Methylation of multiple genes as molecular markers for diagnosis of a small, well-differentiated hepatocellular carcinoma



The current study was conducted to identify robust methylation markers and their combinations that may prove useful for the diagnosis of early hepatocellular carcinoma (HCC). To achieve this, we performed in silico CpG mapping, direct sequencing and pyrosequencing after bisulfite treatment, and quantitative methylation-specific PCR (MSP) in HCC and non-HCC liver tissues. In the filtering group (25 HCCs), our direct sequencing analysis showed that, among the 12 methylation genes listed by in silico CpG mapping, 7 genes (RASSF1A, CCND2, SPINT2, RUNX3, GSTP1, APC and CFTR) were aberrantly methylated in stages I and II HCCs. In the validation group (20 pairs of HCCs and the corresponding non-tumor liver tissues), pyrosequencing analysis confirmed that the 7 genes were aberrantly and strongly methylated in early HCCs, but not in any of the corresponding non- tumor liver tissues (p < 0.00001). The results obtained using our novel quantitative MSP assay correlated well with those observed using the pyrosequencing analysis. Notably, in MSP assay, RASSF1A showed the most robust performance for the discrimination of HCC and non-HCC liver tissues. Furthermore, a combination of RASSF1A, CCND2 and SPINT2 showed 89–95% sensitivity, 91–100% specificity and 89–97% accuracy in discriminating between HCC and non-HCC tissues, and correctly diagnosed all early HCCs. These results indicate that the combination of these 3 genes may aid in the accurate diagnosis of early HCC. © 2009 UICC

Hepatocellular carcinoma (HCC) represents a major international health problem as a result of its increasing incidence in many countries.1, 2 With the advent of hepatic imaging modalities, many small nodular lesions are currently detected in hepatitis B virus (HBV)- and hepatitis C virus (HCV)-infected individuals exhibiting chronic hepatitis and liver cirrhosis.3 However, the pathological diagnosis of HCC from needle biopsy in these individuals presents numerous limitations.4 For example, it is difficult to distinguish nonmalignant hepatic nodules from an early HCC such as a small tumor with well differentiation grade.

Recent reports have suggested that epigenetic inactivation of gene expression by aberrant methylation on CpG islands may be a fundamental contributor to carcinogenesis and cancer progression.5–7 Clinical studies investigating associations between carcinogenesis and abnormal methylation on CpG islands in tumor suppressor genes or genes involved in cell proliferation and death have been conducted in various cancers.8–12 For HCC, it has been reported that some genes undergo hypermethylation in liver tissues,13–21 supporting the hypothesis that determination of methylation in specific genes may be useful for HCC diagnosis. However, to the best of our knowledge, the diagnosis of HCC, in particular early HCC, based on the methylation levels of a single gene has not been clinically investigated to date.13–21

The aim of the current study was to identify a combination of methylated marker genes suitable for use in the diagnosis of a small HCCs less than 3 cm HCC or a well-differentiated HCC as an early HCC, and to establish a user-friendly methylation-specific PCR (MSP) system to measure methylation status of these HCC gene markers. To achieve the aim, we selected 12 genes (APC, CASP8, CCND2, CFTR, CDKN2A, GSTP1, HIC1, POU3F1, RASSF1A, RUNX3, SPINT2 and TP73) with a clear methylation in >50% (actually 68–100%) of HCC cases examined in studies reported previously as genetic marker candidates for diagnosis of early HCC.13–21 We then analyzed the methylation status of these genes in HCC and non-tumor liver tissues using sequencing and MSP, and evaluated their use, individually and in combination, as a diagnostic tool for the detection of early HCC.

Material and methods


In the current study, we isolated genomic DNA from liver samples of 3 healthy individuals and 45 HCC patients. For 20 of the 45 HCC patients, we also isolated genomic DNA from non-tumor liver tissues. These samples were examined histologically and found to exhibit liver cirrhosis (n = 12), chronic hepatitis (n = 7) and no abnormalities (n = 1). All samples were taken from patients who underwent surgical treatment for liver tumors at Yamaguchi University Hospital between April 2001 and May 2003. Among these samples, 25 HCC samples were used as the filtering group (Table I), and the 3 normal liver, 20 HCC and the corresponding 20 non-tumor liver samples were used as the validation group (Table I), for the identification and comparison of the methylation status at CpG positions (methylation profiling). In particular, all of the validation samples were consecutively collected without any bias after collection of the filtering samples. The clinicopathological characteristics of these HCC patients based on the International Union against Cancer TNM classification22 are summarized in Table I. Most cases of small HCCs less than 3 cm and a half of well-differentiated HCCs were assigned to Stage I and II. Written, informed consent was obtained from each patient before surgery. The study protocol was approved in advance by the Institutional Review Board for Human Use at Yamaguchi University School of Medicine. Histopathological diagnosis of HCC for each patient was made post-surgery.

Table I. Patient Characteristics in This Study
PatientFiltering group1Validation group2Reference
HCC (n = 25)HCC (n = 20)HCC (n = 3)
  • 1

    These samples were used for filtering of candidate genes methylated aberantly in HCCs at early stage.

  • 2

    These samples were used for validation of the genes selected by filtering step.

  • 3

    Stage and histological grade were based on the International Union against Cancer TNM classitication.

Age (yrs)65.2 ± 8.467.3 ± 7.162.7 ± 6.1
Viral infection
 HCV positive2416
 HBV positive13
 HCV/HBV negative03
Primary lesion
 Single tumor1815
 Multiple tumors75
Tumor size (cm)2.9 ± 0.83.9 ± 3.3
Histological grade3

CpG mapping

For methylation analysis, we identified CpG islands in the promoter and exon 1 regions (defined as 1,500 bp upstream to 2,200 bp downstream from exon 1 start site), with the exception of CASP8, which does not contain any CpG islands, by CpG mapping analysis as described previously.23 For CASP8, several CpG positions scattered throughout intron 3 that were reported by Yu et al. were used for analysis.16

Analysis of DNA methylation by direct sequencing

Total genomic DNAs were extracted from liver samples using the DNA Isolation Kit for Cells and Tissues (Roche Diagnostics GmbH, Mannheim, Germany). One microgram of extracted DNAs was treated with bisulfite (BIS) for 5 hr at 50°C to convert unmethylated cytosine to uracil (detected as thymine after PCR) as described previously.24, 25 BIS-treated DNA was quantified as described previously,26 and then was prepared with 75 μl of 10 mM Tris/HCl-0.1 mM EDTA solution. After BIS processing, the regions containing short CpG-rich stretches on CpG islands for the 12 genes were amplified by PCR and the methylation status was examined by sequencing using BIS-treated DNA as a template. For the filtering group, 25 HCC tissues were analyzed by direct sequencing using dideoxy chain termination method. For genes surviving in the filtering step, comparative methylation profiles were generated by quantitative sequencing with the 20 tumor and corresponding non-tumor liver samples (validation group; see Table I) and 3 commercially available normal liver samples derived from healthy, Asian individuals (Tissue Transformation Technologies, Edison, NJ). Quantitative sequencing was conducted by pyrosequencing using the Pyrosequencer PSQ96MA and the Pyro Gold Reagents (Biotag AG, Uppsala, Sweden).27 The pyrosequencing results are shown as a Pyrogram (Fig. 1). As outlined in Figure 1, the peak height of cytosine and thymine correspond to the ratio of methylated and unmethylated cytosines at a CpG position. Therefore, the methylation rate (%) at each CpG position is given by the following formula: peak height of Cytosine/(peak height of Cytosine + peak height of Thymine) × 100. Furthermore, the average methylation rate for all of the examined CpG positions on candidate genes for each sample was calculated and subsequently used for the assessment of the methylation of marker genes.

Figure 1.

Representative pyrogram for methylation analysis. HCC and non-HCC tissues from an HCC patient (T7 in Fig. 2) were analyzed for SPINT2 gene. Y- and X-axes are luminescence strength and dispensation order (time), respectively. To discriminate methylated cytosine and unmethylated cytosine at the CpG position, T and C were dispensed continuously. Peak height was calculated by subtracting baseline from luminescence strength.

PCR amplification of short CpG-rich stretches

For sequencing analysis, the region containing short CpG-rich stretches on CpG islands was amplified by PCR using a primer set listed in Table II. The PCR reaction solution was composed of 26.7 ng of BIS-treated DNA, 2 units of rTaq DNA polymerase (TOYOBO, Osaka, Japan), which was pretreated with the equal volume of TaqStart™Antibody (Clontech Laboratories, Mountain View, CA) for 5 min at room temperature, 67 mM Tris/HCl (pH 8.8), 16.6 mM ammonium sulfate, 0.01% Tween 20, 200 μM dNTPs, each 1 μM of a primer pair, and 1.5 or 3 mM magnesium chloride in a final volume of 100 μl. DNA amplification was performed by using the GeneAmp PCR system 9600 (Applied Biosystems, Foster City, CA) by initial denaturation at 95°C for 2 min followed by 5 cycles of denaturation at 95°C for 25 sec, annealing at 70°C for 45 sec, extension at 72°C for 45 sec, and followed by 50 cycles of denaturation at 95°C for 25 sec, annealing at 55°C for 50 sec, extension at 72°C for 45 sec. The PCR products (361–551 bp) were concentrated using the SUPREC™-02 (TAKARA BIO, Otsu, Shiga, Japan) and subjected to electrophoresis on 2% agarose gel. The target bands of amplified products were excised from the gel and isolated using the QIAquick Gel Extraction Kit (QIAGEN GmbH, Hilden, Germany).

Table II. Primers Used for Amplication of the Regions Containing CpG Island
Gene symbol1Gene nameAccession no.2Methylation analysis on CpG islands
Bisulfite sequencing region3PCR primer setAmplified size (bp)
  • 1

    Gene symbols used are based on the data from LocusLink.

  • 2

    Accession number of each gene was obtained from PubMed or the Institute for Genomic Research database.

  • 3

    Position of bisulfite sequencing region was given relative to exon 1 start site.

Aberrantly methylated genes in HCC tissues
 APCAdenomatosis polyposis coliAFO38181−58 to +2215′-TTG TTT GTT GGG GAT TGG GGT-3′479
 CCND2Cyclin D2X68452−1287 to −8615′-TTT TTG GAG TGA AAT ATA TTA AAG G-3′427
 CFTRCystic fibrosis transmembrane conductance regulator, ATP-binding cassette (sub-family C, member 7)M96936−471 to −275′-GAG GAG GAG GAA GGT AGG TT-3′445
 GSTP1Glutathione S-transferase piU21689−223 to +1775′-GGG ATT TGG GAA AGA GGG AAA-3′400
 RASSF1ARas association (RalGDS/AF-6) domain family 1AAFO61836+25 to +4345′-TAG TTT GGA TTT TGG GGG AGG-3410
 RUNX3Runt-related transcription factor 3Z35278−393 to +395′-TGA TTT TGG AGG ATT TGT TTT GG-3′432
 SPINT2Serine protease inhibitor, kunitz-type, 2 (hepatocyte growth factor activator inhibitor 2; bikunin, placental)U78095−256 to +1055′-GGA AGG GTG GTA GGT GTT TAG-3′361
Sporadically methylated genes in HCC tissues
 CASP8Caspase 8, apoptosis-related cystein proteaseX98176+24786 to +252415′-GAG TTA GGG TGG TTA TTG AAA GT-3′456
 HICIHypermethylated in cancer IAI391567−1249 to −8065′-GGG TYG TTT TAG ATA AGA GTG TG-3′444
 POU3FIPOU domain class 3, transcription factor IL26494−749 to −3785′-ATT AGA GGA AGG AGA AGG AAA G-3′372
Unmethylated genes in HCC tissues
 CDKN2ACyclin-dependent kinase inhibitor 2A, p16(INK4)U26727−734 to −2445′-TTT TTA ATT AGG TTA TTA GGT TTA AG-3′491
 TP73Tumor protein p73Y16961−458 to +935′-TTT GGG GGA TAG TAG GGA GTT-3′551

Quantitative methylation-specific PCR

The methylation pattern of genes identified to be aberrantly methylated in HCC tissues by direct sequencing was also assessed by quantitative real-time MSP with primer pairs and TaqMan probes (Table III) specific to nucleotide sequence containing methylated cytosines at CpG positions. For CCND2, CFTR, RASSF1A and SPINT2, the PCR reaction mixture was composed of 5 μl of diluted BIS-treated DNA (1 ng), a primer pair and TaqMan probe in concentrations as shown in Table III in a final volume of 20 μl of 1 × Master mix from the LightCycler® TaqMan Master Kit (Roche Diagnostics GmbH). PCR amplification was undertaken using the PCR parameters shown in Table III on the LightCycler® II (Roche Diagnostics GmbH). For APC, the PCR reaction mixture was composed of 5 μl of diluted BIS-treated DNA (1 ng), 100 mM potassium acetate (pH 7.5), 50 mM Tricine (pH 8.3), 3 mM magnesium acetate, 375 μM dNTPs, 2.5% glycerol, 3 unit ZO5 thermostable DNA polymerase, a primer pair and TaqMan probe in concentrations shown in Table III, in a final volume of 20 μl. The PCR reaction mixture for RUNX3 was identical to that for APC except that 50 mM potassium acetate was used. For GSTP1 amplification, 4 mM magnesium acetate, 750 μM dNTPs and 3.33% glycerol were used. PCR amplification was performed with the PCR parameters as shown in Table III on the LightCycler® 2.0 (Roche Diagnostics GmbH). For all gene analyses, fluorescent signal was detected following the extension reaction for each cycle. Amplification of target gene was assessed by F1/F3 analysis mode in the LightCycler software, and the amount of methylated DNA in 75 μl of BIS-treated DNA solution was quantified using a standard curve constructed with simultaneously measured standards made from a dilution series of artificially methylated DNA (CpGenome™ Universal Methylated DNA, CHEMICON International, CA). The dilution series was 1000, 200, 40 and 4 pg/μl for CCND2, RASSF1A and SPINT2, and 200, 50 and 20 pg/μl for CFTR, APC, RUNX3 and GSTP1.

Table III. Primers, TaqMan Probes and PCR Parameters Used for MSP
GenePrimer pairConcentration (μM)Size of PCR product (bp)TaqMan probeConcentration (μM)PCR parameter
Denaturation (1 cycle)Amplification (50 cycle)Cooling (1 cycle)
APC5′-AGTGCGGGTCGGGAAGC-3′0.51275′-FAM-AAAACGCCCTAATCCGCATCCAACG-TAMRA-3′0.2595°C, 2 min95°C, 10 sec40°C, 30 sec
CCND25′-TTT GAT TTA AGG ATG CGT TAG AGT ACG-3′0.25855′-FAM-AAT CGC CGC CAA CAC GAT CGA CCCTA-TAMRA-3′0.195°C, 10 min95°C, 10 sec40°C, 30 sec
72°C, 5 sec
CFTR5′-AATCGGGAAAGGGAGGTGC-3′0.61395′-FAM- TCCACCCACTACGC ACCCCCG-TAMRA-3′0.195°C, 10 min95°C, 10 sec40°C, 30 sec
GSTP15′-ATAAGGTTCGGAGGTCGCGA-3′0.51425′-FAM-ATCCCCGAAAACGCG AACCGCG-TAMRA-3′0.062595°C, 2 min95°C, 15 sec40°C, 30 sec
RASSF1A5′-GCG TTG AAG TCG GGG TTC-3′0.25755′-FAM-ACA AAC GCG AAC CGA ACG AAA CCA-TAMRA-3′0.195°C, 10 min95°C, 10 sec40°C, 30 sec
72°C, 5 sec
SPINT25′-TCG GTT ATT TTCGGG AGT C-3′0.251185′-FAM-CCA ACG CGC GAA A AT CGC CAA AA-TAMRA-3′0.195°C, 10 min95°C, 10 sec40°C, 30 sec
5′-CGC CTA CGA CAC TCA ACG A-3′63°C, 45 sec
72°C, 5 sec

Statistical analysis

Significant differences in methylation rate between HCC and non-tumor liver tissues in the validation group were analyzed by Student's t-test, Mann-Whitney's U-test and Welch's t-test. Correlations between methylation rates in HCC tissues and patient characteristics, and among the methylation rates for candidate genes, were evaluated with Pearson's correlation coefficient test. Correlations between the methylation status analyzed by pyrosequencing and MSP were also evaluated with Pearson's correlation coefficient test. The ability to discriminate between HCC and non-HCC tissues was evaluated by the Fisher criterion as described previously.28 The Fisher criterion measures the difference between 2 means normalized by the average variance and in this study represents the ability of a gene marker to discriminate between HCC and non-HCC liver samples. Therefore, we calculated the Fisher criterion for the 7 methylated genes in the validation group and ranked these 7 genes in order of decreasing magnitude of the Fisher criterion. We used the Statcel Ver.2 software for statistical analysis and a value of p < 0.05 was considered to be statistically significant.


Filtering of genes hypermethylated in early HCCs by direct sequencing after BIS treatment

For samples in the filtering group (Table I), we performed in silico CpG mapping and direct sequencing after BIS treatment to filter 12 potential marker genes. APC, CASP8, CCND2, CFTR, CDKN2A, GSTP1, HIC1, POU3F1, RASSF1A, RUNX3, SPINT2 and TP73 were the genes filtered and were selected from previously reported data.13–21 With the exception of CASP8, all genes were found to contain CpG islands located in their promoter regions by CpG mapping analysis, and the regions including their short CpG-rich stretches were successfully amplified for methylation analysis by sequencing. For CASP8, the region located in intron 3 that contained several CpG dinucleotide sequences was amplified (Table II). These 12 genes were then subjected to direct sequencing after BIS treatment. Among the candidate genes, 7 genes of APC, CCND2, CFTR, GSTP1, RASSF1A, RUNX3 and SPINT2 were found to be aberrantly methylated in 21 (84%), 13 (52%), 25 (100%), 21 (84%), 23 (92%), 19 (76%), 22 cases (88%) of the 25 cases of Stages I and II HCCs less than 5 cm in the filtering group, respectively. The remaining 5 genes were excluded from the list of potential HCC marker genes as they were found to be either sporadically methylated (CASP8, HIC1 and POU3F1), or were completely unmethylated (CDKN2A and TP73) in HCC tissues (data not shown).

Validation of selected methylated genes by pyrosequencing after BIS treatment

The 7 candidate gene markers were further examined by pyrosequencing (comparative quantitative methylation analysis) in 20 HCC and 20 corresponding non-tumor liver tissues in the validation group. As shown in Figure 2, this methylation profiling revealed that the 7 genes were extensively highly methylated in HCC tissues but not in any of non-HCC liver tissues. In addition, the methylation rate of the CpG positions examined for each gene was significantly higher (APC, GSTP1 and CFTR: p < 0.0001, RUNX3: p < 0.001, SPINT2 and RASSF1A: p < 0.01, CCND2: p < 0.05) in HCC tissues than in non-HCC tissues. One exception to this rule was a single CpG site at position (−1080) on CCND2 that was not significantly hypermethylated in HCCs. In addition, all CpG positions in APC, GSTP1 and CFTR were more highly methylated in small HCCs less than 3 cm than non-tumor liver tissues (p < 0.05). Intriguingly, the CpG methylation rate, with the exception of SPINT2, uniformly indicated a trend to synchronize among HCC, well-differentiated HCC and non-HCC tissues as shown by the line graphs in Figure 2. That is, the identical CpG positions of the 6 genes may be commonly susceptible or insusceptible to methylation through HCC and non-HCC liver tissues. The synchronized patterns of methylation rate on the 6 genes and a lack of synchronization for SPINT2 were similarly observed in small HCCs less than 3 cm and Stages I and II HCCs (data not shown).

Figure 2.

Comprehensive quantitative methylation profiling in HCC and non-tumor liver tissues for 7 genes by pyrosequencing. Methylation patterns for 7 genes with HCC and non-tumor liver tissues in second round methylation analysis are shown; 20 HCC and the corresponding non-HCC tissues from 20 HCC patients in second round methylation analysis (refer in Table I) were used for the analysis. Methylation rates were shown in different colors (red for >75%, orange for 75–50%, pale orange for 50–30% and white for <30%). HCC categories were indicated in light blue, light green and pink for well-differentiated HCCs, small HCCs less than 3 cm and stages I and II HCCs, respectively. Graphs are showing the average of methylation rates at each CpG positions. Open circle (○) and filled circle (•) show the average of methylation rates (±SEM) in HCC and non-HCC liver tissues in all 20 HCC cases, respectively. Open triangle (▵) and filled triangle (▴) show the average of methylation rates (±SEM) in HCC and non-HCC liver tissues in 10 well-differentiated HCCs, respectively.

We next evaluated the diagnostic power of the 7 marker genes using the average value of methylation rates for each patient in the validation group. All the 7 marker genes demonstrated a significant difference in the extent of gene methylation between tumor and non-tumor liver tissues in HCC patients (p < 0.00001, Fig. 3). Notably, RASSF1A, CCND2 and APC were found to be more frequently positive in 19 (95%), 19 (95%) and 18 (90%) of 20 HCC tissues, respectively, as compared with the remaining 4 genes, when a cut-off value of 30% methylation rate as estimated by pyrosequencing analysis was defined. This result was observed despite the fact that several false positive cases were identified in non-HCC tissues. In addition, RASSF1A, CCND2 and APC were frequently positive (91–100% of cases examined) in well-differentiated HCCs, small HCCs less than 3 cm, and Stages I and II HCCs (Fig. 3). The aberrant methylations of GSTP1, CFTR and SPINT2 were positive in 16 (80%), 16 (80%) and 12 (60%) of 20 HCC tissues, respectively, but not for any of the corresponding non-tumor liver tissues. When the cut-off value of 30% was defined, the sensitivity to early HCC was 73–80%, 70–82% and 33–50% for GSTP1, CFTR and SPINT2, respectively. Thus, the 7 genes demonstrated abnormal methylation at higher rates in early HCC tissues. When the total number of positive genes were analyzed in individual cases, all 7 genes were positive in only 1 (11%) of 9 small HCCs less than 3 cm, but in 9 (82%) of 11 large HCCs over 3 cm (p < 0.01). The number of positive genes for large HCCs over 3 cm was significantly larger (p < 0.05) than for small HCCs less than 3 cm (>3 cm HCCs vs. ≤3 cm HCCs (mean): 6.0 vs. 5.2). In particular, all 7 genes were positive for 5 HCCs over 4 cm. There was no significant correlation with the histological grade of differentiation or staging. APC, CCND2, GSTP1 and RASSF1A were also found to be highly methylated in healthy individuals (Fig. 3).

Figure 3.

Comparison of the methylation rates in HCC and non-tumor liver tissues among 7 genes by pyrosequencing. Average values of methylation rates of all the CpG positions in Figure 2 were calculated for each patient. Methylation rates were shown in different colors (red for >75%, orange for 75–50%, pale orange for 50–30% and white for <30%). HCC categories were indicated in light blue, light green and pink for well-differentiated HCCs, small HCCs less than 3 cm and stages I and II HCCs, respectively.

Validation of developed quantitative MSP

The next component of our study involved the establishment of MSP systems based on the methylation profiles identified by direct sequencing to evaluate the ability of marker genes to discriminate between HCC and non-HCC samples. We quantified the levels of methylated DNA by quantitative real-time MSP for 20 pairs of HCC and the corresponding non-tumor liver tissues in the validation group. Methylated DNA levels in 75 μl of BIS-treated DNA solution measured by MSP were shown to correlate highly with the methylation rate estimated by sequencing analysis (Pearson's correlation coefficients, r = 0.830, r = 0.773, r = 0.770, r = 0.742, r = 0.731, r = 0.663 and r = 0.652 for SPINT2, RUNX3, APC, CFTR, RASSF1A, CCND2 and GSTP1, respectively; p < 0.00001 for all genes, Fig. 4). This result suggests that the methylation rate estimated using direct sequencing may be adequately and reliably measured using our easy-to-use MSP system.

Figure 4.

Correlation between the methylated DNA amount by quantitative real-time MSP and the methylation rate by pyrosequencing for 7 genes. Y- and X-axes are the methylated DNA amount (pg/μl) and the methylation rate (%), respectively.

Discrimination between HCC and non-HCC by developed quantitative MSP

We evaluated the ability to discriminate HCC from non-HCC using methylated DNA amounts calculated by the quantitative MSP systems. Firstly, we confirmed the distribution of 20 pairs of HCC and non-HCC liver samples in the validation group. The methylation analysis using quantitative MSP for each gene is summarized in Figure 5. We found that quantitative MSP developed in our laboratory permitted the identification of HCC and non-HCC liver samples with a significantly high degree of separation (p < 0.05). Next, we examined the diagnostic power of our marker genes using 44 HCC and 20 non-HCC liver tissues from the filtering and validation groups. For this step in the evaluation, we eliminated GSTP1, as the MSP system did not adequately amplify the target DNA in the filtering group unlike in the validation group. For the remaining 6 genes, the ability to distinguish between the HCCs and non-HCC samples was assessed using the Fisher criterion (Table IV). Among the 6 genes, RASSF1A yielded the largest Fisher criterion, whereas CCND2 and SPINT2 yielded the second and third largest Fisher criteria. The Fisher criteria of RUNX3, GSTP1 and APC were slightly lower than that for SPINT2. CFTR demonstrated the lowest Fisher criterion among the 6 gene markers.

Figure 5.

Discrimination between HCC and non-tumor liver tissues by quantitative real-time MSP assays for 7 genes. Circles indicate methylated DNA amounts estimated by MSP assays. Bars and dashed lines indicate the median methylated DNA amounts and the cut-off values (non-HCC, mean + 1.96 × SD, two-sided 95% confidence interval), respectively. *p < 0.00001, **p < 0.0001, ***p < 0.001, ****p < 0.05.

Table IV. Diagnostic Ability of 6 Methylated Genes and Their Combinations Based on Quantitative MSP Analysis
 Fisher criterion1Cut-off valueHCC category
All n = 44Well-differentiated n = 20Small size less than 3 cm n = 28TNM stages I and II n = 35
Sensitivity (%)Specificity (%)Accuracy (%)Sensitivity (%)Specificity (%)Accuracy (%)Sensitivity (%)Specificity (%)Accuracy (%)Sensitivity (%)Specificity (%)Accuracy (%)
  • 1

    Fisher criterion was calculated as described previously (Ref. 29).

Single marker
Multiple markers

The clinical performance of each marker gene was then evaluated using the sensitivity in diagnosing HCC with a cut-off value of mean methylated DNA levels in the 20 non-HCC tissues plus 1.96 times the standard deviation (mean + 1.96 × SD, 2-sided 95% confidence interval) as described previously.21 The diagnostic ability of the 6 methylated genes and their combinations based on quantitative MSP analysis is shown in Table IV. The sensitivity of each marker correlated well coincident with the magnitude of the Fisher criterion in early HCCs and the total 44 HCCs. This was especially the case for CCND2, which demonstrated a very high performance in well-differentiated HCCs (85% sensitivity, 100% specificity and 90% accuracy). RASSF1A and SPINT2 also demonstrated relatively high performance levels in well-differentiated HCCs (≥70% sensitivity, 100% specificity and ≥80% accuracy) compared to the other genes. In addition, RASSF1A and CCND2 indicated much higher performance in small HCCs less than 3 cm (≥75% sensitivity, 100% specificity and ≥80% accuracy). The results for all 6 genes were negative in the 3 normal liver samples isolated from healthy individuals when assessed by quantitative MSP analysis.

We further assessed the clinical performance of these potential marker genes in HCC diagnosis by investigating whether a combination of genes may increase the sensitivity of HCC diagnosis. In our study, we found that when a combination of 2 or 3 genes demonstrated positive levels of DNA methylation, samples were finally diagnosed as HCC. In contrast, when all genes were found to be negative for DNA methylation, samples were judged to be non-HCC. In addition, these calculated performances were assessed by Youden's index (sensitivity+specificity−1) and the combinations were ranked in order of decreasing Youden's index values. As shown in Table IV, with 2- or 3-markers combinations of RASSF1A, CCND2 and SPINT2, in that order, yielded the highest Fisher criteria, demonstrated much higher clinical performances than for a single marker alone. Notably, the 3-markers combination of RASSF1A, CCND2 and SPINT2 demonstrated the highest sensitivity and accuracy (89–95% and 89–97%) for all 44 HCCs and 3 categories of early HCCs, and especially correctly diagnosed all HCC cases in the 3 early HCC groups. This 3-markers combination was ranked in the top position for all HCCs and 3 categories of early HCCs. In regards to 2-markers combinations, RASSF1A and CCND2 achieved the most accurate diagnosis for well-differentiated HCCs and small HCCs less than 3 cm (93–95% sensitivity, 100% specificity and 95–97% accuracy). The combination of RASSF1A and SPINT2 showed much higher specificity, and was particularly accurate in the detection of early HCCs (100% specificity). The combination of CCND2 and SPINT2 also resulted in a diagnostic performance that was higher than a single marker. These 2-markers combinations were consistently ranked in the top 3 for all the 4 HCC categories. Although the number of positive genes in MSP analysis was not associated with tumor size, differentiation and staging, it did significantly correlate with that achieved by direct sequencing (p < 0.005).


The current study successfully identified several genes suitable for the diagnosis of early HCC using a specifically developed comparative quantitative methylation profiling system. Among a total of 12 genes tested, 7 genes were able to accurately discriminate between HCC and non-HCC (non-neoplastic) liver samples based on their frequency and rate of hypermethylation. Although previous studies have reported hypermethylation in non-neoplastic liver for some of these genes,13, 15, 16, 19, 29 hypermethylation was not observed in such samples in the present study. This may be due to differences in DNA regions subjected to methylation analysis, or be the result of our filtering step, whereby the identification of candidate marker genes was focused on early HCCs. Indeed, the remaining 5 genes (CASP8, CDKN2A, HIC1, POU3F1 and TP73) were eliminated from our study following this filtration step.

Of the eliminated genes, it has been extensively reported that CDKN2A, the gene encoding the tumor suppressor p16, is frequently and highly methylated in HCC.15, 29–32 However, our results showed that CDKN2A remained unmethylated in 2 independent groups of HCC samples, the filtering and validation groups (data not shown), indicating poor performance in the detection of early HCC. These differences in methylation status may be due to differences in the location of CpG islands analyzed in both the present and previous studies.15, 29–31, 33, 34 In our present study, we analyzed short CpG-rich stretches on CpG islands upstream to those previously reported. Alternatively, CDKN2A may simply be poorly methylated in early HCC, the particular stage of HCC progression featured in our present study. In this regard, our data is supported by the observation that CDKN2A remains largely unmethylated in well-differentiated HCC when assessed using pyrosequencing.34 Furthermore, this result is not surprising given that the majority of previous methylation studies were not quantitative and were not conducted with the aim of identifying practical methylation markers for the diagnosis of early HCC. Thus, our systematic strategy of focusing on small, well-differentiated HCC appears to be more appropriate for the identification of methylation markers that could be potentially used for the diagnosis of early HCC.

Our developed quantitative MSP assay revealed that when RASSF1A and CCND2 were used as a single marker, they demonstrated a superior ability to separate HCC and non-HCC samples compared to other genes, and provided a much greater sensitivity in diagnosis of all HCCs and early HCCs (Table IV). Although PCR amplification of RASSF1A and CCND2 was observed in 16 and 13 cases of 20 non-HCC tissues, respectively (Fig. 5), these genes appeared capable of distinguishing between HCC and non-HCC tissues. When an adequate cut-off value is used, these 2 genes may prove very useful as high-sensitivity molecular markers for diagnosis of early HCC regardless of the stage of HCC progression. Indeed, RASSF1A and CCND2 demonstrated high diagnostic powers when our cut-off values were defined (Table IV). In the present study, however, RASSF1A, CCND2 and APC were also methylated in the DNA from 3 healthy individuals, though the methylation level was lower than that in HCC tissues in pyrosequencing analysis and the methylated DNA amount was lower than the cut-off value in MSP analysis. It could be coming from the difference of sample sources or qualities, and cause false positive results in routine testing for diagnosis of HCC. In contrast, SPINT2, GSTP1 and CFTR were negative in any of the corresponding non-HCC tissues using sequencing analysis, although their sensitivities (33–82%) were lower than RASSF1A, CCND2 and APC. The finding that SPINT2 and CFTR were highly unmethylated in normal liver from healthy individuals (Fig. 3) and in liver from HCV carriers (data not shown) indicates a great potential for molecular markers to be highly specific for HCC diagnosis. In this regard, known methylated genes, including SOCS1 and HIC-1, may prove unsuitable for HCC diagnosis due to their high methylation rate in non-tumor liver tissues.35 In the present study, only SPINT2 showed a complete lack of synchronized methylation trend between HCC and non-tumor tissues, suggesting that it may be highly and more specifically methylated in HCC than the other 6 genes examined in this study. This novel finding obtained from our detailed methylation analysis supports the use of SPINT2 as a molecular marker for HCC. Indeed, quantitative MSP indicated good specificity of SPINT2 in diagnosing HCC (Table IV). In addition, methylation of SPINT2 has not been identified in malignancies other than kidney and liver cancers,18, 36 while the other genes have been found to be hypermethylated in various cancers.8–12 Thus, SPINT2 may be a valuable marker that exhibits a high specificity for HCC and early HCCs.

Distinct hepatitis virus infection might affect gene expression patterns of HCC via altered methylation status.28 This concept is supported by a quantitative methylation study of Nishida et al.35 in which several genes were more frequently methylated in HCV-related HCC than in other types of HCC. In this regard, our current study used a cohort consisting of majority of HCV-related HCC. However, our quantitative MSP showed no significant differences in methylation of the 6 genes between HCV-related HCC (n = 39) and HCV-unrelated HCC (n = 5) (data not shown), and those 6 genes were not included in the list of HCV-related methylation genes by Nishida et al.35 Rather, the 6 methylation genes surviving in our selection process might be related to a common pathogenetic mechanism underlying hepatocarcinogenesis.

Pyrosequencing is an excellent tool to directly and quantitatively estimate methylation status at CpG sites and was therefore the main method of methylation analysis in our study. However, pyrosequencing is not suitable for the comprehensive search of methylation markers or use in routine clinical testing, as it involves many complicated steps and troublesome manipulations. In addition, even completely unmethylated samples were always given around 10–30% methylation rate by pyrosequencing when they were not amplified by MSP (Fig. 4). This means that MSP can quantitatively reflect the methylation status in broad range, but pyrosequencing may show poor quantitative results in low range. Therefore, to address the second aim of this study, we integrated the information obtained from pyrosequencing into a PCR technique and established a user-friendly MSP system that measured the methylation status of our 7 genes. The development of the system was essential, as the results reported in previous studies by sequencing were insufficiently validated in clinical samples, and were limited to examination in a few patients and cell lines by non-quantitative sequencing analysis after cloning into plasmid vector DNA.16–20, 37 Moreover, in most reports the methylation status in HCC and non-HCC tissue samples has been evaluated by MSP assay without validation by sequencing, and is largely based on visual judgment of ambiguous PCR products amplified by non-quantitative MSP on agarose gels.13, 15, 20, 34 Therefore, the methylation status estimated by previous MSP assays may not represent the actual methylation profiles in clinical samples. In that respect, our novel MSP system may provide more accurate and reliable methylation data in the future.

Most HCCs are initiated as minute nodules in which cells strongly retain the phenotype of benign hepatocytes. Dysplastic changes, which mark the transition to small, well-differentiated HCCs, usually occur when the nodules measure between 1 and 2 cm in maximal diameter and herald the ability to proliferate, invade and disseminate.38 At this stage, the tumors are easily detected by imaging techniques including ultrasonography and computed tomography; however, it is not easy to discriminate pathologically between well-differentiated HCC and other benign tumors such as dysplasia. Accordingly, the best way for an accurate diagnosis of this particular cancer may be to identify molecular changes that govern the transition to cancer. In this regard, we have successfully identified multiple-markers combinations of 2 or 3 genes, including RASSF1A, CCND2 and SPINT2, that may prove both sensitive and specific enough for the diagnosis of small, well-differentiated HCC, in addition to advanced HCC.

In conclusion, early HCCs can be accurately diagnosed using real-time MSP systems developed in our laboratory. Our present data also show that multiple-markers combinations can increase diagnostic potential for early HCC. However, it remains unclear whether the optimal combination of RASSF1A, CCND2 and SPINT2 is the most robust for diagnosis of HCC because our current study was not completely blinded and used only a small number of samples that were obtained from the same institute. Further large studies will be required to gain new insights into our current findings and to identify a new combination of methylated genes with higher accuracy for diagnosis of early HCC.


The authors are grateful to Dr. Markus Stark for his help in CpG mapping and designing TaqMan® oligonucleotides. The authors are grateful to Drs. Andrea Johnson and Brian Rhees for their review of the manuscript.