The methylation status of RASSF1A promoter predicts responsiveness to chemotherapy and eventual cure in hepatoblastoma patients

Abstract

Despite the progress of therapy, outcomes of advanced hepatoblastoma patients who are refractory to standard preoperative chemotherapy remain unsatisfactory. To improve the mortality rate, novel prognostic markers are needed for better therapy planning. We examined the methylation status of 13 candidate tumor suppressor genes in 20 hepatoblastoma tumors by conventional methylation-specific PCR (MSP) and found hypermethylation in 3 of the 13 genes. We analyzed the methylation status of these 3 genes (RASSF1A, SOCS1 and CASP8) in 97 tumors and found hypermethylation in 30.9, 33.0 and 15.5%, respectively. Univariate analysis showed that only the methylation status of RASSF1A but not the other 2 genes predicted the outcome, and multivariate analysis showed a weak contribution of RASSF1A methylation to overall survival. Using quantitative MSP, we found RASSF1A methylation in 44.3% of the 97 tumors. CTNNB1 mutation was detected in 67.0% of the 97 tumors. While univariate analysis demonstrated RASSF1A methylation, CTNNB1 mutation and other clinicopathological variables as prognostic factors, multivariate analysis identified RASSF1A methylation (p = 0.043; relative risk 9.39) and the disease stage (p = 0.002; relative risk 7.67) but not CTNNB1 mutation as independent prognostic factors. In survival analysis of 33 patients in stage 3B or 4, patients with unmethylated tumor had better overall survival than those with methylated tumor (p = 0.035). RASSF1A methylation may be a promising molecular-genetic marker to predict the treatment outcome and may be used to stratify patients when clinical trials are carried out. © 2008 Wiley-Liss, Inc.

Hepatoblastoma is a rare malignant neoplasm of the liver, with an incidence of 0.5–1.5 per million children.1 Remarkable progress in clinical outcome has been achieved in the past 20 years due to advances in chemotherapy and surgical procedures; however, the mortality rate remains 20–30% and treatment results in patients in advanced stages who are refractory to standard preoperative chemotherapy regimens are unsatisfactory.2, 3 To improve the mortality of these patients, innovative treatment and potent prognostic markers for better therapy planning are needed. The present clinical factors predicting outcome include the level of alpha-feto protein, histology, disease stage and growth pattern of the tumor.2–4 Chromosomal gains of 2q, 8q and 20 and high expression of telomerase or PLK1 were shown to be molecular-genetic markers predicting poor outcome5–8; however, none have been proven to be independent prognostic factors by multivariate analysis.

We previously reported that RASSF1A (RAS association domain family protein 1) methylation, found in 39% of 39 hepatoblastoma tumors, was correlated with poor outcome by univariate analysis.9 Nevertheless, the article had some limitations that the number of tumors was not enough, the method used to detect the hypermethylation was suboptimal, and the prognostic significance of RASSF1A methylation was ambiguous by multivariate analysis.

CTNNB1 (catenin, beta-1) mutation was reported in the majority of hepatoblastoma tumors, but reports on alterations of other oncogenes or tumor suppressor genes are rare.10–12 Thus, we thought that epigenetic silencing of tumor suppressor genes might be involved in the tumorigenesis of hepatoblastoma and examined the methylation status of 13 candidate tumor suppressor genes, whose aberrant methylation has previously been shown in various cancers.13–22 Conventional methylation-specific PCR (MSP) analysis showed hypermethylation in only 3 of the 13 genes, RASSF1A, SOCS1 (suppressor of cytokine signaling 1) and CASP8 (caspase-8) genes, but not in the remaining 10 genes. We examined the correlation of the methylation status of the 3 genes with various clinical characteristics in a substantial number of hepatoblastoma tumors. Furthermore, we analyzed the methylation status of RASSF1A by more sensitive quantitative MSP and verified the prognostic implication of methylation by multivariate analysis. We suggest that RASSF1A may be a promising molecular-genetic marker predicting treatment outcome that may be used to stratify hepatoblastoma patients when clinical trials are carried out.

CASP8, caspase-8; CI, confidence interval; CR, complete response; CTNNB1, catenin, beta-1; JPLT, the Japanese Study Group for Pediatric Liver Tumor; MSP, methylation-specific PCR; NC, no change; PR, partial response; RASSF1A, RAS association domain family protein 1; RR, relative risk; SOCS1, suppressor of cytokine signaling 1.

Material and methods

Patients and samples

Tumor tissues were obtained from 97 Japanese children with hepatoblastoma and adjacent normal liver tissues were available from 3 patients. Nonmatched normal liver tissues were also obtained from 5 other hepatoblastoma patients who were not included in the present clinicopathological study. Thirty-five of 39 specimens in the previous report were included; 4 were excluded because of the lack of DNA and 62 were supplied by the Tissue Bank of the Japanese Study Group for Pediatric Liver Tumor (JPLT).23 The median age of the 97 patients at diagnosis was 16 months (range, 2–177 months).

The clinical stage of the disease was determined at the time of initial biopsy or surgery according to the classification of the Japanese Society of Pediatric Surgeons.24 While most tumors in stages 1 and 2, and those in 3A, occupying 3 segments of the liver, are completely resectable, tumors in stage 3B, occupying 4 segments of the liver, and those in stage 4 are not. The extent of disease was distributed in stage 1 in 6 tumors, in 2 in 33, in 3A in 25, in 3B in 11 and in 4 in 22. Patients were treated at various hospitals or institutions, mostly under the framework of JPLT-1 (1991–1999) or JPLT-2 (2000–2006) protocols.23, 25 The protocols include pre- and postoperative chemotherapy with cisplatin and THP-adriamycin. Complete response (CR) was defined as the complete disappearance of tumor, and partial response (PR) as at least a 50% reduction of tumor. No change (NC) was defined as a decrease of less than 50% or an increase of tumor.25 Seventy-two patients underwent preoperative chemotherapy, and one underwent salvage liver transplantation. The median follow-up of survivors was 66 months (range, 9–175 months). The PRETEXT system is based on hepatic surgical anatomy, described elsewhere.26 The pathological classifications of hepatoblastoma by Haas et al. and the Japanese Society of Pathology divide hepatoblastoma into 2 major subtypes, namely the well-differentiated (fetal) type and the poorly differentiated (embryonal) type.4, 24

Bisulfite treatment and conventional methylation-specific PCR (MSP) analysis

Genomic DNA from tumor samples was treated with sodium bisulfite, and the methylation status of the promoter region in various genes was analyzed by MSP, as previously described.9, 27 The genes examined were RASSF1A, RASSF2A, NORE1A, SOCS1, CASP8, RUNX3, RIZ1, BLU, HOXA9, HOXB5, p16INK4A, p14ARF and DCR2.13–22 The primer sequences and their location in the original genomic sequences are listed in Table I, and the location of the analyzed fragments for RASSF1A, SOCS1 and CASP8 are shown in Figure 1a. While the primer sequences of RASSF1A are located in the promoter region, those of CASP8 and SOCS1 are derived from the exon 4-intron 4 region and the exon 1, respectively, because the methylation status of these regions is correlated with the expression.15, 20, 30 CpGgenome™ Universal Methylated DNA (Chemicon International, Temecula, CA) and normal lymphocyte DNA were used as controls for methylated or unmethylated templates, respectively. PCR products were run on 2% agarose gels and visualized after staining with ethidium bromide.

Figure 1. (a) The location of the RASSF1A, CASP8 or SOCS1 fragment analyzed by the conventional or quantitative (MethyLyte) MSP method is shown as horizontal arrows. The transcription start site of each gene is shown as a bent arrow. (b) Examples of methylation status using conventional methylation-specific PCR. PCR products of methylated or unmethylated RASSF1A, CASP8 and SOCS1 from hepatoblastoma tumors are shown. M, methylated products; U, unmethylated products.

Table I. Primer Sequence, Genomic Position, MSP Condition and Product Size

Primer name	Primer sequence	Genomic position1	Annealing temp. (°C)	Product size (bp)	Ref.
UF, unmethylated forward primer; UR, unmethylated reverse primer; MF, methylated forward primer; MR, unmethylated reverse primer. 1 The 5′ position of the sense unmethylated or sense methylated primer sequences is numbered relative to the transcription start site of the gene concern. 2 The number indicates the location relative to the transcription start site of CASP8 transcript variant B (NM_033355.2). 3 Designed for bottom strand.
Quantitative MSP
ACTB-F	5′-TGGTGATGGAGGAGGTTTAGTAAGT	−1596	60	133	28
ACTB-R	5′-AACCAATAAAACCTACTCCTCCCTTAA
TaqMan probe	5′-6FAM-TGTGTTTGTTATTGTGTGTTGGGTGGTGGT-TAMRA-3′
RASSF1A-F	5′-GGTTTTGCGAGAGCGCGT	−72	62	168	29
RASSF1A-R	5′-GCTAACAAACGCGAACCGAAC
TaqMan probe	5′-6FAM-GGAGGCGTTGAAGTCGGGGTT-TAMRA-3′
Conventional MSP
RASSF1A-UF	5′-GGGGTTTTGTGAGAGTGTGTTTAG	−74	63	175	30
RASSF1A-UR	5′-TAAACACTAACAAACACAAACCAAAC
RASSF1A-MF	5′-GGGTTTTGCGAGAGCGCG	−73	63	169
RASSF1A-MR	5′-GCTAACAAACGCGAACCG
BLU-UF	5′-TTGTTTGGATTTAGGTGTGAGTT	−73	58	160	18
BLU-UR	5′-CAAAAACAACAAACCCCAACA
BLU-MF	5′-CGTTCGGATTTAGGCGCGAGTT	−72	68	158
BLU-MR	5′-GAAAACGACGAACCCCGACGA
CASP8-UF	5′-TAGGGGATTTGGAGATTGTGA	+3082	55	321	20
CASP8-UR	5′-CCATATATATCTACATTCAAAACAA
CASP8-MF	5′-TAGGGGATTCGGAGATTGCGA	+3082	58	320
CASP8-MR	5′-CGTATATCTACATTCGAAACGA
DCR2-UF	5′-TTGGGGATAAAGTGTTTTGATT	+101	58	146	21
DCR2-UR	5′-AAACCAACAACAAAACCACA
DCR2-MF	5′-GGGATAAAGCGTTTCGATC	+104	59	139
DCR2-MR	5′-CGACAACAAAACCGCG
HOXA9-UF	5′-TAATAGTGTGTGGAGTGATTTAT	−124	56	94	22
HOXA9-UR	5′-TAATAAATTACCAACACCCA
HOXA9-MF	5′-GCGTTTGGTTCGTTCGGTTC	−613	64	123
HOXA9-MR	5′-CAATAAAAACGCGAACGCCG
HOXB5-UF	5′-TGAATTGGTTTTAATGATTTTTGGATT	−217	53	117	19
HOXB5-UR	5′-TTAAAAAATCACATACTTTTATTAACCAATCA
HOXB5-MF	5′-AATCGGTTTTAACGATTTTCGGATC	−215	53	113
HOXB5-MR	5′-AAAAAATCACGTACTTTTATTAACCAATCG
NORE1A-UF	5′-ATTTATATTTGTGTAGATGTTGTTTGGTAT	−176		214	14
NORE1A-UR	5′-ACTTTAACAACAACAACTTTAACAACTACA
NORE1A-MF	5′-CGTCGTTTGGTACGGATTTTATTTTTTTCGGTTC	−159		202
NORE1A-MR	5′-GACAACTTTAACAACGACGACTTTAACGACTACG
p14ARF-UF	5′-GGAATAGGGGAGTGGGGAT	−388	60	144	22
p14ARF-UR	5′-AATAACAACCCAAAAACCAAACA
p14ARF-MF	5′-GGAATAGGGGAGCGGGGAC	−388	60	144
p14ARF-MR	5′-GATAACGACCCAAAAACCGAACG
p16INK4A-UF	5′-TTATTAGAGGGTGGGGTGGATTGT	+133	63	151	27
p16INK4A -UR	5′-CAACCCCAAACCACAACCATAA
p16INK4A -MF	5′-TTATTAGAGGGTGGGGCGGATCGC	+133	63	150
p16INK4A -MR	5′-GACCCCGAACCGCGACCGTAA
RASSF2A-UF	5′-GAAGGTGTTTTTATTTTATTTTTGG	+684	59	156	13
RASSF2A-UR	5′-AAAACCTACCTCTAAAAAATCCACC
RASSF2A-MF	5′-GTTCGTCGTCGTTTTTTAGGCG	+798	60	109
RASSF2A-MR	5′-AAAAACCAACGACCCCCGCG
RIZ1-UF	5′-TGGTGGTTATTGGGTGATGGT	−4782		177	17
RIZ1-UR	5′-ACTATTTCACCAACCCCAAGA
RIZ1-MF	5′-GTGGTGGTTATTGGGCGACGGC	−4781		176
RIZ1-MR	5′-GCTATTTCGCCGACCCCGACG
RUNX3-UF	5′-ATAATAGTGGTTGTTAGGGTGTTG	+2983	60	115	16
RUNX3-UR	5′-ACTTCTACTTTCCCACTTCTCACA
RUNX3-MF	5′-ATAATAGCGGTCGTTAGGGCGTCG	+2983	60	115
RUNX3-MR	5′-GCTTCTACTTTCCCGCTTCTCGCG
SOCS1-UF	5′-TTATGAGTATTTGTGTGTATTTTTAGGTTGGTT	+1072	60	175	15
SOCS1-UR	5′-CACTAACAACACAACTCCTACAACAACCA
SOCS1-MF	5′-TTCGCGTGTATTTTTAGGTCGGTC	+1081	60	160
SOCS1-MR	5′-CGACACAACTCCTACAACGACCG

Quantitative MSP and reverse-transcription (RT)-PCR analyses of RASSF1A

The methylation status of the RASSF1A promoter was also examined in all 97 tumor samples by fluorescence-based, real-time quantitative PCR using a LightCycler (Roche Diagnostics). Primers and probes designed to specifically amplify the promoter of RASSF1A or a reference gene, ACTB, were described elsewhere.28, 29 The primer sequences used for quantitative MSP and those used for conventional MSP share the 17 nucleotides with 1 nucleotide deviation in the forward primer, and the 18 nucleotides with 3 nucleotides deviation in the reverse primer, although they amplified the same RASSF1A CpG islands (CGIs) (Fig. 1a and Table I).31 Each amplification reaction included tumor DNA samples, positive and negative controls and water blank. ACTB was used as a reference gene to determine the relative level of methylated DNA for RASSF1A in each sample. Dividing the methylated RASSF1A/ACTB ratio of template amounts in a sample by the methylated RASSF1A/ACTB ratio of template amounts in a fully methylated control and multiplying this value by 100 calculated the percentage of methylation.

To determine whether the percentage of RASSF1A methylation is correlated with the expression level, we performed RT-PCR analysis of the RASSF1A gene in 7 tumor samples with methylated or unmethylated RASSF1A and 1 normal liver sample available by the method described previously.9

Mutation analysis of the CTNNB1 gene

To detect point mutations and deletions of the CTNNB1 gene, genomic DNA from each tumor sample was amplified using 2 sets of primers, F1, 5′-TGGCTATCATTCTGCTTTTCTTG-3′ and R1, 5′-CTCTTTTCTTCACCACAACATTTT-3′, and BCAT-3, 5′-AAAATCCAGCGTGGACAATGG-3′, and BCAT-4, 5′-TGTGGCA AGTTCTGCATCATC-3′, respectively (Suppl Fig. 1a).10, 32 The PCR products were either directly sequenced or inserted into a vector [pGEM (R)-T Easy Vector System (Promega, Madison, WI)], and 6 or more clones were sequenced.

Statistical analysis

Patients were grouped according to various biological and clinical aspects of disease. Significance of differences in the characteristics between patient's groups was examined using the chi-square or Fisher's exact test. Overall survival for each group of patients was estimated using the Kaplan-Meier method, and compared using the log-rank test. Time to failure was defined as the interval between surgery or preoperative chemotherapy and death from any cause. The influence of various biological and clinical factors on overall survival was estimated using the Cox proportional-hazards model calculated with Stat Flex software for Windows, version 5.0 (Artec Co., Osaka, Japan).

Results

Conventional MSP analysis of various genes in hepatoblastomas

We first examined the methylation status of 13 genes in 20 tumors, including 2 tumors in stage 1, 6 in stage 2, 6 in stage 3 and 6 in stage 4, by conventional MSP and found no methylation in 10 (RASSF2A, NORE1A, RUNX3, RIZ1, BLU, HOXA9, HOXB5, p16INK4A, p14ARF and DCR2); no further analysis was performed on these 10 genes. The remaining 3 genes, including RASSF1A, SOCS1 and CASP8, were methylated in a substantial number of tumors. Therefore, we extended the analysis to all 97 tumors and found hypermethylation of RASSF1A, SOCS1 and CASP8 in 30 (30.9%), 32 (33.0%) and 15 (15.5%) tumors, respectively (Fig. 1b). All 3 genes were methylated in 3 tumors. Two of 3 genes, RASSF1A and SOCS1, RASSF1A and CASP8 and SOCS1 and CASP8, were methylated in 7, 3 and 5 tumors, respectively. Only 1 gene, RASSF1A, SOCS1 or RASSF1A, was methylated in 15, 19 or 4 tumors. Conventional MSP detected unmethylated RASSF1A in all 8 adjacent normal liver tissues.

Correlation of the methylation status of the 3 genes analyzed by conventional MSP with overall survival

When we analyzed the correlation between the methylation status of any 1 of the 3 genes and overall survival, RASSF1A methylation was associated with a poor outcome (p < 0.001), but SOCS1 or CASP8 methylation was not; however, multivariate analysis using the various factors shown in Table III indicated the significant contribution of disease stage [p < 0.001; relative risk (RR) 9.44; 95% confidence interval (CI), 2.51–35.46], but no contribution of RASSF1A methylation to overall survival (p = 0.149; RR 2.38; 95% CI, 0.73–7.72).

Quantitative MSP analysis of RASSF1A methylation and the correlation between the percentage of the RASSF1A methylation and the expression or clinical outcome

To clarify whether RASSF1A methylation is an independent factor predicting outcome, we performed quantitative MSP analysis of RASSF1A in 97 tumors. Tumors were classified by the percentage of RASSF1A methylation, and about one half of tumors (46) had 0–2.5% of the methylation, and others distributed in various percentages of the methylation (Fig. 2a). RT-PCR detected RASSF1A expression in 1 normal liver sample and 2 tumor samples with less than 1% of the methylation, but did not detect the expression in tumors with more than 11% of the methylation; 2 tumors with the intermediate incidence of the methylation (4.2 or 4.8%) showed the ambiguous expression (Fig. 2b). Thus, there is an inverse relationship between the percentage of the RASSF1A methylation and the expression.

Figure 2. (a) Histogram showing the number of tumors categorized by the percentage of RASSF1A methylation. The number of tumors classified by the stage of disease and clinical outcome are shown under the columns. (b) RT-PCR analysis of RASSF1A mRNA in 1 normal liver and 7 tumor samples.

Next, we examined the dose-response relationships between the percentage of RASSF1A methylation and overall survival analyzed by the Kaplan-Meier method and adopted a cutoff value of 4.8%, which gave the smallest p-value (p < 0.00001). We also examined the dose-response relationships between the percentages of RASSF1A methylation and stage of the disease or clinical outcome (Fig. 2a). Patients were classified into 3 groups (0∼<5%, 5∼<30% and 30∼100% of the methylation), and we found that the higher the percentage of the methylation was, the higher the incidence of tumors at advanced stages or with poor outcome was (p < 0.001 and p <0.001). On the basis of this cutoff value, 43 (44.3%) tumors were classified as having methylated RASSF1A and 54 as having unmethylated RASSF1A. In contrast, 30 (30.9%) tumors were classified as having methylated RASSF1A by conventional MSP; therefore, 13 (13.4%) tumors classified as the unmethylated group by conventional MSP changed to the methylated group by quantitative MSP. We used this incidence rate of hypermethylation in subsequent analysis of the correlation between RASSF1A methylation and clinicopathological characteristics in hepatoblastoma.

Mutation and deletion of the CTNNB1 gene

Of 97 tumors, 19 (19.6%) had a point mutation in CTNNB1 and 46 (47.4%) had various sizes of CTNNB1 deletion, ranging from 9 to 1061 bp, always including a region from amino acid 32 to 45, wherein lie 4 serine/threonine residues, which are targeted for phosphorylation. One tumor had both an insertion of 7 bp and a deletion of 19 bp in the same locus.

Incidences of tumors with RASSF1A methylation or CTNNB1 mutation between tumors obtained before or after chemotherapy

CTNNB1 mutation and RASSF1A methylation were found in 47 (65.2%) and 33 (47.2%) of 72 tumors preoperatively treated with chemotherapy and in 18 (72.0%) and 10 (40.0%) of 25 preoperatively untreated tumors. There were no differences in the incidences of CTNNB1 mutation or RASSF1A methylation between tumors that received preoperative chemotherapy and those that did not. The findings indicate that CTNNB1 mutation or RASSF1A methylation did not occur during the period of preoperative chemotherapy, or seem to reject that the normal CTNNB1 or unmethylated RASSF1A status was merely a result of effective chemotherapy for the tumors.

Overall survival of patients classified by clinical and biological characteristics

We evaluated the association of clinical and biological characteristics with overall survival in 97 patients with heptoblastoma (Fig. 3). Patients less than 2 years of age showed better overall survival than those 2 years old or over (p < 0.001), and patients with fetal-type tumor showed better overall survival than those with embryonal-type tumor (p = 0.044). Likewise, patients with a PRETEXT 1, 2 or 3 tumor or a stage 1, 2 or 3A tumor showed better overall survival than those with a PRETEXT 4 (p = 0.003), or a stage 3B or 4 tumor (p < 0.001), respectively. Patients who achieved CR or PR with cisplatin-based chemotherapy had better overall survival than those who did not respond to therapy (NC) (p = 0.011). Finally, patients with a tumor with unmethylated RASSF1A or wild-type CTNNB1 showed better overall survival than those with a tumor with methylated RASSF1A or mutated CTNNB1 (p < 0.001 or p = 0.030), respectively.

Figure 3. Overall survival curves for hepatoblastoma patients based on different variables: (a) age, (b) histological type of tumor, (c) PRETEXT disease stage, (d) disease stage, (e) response to cisplatin-based chemotherapy, (f) methylation status of the RASSF1A gene, (g) mutation status of the CTNNB1 gene.

To clarify the prognostic implication of the RASSF1A status in unfavorable groups, we only included 33 patients with a stage 3B or 4 tumor in the next analysis and found that RASSF1A methylation predicted a poor outcome in this group of tumors (Fig. 4a). Only 1 patient with a tumor with unmethylated RASSF1A died of recurrent brain metastases. When we only included 64 patients in stages 1, 2 and 3A in the next analysis, we also found that RASSF1A methylation predicted a poor outcome in this group of tumors (Fig. 4b). Three (16%) of 19 patients with a RASSF1A-methylated tumor died within 3 years after surgery, while all 45 patients with unmethylated RASSF1A were alive. These findings suggest that the RASSF1A methylation status is useful to identify patients who are likely to suffer recurrence or death from disease, irrespective of a favorable or unfavorable stage of the disease.

Figure 4. (a) Overall survival curves for hepatoblastoma patients in stages 3B and 4 classified by the methylation status of RASSF1A. Dotted line indicates the overall survival curve of all 33 patients. (b) Overall survival curves for hepatoblastoma patients in stages 1, 2 and 3A classified by the methylation status of RASSF1A. Dotted line indicates the overall survival curve of all 64 patients.

Association of RASSF1A, CASP8 or SOCS1 methylation or CTNNB1 mutation with clinical characteristics in hepatoblastoma

RASSF1A methylation was significantly associated with various factors predicting poor outcome except for the histological type and PRETEXT classification (Table II). CASP8 methylation was associated with recurrent disease (p = 0.034), whereas SOCS1 methylation was associated with the fetal histological type (p = 0.020). Nevertheless, the methylation status of both genes was unrelated to other clinical and biological characteristics, including outcome. There was no difference in the overall survival or stage distribution between patients with a tumor with only RASSF1A methylation and those with a tumor with RASSF1A and SOCS1 or CASP8 methylation, or with joint methylation of the 3 genes. CTNNB1 mutation was significantly associated with recurrent disease, poor outcome and RASSF1A methylation. CTNNB1 mutation includes both a point mutation and deletion of various sizes. There was no difference in the clinical characteristics, including outcome, between tumors with the point mutation and those with the deletion.

Table II. Association Between Clinicopathological Factors and the RASSF1A or CTNNB1 Status in 97 Patients with Hepatoblastoma

Factors	Number of tumor	RASSF1A			CTNNB1
		Methylated	Unmethylated	p1	Mutated	Not mutated	p1
CR, complete response; PR, partial response; NC, no change. 1 Chi-square test or Fisher's exact test. 2 Four patients, whose histological type could not be determined, were excluded. 3 One patient who was treated only surgically was excluded. 4 Five patients, whose tumors did not disappear and who died of the disease, were excluded.
Sex				0.761			0.335
Male	57	26	31		36	21
Female	40	17	23		29	11
Age at diagnosis				<0.001			0.262
<2 year	59	11	48		37	22
≥2 year	38	32	6		28	10
Histological type2				0.360			0.216
Fetal	38	15	23		23	15
Non-fetal	55	27	28		40	15
PRETEXT				0.063			0.361
1, 2, 3	80	32	48		28	52
4	17	11	6		4	13
Stage				<0.001			0.686
1, 2, 3A	64	19	45		42	22
3B, 4	33	24	9		23	10
Response to cisplatin-
based chemotherapy3				0.010			0.218
CR, PR	85	34	51		55	30
NC	11	9	2		9	2
Recurrence4				<0.001			0.008
No	68	19	49		39	29
Yes	24	19	5		21	3
Outcome				<0.001			0.020
Alive	78	25	53		48	30
Dead	19	18	1		17	2
CTNNB1				0.007
Not mutated	32	8	24
Mutated	65	35	30

Multivariate Cox proportional hazard regression analysis was performed to clarify whether various factors independently affect overall survival in 92 patients, in whom all variables were available. Disease stage and the RASSF1A methylation status were shown to be independent factors predicting poor outcome, but CTNNB1 mutation was not (Table III). The incidence of CTNNB1 mutation (67.4%, 62/92) was higher than that of RASSF1A methylation (45.7%, 42/92), and tumors with the mutation included the great majority (81.0%, 34/42) of tumors with methylation. Furthermore, while patients with a tumor with mutated CTNNB1 and unmethylated RASSF1A enjoyed excellent prognosis, those with a tumor with mutated CTNNB1 and methylated RASSF1A suffered an unfavorable outcome (p < 0.001). These findings led to different results of the prognostic implication of CTNNB1 mutation by univariate and multivariate analyses.

Table III. Multivariate Analysis on 6 Clinicopathological and Genetic Factors in 92 Patients with Hepatoblastoma

Prognostic factors	Relative risk (95%CI)	p value
95%CI, 95% confidence interval; CR, complete response; PR, partial response; NC, no change.
Age
<2 year versus ≥2 year	1.34 (0.45–3.95)	0.600
Histological type
Fetal versus Nonfetal	2.15 (0.69–6.74)	0.189
Stage
1, 2, 3A versus 3B, 4	7.67 (2.13–27.61)	0.002
Response to chemotherapy
CR, PR versus NC	1.95 (0.65–5.87)	0.234
CTNNB1
Not mutated versus mutated	2.19 (0.47–10.23)	0.321
RASSF1A
Unmethylated versus methylated	9.39 (1.08–82.06)	0.043

Discussion

Hepatoblastoma occupies 90% of childhood liver tumors, although its incidence is relatively low.1 Currently, 20–30% of patients who do not respond to preoperative chemotherapy, or who present with or develop metastatic disease, continue to face a poor outcome.2, 3 To improve the mortality rate, treatment strategies for hepatoblastoma refractory to the standard cisplatin and THP adriamycin regimen or with metastasis should be innovated.33 To achieve a higher complete resection rate, more effective preoperative chemotherapy for refractory hepatoblastoma is mandatory, and such therapy offers a realistic hope for cure. In addition, novel molecular-genetic markers that predict the treatment outcome of patients are needed for better therapy planning.

Because oncogenes or tumor suppressor genes other than CTNNB1 are rarely mutated in hepatoblastoma,10–12 and there are no reports on the prognostic implication of CTNNB1 mutation, we suspected that methylation of tumor suppressor genes may occur, acting as a biomarker to predict treatment outcome. Thus, we analyzed the methylation status of 13 candidate tumor suppressor genes, RASSF1A, RASSF2A, SOCS1, CASP8, NORE1A, RUNX3, RIZ1, BLU, HOXA9, HOXB5, p16INK4A, p14ARF and DCR2, by conventional MSP.13–22 These genes have previously been shown to be aberrantly methylated in various adult and childhood cancers and also represent important elements for several signaling pathways and cell cycle regulation (Table IV). We found that 3 genes, RASSF1A, SOCS1 and CASP8, were methylated in a substantial number of hepatoblastoma tumors. Interestingly, univariate analysis showed that only RASSF1A methylation was correlated with a poor outcome, but not SOCS1 or CASP8 methylation. When we examined the contribution of various prognostic factors to overall survival by multivariate analysis, only the disease stage was identified as an independent factor, but not RASSF1A methylation.

Table IV. Tumor Suppressor Genes and the Incidence of Methylated Hepatoblastoma Tumors Examined by Methylation-Specific PCR

Pathway	Gene	Gene location	Function	Incidence of methylated tumor	References
Signal transduction	RASSF1A	3p21	RAS effector	43/97 (44%)	Present study
				15/39 (39%)	9
				5/27 (19%)	34
	RASSF2A	20p13	RAS effector	0/20 (0%)	Present study
	NORE1A	1q32	RAS effector	0/20 (0%)	Present study
	SOCS1	16p13	Inhibitor of JAK/STAT pathway	32/97 (33%)	Present study
				7/15 (47%)	35
	RUNX3	1p36	TGF-beta pathway	0/20 (0%)	Present study
	RARB	3p24	Retinoic acid receptor	0/27 (0%)	34
	APC	5q21	Wnt signaling pathway	0/27 (0%)	34
	SFRP1	8p12	Secreted frizzled-related protein	0/39 (0%)	9
	SFRP2	4q31	Secreted frizzled-related protein	0/39 (0%)	9
	SFRP4	7p14	Secreted frizzled-related protein	0/39 (0%)	9
	SFRP5	10q24	Secreted frizzled-related protein	0/39 (0%)	9
Cell-cycle regulation	p16INK4A	9p21	Cell cycle regulation	0/20 (0%)	Present study
				0/27 (0%)	34
	p14ARF	9p21	MDM2 inhibitor	0/20 (0%)	Present study
Apoptosis	CASP8	2q33	Activation of effector caspases	15/97 (16%)	Present study
	DCR2	8p22	Antiapoptotic decoy receptor	0/20 (0%)	Present study
	DAPK	9q34	Death-associated protein kinase	0/27 (0%)	34
Chromatin regulation and transcription	RIZ1	1p36	Methyltransferase superfamily	0/20 (0%)	Present study
DNA repair	MGMT	10q24	DNA methyltransferase	0/27 (0%)	34
Detoxification	GSTP1	11q13	Glutathione S-transferase	0/27 (0%)	34
Cell adhesion	CDH1	16q22	E-cadherin	0/27 (0%)	34
	CDH13	16q24	H-cadherin	0/27 (0%)	34
Unknown	BLU	3p21	Suppressor of cell cycle entry (?)	0/20 (0%)	Present study
	HOXA9	7p15-p14	Homeobox protein	0/20 (0%)	Present study
	HOXB5	17q21	Homeobox protein	0/20 (0%)	Present study

Then we analyzed the methylation status of RASSF1A by quantitative MSP because this method gives more reproducible and accurate results than conventional MSP. The accuracy and reliability of quantitative MSP were proved by the inverse relationship found between the percentage of RASSF1A methylation and the expression (Fig. 2b). The incidence of tumors with hypermethylated RASSF1A increased from 30.9 to 44.3%, probably because quantitative MSP is more sensitive than conventional MSP. The low cutoff value of 4.8% and the slight difference in the primer locations may have also contributed to the different incidences of the methylated tumors examined by the 2 MSP methods. In our previous study of RASSF1A methylation in 39 hepatoblastoma tumors, multivariate analysis using the prognostic factors similar to the present ones showed an equivocal p-value of 0.079 with relative risk of 12.84 (95% CI, 0.74–223.13). The present multivariate analysis using the results examined by quantitative MSP and the substantial number of tumors clearly demonstrated that the methylation is an independent factor predicting treatment outcome, and its contribution ranked next to the disease stage (Table III).

RASSF1A is a gene located in the 3p21 chromosomal region where deletions and loss of heterozygosity are frequently reported in small cell lung cancer.31 Previous studies, including ours, have repeatedly shown that promoter hypermethylation of RASSF1A correlated with loss of expression in various cancers, and treatment with a demethylating agent reactivated RASSF1A gene expression in various cancer cell lines, including a hepatoblastoma cell line HepG2.9, 34, 36, 37 RASSF1A inhibits tumor formation by apoptosis, and regulates microtubule dynamics and mitotic arrest via multiple effectors. By dysregulation of the Ras signaling pathway, RASSF1A methylation is correlated with poor differentiation and vascular invasion of cancer cells, and an unfavorable outcome.36

Among the 13 genes examined that were frequently methylated in various cancers, only 3 genes were methylated in hepatoblastoma. The present and previous studies evaluated the methylation status of at least 20 genes in hepatoblastoma and found that only 3 genes were methylated (Table IV).9, 34, 35 The limited number of methylated genes suggest that this profile may be specific for hepatoblastoma,38 and the survival and stage distribution analyses disclosed that combined RASSF1A and SOCS1 or CASP8 methylation, or joint methylation of the 3 genes are not correlated with the advanced stage of disease or a poor outcome, contrary to the findings that methylation of multiple genes were correlated with a poor outcome, reported in neuroblastoma.39

The present multivariate analysis identified unresectable tumor stages of disease (3B and 4) as the most significant factor predicting overall survival, followed by RASSF1A methylation. Down-staging of stage 3B tumors and control of metastatic lesions of stage 4 tumors by preoperative chemotherapy proceeds to subsequent complete resection, and this procedure may be critical to cure patients in such stages. Presently, JPLT or other protocols treat hepatoblastoma patients by a preoperative regimen consisting of cisplatin and adriamycin or its derivatives.2, 23 The present study showed that patients with a RASSF1A-methylated tumor in stage 3B or 4 were less likely to respond to preoperative therapy than those with a RASSF1A-unmethylated tumor in the same stage (Table II and Fig. 4a). In addition, in an analysis of 70 male germ cell tumors, Koul et al. found that the incidence of RASSF1A methylation is higher in cisplatin-resistant tumors than in cisplatin-sensitive tumors.40 Therefore, we propose that patients with a RASSF1A-methylated hepatoblastoma tumor should be treated with a more intensive regimen with anticancer drugs other than cisplatin and adriamycin or its derivatives.

Abnormalities of the Wnt pathway are the genetic hallmark of hepatoblastoma, and CTNNB1 mutation is the most frequent genetic changes found in the pathway10, 41; however, there has been only one study on the prognostic implication of CTNNB1 mutation in hepatoblastoma, which failed to show a correlation between the mutation and outcome.42 The present univariate analysis showed that patients with CTNNB1 mutation had a lower overall survival rate than those without CTNNB1 mutation (Fig. 3); however, multivariate analysis rejected the mutation as an independent factor (Table III). The great majority of tumors with RASSF1A methylation were included in tumors with CTNNB1 mutation, and patients with tumors with the mutation but not with the methylation showed favorable prognosis. These findings suggest that CTNNB1 mutation may be an early genetic event in hepatoblastoma tumorigenesis, whereas RASSF1A methylation may be a later event associated with tumor progression.

In the present study on various candidate tumor suppressor genes, RASSF1A was the most frequently methylated gene in hepatoblastoma and its methylation clearly predicted the poor outcome of patients. We believe that the RASSF1A status is a promising molecular-genetic marker, and we expect that this biomarker may be used to stratify patients treated in clinical trials.

Acknowledgements

The authors are grateful to Dr. K. Hiyama, Hiroshima University, a data administrator for JPLT, for data management. They also express gratitude to the physicians participating in JPLT who supplied samples for this study.