CpG island hypermethylation in progression of esophageal and gastric cancer


  • Fumiaki Sato M.D., Ph.D.,

    Corresponding author
    1. Department of Pathology, University of Maryland School of Medicine, Baltimore, Maryland
    2. Marlene and Stewart Greenebaum Cancer Center, University of Maryland School of Medicine, Baltimore, Maryland
    • Department of Pathology, University of Maryland School of Medicine, 655 West Baltimore St., BRB8012, Baltimore, MD 21201
    Search for more papers by this author
    • Fax: (410) 706-1099

  • Stephen J. Meltzer M.D.

    1. Marlene and Stewart Greenebaum Cancer Center, University of Maryland School of Medicine, Baltimore, Maryland
    2. Gastroenterology Division, Department of Medicine, University of Maryland School of Medicine, Baltimore, Maryland
    3. Gastroenterology Service, Veterans Affairs Medical Center, Baltimore, Maryland
    Search for more papers by this author


The upper gastrointestinal (GI) cancers have various carcinogenic pathways and precursor lesions, such as dysplasia for esophageal squamous cell carcinoma, Barrett esophagus for esophageal adenocarcinoma, and intestinal metaplasia for the intestinal-type of gastric cancer. Recently, many epigenetic events in carcinogenic pathways have been revealed, along with genomic and genetic alterations. This information has provided deeper insight into an understanding of the mechanisms of upper GI carcinogenesis. Moreover, detection methods of aberrant methylation have been applied to clinical fields to stratify high-risk groups, detect early cancer, and to predict clinical outcomes. In this review, a variety of information is summarized regarding gene hypermethylation in esophageal and gastric cancer. Cancer 2006. © 2005 American Cancer Society.

CpG islands are short CpG-rich regions frequently associated with gene promoter regions. Approximately 27,800 CpG islands have been identified in the human genome, according to the UCSC Genome Bioinformatics Site (http://genome.ucsc.edu), and approximately 60% of human genes have CpG islands near their 5′ ends. Most CpG islands are normally unmethylated, and a variety of transcriptional factors control gene expression. In normal cells, cytosine methylation of CpG sites within the gene promoter region is involved in the allele-specific inactivation of certain genes, such as imprinting on the inactive X-chromosome, and CpG methylation in the chromosomal centromeric region that contributes to chromosomal stability.

Aberrant CpG methylation is common in cancer development and may play an important role in the carcinogenic process. Methylation changes occurring in cancer include global hypomethylation in genomic DNA as well as gene-specific promoter hypermethylation. Whereas global hypomethylation increases mutation rates and chromosomal instability, promoter hypermethylation usually results in transcriptional gene inactivation. Thus, promoter hypermethylation, by silencing anticell-proliferation genes, antiapoptosis genes, antiangiogenesis genes, DNA repair genes, and antimetastasis genes, plays an important role in carcinogenesis.

Data on hypermethylation in cancer are progressively increasing. Epigenetic studies vary from single-gene to multigene analyses, from associations with clinicopathologic data to correlations with genetic alterations, and from basic research experiments to preclinical biomarker research. Herein, we summarize accumulated information on gene-specific hypermethylation in esophageal and gastric cancers and describe topics about these methylated genes in carcinogenic steps, microsatellite instability, methylator phenotype, and detection of aberrantly methylated DNA in remote specimens.

Esophageal Cancer

Esophageal cancer (EC) has two common histological types and several uncommon histological types. The two common histological types are squamous cell carcinoma (ESCC) and adenocarcinoma (EAC). No gene methylation study focusing on any particular uncommon esophageal tumor has ever been reported. Besides, the ESCC and EAC have different epidemiology and etiology. Thus, up-to-date knowledge about gene hypermethylation in both types of EC is summarized separately.

Methylated genes in esophageal squamous cell carcinogenesis

ESCC is the most predominant type of EC worldwide, and tobacco is a key factor in ESCC carcinogenesis. Histologically, ESCC develops in sequential steps through normal squamous epithelium, basal cell hyperplasia, dysplasia, and ESCC. Table 1 summarizes the prevalence of gene hypermethylation in EC that has been reported in multiple studies.

Table 1. Hypermethylated Genes in Esophageal Carcinogenesis
  1. NE: normal esophagus; D: dysplasia; ESCC: esophageal squamous cell carcinoma; BE: Barrett esophagus; DBE: dysplasia in Barrett esophagus; EAC: esophageal adenocarcinoma; Blank: not reported.

  2. Numbers represent the median value of reported prevalence (percentage) and range.

APC6.3 5080897021, 30, 42, 113–115
Signal transduction(0–14)  (40–82) (66–92) 
CDH117 701007021, 30, 44, 113, 116, 117
Cell-cycle regulation(0–19)    (0–84) 
DAP-K5.6    2021, 30
GSTP10  00521, 113
DNA repair       
hMLH10  001221, 113, 118
DNA repair(0–0)    (9–14) 
HPP13  44 68115, 118
Growth factor     (64–71) 
MGMT21 3943895930, 113, 115, 119
DNA repair(12–57)  (25–60) (23–73) 
p14/ARF0 18400113–115, 120–122
Cell-cycle regulation(0–0) (15–22)(0–8.3) (0–9.1) 
p15/INK4B3 131204.5113, 120, 122
Cell-cycle regulation   (11–13)   
p16/INK4A1.3715620434512, 21, 30, 44, 113–115, 120, 122–124
Cell-cycle regulation(0–29) (29–86)(3–66)(22–59)(33–82) 
RARβ255863   18, 125
Nuclear receptor(11–38) (55–70)    
RASSF1A3.7 51   19, 21
Signal transduction(3.1–4.3)      
RUNX-30.8  25 4821, 115
Transcriptional factor(0–1.6)      
TIMP34.8  60785630, 113, 115
MMP/angiogenesis(0–20)  (59–60) (20–86)

The genes Adenomatous polyposis coli (APC), E-Cadherin (CDH1), Cyclin-dependent kinase inhibitor 2A (p16/INK4A), Retinoic acid receptor β (RARβ), and Ras association domain family protein 1 (RASSF1A) are highly methylated in ESCC.

APC is regarded as the gatekeeper gene for colorectal cancer, in which it is the most early and frequent target of point mutation and allelic deletion.1 Although frequent loss of heterozygosity (LOH) (55–80%) of the APC locus (5q21) has been found in ESCC,2, 3 mutation of APC in this cancer type is rare.4 Thus, APC hypermethylation is the predominant mechanism of APC inactivation in ESCC.

CDH1 is a calcium-dependent adhesion molecule that plays a major role in the maintenance of intercellular junctions in normal epithelial cells.5 In ESCC, reduced expression of CDH1 is associated with tumor invasiveness, metastasis, and prognosis.6, 7 Although CDH1 protein levels are frequently reduced in ESCC, somatic mutation has not been reported and chromosomal loss at the CDH1 locus (16q22.1) is infrequent.8CDH1 hypermethylation was observed in 70% of ESCC, and could represent a predominant cause of reduced CDH1 expression.

The tumor suppressor gene p16/INK4A is a potent inhibitor of the cyclin-dependent kinase CDK4A. The inactivation of p16/INK4A allows the tumor cells to progress through the G1 checkpoint of the cell cycle.9 Somatic mutation of p16/INK4A is relatively infrequent in ESCC,10 even though allelic deletion at the chromosomal locus containing this gene is frequent (50–65%, 9q21).11, 12 However, p16/INK4A hypermethylation is common (56%) and a principal cause of p16 inactivation in ESCC. Indeed, the loss of the p16 function by either gene deletion or hypermethylation is associated with metastasis and a poor prognosis in ESCC.13, 14

RARβ and RASSF1A are tumor suppressor genes located on chromosome 3p whose loss is common in various malignancies.15 RARβ and Retinoid X receptor (RXR)-dependent signaling block cell proliferation induced by Cyclin D1 degradation.16 RASSF1A protein inhibits the anaphase-promoting complex and prevents cyclin A and cyclin B degradation until the spindle checkpoint becomes fully operational.17RARβ and RASSF1A are frequently methylated in ESCC (63% and 51%, respectively),18, 19 as well as undergoing frequent allelic alteration on chromosome 3p (LOH: 41% of ESCC20)

Among these genes, APC, CDH1, and RARβ hypermethylation have often been detected even in normal esophageal epithelium,21 whereas p16/INK4A and RASSF1A are generally unmethylated in normal cells. Thus, APC, CDH1, and RARβ may constitute age-related as well as cancer-related targets of promoter hypermethylation (type-A; see CpG island methylator phenotype [CIMP] section of gastric cancer). In contrast, p16/INK4A and RASSF1A should only be cancer-specific methylated genes (type-C).

Methylated genes in Barrett esophagus-associated carcinogenesis

Esophageal adenocarcinoma (EAC) is a malignant esophageal neoplasm arising predominantly in metaplastic columnar epithelium (Barrett esophagus: BE) in the setting of chronic gastroesophageal reflux. Histologically, EAC is believed to develop through the dysplasia-carcinoma sequence in BE, whereby specialized columnar epithelium progresses to low-grade dysplasia, high-grade dysplasia, and EAC. Actually, carcinoma foci typically appear adjacent to dysplasia.22 In addition, EAC is chiefly associated with a prior diagnosis of dysplasia by endoscopic surveillance.23 In the past decade the molecular biology underlying BE, including epigenetics, has become better focused.

As in ESCC, EAC is characterized by frequent methylation and a lower mutation rate of APC,4CDH1,24 and p16/INK4A25 compared with frequent allelic alteration at these loci. The reduced expression of CDH1 is associated with lymph node metastasis of EAC.26 Thus, hypermethylation of CDH1 may play a role in the metastatic behavior of EAC.

hMLH1 plays an important role in the DNA mismatch repair system, and hypermethylation of this gene causes frequent microsatellite instability (MSI) in a subset of sporadic colorectal, endometrial, and gastric cancers. However, the prevalence of MSI in EAC is much lower than in sporadic gastric, endometrial, or colorectal cancers.27, 28 The relatively lower frequency of hMLH1 hypermethylation in EAC is consistent with this lower MSI prevalence. O6-methylguanine-DNA methyltransferase (MGMT) removes alkylating lesions at position O6 of guanine and prevents a G:A mutation or a DNA strand break. Thus, MGMT inactivation by promoter hypermethylation increases carcinogenic risk and responsiveness to chemotherapy.29MGMT hypermethylation frequency is markedly elevated in BE (43%) and EAC (59%), although normal esophageal epithelia often show MGMT hypermethylation (21%). Moreover, MGMT hypermethylation is associated with a poor prognosis in EAC patients,30 as well as in lung cancer patients who smoke.31

Meanwhile, HPP1 and Tissue inhibitor of metalloproteinase 3 (TIMP3) are often methylated in both BE and EAC. HPP1 was cloned as a methylated gene in colorectal polyps,32 and it inhibits cell proliferation in prostate cancer.33TIMP3 is an antiangiogenic factor as well as a metalloproteinase inhibitor. TIMP3 inhibits the VEGF signal pathway and angiogenesis.34 Observed HPP1 (44%) and TIMP3 (56%) hypermethylation in BE may increase a malignant potential as a precursor of EAC.

Conversely, p14/ARF and p15/INK4B hypermethylation are uncommon in BE-associated carcinogenesis, whereas hypermethylation of both genes are common in colorectal (including ulcerative-colitis-associated) cancer35–37 and gastric cancer (see below).

Correlation between clinical outcome and hypermethylation in esophageal cancer

Recently, Brocks et al.30 showed that patients with densely methylated EAC in a multiple-gene methylation panel had a worse prognosis after surgery than did patients with less methylated tumors by the Kaplan–Meier method. Furthermore, the Cox Proportional Hazard Model showed that methylation status was associated with short-term (2-year) but not long-term survival. Thus, methylation status was a prognostic factor independent of other clinicopathological parameters.

Schulmann et al.38 applied a multiple-gene methylation panel to correlate the neoplastic progression-free survival of BE. The Cox model revealed that HPP1, p16/INK4A, and RUNX3 were independent prognostic factors for progression-free survival of BE. Moreover, the combined hazard ratios of progressor cases were elevated within 2 years before neoplastic progression in BE. Together in a multigene panel, methylation status could predict short-term outcome, but not long-term outcome.

Detection of DNA methylation in remote specimens of esophageal cancer patients

Conventional serum tumor biomarkers, such as CEA, AFP, and CA19-9, have been used for the early detection of tumors and for the monitoring of therapeutic effect. Squamous cell carcinoma (SCC) antigen and CYFRA21-1 are also clinically applied as EC biomarkers.39 However, these conventional biomarkers have low sensitivity and low specificity. Thus, researchers have sought other types of tumor biomarkers. Aberrant DNA methylation is an exciting target as a tumor biomarker. Recently, methylation detection methods have been applied to various types of remote specimens such as serum or plasma, urine, and stool.40–42

The average concentration of free DNA in serum is 13 ng/mL in noncancer-bearing individuals and 180 ng/mL in various cancer-bearing patients.43 A quantitative methylation-specific polymerase chain reaction (PCR) (MSP) method was applied to detect APC methylation in plasma samples of EAC patients.42 Patients without cancer did not show any false-positive amplification, whereas 25% of EAC patients demonstrated positive results. Moreover, plasma APC methylation status correlated with EAC patients' prognosis. Meanwhile, p16/INK4A MSP of plasma DNA had a sensitivity of 23% in ESCC patients. Because of the lack of specificity data, the clinical significance of this method remains undetermined.44

Gastric Cancer

Gastric cancer (GC) is the second-most frequent cause of cancer death worldwide.45 GC is histologically classified into two subtypes: the intestinal and the diffuse types.46 The precise mechanism underlying both types of gastric carcinogenesis is not fully understood. However, many reports regarding gene hypermethylation in gastric carcinogenesis have been published recently. In this section, several topics regarding gene hypermethylation in GC are reviewed.

Methylated genes in gastric carcinogenesis

Table 2 summarizes the prevalence of promoter hypermethylation in the carcinogenic steps in GC.

Table 2. Hypermethylated Genes in Gastric Carcinogenesis (GC)
  1. NG: normal gastric mucosa; CG: chronic gastritis; IM: intestinal metaplasia; AD: gastric adenoma; GAC: gastric carcinoma; Blank: not reported.

  2. Numbers represent the median value of reported prevalence (percentage) and range. Sequential histologic stages are related mainly in carcinogenesis of intestinal-type GC, not diffuse-type. However, most of the articles did not describe methylation prevalence in two types of GC separately. Thus, GAC includes both types of GC.

APC696581727821, 42, 113, 114, 126
Signal transduction(41–100)   (78–84) 
CDH1388557586121, 50, 52, 64, 67, 80, 81, 91, 92, 107, 113, 126–128
Cell-cycle regulation(12–54) (36–72) (29–81) 
CHFR3.1   37129, 130
Cell-cycle regulation    (35–39) 
COX-, 126, 131
Inflammatory response    (12–46) 
DAP-K553544345421, 78, 79, 81, 91, 107, 126, 132
Apoptosis(54–56) (39–49) (15–71) 
GSTP100001421, 78, 79, 81, 91, 107, 113, 126
DNA repair(0–3.4) (0–0) (0–19) 
hMLH11.7079.82421, 64, 67, 69, 76–81, 90, 91, 113, 126, 133
DNA repair(0–14) (6.3–22)(8.9–13)(5–35) 
HPP13.1   5069, 118
Growth factor    (47–54) 
MGMT25158.8102364, 79, 91, 104, 105, 113, 126
DNA repair    (14–61) 
p14/ARF9.43020313378, 79, 81, 91, 113, 114, 121, 126
Cell-cycle regulation(0–19) (8.3–32) (10–63) 
p15/INK4B9.6 11 5880, 81, 91, 107, 113
Cell-cycle regulation(0–19)   (45–73) 
p16/INK4A4.92.77113821, 67, 74–81, 90, 91, 107, 113, 114, 123, 133–137
Cell-cycle regulation(0–19)(0–72)(2.1–72)(8.8–12)(9–67) 
RASSF1A0.40002321, 67, 78, 79, 81, 90, 126
Signal transduction(0–5.3)   (21–26) 
RUNX398.128286521, 138
Transcriptional factor      
THBS11.91849343075, 76, 78, 79, 90, 113, 126
Angiogenesis(0–37)   (24–68) 
TIMP3201944284378, 79, 91, 113, 126
MMP/angiogenesis(17–23)(15–23)  (21–65) 

The APC gene is mutated in 25% of intestinal-type GC and in 20% of gastric adenomas, but uncommonly in diffuse-type GC.47, 48 In contrast to its mutation prevalence, APC is frequently hypermethylated in sequential carcinogenic steps, as well as in normal gastric mucosae. Thus, APC hypermethylation represents an early event in gastric carcinogenesis.

Defects in CDH1 function are specifically associated with diffuse-type GC, and also with metastasis and prognosis of all GCs.49 Somatic mutations of CDH1 are found in 50% of diffuse-type GCs, but were uncommon in intestinal-type GCs. Similar to the APC methylation pattern, CDH1 is often hypermethylated in any histologic type, including metastatic lesions.50 However, CDH1 is hypermethylated more frequently in diffuse-type GCs than in intestinal-type GCs.51, 52 Thus, hypermethylation of the CDH1 gene plays an important role in the carcinogenesis of diffuse-type GC and its metastasis.

p15/INK4B and p16/INK4A are cell-cycle regulator genes and p14/ARF inhibits MDM2-mediated p53 degradation. These three tumor suppressor genes are mapped to chromosome 9p21, which is frequently (22%) deleted in primary GCs.53 However, somatic mutation of these genes is uncommon in GC.54 Therefore, hypermethylation of these genes is a principal mechanism of inactivation of these genes in GC. There is a contrast in methylation patterns between these genes. The methylation prevalence of p15/INK4B and p16/INK4A is relatively tumor-specific, whereas p14/ARF hypermethylation occurs in precursor lesions as frequently as in tumors. Also, p16/INK4A is hypermethylated largely in intestinal-type GC,55 whereas p14/ARF methylation occurs predominantly in diffuse-type GC.51

RUNX3 is a key molecule in the TGFβ signaling pathway. RUNX3 knockout mice exhibited hyperplastic gastric mucosae in which apoptosis was suppressed.56RUNX3 mutation is rare in GC, and hemizygous deletion and hypermethylation are the main causes of RUNX3 down-regulation.57 In normal gastric mucosa, RUNX3 is hypermethylated more frequently in the antrum than in other parts of the stomach. RUNX3 methylation prevalence was increased in sequential steps of gastric carcinogenesis. Moreover, intestinal cell differentiation was found in gastric epithelial cells of RUNX3 knockout mice. This evidence suggests that RUNX3 hypermethylation could play a role in cell differentiation and proliferation in intestinal-type gastric carcinogenesis.

Microsatellite instability and hMLH1 methylation

Microsatellites are repetitive DNA sequences consisting of oligonucleotide units, which are distributed widely throughout the human genome. Microsatellite instability (MSI) connotes length mutation that occurs especially in microsatellites. MSI is frequently detected in some subsets of cancers with a deficient DNA mismatch repair system, such as hereditary nonpolyposis colorectal cancer and sporadic gastric, colon, and endometrial cancers.

Twenty-one percent (median of 7 large-scale reports; range, 16–50%) of GCs have been categorized as frequent MSI cancers (MSI-H).58 MSI-H cancers have unique clinicopathological characteristics, such as lymphocyte infiltration, tumor location (antrum), low-frequency metastasis, good therapeutic response, and a good prognosis59, 60

MSI within the gene coding region causes frameshift mutations (FSMs), which result in gene inactivation. Coding region MSI is relatively rare even in MSI-H cancer compared with noncoding region MSIs. However, some known FSMs have been described, such as in the TGFβ type II receptor (TGFBR2), IGF2 receptor (IGF2R), BCL2-associated X protein (BAX), hMSH3, and hMSH6.59, 61 Mori et al.62 demonstrated that this MSI-H-specific mutation spectrum forms an MSI-H-specific gene expression pattern in MSI-H colorectal cancer by using microarray analysis. The MSI-H gene expression pattern may cause unique characteristics in MSI-H GC.

A deficiency of the DNA mismatch repair system causes frequent MSI. Normal mismatch repair requires functional hMLH, hMSH, and hPMS protein molecules. In sporadic GC, mutations of the hMLH1, hPMS1, hPMS2, and hMSH2 genes are quite rare.63 Thus, hMLH1 inactivation by promoter hypermethylation is a predominant cause of mismatch repair deficiency and resultant frequent MSI. The majority of MSI-H GC has hMLH1 hypermethylation (median reported prevalence, 92%; range, 63–100%), whereas hMLH1 hypermethylation is uncommon in microsatellite-stable GC (median, 4.3%; range, 0–39%).64–69 Other hMLH and hMSH genes are generally unmethylated.70

Meanwhile, MGMT inactivation by promoter hypermethylation was associated with MSI-low more than MSI-high and stable colorectal carcinomas.71 However, in GC the specific relationship between MSI-low and MGMT hypermethylation has not been described.

CpG island methylator phenotype in gastric cancer

Toyota et al.72 proposed a novel molecular phenotype regarding colorectal carcinogenesis and gene hypermethylation in 1999. The authors identified 26 hypermethylated CpG islands (called “MINT”: methylated in tumor) in colorectal cancers by a methylated CpG island amplification method.73 These MINTs were classified into two types: 19 age-related methylated genes (type-A) and 7 cancer-specific methylated genes (type-C, MINT1, 2, 12, 17, 25, 27, 31). The authors found a frequent hypermethylation of type-C MINTs in a subset of cancer, and designated this phenotype the CpG island methylator phenotype (CIMP). Indeed, a subset of samples was frequently methylated in type-C genes (CIMP+), whereas other samples were mostly unmethylated (CIMP–).

Several authors discovered that 24–47% of GCs had the CIMP+ phenotype.51, 74–77 Other GCs also had low-frequency methylation in type-C MINTs,74, 76 unlike a concurrent methylation pattern in colorectal cancers.72 However, as in colorectal cancer, CIMP+ correlated with hypermethylation of other known cancer-related genes, such as p16, hMLH1, and THBS-1.74, 76, 77 Moreover, CIMP had been detected not only in tumors but also in premalignant lesions (i.e., intestinal metaplasia, 15%; adenoma, 49%).77 This finding suggested that CIMP could represent one of the early molecular events in gastric carcinogenesis.

Alternatively, other studies demonstrated a multigene methylation panel in GC. Because these studies included some type-A genes (e.g., CDH1, TIMP-3, and DAP-kinase),21, 78 the methylated-gene rate among the studied gene sets (sometimes referred to as MI: methylation index) and the CIMP concept are slightly different. MI was also correlated with the histological steps in gastric carcinogenesis.79–81 Moreover, CIMP and MI were associated with hMLH1 methylation and MSI status.64, 67, 74, 77

The cause of CIMP is still unknown. However, some reports suggested possible causes of CIMP.

DNA methyltransferase (DNMT) expression.

To date, several types of DNMT have been identified: DNMT1, DNMT2, DNMT3A, DNMT3B, and DNMT3L. DNMT1 acts as a maintenance-type DNA methyltransferase, and copies the methylation pattern of the parental strand to the daughter strand.82 Some workers proposed that DNMT1 possessed both maintenance and de novo DNA methylase activity. DNMT3A and 3B were identified as a de novo-type of DNA methyltransferase.82 One group demonstrated that DNMT1 overexpression was correlated with CIMP+ status.75, 76 DNMT1 may contribute to the initiation/maintenance of CIMP+ status. One report showed that CpG methylation status did not correlate with DNMT1, 3A, or 3B expression normalized by cell proliferation-related genes.83 However, in this study DNMT expression normalized by conventional housekeeping genes correlated with promoter methylation. Recently, several transcriptional variants of DNMT3A84 and DNMT3B85 have been found, and some of them do not have catalytic activity.86 Thus, a detailed analysis of the correlation between DNMT isoforms expression and the CIMP in GC is needed in the future.

Epstein-Barr virus (EBV) infection.

EBV is an oncovirus whose infection induces malignancies in the human body. Indeed, EBV has been detected in various malignancies87; 6.7% and 16% of GCs were infected by EBV in Japanese and North American series, respectively.88, 89 EBV status was correlated with MI,90, 91CDH1, and p16 hypermethylation92, 93 in GC. Thus, these findings suggested that EBV infection could induce hypermethylation of various human genes, and that EBV infection may be a cause of CIMP in GC. The connection between EBV infection and de novo methylation is still unclear. Whether EBV infection is correlated with DNMT1 overexpression is controversial.76, 91 EBV genomes existing in human cells as episomes are often methylated by a human defense system to silence EBV transcripts.94 Like the virus genome, some unknown mechanisms could induce methylation also in a subset of human genes.


Some articles suggest that diet could affect the methylation patterns of cells. Folate plays a central role in one-carbon metabolism, and folate deficiency leads to global DNA hypomethylation and chromosomal instability.95 Acetaldehyde produced from alcohol destroys folate, resulting in localized folate deficiency.96 In the Netherlands Cohort Study97 the prevalence of promoter hypermethylation was higher in colon cancers of patients with low-folate/high-alcohol intake than in patients with high-folate/low-alcohol intake. Moreover, a dietary methyl supplementation affected the methylation pattern of cells in a mouse model.98 Tea polyphenol, epigallocatechin-3-gallate, binds to the DNMT1 molecule and inhibits its function.99 Furthermore, vitamins (A, B6, and B12) and minor elements (cadmium, nickel, selenium, and zinc) deficiency also influences DNA methylation status.95 Taken together, these findings suggest that dietary intake could affect CIMP.

Helicobacter pylori and gene methylation in stomach

The intestinal-type gastric carcinogenesis has been described in terms of the presumed sequential steps: superficial gastritis, atrophic gastritis, intestinal metaplasia, and dysplasia. Helicobacter pylori infection may be involved in all steps.100 Long-standing inflammation and the release of reactive oxygen and nitrogen species may cause mutations and the development of GC.100 However, whether H. pylori infection causes gene hypermethylation in gastric mucosae is controversial. CDH1 hypermethylation in gastric dysplasia lesions was associated with H. pylori status.50 Nardone et al.101 speculated that H. pylori may cause the epigenetic alteration because of a stepwise increment of MI in the gastric carcinogenesis.79 However, in another study MI was not associated with H. pylori status.78 Also, CDH1 hypermethylation in GC was not associated with H. pylori status.50 Moreover, Tamura et al.21 reported that CDH1 hypermethylation prevalence in normal stomach of autopsy cases was increased in an age-related fashion, and that a similar pattern was observed not only in stomach but also in small and large intestines where H. pylori was unlikely to be involved. He concluded that the age effect is more dominant on CDH1 hypermethylation than H. pylori.

In contrast, aberrant methylation affects the reaction of gastric mucosae to H. pylori. COX-2 promoter hypermethylation regulates H. pylori-stimulated COX-2 expression in GC cells. COX-2 hypermethylation was observed not only in GC (27%) but also in normal (4.1%) and precursor (1.4–8.8%) lesions. Thus, COX-2 hypermethylation may modify the effect of H. pylori on gastric carcinogenesis.

Correlation between clinical outcome and hypermethylation in gastric cancer

According to single-gene analysis, CDH1103 and MGMT104, 105 hypermethylation in primary GC was associated with worse outcomes after surgery. Interestingly, the CIMP-high patients had a better outcome than CIMP-negative and -low patients.106 This result seemed to contradict Brock et al.'s article30 about prognosis and methylation profiling of EAC. However, the genes studied in the two articles were different, and CIMP in EC has not been studied yet. Therefore, further clinical investigation using a multiple-gene methylation panel is required.

Detection of DNA methylation in remote specimens of gastric cancer patients

As mentioned in the EC section, serum DNA is a useful remote material for tumor biomarkers. A single-gene approach cannot obtain sufficient sensitivity and specificity to discriminate the existence of GC. Thus, two groups tried a multiple-marker approach for GC patients. By using five markers (CDH1, DAP-kinase, GSTP1, p15 and p16) for serum MSP, the sensitivity of each gene varied from 15–56%. However, a combination of the five genes improved the sensitivity (83%).107 The serum MSP using three genes (p16, CDH1, and RARβ) obtained a sensitivity of 44%, whereas the RT-PCR method had 23% sensitivity to detect floating cancer cells in the blood.108

The serum MSP method is new and still developing. The significance of serum MSP is uncertain, because there are several unsolved concerns: 1) It is unknown if serum MSP can predict the patients' long-term prognosis. 2) A multi-institutional study is required to enlarge the sample size not only of cancer patients but also of normal age-matched controls. 3) The incorporation of an age-related gene will lead to a false-positive result in older individuals. 4) The value of the method as an early detection has not been shown. 5) The consistency of the result in a clinical time-course of the same patient. Future studies should address these issues.

Future Directions

After Herman and colleagues developed MSP in 1996,37 a large amount of information about DNA hypermethylation has been published. However, there are several forward-looking agendas in the methylation research field of EC and GC.

  • 1Because of a variety of detection methods (e.g., original MSP, quantitative MSP, COBRA, bisulfite sequencing, etc.) and sensitivities, the reported prevalence of DNA hypermethylation varied very much, as shown in the tables. Thus, to prevent confusion a certain degree of uniformity will be required to report the prevalence of gene hypermethylation.
  • 2Compared with the number of CpG islands (about 27,800) or CpG island-associated genes in the human genome, the number of genes studied extensively remains small. New high-throughput methods will be needed. Several researchers established array-based detection methods of DNA methylation.109, 110 However, an extensive study has not been published for EC and GC.
  • 3A demethylating agent, 5′-aza-2′-deoxycytidine, is a commonly used tool in the methylation research field. However, this agent globally demethylates genomic DNA during cell proliferation. This global change of methylation status makes it difficult to interpret the experimental results. In 2004, two articles demonstrated that siRNA targeting a gene promoter sequence induced gene-specific methylation and gene silencing.111, 112 A gene-specific demethylation method has not been discovered yet. The development of gene-specific control of DNA methylation would be beneficial as an experimental and potential therapeutic tool.
  • 4DNA hypermethylation has been used as early-detection biomarkers and prognostic factors. However, most studies remain in a method development phase. For validation of the method, a large-scale prospective study is necessary.