金田 篤志

Accumulation of epigenomic aberrations plays a critical role in tumorigenesis, both in tumor initiation and progression. Development of epigenomic markers can therefore be useful in determination of tumor risk of apparently normal cases, or in classifying tumor cases in relation to etiology, prognosis or therapy response. Loss of imprinting of IGF2 is an epigenetic aberration that exists in normal tissues and modifies tumor risk, and is a possible therapy target for tumor risk management. Thanks to development of quantitative and comprehensive analysis tools for DNA methylation, it becomes possible to epigenotype tumors, i.e., classification of tumors using epigenomic information, by unsupervised hierarchical clustering. Gastric cancer is epigenotyped in relation to etiology such as EB virus infection, and colorectal cancer is epigenotyped in relation to oncogene mutation statuses, suggesting distinct molecular mechanisms in colorectal tumorigenesis.

A subset of colorectal cancer shows significant accumulation of aberrant promoter methylation. Previously, we developed two groups of methylation markers that classified colorectal cancer into three epigenotypes: i) high-, ii) intermediate-, and iii) low-methylation epigenotypes. High-methylation epigenotype, with methylation of both group 1 and group 2 markers, correlates to BRAF-mutation((+)). Intermediate-methylation epigenotype, with methylation of group 2 markers, but not group 1, correlates to KRAS-mutation((+)). To gain insight into epigenotype development in colorectal carcinogenesis, especially intermediate-methylation epigenotype and its correlation to KRAS-mutation((+)) in adenoma, we analyzed methylation levels of group 1 and group 2 markers quantitatively by matrix assisted laser desorption ionization-time of flight mass spectrometry, in 51 adenomas, 13 aberrant crypt foci, and 26 normal mucosa samples, and we compared them to 149 previously analyzed colorectal cancer samples. Three serrated adenomas were all BRAF-mutation((+)), showing great methylation of group 1 and group 2 markers, thus high-methylation epigenotype. Forty-eight conventional adenomas were not methylated in group 1 markers and were classified into two clusters with higher and lower methylation of group 2 markers, thus into intermediate- and low-methylation epigenotypes, respectively. Adenoma with intermediate-methylation epigenotype correlated to KRAS-mutation((+)). Methylation levels of group 2 markers in adenoma were higher than aberrant crypt foci and normal samples, but similar to cancer. These data suggested that epigenotype development occur at an earlier stage than carcinoma formation, and already be completed at the adenoma stage. Intermediate methylation epigenotype and its correlation to KRAS-mutation((+)) were developed in conventional adenoma.

Whole-exome sequencing (Exome-seq) has been successfully applied in several recent studies. We here sequenced the exomes of 15 pancreatic tumor cell lines and their matched normal samples. We captured 162,073 exons of 16,954 genes and sequenced the targeted regions to a mean coverage of 56-fold. This study identified a total of 1517 somatic mutations and validated 934 mutations by transcriptome sequencing. We detected recurrent mutations in 56 genes. Among them, 41 have not been described. The mutation rates varied widely among cell lines. The diversity of the mutation rates was significantly correlated with the distinct MLH1 copy-number status. Exome-seq revealed intensive genomic instability in a cell line with MLH1 homozygous deletion, indicated by a dramatically elevated rate of somatic substitutions, small insertions/deletions (indels), as well as indels in microsatellites. Notably, we found that MLH1 expression was decreased by nearly half in cell lines with an allelic loss of MLH1. While these cell lines were negative in conventional microsatellite instability assay, they showed a 10.5-fold increase in the rate of somatic indels, e.g., truncating indels in TP53 and TGFBR2, indicating MLH1 haploinsufficiency in the correction of DNA indel errors. We further analyzed the exomes of 15 renal cell carcinomas and confirmed MLH1 haploinsufficiency. We observed a much higher rate of indel mutations in the affected cases and identified recurrent truncating indels in several cancer genes such as VHL, PBRM1, and JARID1C. Together, our data suggest that MLH1 hemizygous deletion, through increasing the rate of indel mutations, could drive the development and progression of sporadic cancers.

Introduction Cancer is known to arise through the accumulation of epigenetic alterations and genetic alterations (Feinberg et al., 2006 Jones and Baylin, 2007). Gene silencing is a major epigenetic gene-inactivation mechanism by DNA methylation of its promoter region, and is involved in the initiation and progression of cancer (Feinberg and Tycko, 2004 Jones and Baylin, 2007). Genome-wide approaches to search for aberrantly methylated regions in cancer have been developed since 1993, mainly since 1997 (Hayashizaki et al., 1993 Ushijima, 2005). Identification of novel inactivated genes in cancer using aberrant methylation as markers is useful to identify novel tumor-suppressor genes and methylation markers (Yoshikawa et al., 2001 Kaneda et al., 2004 Suzuki et al., 2004 Yu et al., 2005). Most of these methods utilized methylation-sensitive restriction enzymes, however, and thus only limited regions of genome could be analyzed (Kaneda et al., 2003 Ushijima, 2005), which means that detected methylation markers may not be comprehensive enough. For classification of cancer cases using DNA methylation data, a subset of colorectal cancer was found to show accumulation of CpG island methylation, so-called “CpG island methylator phenotype (CIMP)”, in 1999 by Toyota et al. (1999). CIMP-low was proposed by Ogino et al. (2006) as a subgroup with less extensive methylation of CIMP-related markers, which was associated with KRAS mutation compared with CIMP-high and CIMP(–) subgroups. Shen et al. (2007) reported through genetic and epigenetic analysis that colon cancer was classified into three subsets, CIMP-1, CIMP-2, and CIMP-negative. In Shen’s report, however, methylation levels of 27 regions were analyzed using four different methylation detection methods: 13 regions by pyrosequencing, seven by combined bisulfite restriction analysis (COBRA), six by methylated CpG island amplification, and one by methylation-specific PCR. Moreover, previously reported and frequently analyzed markers such as p16, hMLH1, and MINT (methylated-in-tumor) loci had been analyzed in most studies so far, and Ogino and Goel (2008) suggested that CIMP markers were considered to be specific for CIMP-high and not ideal for the identification of CIMP-low. New specific markers for CIMP-low and/or any other methylation phenotypes therefore needed to be identified if they really existed. To classify cancers using a more comprehensive approach, e.g., hierarchical clustering using highly quantitative methylation data by a single detection method using genome-widely selected novel regions, was required to be performed to classify cancers comprehensively.

Epstein-Barr virus (EBV) is associated with Burkitt lymphoma, nasopharyngeal carcinoma, opportunistic lymphomas in immunocompromised hosts, and a fraction of gastric cancers. Aberrant promoter methylation accompanies human gastric carcinogenesis, though the contribution of EBV to such somatic methylation changes has not been fully clarified. We analyzed promoter methylation in gastric cancer cases with Illumina's Infinium BeadArray and used hierarchical clustering analysis to classify gastric cancers into 3 subgroups: EBV(-)/low methylation, EBV(-)/high methylation, and EBV(+)/high methylation. The 3 epigenotypes were characterized by 3 groups of genes: genes methylated specifically in the EBV(+) tumors (e.g., CXXC4, TIMP2, and PLXND1), genes methylated both in EBV(+) and EBV(-)/high tumors (e.g., COL9A2, EYA1, and ZNF365), and genes methylated in all of the gastric cancers (e.g., AMPH, SORCS3, and AJAP1). Polycomb repressive complex (PRC) target genes in embryonic stem cells were significantly enriched among EBV(-)/high-methylation genes and commonly methylated gastric cancer genes (P = 2 × 10(-15) and 2 × 10(-34), respectively), but not among EBV(+) tumor-specific methylation genes (P = 0.2), suggesting a different cause for EBV(+)-associated de novo methylation. When recombinant EBV was introduced into the EBV(-)/low-methylation epigenotype gastric cancer cell, MKN7, 3 independently established subclones displayed increases in DNA methylation. The promoters targeted by methylation were mostly shared among the 3 subclones, and the new methylation changes caused gene repression. In summary, DNA methylation profiling classified gastric cancer into 3 epigenotypes, and EBV(+) gastric cancers showed distinct methylation patterns likely attributable to EBV infection.

Cellular senescence involves epigenetic alteration, e.g. loss of H3K27me3 in Ink4a-Arf locus. Using mouse embryonic fibroblast (MEF), we here analyzed transcription and epigenetic alteration during Ras-induced senescence on genome-wide scale by chromatin immunoprecipitation (ChIP)-sequencing and microarray. Bmp2 was the most activated secreted factor with H3K4me3 gain and H3K27me3 loss, whereas H3K4me3 loss and de novo formation of H3K27me3 occurred inversely in repression of nine genes, including two BMP-SMAD inhibitors Smad6 and Noggin. DNA methylation alteration unlikely occurred. Ras-activated cells senesced with nuclear accumulation of phosphorylated SMAD1/5/8. Senescence was bypassed in Ras-activated cells when Bmp2/Smad1 signal was blocked by Bmp2 knockdown, Smad6 induction, or Noggin induction. Senescence was induced when recombinant BMP2 protein was added to Bmp2-knocked-down Ras-activated cells. Downstream Bmp2-Smad1 target genes were then analyzed genome-wide by ChIP-sequencing using anti-Smad1 antibody in MEF that was exposed to BMP2. Smad1 target sites were enriched nearby transcription start sites of genes, which significantly correlated to upregulation by BMP2 stimulation. While Smad6 was one of Smad1 target genes to be upregulated by BMP2 exposure, Smad6 repression in Ras-activated cells with increased enrichment of Ezh2 and gain of H3K27me3 suggested epigenetic disruption of negative feedback by Polycomb. Among Smad1 target genes that were upregulated in Ras-activated cells without increased repressive mark, Parvb was found to contribute to growth inhibition as Parvb knockdown lead to escape from senescence. It was revealed through genome-wide analyses in this study that Bmp2-Smad1 signal and its regulation by harmonized epigenomic alteration play an important role in Ras-induced senescence.

A subgroup of colorectal cancer (CRC) shows non-random accumulation of aberrant DNA methylation, so-called CpG island methylator phenotype (CIMP), which was associated with microsatellite instability and BRAF mutation. As just one group of methylation markers was suitable to extract CIMP+/CIMP-high, and had been commonly used in the "one-panel method", it had been unclear whether another cluster of CRC with DNA methylation accumulation exists in microsatellite-stable CRC. We therefore epigenotyped CRC by a comprehensive approach, that is, the two-way unsupervised hierarchical clustering method using highly quantitative methylation data by a single detection method, MALDI-TOF mass spectrometry, on novel regions selected genome-widely through methylated DNA immunoprecipitation on array analysis. CRC was clearly clustered into three DNA methylation epigenotypes, high-, intermediate- and low-methylation epigenotypes (HME, IME, and LME, respectively). Methylation markers are clustered into two distinct groups: Group-1 methylated specifically in HME and including most reported CIMP-related markers; and Group-2 methylated both in HME and IME. While suitable markers to detect a subgroup of CRC with intermediate methylation and correlation to KRAS mutation have been expected to be developed, our data indicated that a "two-panel method" is necessary to properly classify CRC into three epigenotypes, the first panel to extract HME using Group-1 markers, and the second panel to divide the remaining into IME and LME using Group-2 markers. Here we review and compare our recent study and reported CRC classification methods by DNA methylation information, and propose the use of two panels of methylation markers as CRC classifiers.

Chronic infections by hepatitis B virus (HBV) and hepatitis C virus (HCV) appear to be the most significant causes of hepatocellular carcinoma (HCC). Aberrant promoter methylation is known to be deeply involved in cancer, including in HCC. In this study, we analyzed aberrant promoter methylation by methylated DNA immunoprecipitation-on-chip analysis on a genome-wide scale in six HCCs including three HBV-related and three HCV-related HCCs, six matched noncancerous liver tissues, and three normal liver tissues. Candidate genes with promoter methylation were detected more frequently in HCV-related HCC. Candidate genes methylated preferentially to HBV-related or HCV-related HCCs were detected and selected, and methylation levels of the selected genes were validated by quantitative methylation analysis using MALDI-TOF mass spectrometry using 125 liver tissue samples, including 61 HCCs (28 HBV-related HCCs and 33 HCV-related HCCs) and 59 matched noncancerous livers, and five normal livers. Among analyzed genes, preferential methylation in HBV-related HCC was validated in one gene only. However, 15 genes were found to be methylated preferentially in HCV-related HCC, which was independent from age. Hierarchical clustering of HCC using these genes stratified HCV-related HCC as a cluster of frequently methylated samples. The 15 genes included genes inhibitory to cancer-related signaling such as RAS/RAF/ERK and Wnt/beta-catenin pathways. Methylation of dual specificity phosphatase 4 (DUSP4), cytochrome P450, family 24, subfamily A, polypeptide 1 (CYP24A1), and natriuretic peptide receptor A (NPR1) significantly correlated with recurrence-free survival. It was indicated that genes methylated preferentially in HCV-related HCC exist, and that DNA methylation might play an important role in HCV-related HCC by silencing cancer-related pathway inhibitors, and might perhaps be useful as a prognostic marker.

PURPOSE: Whereas the CpG island methylator phenotype (CIMP) in colorectal cancer associates with microsatellite instability (MSI)-high and BRAF-mutation(+), the existence of an intermediate-methylation subgroup associated with KRAS-mutation(+) is controversial, and suitable markers for the subgroup have yet to be developed. Our aim is to clarify DNA methylation epigenotypes of colorectal cancer more comprehensively. EXPERIMENTAL DESIGN: To select new methylation markers on a genome-wide scale, we did methylated DNA immunoprecipitation-on-chip analysis of colorectal cancer cell lines and re-expression array analysis by 5-aza-2'-deoxycytidine/Trichostatin A treatment. Methylation levels were analyzed quantitatively in 149 colorectal cancer samples using matrix-assisted laser desorption/ionization-time-of-flight mass spectrometry. Colorectal cancer was epigenotyped by unsupervised two-way hierarchical clustering method. RESULTS: Among 1,311 candidate silencing genes, 44 new markers were selected and underwent quantitative methylation analysis in colorectal cancer samples together with 16 previously reported markers. Colorectal cancer was clustered into high-, intermediate-, and low-methylation epigenotypes. Methylation markers were clustered into two major groups: group 1 showing methylation in high-methylation epigenotype, and group 2 showing methylation in high- and intermediate-methylation epigenotypes. A two-step marker panel deciding epigenotypes was developed with 95% accuracy: the 1st panel consisting of three group-1 markers (CACNA1G, LOX, SLC30A10) to extract high-methylation epigenotype, and the 2nd panel consisting of four group-2 markers (ELMO1, FBN2, THBD, HAND1) and SLC30A10 again to divide the remains into intermediate- and low-methylation epigenotypes. The high-methylation epigenotype correlated significantly with MSI-high and BRAF-mutation(+) in concordance with reported CIMP. Intermediate-epigenotype significantly correlated with KRAS-mutation(+). KRAS-mutation(+) colorectal cancer with intermediate-methylation epigenotype showed significantly worse prognosis. CONCLUSIONS: Three methylation epigenotypes exist in colorectal cancer, and suitable classification markers have been developed. Intermediate-methylation epigenotype with KRAS-mutation(+) correlated with worse prognosis.

Loss of imprinting (LOI) of the insulin-like growth factor-II gene (IGF2), leading to abnormal activation of the normally silent maternal allele, is a common human epigenetic population variant associated with a 5-fold increased frequency of colorectal neoplasia. Here, we show first that LOI leads specifically to increased expression of proliferation-related genes in mouse intestinal crypts. Surprisingly, LOI(+) mice also have enhanced sensitivity to IGF-II signaling, not simply increased IGF-II levels, because in vivo blockade with NVP-AEW541, a specific inhibitor of the IGF-II signaling receptor, showed reduction of proliferation-related gene expression to levels half that seen in LOI(-) mice. Signal transduction assays in microfluidic chips confirmed this enhanced sensitivity with marked augmentation of Akt/PKB signaling in LOI(+) cells at low doses of IGF-II, which was reduced in the presence of the inhibitor to levels below those found in LOI(-) cells, and was associated with increased expression of the IGF1 and insulin receptor genes. We exploited this increased IGF-II sensitivity to develop an in vivo chemopreventive strategy using the azoxymethane (AOM) mutagenesis model. LOI(+) mice treated with AOM showed a 60% increase in premalignant aberrant crypt foci (ACF) formation over LOI(-) mice. In vivo IGF-II blockade with NVP-AEW541 abrogated this effect, reducing ACF to a level 30% lower even than found in exposed LOI(-) mice. Thus, LOI increases cancer risk in a counterintuitive way, by increasing the sensitivity of the IGF-II signaling pathway itself, providing a previously undescribed epigenetic chemoprevention strategy in which cells with LOI are "IGF-II addicted" and undergo reduced tumorigenesis in the colon upon IGF-II pathway blockade.

DNA methylation is an epigenetic mark crucial in regulation of gene expression. Aberrant DNA methylation causes silencing of tumor suppressor genes and promotes chromosomal instability in human cancers. Most of previous studies for DNA methylation have focused on limited genomic regions, such as selected genes or promoter CpG islands (CGIs) containing recognition sites of methylation-sensitive restriction enzymes. Here, we describe a method for high-resolution analysis of DNA methylation using oligonucleotide tiling arrays. The input material is methylated DNA immunoprecipitated with anti-methylcytosine antibodies. We examined the ENCODE region ( approximately 1% of human genome) in three human colorectal cancer cell lines and identified over 700 candidate methylated sites (CMS), where 24 of 25 CMS selected randomly were subsequently verified by bisulfite sequencing. CMS were enriched in the 5' regulatory regions and the 3' regions of genes. We also compared DNA methylation patterns with histone H3 and H4 acetylation patterns in the HOXA cluster region. Our analysis revealed no acetylated histones in the hypermethylated region, demonstrating reciprocal relationship between DNA methylation and histone H3 and H4 acetylation. Our method recognizes DNA methylation with little bias by genomic location and, therefore, is useful for comprehensive high-resolution analysis of DNA methylation providing new findings in the epigenomics.

INTRODUCTION: Risk prediction of gastric cancers is important to implement appropriate screening procedures. Although aberrant DNA methylation is deeply involved in gastric carcinogenesis, its induction by Helicobacter pylori, a strong gastric carcinogen, is unclear. Here, we analyzed the effect of H. pylori infection on the quantity of methylated DNA molecules in noncancerous gastric mucosae and examined its association with gastric cancer risk. EXPERIMENTAL DESIGN: Gastric mucosae were collected from 154 healthy volunteers (56 H. pylori negative and 98 H. pylori positive) and 72 cases with differentiated-type gastric cancers (29 H. pylori negative and 43 H. pylori positive) by endoscopy. The numbers of DNA molecules methylated and unmethylated for eight regions of seven CpG islands (CGI) were quantified by quantitative PCR after bisulfite modification, and fractions of methylated molecules (methylation levels) were calculated. RESULTS: Among healthy volunteers, methylation levels of all the eight regions were 5.4- to 303-fold higher in H. pylori positives than in H. pylori negatives (P < 0.0001). Methylation levels of the LOX, HAND1, and THBD promoter CGIs and p41ARC exonic CGI were as high as 7.4% or more in H. pylori-positive individuals. Among H. pylori-negative individuals, methylation levels of all the eight regions were 2.2- to 32-fold higher in gastric cancer cases than in age-matched healthy volunteers (P < or = 0.01). Among H. pylori-positive individuals, methylation levels were highly variable, and that of only HAND1 was significantly increased in gastric cancer cases (1.4-fold, P = 0.02). CONCLUSIONS: It was indicated that H. pylori infection potently induces methylation of CGIs to various degrees. Methylation levels of specific CGIs seemed to reflect gastric cancer risk in H. pylori-negative individuals.

Epigenetic alterations in cancer occur at least as commonly as genetic mutations, but epigenetic alterations could occur secondarily to the tumor process itself. To establish a causal role of epigenetic changes, investigators have turned to genetically engineered mouse models. Here, we review a recent study showing that a mouse model of loss of imprinting (LOI) of the insulin-like growth factor II gene (Igf2), which shows aberrant activation of the normally silent maternal allele, modifies the risk of intestinal neoplasia caused by mutations of the adenomatous polyposis coli (Apc) gene. This increased risk corresponds to the apparent increased risk of colorectal cancer in patients with LOI of IGF2. The model suggests that preexisting epigenetic alterations in normal cells increase tumor risk by expanding the target cell population and/or modulating the effect of subsequent genetic alterations on these cells, providing a novel idea for cancer risk management.

RUNX3 has been suggested to be a tumor suppressor of gastric cancer. The gastric mucosa of the Runx3-null mouse develops hyperplasia due to enhanced proliferation and suppressed apoptosis accompanied by a decreased sensitivity to transforming growth factor beta1 (TGF-beta1). It is known that TGF-beta1 induces cell growth arrest by activating CDKN1A (p21(WAF1)(/Cip1)), which encodes a cyclin-dependent kinase inhibitor, and this signaling cascade is considered to be a tumor suppressor pathway. However, the lineage-specific transcription factor that cooperates with SMADs to induce p21 expression is not known. Here we show that RUNX3 is required for the TGF-beta-dependent induction of p21 expression in stomach epithelial cells. Overexpression of RUNX3 potentiates TGF-beta-dependent endogenous p21 induction. In cooperation with SMADs, RUNX3 synergistically activates the p21 promoter. In contrast, RUNX3-R122C, a mutation identified in a gastric cancer patient, abolished the ability to activate the p21 promoter or cooperate with SMADs. Furthermore, areas in mouse and human gastric epithelium where RUNX3 is expressed coincided with those where p21 is expressed. Our results suggest that at least part of the tumor suppressor activity of RUNX3 is associated with its ability to induce p21 expression.

Loss of imprinting (LOI) of the insulin-like growth factor II gene (IGF2) is an epigenetic alteration that results in a modest increase in IGF2 expression, and it is present in the normal colonic mucosa of about 30% of patients with colorectal cancer. To investigate its role in intestinal tumorigenesis, we created a mouse model of Igf2 LOI by crossing female H19+/- mice with male Apc+/Min mice. Mice with LOI developed twice as many intestinal tumors as did control littermates. Notably, these mice also showed a shift toward a less differentiated normal intestinal epithelium, reflected by an increase in crypt length and increased staining with progenitor cell markers. A similar shift in differentiation was seen in the normal colonic mucosa of humans with LOI. Thus, altered maturation of nonneoplastic tissue may be one mechanism by which epigenetic changes affect cancer risk.

Neuroblastoma, one of the most common pediatric solid tumors, is characterized by two extreme disease courses, spontaneous regression and life-threatening progression. Here, we conducted a genome-wide search for differences in DNA methylation that distinguish between neuroblastomas of the two types. Three CpG islands (CGI) and two groups of CGIs were found to be methylated specifically in neuroblastomas with a poor prognosis. By quantitative analysis of 140 independent cases, methylation of all the five CGI (groups) was shown to be closely associated with each other, conforming to the CpG island methylator phenotype (CIMP) concept. The presence of CIMP was sensitively detected by methylation of the PCDHB CGIs and associated with significantly poor survival (hazard ratio, 22.1; 95% confidence interval, 5.3-93.4; P < 0.0001). Almost all cases with N-myc amplification (37 of 38 cases) exhibited CIMP. Even in 102 cases without N-myc amplification, the presence of CIMP (30 cases) strongly predicted poor survival (hazard ratio, 12.4; 95% confidence interval, 2.6-58.9; P = 0.002). Methylation of PCDHB CGIs, located in their gene bodies, did not suppress gene expression or induce histone modifications. However, CIMP was significantly associated with methylation of promoter CGIs of the RASSF1A and BLU tumor suppressor genes. The results showed that neuroblastomas with CIMP have a poor prognosis and suggested induction of silencing of important genes as an underlying mechanism.

The loss of thrombomodulin (TM) expression is associated with tumour growth, infiltration and lymph node metastasis in human tumours. In melanoma cell lines, TM is reported to mediate cell adhesion, and its introduction into TM-negative melanoma cell lines suppresses their growth. In this study, we analysed TM expression in surgical melanoma specimens and the role of its promoter methylation in the loss of its expression. In 15 (75%) of the 20 specimens (five from a primary site and 15 from metastatic sites), melanoma cells lacked TM immunoreactivity. Methylation of the TM promoter region was detected in 10 (67%) of the 15 TM-negative specimens by methylation-specific polymerase chain reaction, whereas methylation was detected in two (40%) of the five TM-positive specimens. In cell lines, complete methylation of the TM promoter CpG island was detected in six (46%) of 13 melanoma cell lines, whereas no methylation was detected in two cultured normal melanocytes. There was a good correlation between the methylated status of the CpG island and the loss of TM messenger RNA (mRNA) expression. Treatment of melanoma cell lines with a demethylating agent, 5-aza-2'-deoxycytidine, induced demethylation of the promoter CpG island and the restoration of mRNA and protein expression. These findings suggest that most human melanomas lack TM expression, and that methylation of the promoter CpG island is one of the mechanisms responsible.

Gastric cancer is the second most common cancer in the world [1,2]. Epigenetic alteration, especially aberrant DNA methylation, is highly involved in gastric carcinogenesis [3-7]. Promoter methylation is the major inactivation mechanism of most cancer-related genes except p53 in gastric cancers. Association of promoter methylation of some specific genes with histology of gastric cancers was reported. There is a subset of gastric cancers displaying frequent promoter methylations, and association of the frequent methylation with histology was also reported. There is another subset of gastric cancers displaying frequent promoter hypomethylation. Here, DNA methylation as involved in gastric carcinoma is reviewed. © 2005 Springer-Verlag Tokyo.

The unmethylated or methylated status of individual CpG sites is faithfully copied into daughter cells. Here, we analyzed the fidelity in replicating their methylation statuses in cancer cells. A single cell was clonally expanded, and methylation statuses of individual CpG sites were determined for an average of 12.5 DNA molecules obtained from the expanded population. By counting the deviation from the original methylation patterns inferred, the number of errors was measured. The analysis was done in four gastric cancer cell lines for five CpG islands (CGI), and repeated six times (total 1,495 clones sequenced). HSC39 and HSC57 showed error rates <1.0 x 10(-3) errors per site per generation (99.90-100% fidelity) for all the five CGIs. In contrast, AGS showed significantly elevated error rates, mainly due to increased de novo methylation, in three CGIs (1.6- to 3.2-fold), and KATOIII showed a significantly elevated error rate in one CGI (2.2-fold). By selective amplification of fully methylated DNA molecules by methylation-specific PCR, those were stochastically detected in KATOIII and AGS but never in HSC39 and HSC57. When methylation of entire CGIs was examined for eight additional CGIs, KATOIII and AGS had frequent methylation, whereas HSC39 and HSC57 had few. KATOIII and AGS had four and eight times, respectively, as high expression levels of DNMT3B as HSC39. These data showed that some cancer cells have decreased fidelity in replicating methylation patterns in some CGIs, and that the decrease could lead to methylation of the entire CGIs.

Disturbances of epigenetic information that result in changes in DNA methylation patterns are involved in carcinogenesis and other human disorders. Detection of agents that can cause epigenetic alterations--i.e. epimutagens--is therefore an important objective. We have developed and now describe the first detection system for demethylating agents that involves an endogenous promoter CpG island (CGI). After screening 10 promoter CGIs of genes silenced in human cancers, a CGI of the FLJ32130 gene was found to respond sensitively to a known demethylating agent, 5-aza-2'-deoxycytidine (5-aza-dC), by abundantly re-expressing its mRNA. After introducing the Hyg(r)-EGFP fusion gene into exon 3 of the FLJ32130 gene by homologous recombination, we isolated one clone that had the expected recombination outcomes and designated it F117. Two subclones (F117-47 and F117-123) of this original clone that did not share its propensity for leaky expression of the Hyg(r)-EGFP mRNA were then isolated, and methylation of their 5' CGI was confirmed. The addition of 5-aza-dC at doses of 0.1 microM or higher led to their 5' CGI being demethylated, and to Hyg(r)-EGFP being expressed; the anticipated fluorescence was readily confirmed by fluorescence microscopy. We believe that this is the first assay system that detects agents that disturb the methylated status of a CGI that regulates an endogenous promoter.

Lysyl oxidase (LOX) and HRAS-like suppressor (HRASLS) are silenced in human gastric cancers and are reported to have growth-suppressive activities in ras-transformed mouse/rat fibroblasts. Here, we analyzed whether or not LOX and HRASLS are tumor suppressor genes in human gastric cancers. Loss of heterozygosity and promoter methylation of LOX were detected in 33% (9 of 27) and 27% (26 of 96) of gastric cancers, respectively. Biallelic methylation and loss of heterozygosity with promoter methylation were also demonstrated in gastric cancers. Silencing of LOX was also observed in colon, lung, and ovarian cancer cell lines. As for mutations, only one possible somatic mutation was found by analysis of 96 gastric cancer samples and 58 gastric and other cancer cell lines. When LOX was introduced into a gastric cancer cell line, MKN28, in which LOX and HRASLS were silenced, it reduced the number of anchorage-dependent colonies to 57 to 61%, and the number of anchorage-independent colonies to 11 to 23%. Sizes of tumors formed in nude mice were reduced to 19 to 26%. Growth suppression in soft agar assay was also observed in another gastric cancer cell line, KATOIII. On the other hand, neither loss of heterozygosity nor a somatic mutation was detected in HRASLS, and its introduction into MKN28 did not suppress the growth in vitro or in vivo. These data showed that LOX is a tumor suppressor gene inactivated by methylation and loss of heterozygosity in gastric cancers, and possibly also in other cancers.

The Arp2/3 complex and filamins play important roles in organization of actin cytoskeleton, and thus in cellular morphology and locomotion. We recently identified decreased expression of a gene for one of seven subunits of the Arp2/3 complex, the p41-Arc gene, and silencing of a filamin gene, the FLNc gene, in human gastric cancers. In this study, gene expressions of the seven subunits of the Arp2/3 complex, including p41-Arc, and their methylation statuses were analyzed in human gastric cancers. Quantitative real-time RT-PCR analysis of 32 primary gastric cancer samples and eight gastric cancer cell lines revealed that expressions of all the seven genes were significantly decreased. All the 32 primary cancer samples showed decreased expression of at least one subunit, and 25 samples showed decreased expressions of four or more of the seven subunits. Methylation-specific PCR analysis showed that none of the CpG islands in the 5' regions of the six genes other than p41-Arc were methylated in primary gastric cancers or cell lines. The consistent decrease of the Arp2/3 complex genes and its important role in actin organization suggested that the decrease could be involved in cancer phenotypes, such as dysplastic morphology.

Hypomethylation of the global genome, considered to be composed mainly of repetitive sequences, is consistently observed in cancers, and aberrant hypo- and hypermethylation of CpG islands (CGIs) in promoter regions are also observed. Since methylation alterations in unique promoter sequences and in other genomic regions have distinct consequences, we analyzed the relationship between the global hypomethylation and the hypomethylation of unique promoter CGIs using human gastric cancers. Seven of ten gastric cancer cell lines showed marked decreases in 5-methylcytosine content, which correlated with hypomethylation of the LINE1 repetitive sequence. Six of the seven cell lines showed hypomethylation in five or all of the six normally methylated CGIs in promoter regions of six genes, and this was associated with induction of aberrant expression. The remaining three cell lines without global hypomethylation showed promoter hypomethylation in one or none of the six CGIs. Frequent promoter hypomethylation, however, did not correlate with frequent promoter hypermethylation. In primary gastric cancers too, global hypomethylation was associated with hypomethylation of LINE1 repetitive sequence and promoter hypomethylation. Of 93 gastric cancers, 33 cancers with frequent promoter hypomethylation and 27 cancers with frequent promoter hypermethylation constituted different groups. These findings represent experimental evidence that frequent hypomethylation of normally methylated promoter CGIs is associated with global hypomethylation, and that these hypomethylations occur independently of frequent promoter CGI hypermethylation.

The methylated or unmethylated status of a CpG site is copied faithfully from parental DNA to daughter DNA, and functions as a cellular memory. However, no information is available for the fidelity of methylation pattern in unmethylated CpG islands (CGIs) or its variation in the genome. Here, we determined the methylation status of each CpG site on each DNA molecule obtained from clonal populations of normal human mammary epithelial cells. Methylation pattern error rates (MPERs) were calculated based upon the deviation from the methylation patterns that should be obtained if the cells had 100% fidelity in replicating the methylation pattern. Unmethylated CGIs in the promoter regions of five genes showed MPERs of 0.018-0.032 errors/site/21.6 generations, and the fidelity of methylation pattern was calculated as 99.85%-99.92%/site/generation. In contrast, unmethylated CGIs outside the promoter regions showed MPERs more than twice as high (P < 0.01). Methylated regions, including a CGI in the MAGE-A3 promoter and DMR of the H19 gene, showed much lower MPERs than unmethylated CGIs. These showed that errors in methylation pattern were mainly due to de novo methylations in unmethylated regions. The differential MPERs even among unmethylated CGIs indicated that a promoter-specific protection mechanism(s) from de novo methylation was present.

Methylation-sensitive representational difference analysis (MS-RDA) was previously established to detect differences in the methylation status of two genomes. This method uses the digestion of genomic DNA with a methylation-sensitive restriction enzyme, HpaII, and PCR to prepare "HpaII-amplicons," followed by RDA. An HpaII-amplicon prepared using betaine and reverse electrophoresis was enriched 3.6-fold (compared with the HpaII-amplicon prepared by the original method) with DNA fragments originating from CpG islands (CGIs). As for the specificity of MS-RDA, it was shown that DNA fragments that are unmethylated in the tester and almost completely methylated in the driver are efficiently isolated. This indicated that genes that are in biallelic methylation or in monoallelic methylation with loss of the other allele are efficiently isolated. Further, by use of two additional methylation-sensitive six-base recognition restriction enzymes, SacII and NarI, more DNA fragments were isolated from CGIs in the 5' regions of genes. After analysis of human lung, gastric, and breast cancers, 12 genes were seen to be silenced and additional genes seen to show decreased expression in association with methylation of genomic regions outside CGIs in the 5' regions of genes. MS-RDA is effective in identifying silenced genes in various cancers.

Aberrant methylations in human gastric cancers were searched for by a genome scanning technique, methylation-sensitive representational difference analysis. Nine CpG islands (CGIs) in the 5' regions of nine genes, LOX, HRASLS, bA305P22.2.3, FLNc (gamma-filamin/ABPL), HAND1, a homologue of RIKEN 2210016F16, FLJ32130, PGAR (HFARP/ANGPTL4/ARP4), and thrombomodulin, were found to be methylated in two gastric cancer cell lines, MKN28 and MKN74, but unmethylated in the normal sample. Their expressions were lost in the two cell lines, and the causal roles of the methylations in gene silencing were shown by treatment with 5-aza-2'-deoxycytidine. In 41 primary gastric cancers, methylations of these CGIs in association with reduced expressions were observed at high incidences (29-41%) for five genes, including LOX and HRASLS, which have been reported to have tumor-suppressive activities. The other four genes were methylated at low incidences (0-7%). By analysis of the numbers of aberrant methylations in individual cancers, a subset of cancers with high prevalence of aberrant methylations was identified, and all of the 11 cancers in the subset were diffuse type. To analyze the possible involvement of decreased fidelity of maintenance methylation in this subset of cancers, aberrant hypomethylations of three normally methylated CGIs were examined. Cancers with high prevalence of hypomethylations of normally methylated CGIs, however, constituted a different subset. It is expected that these nine genes may include important genes in gastric cancer development and would be useful to identify a distinct subset of gastric cancers.

Hypermethylation of the 5' upstream region (5' region) of the human p16(CDKN2A) (p16) gene is known to cause silencing, which is involved in a wide range of human cancers. For the rat p16 gene, its 5' region has not been cloned, and it is uncertain whether surrogate use of exon 1 alpha is adequate for analysis of p16 silencing. In this study, we observed that methylation analysis of exon 1 alpha gave false positive results in three samples of normal rat mammary epithelia and in two of six primary mammary carcinomas. Therefore, we determined the nucleotide sequence of the 5' region of the rat p16 gene. To confirm that methylation status of the 5' region is correlated with p16 expression, the methylation status was analyzed by bisulfite sequencing and methylation-specific PCR in three samples of normal mammary glands, six samples of mammary carcinomas and four cell lines. The 5' region was demethylated in all of the three normal and six carcinoma samples that fully expressed p16. On the other hand, the 5' region was highly methylated in the 3Y1 cell line, which lacked p16 expression, but without deletion. These results showed that the methylation status of the 5' region was more closely correlated with p16 expression than that of the exon 1 alpha and analysis of the methylation status is useful in examining p16 silencing in various rat tumors.

Aberrantly methylated DNA fragments in a human gastric cancer were searched for by a genome-scanning method, methylation-sensitive-representational difference analysis (MS-RDA). Six DNA fragments flanked by CpG islands (CGIs) and hypermethylated in the cancer were isolated. Four of the 6 fragments possessed genes in their vicinities. Quantitative RT-PCR analysis of the 4 genes showed reduced expression of 2 genes in cancers: Insulin-induced protein 1 (INSIG1/CL-6) and p41 Arp2/3 complex (p41-Arc). As for INSIG1, a DNA fragment was derived from the edge of a CGI in the promoter region. The edge was methylated in 11 of 22 primary gastric cancers, whereas the center was not methylated in any cancer. INSIG1 expression was markedly reduced in 19 cancers, including the 11 cancers with the methylation. By 5-aza-2'-deoxycytidine treatment of 5 cell lines with the methylation of the edge, partial restoration of INSIG1 expression was detected only in 2 of them. These data indicated that, although the reduced INSIG1 expression in cancers was associated with the methylation at the edge of the CGI in the promoter region, the methylation was likely to be a secondary change. As for p41-Arc, a DNA fragment was derived from a CGI overlapping exon 8, and its methylation did not correlate with its expression. However, methylation of a CGI in the promoter region with a marked reduction of its expression was observed in 1 of the 22 primary cancers. INSIG1 and p41-Arc are known to be involved in cellular differentiation and morphology, respectively, and it was suggested that their reduced expressions might be involved in gastric cancer development or progression.

Runx3/Pebp2alphaC null mouse gastric mucosa exhibits hyperplasias due to stimulated proliferation and suppressed apoptosis in epithelial cells, and the cells are resistant to growth-inhibitory and apoptosis-inducing action of TGF-beta, indicating that Runx3 is a major growth regulator of gastric epithelial cells. Between 45% and 60% of human gastric cancer cells do not significantly express RUNX3 due to hemizygous deletion and hypermethylation of the RUNX3 promoter region. Tumorigenicity of human gastric cancer cell lines in nude mice was inversely related to their level of RUNX3 expression, and a mutation (R122C) occurring within the conserved Runt domain abolished the tumor-suppressive effect of RUNX3, suggesting that a lack of RUNX3 function is causally related to the genesis and progression of human gastric cancer.