|12 August 2002, Volume 21, Number 35, Pages 5400-5413|
|Table of contents Previous Article Next [PDF]
|DNA methylation in cancer: too much, but also too little|
1Human Genetics Program/SL31, Department of Biochemistry, Tulane Medical School, New Orleans, Louisiana, LA 70122, USA
2Tulane Cancer Center, Tulane Medical School, New Orleans, Louisiana, LA 70122, USA
Correspondence to: M Ehrlich, E-mail: email@example.com
Cancer-associated DNA hypomethylation is as prevalent as cancer-linked hypermethylation, but these two types of epigenetic abnormalities usually seem to affect different DNA sequences. Much more of the genome is generally subject to undermethylation rather than overmethylation. Genomic hypermethylation in cancer has been observed most often in CpG islands in gene regions. In contrast, very frequent hypomethylation is seen in both highly and moderately repeated DNA sequences in cancer, including heterochromatic DNA repeats, dispersed retrotransposons, and endogenous retroviral elements. Also, unique sequences, including transcription control sequences, are often subject to cancer-associated undermethylation. The high frequency of cancer-linked DNA hypomethylation, the nature of the affected sequences, and the absence of associations with DNA hypermethylation are consistent with an independent role for DNA undermethylation in cancer formation or tumor progression. Increased karyotypic instability and activation of tumor-promoting genes by cis or trans effects, that might include altered heterochromatin-euchromatin interactions, may be important consequences of DNA hypomethylation which favor oncogenesis. The relationship of DNA hypomethylation to tumorigenesis is important to be considered in the light of cancer therapies involving decreasing DNA methylation. Inducing DNA hypomethylation may have short-term anticancer effects, but might also help speed tumor progression from cancer cells surviving the DNA demethylation chemotherapy.
Oncogene (2002) 21, 5400-5413. doi:10.1038/sj.onc.1205651
DNA methylation; cancer; satellite DNA; LINE-1; imprinting; MAGE genes
Recently, there has been much emphasis on the critical role of DNA hypermethylation in human carcinogenesis (Baylin and Herman, 2000; Issa, 2000). In comparison, there has been inadequate attention paid to decreased methylation of DNA in cancer. The discovery of extensive cancer-associated DNA hypomethylation in the human genome (Feinberg and Vogelstein, 1983a,b; Gama-Sosa et al., 1983b) preceded that of cancer-linked DNA hypermethylation. DNA hypomethylation in cancer often affects more of the genome than does hypermethylation so that net losses of genomic 5-methylcytosine (m5C) are seen in many human cancers (Gama-Sosa et al., 1983b).
The role in carcinogenesis of cancer-linked hypermethylation of transcription control regions is clear because of the consequent transcriptional silencing of genes important for prevention of cancer (see other reviews in this issue). The biological significance of DNA hypomethylation in cancer is less understood. Nonetheless, experiments involving DNA methylation inhibitors in vivo and in vitro (Carr et al., 1984; Denda et al., 1985; Thomas and Williams, 1992) or analysis of DNA methyltransferase-deficient mice (unpublished data, L Jackson-Grusby and R Jaenisch cited in Ehrlich, 2000) indicate the importance of induced DNA hypomethylation to oncogenesis. Studies of tumorigenesis in rodents fed methyl-deficient diets (reviewed in Poirier, 1994 and Ehrlich, 2000) are in accord with a causative role for DNA hypomethylation in cancer, although the contribution of carcinogenic effects of these diets other than changes in DNA methylation is unclear (Pogribny et al., 1995). As described below, the high frequency of cancer-associated DNA hypomethylation, the nature of the affected sequences, and the absence of associations with DNA hypermethylation are consistent with an independent role for undermethylation of certain DNA sequences in cancer formation or tumor progression.
DNA modification in cancer: hypo- or hypermethylated relative to what?
Hypomethylation and hypermethylation of DNA are relative terms and denote less or more methylation than in some standard DNA. When applied to cancer epigenetics, that standard is normal tissue. However, there are considerable differences in the amounts and distribution of DNA methylation among different vertebrate tissues because DNA methylation is not only species-specific but also tissue-specific (Ehrlich et al., 1982; Gama-Sosa et al., 1983a). For example, at the extremes of 5-methylcytosine (m5C) contents of DNA from normal human postnatal somatic tissues, cerebellum DNA has 17% more m5C than heart DNA. The respective percentages of genomic C methylated in these tissues are 4.03±0.14 and 3.43±0.12 (from 12 determinations each) as determined by high performance liquid chromatography (HPLC) of DNA digested to deoxynucleosides (Ehrlich et al., 1982; Ehrlich et al., submitted). Moreover, these analyses are for organs containing mixtures of cell types. Pure populations of uncultured cells may differ even more in their genomic m5C content. Therefore, unless the studied tumor DNA is being compared to a relatively pure population of cells which is known to be the cell of origin of the tumor, and that is usually not the case, it is best to use DNA from a variety of normal tissues as the control. Another consideration in identifying cancer-specific differences in genome methylation is that the studied cells should be uncultured cell populations whenever possible because of the frequent changes in DNA methylation that occur upon cell culture (Smiraglia et al., 2001). In this review, only cancer-specific DNA methylation determined from uncultured cell populations will be discussed. Most of these changes, like most of the normal tissue-specific differences in DNA methylation patterns, involve CpG sites, the predominant, but not only, location of m5C in the vertebrate genome (Ramsahoye et al., 2000).
How do cancer-associated hypomethylation and hypermethylation differ in extent, frequency, and types of cancers affected?
Extent and frequency of human cancer-associated hypomethylation of DNA
As we first demonstrated by HPLC in a collaboration with Charles Gehrke (Gama-Sosa et al., 1983b), various cancers often have a low overall m5C content compared to many different kinds of postnatal tissues. Although both malignant and normal samples display considerable diversity in their DNA's m5C content, a high percentage of malignant tumors, especially metastases, have DNA with unusually low m5C contents relative to the normal tissues (Table 1). Differences in DNA methylation levels between tumors and normal tissues were probably underestimated in this study and many other cancer studies because of contamination with normal cells. In collaboration with Andrew Feinberg and Charles Gehrke (Feinberg et al., 1988), we compared colon adenocarcinomas and adjacent normal mucosa, which contains the cell type that gives rise to the malignancy. Significant reductions of 10 and 8% in the average genomic m5C levels were observed in the cancers and benign polyps compared with the normal mucosa. Furthermore, all of these tumors were less methylated than the adjacent, non-tumorous tissue. Therefore, how early DNA hypomethylation can be detected during tumorigenesis may depend on the type of tumor as well as the individual tumor sample.
In collaboration with Emerich Fiala, we have recently shown that about 60% of Wilms tumors, a pediatric cancer, were globally hypomethylated in their DNA compared to all five analysed normal postnatal somatic tissues (Ehrlich et al., submitted). Also, 17 out of 19 ovarian epithelial carcinomas examined displayed global DNA hypomethylation relative to microdissected ovarian epithelial cystadenomas (L Dubeau, E Fiala, and M Ehrlich, unpublished data). Our results on the ovarian tumors are similar to those from an earlier study (Cheng et al. 1997) in which the ovarian carcinomas had on the average 25% less m5C in their DNA than did the cystadenomas, which had not been microdissected. Wilms tumors and ovarian epithelial carcinomas are well suited for this kind of analysis because they are mostly composed of tumor cells with relatively small amounts of normal cells. Therefore, unlike many types of tumors, such as breast cancers, the percentage of normal cells is not expected to vary widely between samples.
These findings on overall genomic hypomethylation in human malignancies are in accord with early studies of rat hepatocellular carcinomas (Lapeyre and Becker, 1979; Lu et al., 1983). Those investigations showed that the m5C content of DNA from hepatocellular carcinomas was about 20-40% less than that of normal liver. Furthermore, these results could not simply be explained by increased cell cycling. This is consistent with our comparison of overall genomic methylation in various normal human tissues (Ehrlich et al., 1982).
Variety of cancers displaying DNA hypomethylation
Global genomic hypomethylation has been found in many types of human cancer, including prostate metastatic tumors vs normal prostate (Bedford and van Helden, 1987); leukocytes from B-cell chronic lymphocytic leukemia vs normal leukocytes (Wahlfors et al., 1992); hepatocellular carcinomas vs matched non-hepatoma liver tissue (Lin et al., 2001); and cervical cancer or high-grade dysplastic cervical lesions vs normal cervical tissue or low-grade dysplasia of the cervix (Kim et al., 1994). An exception is a report of higher levels of global DNA methylation in breast cancer (Hakkarainen et al., 1996) in a comparison of acid-hydrolyzed DNA from seven lobular carcinomas and that from 10 benign breast tumors. A possible complicating factor in the latter study is that DNA samples that were digested to the free bases by acid might have had varying amounts of RNase degradation products from contaminating RNA, which could influence the results. For this reason, hydrolysis of DNA to deoxynucleosides for m5C determination is preferable unless the DNA samples are carefully evaluated for lack of contaminating RNA or ribo-oligonucleotides. In the latter study, a small, but not statistically significant, increase in global m5C levels was observed in ductal carcinomas. In contrast, two other large-scale analyses of breast ductal carcinomas found significant global hypomethylation relative to normal breast tissue or benign breast lesions by analysis of patterns of HpaII- or MspI-digested DNA or in an in vitro methyl acceptor assay with SssI methyltransferase (Bernardino et al., 1997; Soares et al., 1999). In summary, global deficiencies in the genomic m5C content in various cancers vs control tissues are seen very much more often than overall increases in m5C levels in DNA, despite the frequent findings of localized hypermethylation in all the above types of cancers.
DNA hypermethylation in human cancer: extent, frequency, and types of cancers
Cancer-associated localized hypermethylation has usually been described for CpG islands, DNA sequences of about 1-2 kb that are (C+G)-rich and do not have the typical underrepresentation of CpG seen in vertebrate genomes (a much lower CpG frequency than predicted for a random sequence of that base composition). When they overlap promoters, CpG islands are usually, but not always (e.g. Lethe et al., 1998), constitutively unmethylated in normal tissues. Frequent and extensive hypermethylation of certain CpG islands is observed in a very wide variety of cancers (Figure 1). In a study of methylation of almost 1200 unselected CpG islands in 98 primary cancers (breast, colon, brain, head and neck, testicular, neuroectodermal tumors and acute myeloid leukemia) by restriction landmark genomic scanning (RLGS) of NotI/EcoRV fragments, it was estimated that an average of ~600 of ~45 000 CpG islands in the genome are hypermethylated in cancer vs control tissue with a maximum of ~4500 and a minimum of 0 (Costello et al., 2000). Among the tumors with the highest frequency of CpG island hypermethylation were colon cancers. Testicular cancers had the lowest frequency of this hypermethylation. Hypermethylation of 12 CpG islands at the 5' ends of tumor suppressor genes was investigated in a methylation-sensitive PCR analysis of DNA from 600 cancers, including most of the types of tumors mentioned above plus kidney, bladder, esophageal, pancreatic, and liver cancers. Cancer-associated hypermethylation was found in most tumors of a given type in at least a few of these CpG islands (Esteller et al., 2001). In both of these studies, the pattern of which CpG islands are most often hypermethylated varied with the kind of cancer (Costello et al., 2000; Esteller et al., 2001).
How do cancer-associated DNA hypomethylation andhypermethylation differ in the types of sequences affected?
Types of sequences affected by cancer-associated hypermethylation
An RLGS search for cancer-hypermethylated sequences indicated that most of the affected sequences in non-small cell lung cancers were CpG islands (Dai et al., 2001). However, the results were probably influenced by the use of the CpG methylation-sensitive NotI for the DNA digestion because NotI preferentially cleaves in CpG-rich regions of the mammalian genome. Cancer-associated hypermethylation is seen in CpG islands distant from the 5' end of the gene (Dai et al., 2001; Nguyen et al., 2001a) as well as in those that straddle the transcription start site. Therefore, the observed prevalence of cancer-linked hypermethylation in 5' CpG islands may be due to an ascertainment bias. Cancer-associated hypermethylation occurs not only in genes with a known or likely role in carcinogenesis, but also in genes not implicated in cancer, such as, MYOD at its 5' CpG island (Issa, 2000). Hypermethylation of ribosomal DNA genes was found in breast cancer (Yan et al., 2000), but reports of cancer-associated hypermethylation of repeated sequences in human cancer are rare.
Hypomethylation of highly repeated, interspersed DNA sequences in cancer
Hypomethylation has been observed very often in DNA repeats in diverse cancers (Figure 1). The phenomenon of repeat-induced gene silencing, which has been seen in mammals as well as plants and fungi (Garrick et al., 1998), is probably related to the finding that mammalian DNA repeats tend to be highly methylated in postnatal somatic tissues. The repeats that display tumor-associated hypomethylation include endogenous retrotransposons. Retrotransposons or retroviral-derived elements can have their transcription upregulated in vivo by DNA demethylation. This was concluded from studies of Dnmt1 knockout mouse embryos, interspecies mammalian hybrids, and mice with an inherited epigenetically controlled phenotype whose expression is regulated by a genetically linked retrotransposon (IAP) (Walsh et al., 1998; O'Neill et al., 1998; Morgan et al., 1999). Also, there is evidence for frequent activation of expression of full-length transcripts from retrotransposons in certain types of murine cancer (Dupressoir and Heidmann, 1997).
Several studies have demonstrated cancer-associated hypomethylation of LINE-1 (or L1), a highly repeated, interspersed human retrotransposon that is up to 6 kb in length. LINE-1 repeats constitute about 15% of the human genome, but of the ~4´105 copies of LINE-1 elements in the human genome, only about 30-60 are estimated to be competent for transposition (Sassaman et al., 1997). There have been occasional reports of cancer-associated retrotransposition-like insertions involving LINE-1 sequences (Miki et al., 1992; Morse et al., 1988), and they may mobilize cellular RNAs at low frequencies (Wei et al., 2001). Their activation can also lead to transcriptional interference involving neighboring genes (Whitelaw and Martin, 2001). However, retrotransposition of endogenous elements is implicated in disease much less frequently for humans than for mice (Kazazian and Moran, 1998).
LINE-1 hypomethylation was observed in chronic lymphocytic leukemia vs normal mononuclear blood cells (Dante et al., 1992), urinary bladder carcinomas compared to normal bladder (Jurgens et al., 1996), hepatocellular carcinomas vs non-tumorous 'normal' or cirrhotic tissue (Takai et al., 2000), and prostate carcinomas vs normal prostate and other normal tissues (Santourlidis et al., 1999). In the prostate carcinoma study, the important question of the percentage of tumor cells in the cancer samples was addressed by microscopic examination of paraffin-embedded sections of tumor adjacent to the frozen sample used for DNA isolation. The samples ranged from having ~10-90% carcinoma cells. The tumor DNAs with the highest levels of hypomethylation of LINE-1 sequences all had high percentages of tumor cells, but some of the tumor specimens that consisted of mostly cancer cells showed no detectable LINE-1 hypomethylation (Santourlidis et al., 1999). In a chemically induced mouse hepatoma, LINE-1 hypomethylation was also seen (Ushijima et al., 1997). That study involved methylation-sensitive representational difference analysis (MS-RDA) to survey HpaII-sensitive fragments in the tumor vs normal liver, with confirmation of the results by Southern blotting (Ushijima et al., 1997). LINE-1 hypomethylation was the only alteration seen repeatedly in the MS-RDA banding patterns from different tumors. Hypomethylation, including of single-copy DNA sequences, was observed more frequently than hypermethylation. In these studies of murine and human tumors, it was not demonstrated that hypomethylation of LINE-1 sequences increased the use of these repeats as transcription units in cancer, but it has been hypothesized that this hypomethylation might sometimes lead to the retrotransposition of the very small percentage of retrotransposition-competent copies of LINE-1 (Santourlidis et al., 1999).
Of even higher copy number than the LINE-1 repeats in the human genome are the Alu repeats (size, ~0.3 kb; copy number, >106), which also can be mobilized, thus leading occasionally to cancer-associated gene insertions (Rothberg et al., 1997; Schichman et al., 1994). These repeats constitute >10% of the genome. Frequent loss of methylation from the normally highly methylated Alu repeats in human cancer is a possibility, but this has not been investigated in more than a few tumors, which were from the germline (Rubin et al., 1994). Analysis of such tumor samples is complicated by the hypomethylated state of Alu repeats in normal sperm compared to postnatal somatic tissues (Rubin et al., 1994).
Hypomethylation of moderately repeated DNA sequences in cancer
Another class of human retrotransposons that is usually highly methylated and may become hypomethylated specifically in association with oncogenesis are endogenous retroviruses, especially, the HERV-K family. While there are only about 30-50 full-length HERV-K sequences in the human genome, there are an estimated 10 000 solitary long terminal repeats (LTRs) from HERV-K (Leib-Mosch et al., 1993). At least some of the solitary LTRs can drive reporter gene expression as determined from an HERV-K LTR-containing transgene (Casau et al., 1999). Expression levels from the transgene were high in adult mouse testes and very low or undetectable in other tissues as determined by RT-PCR and Northern blotting. In the previously mentioned study of urinary bladder cancers (Florl et al., 1999), cancer-associated hypomethylation of the HERV-K sequences was seen by Southern blotting with an HERV-K gag probe. Although analysis of the presence of HERV-K transcripts in these bladder cancers was not reported, evidence for demethylation of HERV-K gag sequences and correlated expression of Gag at the protein level in human testicular tumor samples was provided in a small-scale study involving Southern and Western blotting (Gotzinger et al., 1996). Also HERV-K env transcripts were detected by RT-PCR in 25 out of 55 breast cancer tissues and not in any of the 35 breast tissue samples from cancer-free patients, although DNA methylation was not assayed in that study (Wang-Johanning et al., 2001). Expression of HERV-K env was seen in 7 out of 40 samples of tissue adjacent to the cancer that were judged normal by gross pathology, but that probably contained some malignant cells as determined histologically. The presence of these transcripts specifically in breast tissue from the cancer patients was confirmed by RNA in situ hybridization and Northern blot analysis. The transcripts included HERV-K env RNA capable of encoding the HERV-K envelope protein. Moreover, high-level expression of HERV-K gag and env RNA is associated with germ cell and trophoblastic tumors (Herbst et al., 1996). That env RNA was detected in most assayed term placental samples (Wang-Johanning et al., 2001) may be related to much of term placental tissue being of trophoblast origin and placental DNA displaying global genomic hypomethylation (Ehrlich et al., 1982). With respect to methylated integrated viral DNA sequences, a number of reports examining latent human DNA viruses and the proviral form of exogenous retroviruses indicate that, sometimes, these can be activated by demethylation of critical control elements (Gutekunst et al., 1993; Mikovits et al., 1990; Robertson, 2000; Takacs et al., 1998; Tanaka et al., 1995).
Several types of non-viral, moderately repeated DNA sequences have been shown to be hypomethylated in certain cancers. These include sequences tandemly repeated at a small number of different regions of the genome. For example, a 9-kb DNA repeat (NBL1) in the short arm of the acrocentric chromosomes, the pericentromeric region (vicinity of the centromere) of chromosomes 3 and 4, and subtelomeric locations was hypomethylated in some human neuroblastomas as determined by RLGS on tumor DNA vs blood DNA followed by Southern blot analysis (Thoraval et al., 1996a). Also hypomethylated in many neuroblastomas was another non-satellite repeat (NBL2,1.2-kb) located in the centromeric regions of chromosomes 9, 13, 14, and 21 (Thoraval et al., 1996b). This sequence is about 60% homologous to the SstI repeat, a moderately repeated DNA sequence found in tandem copies at 4q31 and 19q13 (Epstein et al., 1987). In another RGLS study (Nagai et al., 1999), a variant CpG-rich SstI repeat that was present in tandem 1.4-kb repeats was found to be hypomethylated in 75% of hepatocellular carcinomas relative to apparently normal liver from the same patients. The repeat was also hypomethylated in sperm compared with various normal somatic tissues. Therefore, this sequence displays the type of sperm-specific hypomethylation that we previously defined for a heterogeneous group of CpG-rich sequences in human DNA (Zhang et al., 1985, 1987). Moreover, part of the sequence displaying cancer- and sperm-specific hypomethylation is 70% homologous to one of the sequences that we previously cloned on the basis of its sperm-specific hypomethylation (Sp-12; Accession number X07490) (Nagai et al., 1999; Zhang et al., 1987). Another of the sperm-specific hypomethylated sequences (Sp23; Accession number X07493) that we previously isolated (Zhang et al., 1987) is implicated in cancer-linked hypomethylation. It is part of the non-transcribed intergenic spacer for the 45S ribosomal RNA gene repeat. In 5 out of 10 analysed Wilms tumors, it was hypomethylated compared with a variety of normal postnatal somatic tissues (Qu et al., 1999b). A different portion of this rDNA non-transcribed spacer was found to be undermethylated in one of two examined human lung cancers relative to adjacent 'normal' tissue in another study (Shiraishi et al., 1999). The authors of the latter study propose that there may have been spreading of hypomethylation from the relatively unmethylated 45S rDNA repeat to the intergenic spacer during carcinogenesis.
Hypomethylation of satellite DNA in cancer
Satellite DNAs are tandem, high-copy-number repeats, composed of variations of oligonucleotide sequences and are usually found in constitutively heterochromatic chromosome regions. We observed hypomethylation at BstBI sites in the main DNA component of the centromere (satellite ; Sat) in chromosome 1 (Chr1) or in the unusually long region of juxtacentromeric (centromere-adjacent) heterochromatin of Chr1 and Chr16 (satellite 2; Sat2) in 40-90% of human breast adenocarcinomas, ovarian epithelial carcinomas, and Wilms tumors (Narayan et al., 1998; Qu et al., 1999a,b). Hypomethylation in these studies was defined in comparison with normal postnatal somatic tissues, all of which are highly methylated in these sequences. Hypomethylation of Chr1 Sat2 and of Chr16 Sat2 was usually concordant, which is consistent with their sequence homology. However, undermethylation of Chr1 Sat2 was also paralleled by hypomethylation of the above-described heterologous rDNA spacer in the acrocentric chromosomes' p-arms in the ten Wilms tumors analysed for methylation of both repeats (Qu et al., 1999b). In human hepatocellular carcinomas, the frequent hypomethylation of Sat2 DNA (Saito et al., 2001; Wong et al., 2001) was highly concordant with that of Sat3, the main DNA component of the long juxtacentromeric heterochromatin of Chr9 (Saito et al., 2001). Sat2, Sat3, and Sat DNA sequences are normally hypomethylated in human sperm compared with a wide variety of normal postnatal somatic tissues just as the rDNA spacer sequence is (Qu et al., 1999b; Zhang et al., 1987). Moreover, in mice it has been shown that both oocyte and sperm DNAs are hypomethylated in satellite sequences (Chapman et al., 1984). Therefore, hypomethylation of various DNA sequences in both cancers and germ cells seems to be a common theme.
We recently extended our initial Wilms and ovarian tumor studies to demonstrate that most of these cancers are hypomethylated in Sat (centromeric) sequences throughout the genome's chromosomes and not just in Chr1 (Ehrlich et al., submitted). We have obtained similar results for ovarian epithelial carcinomas (L Dubeau and M Ehrlich, unpublished results), indicating that the hypomethylation of Wilms tumor DNA is cancer-specific and not just related to the embryonic nature of the tissue from which these pediatric cancers arise. Furthermore, the global DNA demethylation, including satellite DNA demethylation, that occurs from the one-cell stage through the morula stage is reversed by genome-wide remethylation throughout the early murine embryoblast by 6.5-7.5 d.p.c. (Chapman et al., 1984; Oswald et al., 2000; Razin and Kafri, 1994). Therefore, the highly methylated state of satellite DNA in tissues other than extraembryonic tissues is established much before the appearance of metanephric blastema, the embryonic precursor tissue for Wilms tumor formation. Although there are methylation changes later in embryogenesis, they are much less wide-scale (Razin and Kafri, 1994). It is possible that during oncogenesis there was selection for rare cells in the untransformed precursor tissue that had pre-existing hypomethylation of satellite DNA. However, this type of caveat is generally applicable to examples of cancer-linked DNA methylation changes.
Our findings on cancer-associated centromeric satellite DNA hypomethylation are consistent with the above-described hypomethylation of moderately repeated, non-satellite sequences that hybridize to the centromeric region of a few chromosomes (Thoraval et al., 1996a,b). The importance of satellite DNA hypomethylation is suggested by the highly significant relationship between it and global m5C deficiency in Wilms tumors as well as in ovarian epithelial tumors (Qu et al., 1999a; Ehrlich et al., submitted). Therefore, one of the important consequences of global DNA hypomethylation that may favor oncogenesis might be the strongly associated satellite DNA hypomethylation.
Hypomethylation of single-copy genes in cancer
To study whether cancer-linked hypomethylation is found in DNA sequences of various degrees of repetitiveness, we examined methylation levels in randomly fragmented DNA fractionated by reassociation kinetics using DNA from a human gastric adenocarcinoma that had an abnormally low global m5C content (Gama-Sosa et al., 1983b). Hypomethylation was present in highly repetitive, middle repetitive, and single-copy DNA fractions. Correspondingly, many reports indicate that part of the cancer-associated hypomethylation occurs in single-copy gene regions (Figure 1). Some of these genes have no apparent relationship to carcinogenesis. For example, the genes encoding growth hormone, -globin, -crystallin, and -chorionic gonadatropin displayed hypomethylation at HpaII or HhaI sites in almost all of 12 examined colon adenocarcinomas and seven polyps relative to adjacent apparently normal colon tissue in a Southern blot analysis with cDNA or gene region probes (Goelz et al., 1985). Frequent hypomethylation of the -globin gene and less frequent undermethylation of the parathyroid hormone and catalase genes was seen in breast and colon adenocarcinomas (Ribieras et al., 1994). Hypomethylation within certain gene regions in cancer might have an effect on gene expression but is less likely to do so than when the undermethylation is in the promoter or known enhancer sequences. However, it is possible that hypomethylation within gene regions is sometimes coordinated with hypomethylation in upstream transcription regulatory sequences.
Several tumor- or proliferation-associated genes have been found to be hypomethylated in human cancers. For example, cancer-associated hypomethylation concordant with expression was observed for the pS2 gene (Martin et al., 1997), which encodes a pleiotropic factor implicated in the control of cell proliferation. Normal stomach tissue and five breast adenocarcinomas that transcribe pS2 were hypomethylated at HpaII sites at the 5' end of the gene relative to seven non-expressing breast carcinomas and normal, non-expressing breast and endometrium. In another study, methylation of the CpG-rich promoter of the HOX11 proto-oncogene in three T-cell lymphoblastic leukemia samples, a normal bone marrow sample, a sample from a patient in remission, a leukemic T-cell line, and a control T-cell line was compared by genomic sequencing. This promoter was hypomethylated in the three at-diagnosis leukemic samples and the leukemic cell line, all of which expressed the gene, and not in the control cell line, the normal bone marrow, or the sample from the patient in remission, all of which were non-expressing (Watt et al., 2000).
Genomic sequencing of the promoter of MN/CA9, a tumor antigen gene, from 13 renal cell carcinomas and normal kidney tissue revealed tumor-linked hypomethylation of 6 CpGs out of 10 (Cho et al., 2001). Hypomethylation at one of these sites (GCGC) was correlated with expression of the gene. A caveat of such correlative results is that hypomethylation of transcription control sequences can sometimes be an effect, rather than a cause, of transcriptional activation. In the above study, evidence was provided for a causative role for the cancer-associated hypomethylation in gene expression. A plasmid containing the MN/CA9 promoter driving reporter gene expression was methylated in vitro with HhaI methylase or SssI methylase before being used for transient transfection of a renal cell carcinoma cell line. Remarkably, even though there is only a single HhaI site (the above-mentioned GCGC) compared with many SssI sites (CpG) in the promoter region, both enzymes gave equally strong inhibition of reporter gene expression. More testing is needed for cause-and-effect relationships between cancer-associated promoter/enhancer hypomethylation and cancer-linked expression of genes, e.g. transfection experiments with in vitro methylation of reporter plasmids that mimics in vivo methylation patterns and controls for the effects of methylation of vector and reporter sequences.
In the third exon of c-MYC, hypomethylation of HpaII sites was observed in colorectal cancers and hepatocellular carcinomas compared to the corresponding normal tissues (Sharrard et al., 1992; Shen et al., 1998). Although the relationship of this hypomethylation to transcription remains to be demonstrated, it might be associated with the elevated c-MYC expression often observed in rapidly proliferating cells. A study of murine tumors induced by chlorinated acetic acids gave evidence for an association between DNA hypomethylation and c-Myc or c-Jun expression. Hypomethylation of the c-Myc and c-Jun promoters at HpaII sites and increased expression of the corresponding gene at the RNA and protein levels was seen in liver tumors relative to liver from untreated mice or non-tumorous liver from treated mice (Tao et al., 2000).
Some tumor marker genes display germ cell-specific and cancer-associated expression and can be activated by DNA demethylating agents. A group of genes, including members of the MAGE, LAGE, and GAGE families, encode tumor-specific antigens presented by Class I major histocompatibility proteins and are expressed in testes and certain types of cancers with no detectable expression in a wide variety of normal postnatal somatic tissues (de Backer et al., 1999; de Smet et al., 1996; Lethe et al., 1998). Expression of these genes can be induced in non-expressing cancer cell lines or normal cell lines by treatment with 5-azadeoxycytidine, and methylation of Ets motifs in the promoter of MAGE-1 interferes with binding of an Ets family transcription factor (de Backer et al., 1999; de Smet et al., 1996; Lethe et al., 1998). The MAGE-1 promoter is hypomethylated in sperm and MAGE-1-expressing cell lines, but not in non-expressing cell lines and leukocytes (de Smet et al., 1996; Serrano et al., 1996) Unmethylated promoter sequences were found for three MAGE family genes in lung cancers and, to a lesser extent, in surrounding tissue, but lung from non-cancer patients was not examined (Jang et al., 2001). Expression was significantly correlated with promoter hypomethylation. In ~40% of breast cancers, hypomethylation at the 3' end of TSP50, a protease-like gene expressed specifically in testes, was seen (Yuan et al., 1999). Hypomethylation in breast cancer and testes relative to normal somatic tissues was demonstrated by cleavage with HpaII (at unmethylated CCGG sites) and MspI (specifically at unmethylated GGCCGG sites).
Another group of genes that often have disruptions in their DNA methylation and expression patterns in cancer are imprinted genes (Malik and Brown, 2000). Imprinted genes are characterized by parent-of-origin specific expression, maintenance of which is crucial to normal development (Tycko, 2000). DNA methylation acts as a primary genomic signal for monoallelic expression, such that in the majority of cases, a methylated allele is switched off and an unmethylated allele switched on. Across a cluster of imprinted genes, differential methylation of alleles is focused at CpG islands known as differentially methylated regions (DMRs). The best studied pathological model of the consequences of disrupted imprinting is the Beckwith-Wiedemann syndrome (BWS) (Tycko, 2000). Aberrant imprinting of IGF2 at 11p15.5 is a key factor in the development of this overgrowth syndrome. IGF2 encodes a mitogen and fetal growth promoting protein, which is one of the most frequently dysregulated genes in human cancers (Schofield et al., 2001). Evidence indicates that its overexpression may contribute to the early stages of tumorigenesis (Malik and Brown, 2000). In some cases of BWS, a tumor-predisposing syndrome, there is paternal uniparental disomy of 11p15.5 (as a somatic mosaic) that results in an extra copy of the paternally transmitted, active IGF2 gene replacing the maternally inherited, inactive copy of this gene. Similarly, cancers frequently display relaxation of imprinting (sometimes called loss of imprinting) of imprinted genes, including IGF2. When imprinting is relaxed, the inactive imprinted allele becomes activated to give biallelic, and increased, expression. In most cancers displaying relaxation of imprinting, imprinting is altered by one of two pathways (Tycko, 2000). One is the loss of heterozygosity (LOH) of imprinted genes, i.e., the loss of maternal or paternal alleles. LOH can occur by mitotic recombination, deletion, or chromosome loss. When LOH involves mitotic recombination and the loss of the inactive allele is accompanied by the reciprocal gain of an extra copy of the active allele, relaxation of imprinting can result. Because of differential allelic DMR methylation, LOH usually gives changes in the DNA methylation pattern, but just because of selective loss of maternal or paternal alleles. The other pathway for relaxation of imprinting involves LOH-independent alterations resulting in alteration of normal allele-specific methylation patterns. It is usually due to abnormal de novo methylation at an imprint-associated DMR, such as hypermethylation of the DMR immediately upstream of H19, which acts as a chromatin insulator preventing IGF2 expression only when unmethylated (Sasaki et al., 2000). H19 and IGF2 genes are part of a 1-Mb imprinted domain including KvLQT1 and CDKN1C (see below). The H19 upstream DMR makes H19 and IGF2 reciprocally expressed at a given 11p15.5 locus. Hypermethylation of this DMR leads to relaxation of imprinting of IGF2 (biallelic expression instead of monoallelic expression) and loss of expression of H19 in a variety of cancers (Takai et al., 2001; Tycko, 2000).
Recently, several examples of cancer-related hypomethylation of the normally hypermethylated allele of an imprinted gene have been reported, although it remains to be demonstrated how much this hypomethylation contributes to changes in allele-specific expression. More than one DMR in the H19/IGF2 region (Ishihara et al., 2000; Sullivan et al., 1999) might play a role in imprinting because hypermethylation of the H19 upstream DMR (methylation of the normally unmethylated maternal allele) does not suffice for relaxation of imprinting (Cui et al., 2001). Another DMR was found to overlap exons 2 and 3 in human IGF2 (Sullivan et al., 1999). This gene is paternally expressed and maternally repressed in normal kidney. IGF2 displays maternal allele-specific methylation in normal kidney that was absent in the four examined LOH-negative Wilms tumors which exhibited biallelic expression of IGF2; however, methylation of the H19 upstream DMR was not examined in this study (Sullivan et al., 1999). Decreased DNA methylation by homozygous Dnmt1 knockout, which leads to embryonic death, results in activation of only the normally silent paternal allele of H19 (Li et al., 1993). This suggests that if IGF2 DMR methylation helps to silence the maternal allele, demethylation of this DMR may not suffice to activate it.
Hypomethylation of the paternally transmitted H19 upstream DMR (which is normally paternally methylated) was seen in two of six analysed bladder cancers (Takai et al., 2001). The authors were unable to test whether this hypomethylation resulted in relaxation of imprinting because of the lack of heterozygous markers in the appropriate samples. However, examination of somatic cell hybrids suggests that this promoter hypomethylation results in activation of expression (Takai et al., 2001), which is consistent with results from the Dnmt1 knockout study (Li et al., 1993). Hypomethylation of the promoter region of H19 in the paternally transmitted allele was associated with relaxation of imprinting in a few LOH-negative tumors in a study of lung cancer (Kondo et al., 1995). Frequent hypomethylation of the maternally methylated DMR in the intron downstream of exon 10 of KvLQT1 was seen in Wilms tumors (Tycko, 2000) and hepatocellular carcinomas (Schwienbacher et al., 2000). This intronic DMR may be an alternative imprint regulatory element for the H19/IGF2-containing imprinted domain working through an antisense transcript from the paternally expressed allele (Smilinich et al., 1999). In a study of hepatocellular carcinoma and adjacent non-tumorous liver without LOH in the CDKN1C/IGF2 region, evidence was found for tumor-specific hypomethylation of this KvLQT1 DMR being associated with gain of imprinting (reduced expression of one allele) and, sometimes, loss or relaxation of imprinting for CDKN1C or IGF2 (Schwienbacher et al., 2000).
There is another gene domain on 11p that appears to be imprinted and is frequently hypomethylated in Wilms tumors. At 11p13 an antisense gene, WT1-AS, overlaps the Wilms tumor suppressor gene WT1. In normal kidney, WT1-AS displays allele-specific expression and differential methylation in the antisense regulatory region immediately upstream of the antisense promoter (Malik et al., 2000). In contrast, sense WT1 is biallelically expressed in kidney. Unlike normal kidney, eight out of 10 studied LOH-negative Wilms tumors had only the unmethylated form of the allele and so apparently lost methylation from the maternally inherited allele, as detected by genomic sequencing and Southern blot analysis. One of the tumors displaying this hypomethylation was informative for allele-specific expression and shown to have biallelic expression of WT1-AS, unlike the adjacent, histologically normal kidney. Recently this analysis has been extended to include Wilms tumor pre-malignant lesions known as nephrogenic rests. These also display hypomethylation and biallelic expression, suggesting that relaxation of imprinting at this locus is an early event in tumorigenesis (K Malik, personal communication). Therefore, demethylation of the maternal allele during tumorigenesis might lead to activation of its expression. Interestingly, tumors that do have LOH in this region preferentially lose the silent, hypermethylated maternal allele. These results suggest a role in Wilms tumor formation for hypomethylation of the WT1-AS upstream regulatory region, upregulation of WT1-AS expression, and, perhaps also, increased WT1 sense expression (Malik et al., 2000).
Cancer-associated hypermethylation of promoter regions will probably be involved in gene dysregulation in tumors more often than cancer-associated hypomethylation because these regions are usually constitutively unmethylated in normal tissues. Nonetheless, as illustrated by the above-mentioned studies, it is likely that there are many unelucidated examples of cancer-associated gene-region hypomethylation that affect gene expression. Impediments to their discovery are that studies designed to look for alterations in methylation in gene regions in cancer are usually set up specifically to detect increased methylation, generally involve non-quantitative PCR techniques, and usually do not employ multiple types of control tissue. The cancer-related hypermethylation of CpG islands overlapping the 5' ends of tumor suppressor genes can often be analysed by non-quantitative methods. These assays are designed to detect methylation of multiple CpGs, and those islands tend to have very little or no methylation in normal tissues (although not always, and this could lead to erroneous conclusions). In collaboration with Peter Laird, we quantitatively analysed DNA from 31 Wilms tumors vs five types of normal tissues for hypermethylation at 13 CpG-rich regions by MethyLight, a bisulfite modification-dependent real-time PCR assay (Eads et al., 2000; Ehrlich et al., submitted). Almost all of the assayed regions were standard types of CpG islands, several of which displayed significant hypermethylation in the Wilms tumors. We found an unexpected example of cancer-associated hypomethylation at the 5' end of MTHFR (methylene tetrahydrofolate reductase). This region is CpG-rich, but not as much as in a typical island. The MethyLight assay revealed that there were varying levels of methylation of the 5' region of MTHFR in five different normal tissues, but nonetheless there was a statistically significant cancer-related undermethylation of this region (P<0.0005). The only previous report on methylation of MTHFR described this CpG-rich region as highly methylated in both normal and malignant esophageal tissue (Eads et al., 2001). MTHFR hypomethylation might be only a marker of Wilms tumors or it might contribute to altered DNA metabolism by affecting the expression of the gene and, hence, the pools of dNTPs and S-adenosylmethionine, the donor for methylating C residues in DNA. This hypomethylation probably would have been missed if the analysis for methylation had not been quantitative and had not included a variety of normal tissue DNAs.
Is cancer-associated DNA hypomethylation just aprelude to or byproduct of cancer-linked DNAhypermethylation?
In DNA from breast adenocarcinomas, we found no apparent relationship between the often-observed hypermethylation of the 5' CpG island of the E-cadherin gene and hypomethylation of Chr1 Sat2 (Narayan et al., 1998). In our recent study of Wilms tumors, 31 of the samples were analysed for DNA hypermethylation as well as hypomethylation. Four of the CpG islands examined for hypermethylation were hypermethylated in about 30 to 90% of the tumors compared to various postnatal tissues (Ehrlich et al., submitted). We found no significant positive or negative association between hypermethylation of any of the studied CpG islands and global DNA hypomethylation or satellite DNA hypomethylation (Ehrlich et al., submitted). A negative association between CpG island methylation and global or satellite DNA hypomethylation would have been predicted if tumors with much CpG island hypermethylation were less likely to also display DNA hypomethylation. A positive association might have been seen if CpG island hypermethylation preceded and provoked global genomic hypomethylation of vice versa. The absence of any such relationship suggests that tumor-associated DNA hypomethylation contributes to carcinogenesis or tumor progression separately from aberrant DNA hypermethylation and its attendant silencing of tumor suppressor genes.
Does cancer-associated DNA hypomethylation contribute to karyotypic instability?
The above-described highly methylated state of many retrotransposons and endogenous retroviruses in normal tissues, their frequent hypomethylation in cancer, and the repressive effects of DNA methylation on their promoter function suggest that hypomethylation of these types of sequences in cancer could contribute toward karyotypic instability, which facilitates tumor progression. Another type of repeated DNA sequence whose hypomethylation in cancer may foster chromosome instability is satellite DNA. As described above, cancers often display hypomethylation of the normally highly methylated satellite DNA sequences, including centromeric Sat and juxtacentromeric Sat2, both of which are in the pericentromeric heterochromatin. Sat2, which is found mostly in Chr1 and Chr16, is hypomethylated in all studied tissues and cell cultures from patients with the ICF syndrome (immunodeficiency, centromeric region instability, facial anomalies) (Jeanpierre et al., 1993; Tuck-Muller et al., 2000), a rare recessive disease. This hypomethylation is usually associated with mutations specifically affecting the catalytic domain of DNMT3B, the DNA methyltransferase 3B gene (Hansen et al., 1999; Okano et al., 1999; Xu et al., 1999). While Sat2 sequences are not the only ones that are hypomethylated in ICF cells (Kondo et al., 2000), there is only a small percentage decrease in DNA methylation in ICF genomes. We showed that DNA from the one assayed uncultured ICF tissue, ICF brain DNA, had only about 7% less m5C than normal brain DNA samples (Table 1) (Tuck-Muller et al., 2000). Much of this hypomethylation is in repeated DNA sequences (Kondo et al., 2000). Sat2 and, sometimes, adjacent centromeric regions of Chr1 and Chr16 are specifically targeted for almost all of the rearrangements seen in mitogen-stimulated ICF lymphocytes or untreated ICF lymphoblastoid cell lines (LCLs) (Jeanpierre et al., 1993; Tuck-Muller et al., 2000). In detailed studies of G-banded chromosomes of ICF LCLs from five unrelated patients with DNMT3B mutations, we found that decondensation of the Chr1 juxtacentromeric Sat2-rich region was the most frequent anomaly, and that this decondensation was often seen at the break-points of whole-arm deletions or multiradial chromosomes (three or more Chr1 and/or Chr16 arms fused in the pericentromeric region) (Ehrlich et al., 2001; Tuck-Muller et al., 2000). These results and our findings that 5-azadeoxycytidine or 5-azacytidine treatment of normal lymphoid cultures induces chromosome rearrangements identical to those of ICF lymphoid cells (Hernandez et al., 1997; Ji et al., 1997) suggest that DNA hypomethylation predisposes to heterochromatin decondensation, which in turn, facilitates recombination between repeated DNA sequences.
Pericentromeric rearrangements of Chr1 and Chr16 are overrepresented in diverse types of cancers, including hepatocellular carcinomas and Wilms tumors, and generally lead to gains of 1q and/or losses of 16q (Brito-Babapulle and Atkin, 1981; Le Baccon et al., 2001; Mitelman et al., 2001). In a study of hepatocellular carcinomas by comparative genomic hybridization, 23 out of the 36 tumors examined showed 1q gain; 22 of these displayed Chr1 Sat2 hypomethylation at BstBI sites (Wong et al., 2001). In the above-mentioned study of 35 Wilms tumors involving Southern blotting and karyotype analysis, we also saw overrepresentation of Chr1 rearrangements in the pericentromeric region, all of which were accompanied by 1q gain, but only five of the tumors had pericentromeric rearrangements of Chr1 (Ehrlich et al., submitted). Our examination of karyotypic changes by LOH analysis of Chr16 or karyotype analysis are consistent with hypomethylation of Chr1 Sat2, Chr16 Sat2, and general centromeric hypomethylation favoring pericentromeric chromosome rearrangements, including isochromosome formation, in cis (Ehrlich et al., submitted; Qu et al., 1999b), although our data are not as convincing about this relationship in Wilms tumors as those from the hepatocellular carcinoma study. From this Wilms tumor study, we could detect no relationship of hypomethylation of centromeric or juxtacentromeric satellite DNA and aneuploidy (Ehrlich et al., submitted). This is in agreement with the lack of aneuploidy in ICF cells, which always display juxtacentromeric hypomethylation of Chr1 and Chr16 Sat2 DNA (Jeanpierre et al., 1993; Tuck-Muller et al., 2000), although they usually do not exhibit centromeric hypomethylation (Miniou et al., 1997; Tuck-Muller et al., 2000).
Hypomethylation of DNA sequences other than satellite DNAs may also predispose to chromosome recombination. In support of the postulated involvement of abnormal DNA hypomethylation in aberrant recombination, Chen et al. (1998) recently showed in murine embryonic stem cells that global DNA hypomethylation induced by homozygous knockout of the main DNA methyltransferase gene (Dnmt1) increased abnormal DNA recombination and, thereby, mutagenesis at Hprt and tk genes. However, using different stably transfected murine embryonal stem cells, Chan et al. (2001) unexpectedly found that homozygous knockout of Dnmt1 decreased gene loss and point mutagenesis from a chimeric tk-neo transgene. The differences between these studies might be due to chromosome position effects. Analysis of another model system, an interspecies hybrid of mammalian origin, provided more evidence for DNA undermethylation favoring chromosome instability. In the sterile hybrid obtained from two species of kangaroos, genome-wide hypomethylation was accompanied by several types of chromosomal rearrangements including elongation of the centromeres from one of the parents and various translocations (O'Neill et al., 1998, 2001). The centromeric extension resulted from amplification of a retroviral element in the centromeric region and the amplified element was hypomethylated.
More studies are needed to elucidate the types of DNA sequences and cells that are more susceptible to rearrangements when the DNA is hypomethylated and the extent of the contribution of DNA hypomethylation to karyotypic instability in cancer. However, there is a mechanistic basis for DNA hypomethylation favoring recombination, especially at repeated DNA sequences. The significant association of global DNA hypomethylation with satellite DNA hypomethylation in human cancer (Ehrlich et al., submitted; Qu et al., 1999a) and the frequent hypomethylation of both tandem and interspersed repeated DNA sequences in cancer (see above) suggest that these repeated sequences are playing a special role in carcinogenesis when undermethylated. Repeated DNA sequences should be more recombinogenic than unique sequences, but normally in eucaryotes such recombination is largely repressed. The repression may involve a more heterochromatic structure making the DNA sequences less accessible for recognition of interchromosomal and intrachromosomal DNA homology. For example, the very repetitious satellite DNAs are located in constitutive heterochromatin and are often highly methylated. DNA methylation has been linked to histone hypoacetylation and heterochromatinization (Garrick et al., 1998; Nguyen et al., 2001b). Therefore, it is likely that loss of DNA methylation favors recombination between repeated DNA sequences partly by loosening their packaging in chromatin.
Does cancer-associated DNA hypomethylation lead to altered gene expression that favors oncogenesis? Cis and possible trans effects
Among the previously mentioned cancer or proliferation-associated genes for which increased mRNA levels were correlated with hypomethylation in the promoter or the 5' end of the gene were pS2, HOX11, and MN/CA9 (Cho et al., 2001; Martin et al., 1997; Watt et al., 2000). Also, hypomethylation in transcription control regions of imprinted genes (see above) may promote carcinogenesis by increasing expression of paternally expressed imprinted genes, which usually contribute to increased cell growth. There is evidence that cancer-associated hypomethylation of proto-oncogene promoters is correlated with activation of expression (Tao et al., 2000). While more studies are needed to determine if hypomethylation of retrotransposons or endogenous retroviruses in human cancer causes increased transcription, there is evidence, as mentioned above, that abnormal DNA hypomethylation in model mammalian systems is associated with upregulation of expression of some of these elements (Morgan et al., 1999; O'Neill et al., 1998; Walsh et al., 1998).
Possible trans effects
The very high frequencies at which cancer-linked hypomethylation of DNA repeats is seen, its lack of a positive or negative association with CpG island hypermethylation, and its association with global DNA hypomethylation suggest that this hypomethylation aids tumorigenesis. Most of these repeats do not function as transcription units or do so only very poorly in vivo (e.g., satellite DNAs and LINE-1 repeats). In particular, the above-described cancer-associated satellite DNA hypomethylation (including Sat, Sat2, and Sat3) is unlikely to have a cis effect on transcription unless the hypomethylation spreads from these heterochromatic sequences into neighboring euchromatin. However, this satellite DNA hypomethylation might promote tumorigenesis by interfering with negative regulation of gene expression in trans. There are precedents for centromeric heterochromatin interacting with early lymphogenesis genes and the -globin locus in a manner that downregulates expression of those genes either directly or by acting as a reservoir for transcription control proteins that preferentially bind to them as well as to certain gene promoters (Cobb et al., 2000; Francastel et al., 2001; Sabbattini et al., 2001). In parallel, it has been suggested that the hypomethylation of juxtacentromeric heterochromatin of ICF syndrome lymphoid cells interferes with normal downregulation of certain genes by disrupting heterochromatin-euchromatin interactions in trans (Ehrlich et al., 2001). Therefore, centromeric and/or juxtacentromeric satellite DNA hypomethylation in tumors might be a new source of gene dysregulation in cancer from hypomethylated constitutive heterochromatin, perhaps in conjunction with cancer-linked alterations in expression from elsewhere in the genome.
Outlook and applications
While there is only a poor understanding of the pathways that lead to cancer-associated hypomethylation of some sequences and hypermethylation of others, there are some clues as to the types of sequences generally affected. For cancer-linked DNA hypermethylation, it seems to be usually CpG islands, and for hypomethylation, it is often, but not only, repeated DNA sequences. Furthermore, sequences that are known to be specifically undermethylated or expressed in germ cells or their precursors are often hypomethylated in cancer (Figure 1, satellite DNA, certain moderate-copy repeats, and testes-specific genes). Evidence suggests that this hypomethylation of DNA in cancer is not a consequence of or a prelude to hypermethylation in cancer. The hypomethylation in repeated DNA sequences may play special roles in carcinogenesis, such as, increasing karyotypic instability or affecting expression of genes indirectly. While there are some cancer-related genes with evidence for expression-linked hypomethylation in cancer, the number of studies looking for such associations and testing the effects of altering methylation experimentally has been very limited compared to analysis of repression-linked hypermethylation of tumor suppressor genes in cancer.
DNA hypermethylation is emerging as an important diagnostic and prognostic tool (see other articles in this issue). DNA hypomethylation analysis may also be useful in detecting cancer and managing the disease. Hypomethylation of repeated or single-copy DNA sequences has been significantly correlated with disease progression for some tumors (Qu et al., 1999a; Santourlidis et al., 1999; Shen et al., 1998). Furthermore, a number of studies have shown significant decreases in global DNA methylation levels with progressing tumor stage, tumor grade, or various other indicators of poor prognosis (Lin et al., 2001; Shen et al., 1998; Soares et al., 1999), and global DNA hypomethylation in cancer is significantly associated with hypomethylation of certain DNA repeats (Ehrlich et al., submitted; Qu et al., 1999a). In addition, for some types of cancer, DNA hypomethylation is seen as an early indicator of tumorigenesis (Feinberg et al., 1988; Qu et al., 1999a; Ribieras et al., 1994). Therefore, assays for DNA hypomethylation may become a clinically useful addition to hypermethylation analyses of CpG islands.
Studies of cancer-associated DNA hypomethylation sound a note of caution in the current development of cancer therapies aimed at decreasing DNA methylation in a non-targeted manner. As for cancer-associated DNA hypermethylation, cancer-linked DNA hypomethylation is generally unlikely to suffice for carcinogenesis. For example, normal placental DNA has an overall deficiency in m5C and hypomethylation of satellite DNA relative to normal postnatal somatic tissues. Furthermore, children with the ICF syndrome, a very rare DNA methylation-deficiency and chromosome breakage syndrome, have not been reported to have cancer. Nonetheless, the evidence described above suggests a frequent contributory role of cancer-linked genomic hypomethylation to tumorigenesis or tumor progression. Therefore, while suppressing CpG methylation may have a short-term beneficial effect in cancer treatment, it might help speed tumor progression from residual cancer cells.
This work was supported in part by NIH Grant CA81506. I thank Drs Karim Malik and Anthony Reeves for critical reading of the section on imprinted genes.
Baylin SB, Herman JG. (2000). DNA Alterations in Cancer: Genetic and Epigenetic Alterations Ehrlich M (ed) Natick: Eaton Publishing, pp. 293-309.
Bedford MT, van Helden PD. (1987). Cancer Res., 47: 5274-5276. MEDLINE
Bernardino J, Roux C, Almeida A, Vogt N, Gibaud A, Gerbault-Seureau M, Magdelenat H, Bourgeois CA, Malfoy B, Dutrillaux B. (1997). Cancer Genet. Cytogenet., 97: 83-89. MEDLINE
Brito-Babapulle V, Atkin NB. (1981). Cancer Genet. Cytogenet., 4: 215-225. MEDLINE
Carr BI, Reilly G, Smith SS, Winberg C, Riggs A. (1984). Carcinogenesis, 5: 1583-1590. MEDLINE
Casau AE, Vaughan JE, Lozano G, Levine AJ. (1999). J. Virol., 73: 9976-9983. MEDLINE
Chan MF, van Amerongen R, Nijjar T, Cuppen E, Jones PA, Laird PW. (2001). Mol. Cell. Biol., 21: 7587-7600. MEDLINE
Chapman V, Forrester L, Sanford J, Hastie N, Rossant J. (1984). Nature, 307: 284-286. MEDLINE
Chen RZ, Pettersson U, Beard C, Jackson-Grusby L, Jaenisch R. (1998). Nature, 395: 89-93. Article MEDLINE
Cheng P, Schmutte C, Cofer KF, Felix JC, Yu MC, Dubeau L. (1997). Br. J. Cancer, 75: 396-402. MEDLINE
Cho M, Uemura H, Kim SC, Kawada Y, Yoshida K, Hirao Y, Konishi N, Saga S, Yoshikawa K. (2001). Br. J. Cancer, 85: 563-567. MEDLINE
Cobb BS, Morales-Alcelay S, Kleiger G, Brown KE, Fisher AG, Smale ST. (2000). Genes Dev., 14: 2146-2160. MEDLINE
Costello JF, Fruhwald MC, Smiraglia DJ, Rush LJ, Robertson GP, Gao X, Wright FA, Feramisco JD, Peltomaki P, Lang JC, Schuller DE, Yu L, Bloomfield CD, Caligiuri MA, Yates A, Nishikawa R, Su Huang H, Petrelli NJ, Zhang X, O'Dorisio MS, Held WA, Cavenee WK, Plass C. (2000). Nat. Genet., 24: 132-138. Article MEDLINE
Cui H, Niemitz EL, Ravenel JD, Onyango P, Brandenburg SA, Lobanenkov VV, Feinberg AP. (2001). Cancer Res., 61: 4947-4950. MEDLINE
Dai Z, Lakshmanan RR, Zhu WG, Smiralia DJ, Rush LJ, Fruhwald MC, Brena RM, Li B, Wright FA, Ross P, Otterson GA, Plass C. (2001). Neoplasia, 3: 314-323. MEDLINE
Dante R, Dante-Paire J, Rigal D, Roizes G. (1992). Anticancer Res., 12: 559-563. MEDLINE
De Backer O, Arden KC, Boretti M, Vantomme V, De Smet C, Czekay S, Viars CS, De Plaen E, Brasseur F, Chomez P, Van den Eynde B, Boon T, van der Bruggen P. (1999). Cancer Res., 59: 3157-3165. MEDLINE
De Smet C, De Backer O, Faraoni I, Lurquin C, Brasseur F, Boon T. (1996). Proc. Natl. Acad. Sci. USA, 93: 7149-7153. MEDLINE
Denda A, Rao PM, Rajalakshmi S, Sarma DS. (1985). Carcinogenesis, 6: 145-146. MEDLINE
Dupressoir A, Heidmann T. (1997). Oncogene, 14: 2951-2958. MEDLINE
Eads CA, Danenberg KD, Kawakami K, Saltz LB, Blake C, Shibata D, Danenberg PV, Laird PW. (2000). Nucleic Acids Res., 28: E32. MEDLINE
Eads CA, Lord RV, Wickramasinghe K, Long TI, Kurumboor SK, Bernstein L, Peters JH, DeMeester SR, DeMeester TR, Skinner KA, Laird PW. (2001). Cancer Res., 61: 3410-3418. MEDLINE
Ehrlich M. (2000). DNA Alterations in Cancer: Genetic and Epigenetic Changes. Ehrlich M (ed) BioTechniques Books. Natick: Eaton Publishing, pp. 273-291.
Ehrlich M, Buchanan K, Tsien F, Jiang G, Sun B, Uicker W, Weemaes C, Smeets D, Sperling K, Belohradsky B, Tommerup N, Misek D, Rouillard J-M, Kuick R, Hanash S. (2001). Hum. Mol. Genet., 10: 2917-2931. MEDLINE
Ehrlich M, Gama-Sosa M, Huang L-H, Midgett RM, Kuo KC, McCune RA, Gehrke C. (1982). Nucleic Acids Res., 10: 2709-2721. MEDLINE
Epstein ND, Karlsson S, O'Brien S, Modi W, Moulton A, Nienhuis AW. (1987). Nucleic Acids Res., 15: 2327-2341. MEDLINE
Esteller M, Corn PG, Baylin SB, Herman JG. (2001). Cancer Res., 61: 3225-3229. MEDLINE
Feinberg AP, Gehrke CW, Kuo KC, Ehrlich M. (1988). Cancer Res., 48: 1159-1161. MEDLINE
Feinberg AP, Vogelstein B. (1983a). Nature, 301: 89-92. MEDLINE
Feinberg AP, Vogelstein B. (1983b). Biochem. Biophys. Res. Commun., 111: 47-54. MEDLINE
Florl AR, Lower R, Schmitz-Drager BJ, Schulz WA. (1999). Br. J. Cancer, 80: 1312-1321. Article MEDLINE
Francastel C, Magis W, Groudine M. (2001). Proc. Natl. Acad. Sci. USA, 98: 12120-12125. MEDLINE
Gama-Sosa MA, Midgett RM, Slagel VA, Githens S, Kuo KC, Gehrke CW, Ehrlich M. (1983a). Biochim. Biophys. Acta, 740: 212-219.
Gama-Sosa MA, Slagel VA, Trewyn RW, Oxenhandler R, Kuo KC, Gehrke CW, Ehrlich M. (1983b). Nucleic Acids Res., 11: 6883-6894. MEDLINE
Garrick D, Fiering S, Martin D, Whitelaw E. (1998). Nat. Genet., 18: 56-59. MEDLINE
Goelz SE, Vogelstein B, Hamilton SR, Feinberg AP. (1985). Science, 228: 187-190. MEDLINE
Gotzinger N, Sauter M, Roemer K, Mueller-Lantzsch N. (1996). J. Gen. Virol., 77: 2983-2990. MEDLINE
Gutekunst KA, Kashanchi F, Brady JN, Bednarik DP. (1993). J. Acquir. Immune Defic. Syndr., 6: 541-549. MEDLINE
Hakkarainen M, Wahlfors J, Myohanen S, Hiltunen MO, Eskelinen M, Johansson R, Janne J. (1996). Int. J. Cancer, 69: 471-474. Article MEDLINE
Hansen RS, Wijmenga C, Luo P, Stanek AM, Canfield TK, Weemaes CM, Gartler SM. (1999). Proc. Natl. Acad. Sci. USA, 96: 14412-14417. Article MEDLINE
Herbst H, Sauter M, Mueller-Lantzsch N. (1996). Am. J. Pathol., 149: 1727-1735. MEDLINE
Hernandez R, Frady A, Zhang X-Y, Varela M, Ehrlich M. (1997). Cytogenet. Cell. Genet., 76: 196-201. MEDLINE
Ishihara K, Hatano N, Furuumi H, Kato R, Iwaki T, Miura K, Jinno Y, Sasaki H. (2000). Genome Res., 10: 664-671. Article MEDLINE
Issa JP. (2000). DNA Alterations in Cancer: Genetic and Epigenetic Alterations. Ehrlich M (ed) Natick: Eaton Publishing, pp. 311-322.
Jang SJ, Soria JC, Wang L, Hassan KA, Morice RC, Walsh GL, Hong WK, Mao L. (2001). Cancer Res., 61: 7959-7963. MEDLINE
Jeanpierre M, Turleau C, Aurias A, Prieur M, Ledeist F, Fischer A, Viegas-Pequignot E. (1993). Hum. Mol. Genet., 2: 731-735. MEDLINE
Ji W, Hernandez R, Zhang X-Y, Qu G, Frady A, Varela M, Ehrlich M. (1997). Mutat. Res., 379: 33-41. Article MEDLINE
Jurgens B, Schmitz-Drager BJ, Schulz WA. (1996). Cancer Res., 56: 5698-5703. MEDLINE
Kazazian JrHH, Moran JV. (1998). Nat. Genet., 19: 19-24. MEDLINE
Kim Y-I, Giuliano A, Hatch KD, Schneider A, Nour MA, Dallal GE, Selhub J, Mason JB. (1994). Cancer, 74: 893-899. MEDLINE
Kondo M, Suzuki H, Ueda R, Osada H, Takagi K, Takahashi T. (1995). Oncogene, 10: 1193-1198. MEDLINE
Kondo T, Comenge Y, Bobek MP, Kuick R, Lamb B, Zhu X, Narayan A, Bourc'his D, Viegas-Pequinot E, Ehrlich M, Hanash S. (2000). Hum. Mol. Gen., 9: 597-604. MEDLINE
Lapeyre JN, Becker FF. (1979). Biochem. Biophys. Res. Commun., 87: 698-705. MEDLINE
Le Baccon P, Leroux D, Dascalescu C, Duley S, Marais D, Esmenjaud E, Sotto JJ, Callanan M. (2001). Genes Chrom. Cancer, 32: 250-264. MEDLINE
Leib-Mosch C, Haltmeier M, Werner T, Geigl EM, Brack-Werner R, Francke U, Erfle V, Hehlmann R. (1993). Genomics, 18: 261-269. MEDLINE
Lethe B, Lucas S, Michaux L, De Smet C, Godelaine D, Serrano A, De Plaen E, Boon T. (1998). Int. J. Cancer, 76: 903-908. MEDLINE
Li E, Beard C, Jaenisch R. (1993). Nature, 366: 362-365. MEDLINE
Lin CH, Hsieh SY, Sheen IS, Lee WC, Chen TC, Shyu WC, Liaw YF. (2001). Cancer Res., 61: 4238-4243. MEDLINE
Lu LJ, Randerath E, Randerath K. (1983). Cancer Lett., 19: 231-239. MEDLINE
Malik K, Brown KW. (2000). Br. J. Cancer, 83: 1583-1588. MEDLINE
Malik K, Salpekar A, Hancock A, Moorwood K, Jackson S, Charles A, Brown KW. (2000). Cancer Res., 60: 2356-2360. MEDLINE
Martin V, Ribieras S, Song-Wang XG, Lasne Y, Frappart L, Rio MC, Dante R. (1997). J. Cell. Biochem., 65: 95-106. Article MEDLINE
Miki Y, Nishisho I, Horii A, Miyoshi Y, Utsunomiya J, Kinzler KW, Vogelstein B, Nakamura Y. (1992). Cancer Res., 52: 643-645. MEDLINE
Mikovits JA, Raziuddin Gonda M, Ruta M, Lohrey NC, Kung HF, Ruscetti FW. (1990). J. Exp. Med., 171: 1705-1720. MEDLINE
Miniou P, Jeanpierre M, Bourc'his D, Coutinho Barbosa AC, Blanquet V, Viegas-Pequignot E. (1997). Hum. Genet., 99: 738-745. Article MEDLINE
Mitelman F, Johansson B, Mertens F. (2001). http://cgap.nci.nih.gov/Chromosomes/Mitelman.
Morgan HD, Sutherland HG, Martin DI, Whitelaw E. (1999). Nat. Genet., 23: 314-318. Article MEDLINE
Morse B, Rotherg PG, South VJ, Spandorfer JM, Astrin SM. (1988). Nature, 333: 87-90. MEDLINE
Nagai H, Baba M, Konishi N, Kim YS, Nogami M, Okumura K, Emi M, Matsubara K. (1999). DNA Res., 6: 219-225. MEDLINE
Narayan A, Ji W, Zhang X-Y, Marrogi A, Graff JR, Baylin SB, Ehrlich M. (1998). Int. J. Cancer, 77: 833-838. Article MEDLINE
Nguyen C, Liang G, Nguyen TT, Tsao-Wei D, Groshen S, Lubbert M, Zhou JH, Benedict WF, Jones PA. (2001a). J. Natl. Cancer Inst., 93: 1465-1472. MEDLINE
Nguyen CT, Gonzales FA, Jones PA. (2001b). Nucleic Acids Res., 29: 4598-4606. MEDLINE
O'Neill RJ, Eldridge MD, Graves JA. (2001). Mamm. Genome, 12: 256-259. MEDLINE
O'Neill R, O'Neill M, Graves JA. (1998). Nature, 393: 68-72. Article MEDLINE
Okano M, Bell DW, Haber DA, Li E. (1999). Cell, 98: 247-257.
Oswald J, Engemann S, Lane N, Mayer W, Olek A, Fundele R, Dean W, Reik W, Walter J. (2000). Curr. Biol., 10: 475-478. Article MEDLINE
Pogribny IP, Basnakian AG, Miller BJ, Lopatina NG, Poirier LA, James SJ. (1995). Cancer Res., 55: 1894-1901. MEDLINE
Poirier LA. (1994). Drug Metab. Rev., 26: 185-199. MEDLINE
Qu G, Dubeau L, Narayan A, Yu M, Ehrlich M. (1999a). Mut. Res., 423: 91-101.
Qu G, Grundy PE, Narayan A, Ehrlich M. (1999b). Cancer Genet. Cytogenet., 109: 34-39. MEDLINE
Ramsahoye BH, Biniszkiewicz D, Lyko F, Clark V, Bird AP, Jaenisch R. (2000). Proc. Natl. Acad. Sci. USA, 97: 5237-5242. MEDLINE
Razin A, Kafri T. (1994). Prog. Nucleic Acids Res. Mol. Biol., 48: 53-81.
Ribieras S, Song-Wang XG, Martin V, Lointier P, Frappart L, Dante R. (1994). J. Cell. Biochem., 56: 86-96. MEDLINE
Robertson KD. (2000). Curr. Opin. Microbiol. Immunol., 249: 21-34.
Rothberg PG, Ponnuru S, Baker D, Bradley JF, Freeman AI, Cibis GW, Harris DJ, Heruth DP. (1997). Mol. Carcinog., 19: 69-73. Article MEDLINE
Rubin CM, VandeVoort CA, Teplitz RL, Schmid CW. (1994). Nucleic Acids Res., 22: 5121-5127. MEDLINE
Sabbattini P, Lundgren M, Georgiou A, Chow C, Warnes G, Dillon N. (2001). EMBO J., 20: 2812-2822. Article MEDLINE
Saito Y, Kanai Y, Sakamoto M, Saito H, Ishii H, Hirohashi S. (2001). Hepatology, 33: 561-568. MEDLINE
Santourlidis S, Florl A, Ackermann R, Wirtz HC, Schulz WA. (1999). Prostate, 39: 166-174. Article MEDLINE
Sasaki H, Ishihara K, Kato R. (2000). J. Biochem. (Tokyo), 127: 711-715. MEDLINE
Sassaman DM, Dombroski BA, Moran JV, Kimberland ML, Naas TP, DeBerardinis RJ, Gabriel A, Swergold GD, Kazazian JrHH. (1997). Nat. Genet., 16: 37-43. MEDLINE
Schichman SA, Caligiuri MA, Strout MP, Carter SL, Gu Y, Canaani E, Bloomfield CD, Croce CM. (1994). Cancer Res., 54: 4277-4280. MEDLINE
Schofield PN, Joyce JA, Lam WK, Grandjean V, Ferguson-Smith A, Reik W, Maher ER. (2001). Toxicol. Lett., 120: 151-160. MEDLINE
Schwienbacher C, Gramantieri L, Scelfo R, Veronese A, Calin GA, Bolondi L, Croce CM, Barbanti-Brodano G, Negrini M. (2000). Proc. Natl. Acad. Sci. USA, 97: 5445-5449. MEDLINE
Serrano A, Garcia A, Abril E, Garrido F, Ruiz-Cabello F. (1996). Int. J. Cancer, 68: 464-470. Article MEDLINE
Sharrard RM, Royds JA, Rogers S, Shorthouse AJ. (1992). Br. J. Cancer, 65: 667-672. MEDLINE
Shen L, Fang J, Qiu D, Zhang T, Yang J, Chen S, Xiao S. (1998). Hepatogastroenterology, 45: 1753-1759. MEDLINE
Shiraishi M, Sekiguchi A, Chuu YH, Sekiya T. (1999). Biol. Chem., 380: 81-84. MEDLINE
Smilinich NJ, Day CD, Fitzpatrick GV, Caldwell GM, Lossie AC, Cooper PR, Smallwood AC, Joyce JA, Schofield PN, Reik W, Nicholls RD, Weksberg R, Driscoll DJ, Maher ER, Shows TB, Higgins MJ. (1999). Proc. Natl. Acad. Sci. USA, 96: 8064-8069. Article MEDLINE
Smiraglia DJ, Rush LJ, Fruhwald MC, Dai Z, Held WA, Costello JF, Lang LC, Eng C, Li B, Wright FA, Caligiuri MA, Plass C. (2001). Hum. Mol. Genet., 10: 1413-1419. MEDLINE
Soares J, Pinto AE, Cunha CV, Andre S, Barao I, Sousa JM, Cravo M. (1999). Cancer, 85: 112-118. Article MEDLINE
Sullivan MJ, Taniguchi T, Jhee A, Kerr N, Reeve AE. (1999). Oncogene, 18: 7527-7534. MEDLINE
Takacs M, Myohanen S, Altiok E, Minarovits J. (1998). Biol. Chem., 379: 417-422. MEDLINE
Takai D, Gonzales FA, Tsai YC, Thayer MJ, Jones PA. (2001). Hum. Mol. Genet., 10: 2619-2626. MEDLINE
Takai D, Yagi Y, Habib N, Sugimura T, Ushijima T. (2000). Jpn. J. Clin. Oncol., 30: 306-309. MEDLINE
Tanaka Y, Fukudome K, Hayashi M, Takagi S, Yoshie O. (1995). Int. J. Cancer, 60: 554-561. MEDLINE
Tao L, Yang S, Xie M, Kramer PM, Pereira MA. (2000). Cancer Lett., 158: 185-193. MEDLINE
Thomas GA, Williams ED. (1992). Carcinogenesis, 13: 1039-1042. MEDLINE
Thoraval D, Asakawa J, Kodaira M, Chang C, Radany E, Kuick R, Lamb B, Richardson B, Neel JV, Glover T, Hanash S. (1996a). Proc. Natl. Acad. Sci. USA, 93: 4442-4447.
Thoraval D, Asakawa J, Wimmer K, Kuick R, Lamb B, Richardson B, Ambros P, Glover T, Hanash S. (1996b). Genes Chromosomes Cancer, 17: 234-244.
Tuck-Muller CM, Narayan A, Tsien F, Smeets D, Sawyer J, Fiala ES, Sohn O, Ehrlich M. (2000). Cytogen. Cell. Genet., 89: 121-128.
Tycko B. (2000). DNA and Alterations in Cancer: Genetic and Epigenetic Alterations. Ehrlich M (ed) Natick: Eaton Publishing, pp. 333-349.
Ushijima T, Morimura K, Hosoya Y, Okonogi H, Tatematsu M, Sugimura T, Nagao M. (1997). Proc. Natl. Acad. Sci. USA, 94: 2284-2289. Article MEDLINE
Wahlfors J, Hiltunen H, Heinonen K, Hamalainen E, Alhonen L, Janne J. (1992). Blood, 80: 2074-2080. MEDLINE
Walsh CP, Chaillet JR, Bestor TH. (1998). Nat. Genet., 20: 116-117. Article MEDLINE
Wang-Johanning F, Frost AR, Johanning GL, Khazaeli MB, LoBuglio AF, Shaw DR, Strong TV. (2001). Clin. Cancer Res., 7: 1553-1560. MEDLINE
Watt PM, Kumar R, Kees UR. (2000). Genes Chromosomes Cancer, 29: 371-377. MEDLINE
Wei W, Gilbert N, Ooi SL, Lawler JF, Ostertag EM, Kazazian HH, Boeke JD, Moran JV. (2001). Mol. Cell. Biol., 21: 1429-1439. MEDLINE
Whitelaw E, Martin DI. (2001). Nat. Genet., 27: 361-365. Article MEDLINE
Wong N, Lam WC, Lai PB, Pang E, Lau WY, Johnson PJ. (2001). Am. J. Pathol., 159: 465-471. MEDLINE
Xu G, Bestor TH, Bourc'his D, Hsieh C, Tommerup N, Hulten M, Qu S, Russo JJ, Viegas-Péquignot E. (1999). Nature, 402: 187-191. Article MEDLINE
Yan PS, Rodriguez FJ, Laux DE, Perry MR, Standiford SB, Huang TH. (2000). Br. J. Cancer, 82: 514-517. MEDLINE
Yuan L, Shan J, De Risi D, Broome J, Lovecchio J, Gal D, Vinciguerra V, Xu HP. (1999). Cancer Res., 59: 3215-3221. MEDLINE
Zhang X-Y, Loflin PT, Gehrke CW, Andrews PA, Ehrlich M. (1987). Nucleic Acids Res., 15: 9429-9449. MEDLINE
Zhang XY, Wang RY, Ehrlich M. (1985). Nucleic Acids Res., 13: 4837-4851. MEDLINE
Figure 1 A summary of the most frequent types of sequences affected by cancer-specific DNA hypermethylation or hypomethylation. TSG, tumor suppressor gene
Table 1 Comparison of global DNA methylation levels in various tumours and normal somatic tissues as well as in normal vs ICF brain
|12 August 2002, Volume 21, Number 35, Pages 5400-5413|
|Table of contents Previous Article Next [PDF]