Correlation of postoperative recurrence in hepatocellular ... - Nature

Oncogene (2002) 21, 789 ± 797 ã 2002 Nature Publishing Group All rights reserved 0950 ± 9232/02 $25.00 www.nature.com/onc

Correlation of postoperative recurrence in hepatocellular carcinoma with demethylation of repetitive sequences Osamu Itano1, Masakazu Ueda*,1, Kiyoshi Kikuchi2, Osamu Hashimoto1, Shigeo Hayatsu1, Masaharu Kawaguchi1, Hiroaki Seki1, Kouichi Aiura1 and Masaki Kitajima1 1

Department of Surgery, Keio University School of Medicine, 35 Shinanomachi, Shinjuku-ku, Tokyo 160-8582, Japan; 2Department of Surgery, Tokyo Denryoku Hospital, 9-2 Shinanomachi, Shinjuku-ku, Tokyo 160-0016, Japan

Restriction landmark genomic scanning (RLGS) was utilized to identify novel genomic alterations in hepatocellular carcinoma (HCC). Thirty-one HCC samples were examined by RLGS. Two high intensity spots were common to several RLGS pro®les of dierent HCCs. Nucleotide sequencing and homology search analysis showed that these spots represented repetitive sequences, Human tandem repeat sequence (Genbank, L09552) and centromeric NotI cluster (Genbank, Y10752). These intensi®ed signals were attributable to the occurrence of demethylated areas in the recognition sequence of the NotI site of the corresponding fragments. The intensity of these spots in the RLGS pro®le re¯ects their degree of demethylation, which was signi®cantly correlated with postoperative recurrence, even in patients regarded as belonging to the good prognosis group by conventional prognostic factors. Multivariate analysis showed that the intensities of the two spots retained independent prognostic value. This is a new type of predictive factor for HCC based on epigenetic changes in hepatocarcinogenesis, and in the future it is expected to be of great value in making preoperative diagnosis and selecting postoperative therapy. Oncogene (2002) 21, 789 ± 797. DOI: 10.1038/sj/onc/ 1205124 Keywords: demethylation; repetitive sequence; hepatocellular carcinoma; postoperative recurrence; restriction landmark genomic scanning Introduction Several factors have been employed to predict the postoperative recurrence of hepatocellular carcinoma (HCC). Thus, it has been reported that factors such as tumor size, portal venous invasion, and intrahepatic metastasis are signi®cantly correlated with recurrence (Hsu et al., 1988; Arii et al., 1992; Sirabe et al., 1991). In addition, liver function data and severity of cirrhosis have also been utilized as prognostic markers for postoperative recurrence (Lise et al., 1998; Kubo et al.,

*Correspondence: M Ueda Received 24 June 2001; revised 12 October 2001; accepted 29 October 2001

1998). However, no completely satisfactory predictive factors have been identi®ed. Furthermore, no genetic markers for recurrence of HCC been delineated either (Himeno et al., 1988; Zhang et al., 1987). We therefore sought genomic alterations inherent to HCC using the technique of restriction landmark genomic scanning (RLGS). We have reported that the dierence between non-cancerous liver tissue and HCC in the number of altered RLGS spots was signi®cantly correlated with postoperative disease recurrence. This had been found to be a more useful marker for HCC prognosis than the conventional factors, as documented in our previous report (Itano et al., 2000). This is interpreted to imply that such markers re¯ect genomic changes more closely associated with the actual process of hepatocarcinogenesis than the changes detected by conventional methods. However, the mechanism by which this factor contributes to carcinogenesis is unknown. In the present study eight RLGS spots have been identi®ed which are commonly altered in HCC pro®les. The intensity of two of these spots has been found to be signi®cantly correlated with the number of altered RLGS spots. Here, we have cloned the DNA fragments of these spots from the RLGS gel, established the nature of the genomic alterations, and shown them to represent novel prognostic factors for postoperative course in HCC.

Results Isolation of clones from the RLGS spots and DNA sequence analysis Genomic DNAs from 31 HCCs were analysed by RLGS. The HCC pro®le was compared with that of the adjacent non-cancerous liver tissue from each patient. The spots which diered between HCC and non-cancerous tissue were marked and counted (Figure 1). Eight changes were common to the RLGS pro®les of 2 ± 20 of the 31 cases (A, 20/31; B, 18/31; others, 2 ± 14/ 31). All eight of these spots were seen only in HCC and not in non-cancerous liver tissue. None of the altered spots was speci®c to any particular virus group (HBV group or HCV group). To further characterize the

HCC and demethylation in repetitive sequences O Itano et al

790

B A

Figure 1 Comparison of the RLGS pro®les of HCC and normal surrounding liver tissue from the same patient (case C1). Large arrows point to the spots appearing on only one side of the two pro®les, small arrows point to the spots with greater density than the corresponding spot on the other side. Eight spots were altered in several cases, all in the HCC pro®le. Two of these spots were detected in 20 and 18 of the 31 cases (64.5 and 58.1%) and are labeled (A) and (B)

nature of the genomic alterations re¯ected in these HCC-speci®c spots, we cloned NotI ± PstI fragments corresponding to spot A and B (Figure 1), which were intensi®ed and detected in 20 and 18 (64.5 and 58.1%) of the 31 cases respectively. The nucleotide sequences of the cloned fragments were determined, and a homology search was carried out. The nucleotide sequence of the cloned fragment corresponding to spot A was completely identical to part of a repetitive 13 kb unit submitted to Genbank in 1993 under accession number L09552, designated Human tandem repeat sequence (HTRS) by S Wong, MJ Wagner and D Wells. The unit was reported to be repeated about 200 times in tandem on human chromosome 8q21, and the fragment has also been identi®ed in other human cancers by RLGS (Miwa et al., 1995; Thoraval et al., 1996). The nucleotide sequence of the cloned fragment corresponding to spot B was identical to part of a repetitive unit of the 1415 bp NotI ± NotI fragment submitted to Genbank under accession number Y10752, designated centromeric NotI cluster (CNIC). This fragment had already been identi®ed in HCC by RLGS and mapped to the pericentromeric regions of acrocentric chromosomes such as 3q11.1 ± 11.2, 4p11.1 ± 12, 13p13, 14p11.1, 14p13, 15p13, 19q13.2, 20p11.1, 21p13, 22p11.1 and 22p13 by using ¯uorescence in situ hybridization. (Nagai et al., 1999a,b). According to the homology assessment, it is the complementary sequence to the fragment that Thoraval et al. (1996) isolated and cloned from neuroblastoma by RLGS. Comparison of the intensity of the HTRS and CNIC spots in a variety of normal tissues No HTRS and CNIC spots were detected by RLGS in ®ve hepatitis virus-negative non-cancerous liver tissues resected because of liver metastasis of colon cancer. Oncogene

Other controls in which the intensity of the HTRS and CNIC spots were examined included 10 gastric mucosa, 20 colonic mucosa, 15 mammary tissues and 10 leucocytes, in addition to the liver samples. Neither HTRS nor CNIC spots were detected in any of these types of normal tissue. Correlation between the intensity of the HTRS and CNIC spots, and correlations between these spots and the number of changed RLGS spots in any one profile Whether there was a correlation between the intensity of the HTRS and CNIC spots was examined in all cases by testing for the Spearman rank correlation coecient (Figure 2). In this way, it was found that the intensities of the two spots were signi®cantly correlated with each other. In addition, the numbers of dierent RLGS spots altered in HCC compared to noncancerous liver tissue were compared with the intensity of the HTRS and CNIC spots in each case. The intensity of the latter two spots was found to be signi®cantly correlated with the total number of altered RLGS spots (Figure 3a,b) Correlation between the Intensity of the HTRS and CNIC spots and the postoperative outcome Whether there was a correlation between intensities of the HTRS and CNIC spots and the postoperative outcome was examined. There appeared to be a trend towards an association between high intensity HTRS and CNIC spots, and more episodes of recurrence. To examine this more closely, the patients were divided into two groups: in one (the high intensity group) the spot intensity was greater than the diploid (two-copy) level, and in the other (the low intensity group), it was equal to or less than the diploid level.


791

Figure 2 Correlation between the intensity of the HTRS spot and the CNIC spot was examined in all cases. The correlation was signi®cant, with a Spearman rank correlation coecient of 0.813, P50.0001

The postoperative recurrence rate was then compared between these two groups (Figure 4a,b). The disease-free survival rate of the high intensity group for each spot was signi®cantly lower than that of the low intensity group. The postoperative recurrence rates were then compared between the two groups in patients subdivided according to what are conventionally regarded as good prognostic factors, i.e. without portal vein invasion at operation (Figure 5a,b), in patients with tumor diameters of 3 cm or less (Figure 5c,d), and in patients with a single tumor (Figure 5e,f). The disease-free survival rates of the high intensity group were signi®cantly lower than in the low intensity group for both HTRS and CNIC spots, independent of conventional prognostic markers. Clinical parameters, host factors and tumor factors have been assessed for their utility as predictive factors for postoperative recurrence. We therefore examined whether basic clinicopathological features diered between the high and low intensity HTRS and CNIC groups. The following host factors and test values were considered: sex, age, total bilirubin, serum alanine aminotransferase, serum asparatate aminotransferase, albumin, prothrombin activity, total protein, cholinesterase, total cholesterol, zinc sulfate turbidity test, indocyanine green retention test, level of liver cirrhosis, and Clinical Stage Grouping Classi®cation of Primary Liver Cancer (Liver Cancer Study Group of Japan, 1997). It was found that there was a signi®cant dierence in: (i) serum alanine aminotransferase level which was higher in the HTRS high intensity than low intensity group (Student's t-test; P=0.0223); (ii) the results of the indocyanine green retention test, which were higher in the HTRS high intensity than low intensity group (Student's t-test; P=0.0223); and (iii) albumin, which was lower in the CNIC high intensity than low intensity group (Student's t-test; P=0.0201). There were no signi®cant dierences in other host

Figure 3 The number of altered RLGS spots in HCC compared to non-cancerous liver tissue was compared with the relative density versus the diploid level of the HTRS and CNIC spots in each case by using the Spearman rank correlation coecient. The intensity of the spots was signi®cantly correlated with the number of altered RLGS spots. (a) Spearman rank correlation coecient: 0.650, P50.0001. (b) 0.712, P50.0001

factors, but there was a trend toward worse liver function in the high intensity HTRS and CNIC groups (although this did not achieve statistical signi®cance). Regarding tumor factors, including the number of tumors, tumor size, histological grading, growth type, capsule formation, capsule in®ltration, serosa invasion, vascular and bile duct invasion, intrahepatic metastasis, and the Stage Grouping of Classi®cation of Primary Liver Cancer (Liver Cancer Study Group of Japan, 1997), no dierences were found between the high and low intensity groups for either HTRS or CNIC spots. Correlations between the intensity of the HTRS and CNIC spots and viral infection, which is one of the conventional predictive factors for HCC, were also sought in three subgroups of patients, namely, the nonB/non-C group, the HBV group and the HCV group. Oncogene


792

ground factors (the host and tumor factors mentioned above) was assessed by univariate analysis. Portal vein invasion (relative risk 2.383: P=0.0037), the results of the zinc sulfate turbidity test (relative risk 1.057: P=0.0416), the results of the indocyanine green retention test (relative risk 1.053: P=0.0282), and subdivision into groups according to the intensity of the HTRS spot (relative risk 5.238: P=0.0099) and the CNIC spot (relative risk 4.348: P=0.0078) all emerged as signi®cant prognostic factors. These signi®cant factors were assessed for their independence by multivariate analysis. Only portal vein invasion and the intensity of either spot retained independent prognostic factor status among the factors identi®ed as signi®cant by univariate analysis (Table 1a,b). Discussion

Figure 4 (a) Postoperative recurrence was compared between the group of patients with a low intensity HTRS spot (& ± &: spot intensity was equal to or less than the diploid level, n=12) and the high intensity group (*_*: spot intensity was greater than diploid level, n=19). *P=0.0041. (b) Postoperative recurrence was compared between the group of patients with a low intensity CNIC spot (& ± &: n=17) and the high intensity group (*_*: n=14). *P=0.0036. The disease-free survival rate of the high intensity group was signi®cantly lower than that of the low intensity group. (The disease-free survival curves of patients in the dierent groups were calculated by the Kaplan-Meier method and compared by the log-rank test)

Serological tests showed that eight patients were hepatitis B surface antigen-positive and hepatitis C antibody-negative (HBV group: cases B1 ± B8); 15 patients were hepatitis B surface antigen-negative and hepatitis C antibody-positive (HCV group: cases C1 ± C15); and eight patients were both hepatitis B surface antigen-negative and hepatitis C antibody-negative (non-B/non-C group: cases NBNC1 ± NBNC8). Both spots were signi®cantly more intense in the HBV and HCV groups than in the non-B/non-C group. There was no signi®cant dierence in the intensity of either HTRS or CNIC spots between the HBV and HCV groups (Figure 6a,b). The prognostic value of the intensity of the HTRS and CNIC spots and the clinicopathological backOncogene

The present study documents that two high intensity spots detected by RLGS, the HTRS spot and the CNIC spot, are associated with hepatocarcinogenesis, and that the intensity of these spots is signi®cantly correlated with postoperative disease-free survival of HCC patients. DNA fragments corresponding to these two spots were identi®ed as parts of repetitive sequences. The intensity of the HTRS and CNIC signals has been reported to be due to demethylation in the recognition sequence of the NotI site of the corresponding repeating units (Miwa et al., 1995; Thoraval et al., 1996; Nagai et al., 1999a,b). The signal intensity re¯ects the degree of demethylation, which, in repetitive sequences, is thought to correlate with the malignant potential of HCC. Alteration of DNA methylation status is known to be related to the control of gene expression, such as Xchromosome inactivation, genomic imprinting, control of genes involved in dierentiation and maintenance of chromosome stability (Laird and Jaenisch, 1996; Baylin et al., 1998). It has been reported that hyper- or hypomethylation is an important mechanism of carcinogenesis in several types of cancer (Laird and Jaenisch, 1996) and in HCC (Liew et al., 1999). Demethylation in repetitive sequences is also thought to be associated with the genesis or progression of cancers (ZrihanLicht et al., 1995; Shiraishi et al., 1999; Teubner and Schulz, 1994). In eukaryotic genomes, methylation of cytosine residues commonly occurs in repetitive sequences, and this methylation correlates with reduced gene expression and suppression of recombination. It is therefore thought to serve as a genome-defense mechanism that guards against the deleterious eects of multicopy transposable elements and aberrant gene duplications (Bender, 1998). The sequences that we report here are both repetitive, and such demethylations have already been reported to be correlated with HCC and some other malignant tumors. It is particularly noteworthy that CNIC is located in the pericentromeric region of human acrocentromeric chromosomes and is characterized by a major class


793

Figure 5 Postoperative recurrence was compared in patients without portal vein invasion (a) between the group with a low intensity HTRS spot (& ± &: n=10) and the high intensity group (*_*: n=16) (*P=0.0010) and (b) between the group of patients with a low intensity CNIC spot (& ± &: n=15) and the high intensity group (*_*: n=11) (*P=0.0028). Postoperative recurrence was compared in patients with tumor diameters of 3 cm or less (c) between the group with a low intensity HTRS spot (& ± &: n=7) and the high intensity group (*_*: n=11) (*P=0.0203) and (d) between the group of patients with a low intensity CNIC spot (& ± &: n=11) and the high intensity group (*_*: n=7) (*P=0.0376). Postoperative recurrence was compared in patients with a single tumor (e) between the group with a low intensity HTRS spot (& ± &: n=10) and the high intensity group (*_*: n=11) (*P=0.0332) and (f) between the group with a low intensity CNIC spot (& ± &: n=11) and the high intensity group (*_*: n=10) (*P=0.03). The disease-free survival rate of the patients without portal vein invasion, with tumor diameters of 3 cm or less, and with a single tumor at the time of operation was signi®cantly lower in the high intensity group than in the low intensity group. (The disease-free survival curves of patients in the dierent groups were calculated by the Kaplan ± Meier method and compared by the log-rank test) Oncogene


794

Figure 6 The intensities of the HTRS and CNIC spots was compared among the non-B/non-C, HBV and HCV groups. Each point with a bar represents the mean+s.d. of the relative density of the spot against diploid level. The HBV and HCV groups had signi®cantly more high intensity HTRS and CNIC spots than the non-B/non-C group. There were no signi®cant dierences between the number of either of the spots between the HBV and HCV groups (*Fisher's PLSD P50.05)

of repetitive DNA sequences known as alpha satellite DNA (although CNIC does not contain the typical fundamental repeated alpha satellite monomer units of approximately 170 bp) (Nagai et al., 1999b). Hypomethylation of satellite 2 DNA in the centromere of chromosomes 1 and 16 is a characteristic feature of patients with ICF (immunode®ciency, centromeric region instability, and facial anomalies) and is seen in many types of cancer (Qu et al., 1999; Narayan et al., 1998). It is thought to cause heterochromatin rearrangements in the vicinity of the centromeres of each chromosome, which leads to abnormal chromosomal structures and instability associated with carcinogenesis. It is thought that the mechanism of action of CNIC may be the same as the mechanism of satellite 2 DNA. Oncogene

A new ®nding in this study is that the degree of demethylation of repetitive sequences is signi®cantly correlated with the postoperative recurrence of HCC. Qu et al. (1999) reported a statistically signi®cant correlation between the extent of satellite 2 DNA hypomethylation in chromosomes 1 and 16 and the degree of malignancy in three types of ovarian neoplasms as well as a statistically signi®cant correlation between genome-wide hypomethylation and satellite 2 DNA hypomethylation. Global DNA hypomethylation has been frequently observed in many types of cancers and is correlated with prognostic factors in several of these Kim et al. (1994); Cravo et al. (1996); Soares et al. (1999) and Shen et al. (1998) reported that decreased levels of global DNA methylation in HCC are correlated with tumor characteristics. In our previous paper we showed that the number of altered RLGS spots in HCC compared to non-cancerous liver tissue was signi®cantly correlated with postoperative recurrence of HCC (Itano et al., 2000), and here we show that the intensity of the HTRS and CNIC spots, in other words, the degrees of demethylation in HTRS and CNIC, were signi®cantly correlated with the number of altered RLGS spots (Figure 3a,b). The above considerations strongly suggest that the number of altered RLGS spots and demethylation in repetitive sequences might be correlated with global DNA hypomethylation. Moreover, although unproven, it may be hypothesized that these three factors in¯uence HCC malignancy via a common mechanism. This hypothesis requires further investigation and may reveal important information on the mechanisms of malignancy. The reason why diploid intensity was taken as the cuto was because the intensity of the majority of the RLGS spots (about 80%) was equal to or less than the diploid intensity, as described previously; therefore, a greater than diploid intensity indicates a spot more intense than average. In the present study, there appeared to be a trend towards a positive correlation between high intensity HTRS and CNIC spots, and more episodes of recurrence. The rigor of the analysis was con®rmed by the ®nding that even if the cuto level was increased to ®ve times the diploid intensity for HTRS, or eight times the diploid intensity for CNIC, there was essentially no change in statistical signi®cance. Thus, a dierent level of intensity may possibly be more appropriate for the cuto and it remains necessary to establish the optimal point for this. The HTRS and CNIC spots were signi®cantly correlated with viral infection, although there was no signi®cant dierence between the HBV and HCV groups (Figure 6a,b) and none of the altered spots was speci®c to any particular virus. In addition, virus infection was not a signi®cant prognostic factor according to univariate analysis. These results suggest that viral infection is associated with genomic changes including the demethylation of HTRS and CNIC, which in¯uence the tumor's malignant potential and recurrence. However viral infection itself is not associated with malignant potential. There are dierences among virus types regarding relationships of


Table 1A Multivariate analysis with subdivision according to the intensity of HTRS spot and other signi®cant factors Factors

Partial regression coefficient

Relative risk

Probability rate

1.41 72.248 0.016 70.23 2.7186104

4.096 0.106 1.016 0.794 4.14361075

0.00414 0.0085 0.6733 0.8017 0.9949

Portal vein invasion (vp0 or vp1 or vp2 or vp3) Grouping by the intensity of the HTRS spot (high or low) The results of the indocyamine green retention test Sex (male or female) Age (year)

795

Table 1B Multivariate analysis with subdivision according to the intensity of CNIC spot and other significant factors Portal vein invasion (vp0 or vp1 or vp2 or vp3) Grouping by the intensity of the CNIC spot (high or low) The results of the indocyamine green retention test Sex (male or female) Age (year)

1.046 71.759 0.048 70.062 0.034

2.847 0.172 1.05 0.94 1.034

0.0054 0.053 0.1624 0.9394 0.4654

vp0, no tumor thrombus in the portal vein; vp1, tumor thrombus distal to the second branches of the portal vein; vp2, tumor thrombus in the second branches of the portal vein; vp3, tumor thrombus in the ®rst branch with or withour extension to the trunk or the opposite side branch of the portal vein

HTRS and CNIC, but these dierences do not in¯uence the correlation of HTRS and CNIC with malignant potential. In this study, portal vein invasion, the results of the zinc sulfate turbidity test and the indocyanine green retention test emerged as signi®cant prognostic factors by univariate analysis. These are the conventional factors used for predicting postoperative recurrence of HCC, but have been found to be unsatisfactory in clinical settings. Portal vein invasion has already been recognized as having strong prognostic value, however, a recurrence rate of over 50% within 5 years has been reported in cases without portal vein invasion (Arii et al., 1992). The HTRS and CNIC spots were signi®cantly associated with recurrence even in the cases without portal vein invasion (Figure 5a,b). In addition, tumor diameter and intrahepatic metastasis, which are representative conventional factors, had no in¯uence, as shown in Figure 5c ± f. Therefore, we conclude that these new genomic markers, the HTRS and CNIC spots, were more useful for predicting postoperative recurrence of HCC than conventional factors. Although in our previous paper we reported that the number of changed RLGS spots was more useful as a novel genomic marker than the conventional factors, HTRS and CNIC are more suitable for practical use in clinical settings than the number of altered RLGS spots, because the technique of RLGS may be too labor-intensive and time-consuming for routine use and the degree of demethylation in HTRS and CNIC can be measured by a simpler and easier method, such as southern blotting with NotI digestion or real time PCR. We are now developing a measurement system that can be used in clinical settings. In conclusion, the methylation status of repetitive sequences was found to be signi®cantly correlated with postoperative recurrence of HCC. This is a new type of predictive factor based on epigenetic changes in hepatocarcinogenesis, and in the future it is expected to overcome the shortcomings of conventional factors and to be of great value in making preoperative diagnosis and selecting postoperative therapy.

Materials and methods Patients and specimens In this study informed consent for any treatment was obtained from the patients or their nearest relatives. A total of 31 patients with primary HCC (mean age, 59 years; range, 35 ± 72 years; 29 males and two females) who had undergone curative hepatectomy at Keio University Hospital (Tokyo, Japan) between May 1995 and December 1997 were studied. Samples for DNA extraction were obtained from primary HCC tissue and adjacent non-cancerous tissue in the resected liver, and all samples were kept frozen at 7708C until DNA preparation. The clinicopathological background factors of the patients were assessed according to the Classi®cation of Primary Liver Cancer, ®rst English edition, of the Liver Cancer Study Group of Japan. Patients were examined every 3 months in the outpatient clinic. Recurrences were diagnosed based on the results of diagnostic imaging, including routine total body computed tomography, magnetic resonance imaging, and ultrasonography. The maximum follow-up period was 50 months, with a mean observation time of 32 months. None of the patients had received postoperative adjuvant therapy before recurrence. DNA preparation and RLGS High molecular weight genomic DNA was isolated from all samples (Blin and Staord, 1976). RLGS was performed according to published protocols (Hatada et al., 1991). Brie¯y, genomic DNA was digested with NotI and PvuII restriction enzymes, and the NotI-derived 5' protruding ends were 32P-labeled. After ®rst dimension separation of fragments in agarose disc gels, a second digestion was done in situ with PstI. The resulting fragments were separated perpendicularly in a polyacrylamide gel, and the separated DNA fragments were visualized by autoradiography. Analysis RLGS pro®les were digitized and the positions and intensities of spots were analysed, as described previously. The positions of RLGS spots relative to the surrounding spots were examined visually and the intensity of spots was measured as the pixel value6inches2 on the NIH image (NIH's Division of Computer Research and Technology, MD, USA). As reported in a previous paper, the intensity levels Oncogene


796

of all 1214 spots in a RLGS pro®le of a hepatitis virusnegative non-cancerous liver tissue resected because of liver metastasis of colon cancer were quanti®ed and on this basis the spots fell into several discrete peaks. The minimum peak represented the haploid level (1500 ± 3000 pixel value6 inches2) and the second peak, the intensity of which was two-fold higher than that of the minimum peak, represented the diploid level (4000 ± 5500 pixel value6inches2) (Itano et al., 2000). This `diploid level' was then taken as the basic standard against which to estimate spot intensities in this study. Statistical analysis was performed by using correlation coecients and the unpaired Student's t-test for measured data and the w2 test for categorical data. Patient groups were compared by using the w2 test or Mann ± Whitney test. The disease-free survival curves for the patient groups were calculated by the Kaplan ± Meier method and compared by the log-rank test. Cox's proportional hazards model was used for the univariate and multivariate analyses. P values less than 0.05 were considered to indicate statistical signi®cance.

was labeled (and co-electrophoresed with the remaining threefourths of unlabeled DNA that retained unmodi®ed NotI ends). The NotI/PstI ampli®ed fragment corresponding to the spot of interest was cut from the gel, eluted, and ligated to DNA adapters with NotI and PstI ends. PCR was then performed with adapter-speci®c primers (30 cycles, respectively), and the PCR products were cloned by using the vector pBluescript II. DNA sequence and homology analysis The nucleotide sequences of the cloned fragments were determined by using dye-terminator reaction chemistry and a ABI Prism 310 Genetic Analyser (Applied Biosystems, Foster City, CA, USA). Nucleotide sequences were analysed by using GENEWORKS software, and a homology search was carried out using the BLAST Network Service of the National Center of Biotechnology Information (Aitschul et al., 1990). Homology between two nucleotide sequences was assessed by using Genetyx-Mac Version 10.1 (Software Development Co., Ltd., Tokyo, Japan).

Cloning of the DNA fragment from the RLGS spot of interest To further characterize the nature of the genomic, we cloned RLGS spots by the restriction trapper-based RLGS spotcloning method (Ohsumi et al., 1995). To facilitate cloning of the RLGS fragments, NotI fragments were enriched from a 100 mg DNA sample using a restriction trapper (AGENE, Kamakura, Japan) that consisted of latex beads with covalently attached double-stranded DNA adapters having a free NotI restriction sequence at the end. DNA samples were digested sequentially with NotI and PvuII, extracted with phenol/chloroform/isoamyl alcohol, precipitated, and resuspended in double distilled H2O. The DNA was ligated to the NotI restriction trapper and then centrifuged to remove non-ligated DNA. The trapper-ligated DNA was digested with PvuII, centrifuged to remove PvuII fragments, and then digested with NotI. This NotI-enriched DNA was then subjected to RLGS, except that only one-fourth of it

Acknowledgments The authors would like to thank T Muramatsu and Y Inaba for their expert technical assistance and N Sugimoto (Sanwa Kagaku Kenkyusho Co Ltd) for his valuable advice on statistical analysis. This work was supported by the a Grant-in-Aid from the Ministry of Education, Science and Culture and the Ministry of Health and Welfare of Japan and by a National Grant-in-Aid for the Establishment of a High-Tech Research Center in a Private Japanese University.

References Aitschul SF, Gish W, Miller W, Myers EW and Lipman DJ. (1990). J. Mol. Biol., 215, 403 ± 410. Arii S, Tanaka J, Yamazoe Y, Minematsu S, Morino T, Fujita K, Maetani S and Tobe T. (1992). Cancer, 69, 913 ± 919. Baylin SB, Herman JG, Gra JR, Vertino PM and Issa JP. (1998). Adv. Cancer Res., 72, 141 ± 196. Bender J. (1998). Trends Biochem. Sci., 23, 252 ± 256. Blin N and Staord DW. (1976). Nucleic Acids Res., 3, 2303 ± 2308. Cravo M, Pinto R, Fidalgo P, Chaves P, Gloria L, Leitao CN and Mira FC. (1996). Gut, 39, 434 ± 438. Hatada I, Hayashizaki Y, Hirotsune S, Komatsubara H and Mukai T. (1991). Proc. Natl. Acad. Sci. USA, 88, 9523 ± 9527. Himeno Y, Fukuda Y, Hatanaka M and Imura H. (1988). Liver, 8, 208 ± 212. Hsu H-C, Wu T-T, Wu M-Z, Sheu J-C, Lee C-S and Chen DS. (1988). Cancer, 61, 2095 ± 2099. Itano O, Ueda M, Kikuchi K, Shimazu M, Kitagawa Y, Aiura K and Kitajima M. (2000). Oncogene, 19, 1676 ± 1683.

Oncogene

Kim Y-I, Giuliano A, Hatch KD, Schneider A, Nour MA, Dallel GE, Selhub J and Mason JB. (1994). Cancer, 74, 893 ± 899. Kubo S, Nishiguchi S, Hirohashi K, Shuto T, Kuroki T, Minamitani S, Ikebe T, Yamamoto T, Wakasa K and Kinoshita H. (1998). Jpn. J. Cancer Res., 89, 419 ± 426. Laird PW and Jaenisch R. (1996). Annu. Rev. Genet., 30, 441 ± 464. Liew CT, Li H-M, Lo K-W, Leow CK, Chan JYH, Hin LY, Lau WY, Lai PBS, Lim BK, Huang J, Leung WT, Wu S and Lee JCK. (1999). Oncogene, 18, 789 ± 795. Lise M, Bacchetti S, Da Pian P, Nitti D, Pilati PL and Pigato P. (1998). Cancer, 82, 1028 ± 1036. Liver Cancer Study Group of Japan. (1997). Classi®cation of Primary Liver Cancer; ®rst English edition. Kanehara and Co., Ltd.: Tokyo, pp. 1 ± 40. Miwa W, Yashima K, Sekine T and Sekiya T. (1995). Electrophoresis, 16, 227 ± 232. Nagai H, Kim YS, Yasuda T, Ohmachi Y, Yokouchi H, Monden M, Emi M, Konishi N, Nogami M, Okumura K and Matsubara K. (1999a). Gene, 237, 15 ± 20.


Nagai H, Baba M, Konishi N, Kim YS, Nogami M, Okumura K, Emi M and Matsubara K. (1999b). DNA Res., 6, 219 ± 225. Narayan A, Ji W, Zhang XY, Marrogi A, Gra JR, Baylin SB and Ehrlich M. (1998). Int. J. Cancer, 776, 833 ± 838. Ohsumi T, Okazaki Y, Hirotsune S, Shibata H, Muramatsu M, Suzuki H, Taga C, Watanabe S and Hayashizaki Y. (1995). Electrophresis, 16, 203 ± 209. Qu G, Dubeau L, Narayan A, Yu MC and Ehrlich M. (1999). Mutat. Res., 42312, 91 ± 101. Shen L, Fang J, Qiu D, Zhang T, Yang J, Chen S and Xiao S. (1998). Hepato-Gastroenterology, 45, 1753 ± 1759. Shiraishi M, Sekiguchi A, Chuu YH and Sekiya T. (1999). Biol. Chem., 3801, 81 ± 84.

Sirabe K, Kanematsu T, Matsumata T, Adachi E, Akazawa K and Sugimachi K. (1991). Hepatology, 14, 913 ± 919. Soares J, Pinto AE, Cunha CV, Andre S, Barao I, Sousa JM and Cravo M. (1999). Cancer, 85, 112 ± 118. Teubner B and Schulz WA. (1994). Exp. Cell Res., 2102, 192 ± 200. Thoraval D, Asakawa J, Wimmer K, Kuick R, Lamb B, Richardson B, Ambros P, Glover T and Hanash S. (1996). Gene, Chrom. Cancer, 17, 234 ± 244. Zhang X-K, Huang D-P, Chiu D-K and Chiu J-F. (1987). Biochem. Biophys. Res. Commun., 142, 932 ± 938. Zrihan-Licht S, Weiss M, Keydar I and Wreschner DH. (1995). Int. J. Cancer, 28623, 245 ± 251.

797

Oncogene