosis into that which may generate HCCs and that which will not.

During surveillance at the precancerous stage, to reveal the baseline liver histology, microscopic examination of liver biopsy specimens is performed in patients with HBV or HCV infection before interferon therapy.27, 28 Therefore, carcinogenetic risk estimation using such liver biopsy specimens will be advantageous for close follow-up of patients who are at high risk of HCC development. We have confirmed that pyrosequencing can be performed using a very small amount of degraded DNA extracted from formalin-fixed and paraffin-embedded liver biopsy specimens (unpublished data). We now intend to prospectively validate the reliability of risk estimation based on the revised criteria using pyrosequencing in liver biopsy specimens obtained before interferon therapy from a large cohort of patients with HBV or HCV infection.

Acknowledgements

Author R.N. received a Research Resident Fellowship from the Foundation for Promotion of Cancer Research in Japan.

Ancillary