¹Sydney Cancer Centre, Kanematsu Laboratories, Royal Prince Alfred Hospital, Missenden Road, Camperdown, NSW 2050, Australia

²Faculty of Medicine, University of Sydney, Sydney, NSW 2050, Australia

³CSIRO, Molecular Science, Sydney Laboratory, PO Box 184, North Ryde, NSW 1670, Australia

Correspondence to: S J Clark, CSIRO Molecular Science, Sydney Laboratory, PO Box 184, North Ryde, NSW 1670, Australia; E-mail: susan.clark@molsci.csiro.au

The DNA methylation pattern of a cell is exquisitely controlled during early development resulting in distinct methylation patterns. The tight control of DNA methylation is released in the cancer cell characterized by a reversal of methylation states. CpG island associated genes, in particular tumour suppressor or related genes, are often hypermethylated and this is associated with silencing of these genes. Therefore methylation is commonly convicted as a critical causal event in silencing this important class of genes in cancer. In this review, we argue that methylation is not the initial guilty party in triggering gene silencing in cancer, but that methylation of CpG islands is a consequence of prior gene silencing, similar to the role of methylation in maintaining the silencing of CpG island genes on the inactive X chromosome. We propose that gene silencing is the critical precursor in cancer, as it changes the dynamic interplay between de novo methylation and demethylation of the CpG island and tilts the balance to favour hypermethylation and chromatin inactivation.

Oncogene (2002) 21, 5380-5387. doi:10.1038/sj.onc.1205598

DNA methylation; gene expression; silencing; trigger; cancer

Introduction

Methylation of cytosine residues at CpG dinucleotides occurs by the addition of a methyl group to the carbon-5 position of cytosine through the action of the de novo DNA methyltransferase enzymes. So far three active DNA methyltransferase genes, DNMT1, DNMT3a and DNMT3b have been identified in mammalian cells (Bestor et al., 1988; Okano et al., 1998). Deletion of any of these genes in mice is lethal suggesting that methylation has an indispensable function in mammals (Li et al., 1992; Okano et al., 1999). The methylation profile of the cell is exquisitely controlled during development. The methylation pattern is transmitted to the daughter DNA strand during DNA replication by the action of the maintenance methyltransferase activity (Holliday and Pugh, 1975; Riggs, 1975). Methylation patterns are established in the early embryo with initial demethylation of the parental DNA in the first few cell divisions after fertilization, followed by de novo methylation of specific CpG sites between the eight cell stage and blastocyst implantation (Howlett and Reik, 1991; Mayer et al., 2000; Monk, 1990; Reik et al., 2001) (Global methylation changes in the early embryo are summarized in Figure 1). It was initially thought that methylation patterns, once established, were faithfully maintained at each cell division but with the advent of genomic sequencing it has become clear that the methylation state of any one CpG site is not always maintained. An apparent interplay between de novo methylation and demethylation at each cell division gives rise to a heterogenous pattern of methylation for any one molecule (Warnecke et al., 1998; Warnecke and Clark, 1999). However, even though heterogeneity can occur at individual CpG sites within a region, the actual regions of DNA that become methylated in the early embryo are shared.

Within the global DNA methylation changes that occur in early development there are regional specific differences in the methylation patterns that are established (Figure 1). In fact not all CpG sites are susceptible to methylation in the early embryo. The genome appears to be compartmentalized with respect to CpG methylation (Bird, 1992). The methylated compartments appear to coincide with regions of gene inactivity, whereas the unmethylated compartments coincide with regions of gene activity. Repeated sequences which comprise up to 35% of the genome are generally hypermethylated, as are gene coding regions and promoters of tissue specific genes (Yoder et al., 1997). In contrast, CpG sites in CpG islands remain unmethylated through development. CpG islands have a frequency of CpG dinucleotides approximately five times greater than the genome and as a whole comprise 1-2% of the genome. They are approximately 200 bp to several kb in length, spanning the promoter and first few exons of most house-keeping genes (Gardiner-Garden and Frommer, 1987). It has been estimated that there are 29 000 CpG islands in the genome and approximately 50-60% of all genes contain a promoter-associated island (Antequera and Bird, 1993). Even though most CpG islands are protected from methylation in the early embryo, some CpG islands undergo differential allelic methylation, for example CpG islands on the inactive X chromosome are methylated and CpG islands of many imprinted genes are also differentially methylated (Bird, 1986). Therefore in the one cell the DNA methylation profile for each sequence is finely regulated.

One of the major roles of DNA methylation in mammals is thought to be in control of gene regulation. This is because methylation within gene regulatory regions such as promoters and enhancers generally suppresses their function. For example, most promoter regions that are methylated in vitro, either from tissue-specific or CpG island associated genes, show reduced expression after transfection. Methylation induced suppression is thought to occur either by the blocking of transcription factor binding (Iguchi-Ariga and Schaffner, 1989; Molloy and Watt, 1990) and/or by formation of an inactive chromatin state (Bird and Wolffe, 1999; Nan et al., 1998). However, it is still unclear whether methylation directly elicits gene inactivation or is a consequence of gene silencing. For example, CpG islands on the inactive X chromosome are methylated subsequent to gene silencing.

Methylation changes in cancer

In contrast to the normal cell, the methylation pattern in a cancer cell is disrupted with major changes in the methylation compartments of the cells (Baylin et al., 1998; Jones, 1999) (Global DNA methylation changes that occur in a cancer cell are summarized in Figure 2). The bulk of the genome becomes hypomethylated, in particular the normally hypermethylated and silent regions containing the repetitive elements are substantially demethylated. Conversely the normally unmethylated CpG island-containing genes often become hypermethylated and inactivated.

The mechanism responsible for genomic hypomethylation in cancer is unclear. Several possibilities have been proposed including insufficient dietary folate or genetic lesions in the folate metabolic pathway, since the livers of rats fed folate deficient diets exhibit genome hypomethylation and increased DNA strand breaks and the rats typically develop liver cancer (Pogribny et al., 1997; Stern et al., 2000). In addition it is also possible that hypomethylation is a result of deregulation of putative demethylase enzymes or conversely due to dysfunction of the maintenance activity of the DNA methyltransferase enzymes. Despite the fact that the pathway leading to hypomethylation in cancer is unknown it is clear that hypomethylation is a hallmark of most cancer genomes. It has been proposed that hypomethylation contributes to malignancy by either contributing directly to the activation of oncogenes (Feinberg and Vogelstein, 1983; Holliday and Pugh, 1975), activation of latent retrotransposons (Bestor and Tycko, 1996) and/or chromosome instability (Ehrlich, 2000). Evidence for activation of oncogenes by specific gene demethylation in cancer is poor, however, hypomethylation has been reported in the body of some oncogenes including cMYC and H-RAS (Vachtenheim et al., 1994). Hypomethylation however has been reported to be responsible for the activation of the MAGE and related genes (de Smet et al., 1999). These genes are germline-specific and the promoters of the genes are normally methylated and silent in all adult somatic tissue but can become aberrantly activated in a number of tumours. The most abundant retrotransposons in the human genome are LINE elements or long interspersed nuclear elements. Loss of promoter methylation and transcriptional activation of LINE elements have been reported in a variety of cancers (Alves et al., 1996; Thayer et al., 1993). The deleterious effects of hypomethylation of LINE retrotransposons in cancer has been linked with both mutational disruption of genes as well as the potential disruption of nearby gene expression. The link between chromosome instability and genome hypomethylation in cancer has come from studies in mouse ES cells that have a homozygous deletion of the methyltransferase Dnmt1 (Chen et al., 1998). The mutant ES cells have a significantly increased mutation rate primarily involving genomic deletion. Moreover chromosome instability is a feature of ICF patients, a genetic disorder in humans caused by mutations in the DNA methyltransferase DNMT3b (Xu et al., 1999).

In the milieu of general hypomethylation, the cancer cell also gains regional specific areas that are hypermethylated. Hypermethylation is thought to contribute to cancer progression because of its potential for induction of mutations particularly in the coding regions of genes as well as its potential for gene inactivation (Figure 2). The regions that are often hypermethylated are the CpG island regions that span the promoters of house-keeping genes and tumour suppressor genes. It is now well established that multiple genes are concurrently methylated in the one cancer cell (Melki et al., 1999). Genes involved in cell cycle regulation, DNA repair, drug resistance, detoxification, differentiation, apoptosis, angiogenesis and metastasis have all been identified as being susceptible to hypermethylation in different cancers (Costello and Plass, 2001). Over half of these genes cause familial forms of human cancer when mutated in the germline and thus the selective advantage for loss of function is very clear (Baylin and Herman, 2000). However many of the hypermethylated genes in cancer are not defined tumour suppressor genes and for some of these genes the promoter methylation may be the only type of gene inactivation found in cancers, since mutations have not been observed (Baylin et al., 2001). Since methylation of the associated CpG islands corresponds with inactivation of these genes in the tumours, hypermethylation has been included as an alternative mechanism to elicit allelic gene silencing of tumour suppressor genes in cancer, in addition to silencing caused by genetic mutation and/or deletion (Jones and Laird, 1999). It is clear that hypermethylation is a significant alteration in the cancer genome, however the mechanism responsible for eliciting this change is not well understood.

Not only does the methylation profile of a cancer cell change but expression of the DNA methyltransferase enzymes is often modulated. DNMT1 expression is elevated 2.5-3.7-fold in colon tumours (Lee et al., 1996); Schmutte et al., 1996) and 4.2-fold in leukaemia (Melki et al., 1998) and DNMT1 activity appears to increase progressively with advancing stages of both human colon and lung cancer (Belinsky et al., 1996; Issa et al., 1993). DNMT3A and DNMT3B showed 4.4- and 11.7-fold elevated transcription in acute myeloid leukaemia (AML) (Mizuno et al., 2001) and significant over expression of DNMT3B was reported in solid tumours (Robertson et al., 1999). The increased expression of the DNMTs may be functionally important in cancer, as even a 2.2-fold average increase in expression of DNMT1 has been shown to be associated with the transformation of NIH3T3 fibroblasts (Wu et al., 1993). Furthermore, reduced DNMT1 activity in mice that are genetically predisposed to colonic adenomas markedly decreases the frequency of colon tumours (Laird et al., 1995). The relationship of the various DNA methyltransferases and their exact role in directing the methylation pattern of a normal cell is yet to be addressed, therefore it is still unclear what role a change in expression may have on modulating the methylation profile of the cancer cell.

Why are some genes methylated more than others?

Recent cancer 'methylomic' studies predict that hundreds of CpG islands could be methylated in a tumour cell (Costello et al., 2000). However, it is clear from both the genome-wide methylation studies and candidate gene approaches that each tumour type may have its own set of cancer cell-type specific genes that are more susceptible to methylation. Thus each cancer-type may have the potential to be typed or classified according to its methylation profile (Gitan et al., 2002). Cancer-type methylation differences are highlighted using a candidate gene approach in Figure 3, which shows the methylation profile of eight genes between two distinct cancer types, namely Acute Myeloid Leukaemia (AML) and prostate cancer. In prostate cancer GSTP1 is methylated in over 95% of tumours (Lee et al., 1994; Millar et al., 1999), but remains unmethylated in AML (Melki et al., 1999). In contrast p15, E-cadherin and calcitonin are commonly methylated in AML but are rarely methylated in prostate cancer. Similarly, Rb is unmethylated in both prostate cancer and AML but is commonly methylated in Retinoblastoma tumours (Stirzaker et al., 1997). The different methylation specificities according to cancer type poses the question: 'Why are some genes more methylated than others in different cancers?' To address this question we need to address the converse question: 'What is the mechanism that protects CpG islands from methylation in the normal cell and therefore what goes wrong in the cancer cell?'

CpG islands commonly span the promoter region of house-keeping genes and it is important that these genes remain transcriptionally active throughout development and differentiation. Since methylation of CpG islands is associated with gene silencing it makes sense that the CpG islands remain methylation free. However, CpG islands are rich in CpG sites, the target for DNA methyltransferase, and yet during early development these CpG sites remain unmethylated, even when the bulk of the genome is undergoing de novo methylation, including the CpG islands spanning genes on the inactive X or imprinted alleles. It is thought that the CpG islands spanning the housekeeping type genes are protected from methylation through either (1) the process of active transcription, including the possible steric hindrance of transcription factor or DNA polymerase binding, (2) active demethylation via RNA transcripts from CpG islands possibly in combination with 5MeC-glycosylase (Jost et al., 1997), (3) replicating timing or (4) local chromatin structure that may inhibit access to the DNA methyltransferase (Kass et al., 1997). The importance of gene activity, however as the mode of protection is exemplified by the fact that the CpG islands on the inactive X chromosome are silenced prior to methylation (Pfeifer et al., 1990). Therefore, CpG islands are not intrinsically 'unmethylatable' in the early embryo but active transcription appears to be a major protective factor.

So why are some CpG island genes more susceptible to hypermethylation than others in different cancer types? If in a cancer cell there is a general deregulation of the entire methylation and/or protective machinery we might expect a random profile of genes methylated in any one cancer and not the apparent cancer specific subsets. It is possible that cancer specific subsets arise through the activation of DNA methyltransferase enzymes or co-factors with sequence specificity, however such specificity in the DNMTs has not yet been identified. A more attractive proposal is that down regulation or silencing of particular genes in different somatic cell types, renders them more susceptible to de novo methylation, especially in a cancer cell which often has elevated DNA methyltransferase activity.

What triggers hypermethylation?

It is not easy to address this question in tumour tissue because DNA hypermethylation is an early event in tumourgenesis and therefore, once the tumour is large enough to detect, the aberrant methylation process has already occurred and the hypermethylated genes are already silenced. What is the evidence to support the proposal that gene inactivation may precede and in fact trigger hypermethylation in cancer?

1. Hypermethylation of the CpG island promoter is associated with gene silencing in tumours. When the methylation pattern of individual cells from the tumour are analysed, a variable pattern of methylation is often observed (Melki et al., 1999, 2000; Millar et al., 2000; Stirzaker et al., 1997). Some molecules show very little methylation whereas other molecules can be extensively methylated. However regardless of the extent of the methylation, the gene appears to be silenced in the tumour. For example, variable levels of methylation are seen in the Rb CpG island in retinoblastoma tumours (Stirzaker et al., 1997), in the p15 gene in leukaemia (Aggerholm et al., 1999; Cameron et al., 1999; Melki et al., 1999) and GSTP1 in prostate cancer (Millar et al., 1999, 2000) and yet all of these genes are commonly silenced in these specific tumours. One explanation would be that these genes are often inactivated in these cell types, possibly due to variable levels of regulatory factors, and that inactivation renders them susceptible to de novo methylation. However de novo methylation in cancer does not appear to be processive, it appears to occur randomly and accumulates with passage number, thus leading to a variable pattern of methylation both in individual cells and in the tumour.
2. Silenced genes are often associated with mutation, gene deletion and hypermethylation of the promoter region of the gene. However, genes can also show variable levels of expression in a cell without any apparent genetic or epigenetic mutation, possibly due to variable levels of regulatory factors. Mosaic or variable expression is often observed when normal tissues or tumour tissues or cell lines are stained for gene expression (Morgan et al., 1999). It is possible that in a cancer cell the subset of genes that are susceptible to methylation are those genes that show variability in expression. For example, the E-cadherin CpG island-associated promoter is commonly silent in leukaemic blast cells, however only some cells show methylation across the E-cadherin CpG island (Melki et al., 2000).
3. The p16^INK4 gene encodes a tumour suppressor gene that is often inactivated and methylated in many human tumour cell lines and primary tumours. The p16^INK4 gene is also inactivated in human mammay epithelial cells (HMECs) that escape selection and is necessary for in vitro life span extension (Noble et al., 1996). We and others have shown that p16^INK4 silencing in HMECs that have escaped selection is associated with hypermethylation of the CpG island spanning the promoter of the p16^INK4 gene (Huschtscha et al., 1998). Interestingly, the cells in colonies that initially escape selection, and stain negative for p16^INK4 expression, show very little methylation across p16^INK4 promoter (Melki et al., manuscript in preparation). Extensive hypermethylation occurs only after multiple passages of the HMECs. In this example, p16^INK4 expression appears to be silenced prior to methylation. However, even though the gene appears to be silenced prior to hypermethylation the gene can be reactivated by treatment with 5 deoxy-aza-cytidine (5-AzaC). This suggests that hypermethylation of CpG islands may not initiate gene silencing in cancer, but instead is critical for maintenance of the inactive stage of genes, similar to methylation of genes on the inactive X chromosome (Csankovszki et al., 2001; Mermoud et al., 2002).
4. The CpG island promoter of the gene glutathione-S-transferase (GSTP1) is silenced and hypermethylated in over 95% of all prostate cancers (Lee et al., 1994). In contrast, to other tumour suppressor genes which are often silenced by mutation or deletion, both alleles of GSTP1 are inactivated and hypermethylated (Millar et al., 1999). The GSTP1 gene is a good example of a cancer-type specific gene because it is not often methylated in other tumours. To address why the GSTP1 CpG island is uniquely susceptible to hypermethylation in prostate cancer but remains unmethylated in the majority of other cancers we have investigated a number of factors that are purported to be involved in triggering hypermethylation (Song et al., 2002). We found that the GSTP1 gene was not specifically targeted for methylation in prostate cancer cells and unlike the APRT gene in CHO cells (Brandeis et al., 1994; Macleod et al., 1994) removal of the Sp1 sites, or indeed other critical transcription factor binding sites, did not trigger hypermethylation of the GSTP1 island in prostate cancer cells. Intriguingly, silencing of the GSTP1 promoter was also insufficient to trigger hypermethylation in the cancer cells. We found that a low level of de novo methylation at random CpG sites, we termed 'seeding methylation', was required in combination with gene silencing, to initiate hypermethylation of the island. 'Seeding methylation' of the actively transcribed gene promoted demethylation, whereas 'seeding methylation' of the inactive gene promoted extensive de novo methylation of the GSTP1 CpG island. Our results support the proposal that active transcription can play an important role in maintaining the island in an unmethylated state, possibly by promoting demethylation of any CpG sites that become spuriously methylated. Demethylation of these sites could be achieved either through an active process (Schwarz et al., 2000), or through a passive process, for example lack of maintenance during replication.

Hypermethylation model

'Seeding methylation' occurs randomly at a low level in a normal somatic cell. Analysis by bisulphite sequencing has revealed that not all CpG sites in a CpG island are unmethylated in a normal cell (Cameron et al., 1999; Melki et al., 1999). In any one molecule there is often a few random sites of methylation in an otherwise unmethylated region. We propose, as summarized in Figure 4, that in a normal cell there is dynamic interplay between a low but persistent level of de novo methylation of CpG sites within the CpG island versus a persistent process of demethylation promoted by active transcription or indeed active chromatin (Cervoni and Szyf, 2001). However, in the cancer cell, since active transcription can no longer promote demethylation, the dynamic interplay shifts to favour de novo methylation of the island and consolidation of gene silencing.

Conclusion

We propose that active transcription protects CpG island from methylation in the normal cell and that the sub-set of genes that are prone to hypermethylation in a specific cancer type are those genes that have previously been silenced, either transiently through lack of factors or via chromatin remodelling. The jury is not unanimous, but the evidence suggests that methylation is not responsible for causing the gene silencing in cancer, but rather an accessory to the crime with its primary role in maintenance of silencing. We suggest that gene silencing is the critical precursor in cancer as it changes the dynamic interplay between de novo methylation and demethylation of the CpG island and tilts the balance to favour hypermethylation and chromatin inactivation.

Acknowledgements

We would like to thank Clare Stirzaker, Jenny Song, Peter Warnecke and Peter Molloy for valuable comments. This work was supported in part by grants from NH&MRC (9937987, 000454 and 202907).

Aggerholm A, Guldberg P, Hokland M, Hokland P. (1999). Cancer Res., 59: 436-441. MEDLINE

Alves G, Tatro A, Fanning T. (1996). Gene, 176: 39-44. Article MEDLINE

Antequera F, Bird A. (1993). Proc. Natl. Acad. Sci. USA, 90: 11995-11999. MEDLINE

Baylin SB, Esteller M, Rountree MR, Bachman KE, Schuebel K, Herman JG. (2001). Hum. Mol. Genet., 10: 687-692. MEDLINE

Baylin SB, Herman JG. (2000). Trends Genet., 16: 168-174. MEDLINE

Baylin SB, Herman J, Graff JR, Vertino PM, Issa JP. (1998). Advances in Cancer Research In Vandewoude GF and Klein G (eds) Academic Press Inc,: San Diego, CA, pp. 141-196.

Belinsky SA, Nikula KJ, Baylin SB, Issa JP. (1996). Proc. Natl. Acad. Sci. USA, 93: 4054-4060.

Bestor T, Laudano A, Mattaliano R, Ingram V. (1988). J. Mol. Biol., 203: 971-983. MEDLINE

Bestor TH, Tycko B. (1996). Nature Genet., 12: 363-367. MEDLINE

Bird A. (1992). Cell, 70: 5-8. MEDLINE

Bird AP. (1986). Nature, 321: 209-213. MEDLINE

Bird AP, Wolffe AP. (1999). Cell, 99: 451-454. MEDLINE

Brandeis M, Frank D, Keshet I, Siegfried Z, Mendelsohn M, Nemes A, Temper V, Razin A, Cedar H. (1994). Nature, 371: 435-438. MEDLINE

Cameron EE, Baylin SB, Herman JG. (1999). Blood, 94: 2445-2451. MEDLINE

Cervoni N, Szyf M. (2001). J. Biol. Chem., 276: 40778-40787. MEDLINE

Chen RZ, Pettersson U, Beard C, Jackson-Grusby L, Jaenisch R. (1998). Nature, 395: 89-93. Article 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, Huang HJS, Petrelli NJ, Zhang XL, O'Dorisio MS, Held WA, Cavenee WK, Plass C. (2000). Nature Genet., 24: 132-138. Article MEDLINE

Costello JF, Plass C. (2001). J. Med. Genet., 38: 285-303. MEDLINE

Csankovszki G, Nagy A, Jaenisch R. (2001). J. Cell. Biol., 153: 773-784. MEDLINE

De Smet C, Lurquin C, Lethe B, Martelange V, Boon T. (1999). Mol. Cell. Biol., 19: 7327-7335. MEDLINE

Ehrlich M. (2000). J. Clin. Ligand Assay, 23: 144-146.

Feinberg AP, Vogelstein B. (1983). Nature, 301: 89-92. MEDLINE

Gardiner-Garden M, Frommer M. (1987). J. Mol. Biol., 196: 261-282. MEDLINE

Gitan RS, Shi H, Chen CM, Yan PS, Huang TH. (2002). Genome Res., 12: 158-164. MEDLINE

Holliday R, Pugh JE. (1975). Science, 187: 226-232. MEDLINE

Howlett SK, Reik W. (1991). Development, 113: 119-127. MEDLINE

Huschtscha LI, Noble JR, Neumann AA, Moy EL, Barry P, Melki JR, Clark SJ, Reddel RR. (1998). Cancer Res., 58: 3508-3512. MEDLINE

Iguchi-Ariga SMM, Schaffner W. (1989). Genes Dev., 3: 612-619. MEDLINE

Issa J, Vertino PM, Wu JJ, Sazawal S, Celano P, Nelkin BD, Hamilton SR, Baylin SB. (1993). J. Nat. Cancer Inst., 85: 1235-1240.

Jones PA. (1999). Trends Genet., 15: 34-37. Article MEDLINE

Jones PA, Laird PW. (1999). Nature Genet., 21: 163-167. Article MEDLINE

Jost J-P, Frémont M, Siegmann M, Hofsteenge J. (1997). Nucleic Acids Res., 25: 4545-4550. MEDLINE

Kass SU, Pruss D, Wolffe AP. (1997). Trends Genet., 13: 444-449. Article MEDLINE

Laird PW, Jackson-Grusby L, Fazeli A, Dickinson SL, Jung WE, Li E, Weinberg RA, Jaenisch R. (1995). Cell, 81: 197-205. MEDLINE

Lee PJ, Washer LL, Law DJ, Boland CR, Horon IL, Feinberg AP. (1996). Proc. Natl. Acad. Sci. USA, 93: 10366-10370. Article MEDLINE

Lee W-H, Morton RA, Epstein JI, Brooks JD, Campbell PA, Bova GS, Hsieh W-S, Isaacs WB, Nelson WG. (1994). Proc. Natl. Acad. Sci. USA, 91: 11733-11737. MEDLINE

Li E, Bestor TH, Jaenisch R. (1992). Cell, 69: 915-926. MEDLINE

Macleod D, Charlton J, Mullins J, Bird AP. (1994). Gene Dev., 8: 2282-2292.

Mayer W, Niveleau A, Walter J, Fundele R, Haaf T. (2000). Nature, 403: 501-502. Article MEDLINE

Melki JM, Vincent PC, Clark SJ. (1999). Cancer Res., 59: 3730-3740. MEDLINE

Melki JR, Vincent PC, Brown RD, Clark SJ. (2000). Blood, 95: 3208-3213. MEDLINE

Melki JR, Warnecke P, Vincent PC, Clark SJ. (1998). Leukemia, 12: 311-316. MEDLINE

Mermoud JE, Popova B, Peters AH, Jenuwein T, Brockdorff N. (2002). Curr. Biol., 12: 247-251. MEDLINE

Millar DS, Ow KK, Paul CL, Russell PJ, Molloy PL, Clark SJ. (1999). Oncogene, 18: 1313-1324. MEDLINE

Millar DS, Paul CL, Molloy PL, Clark SJ. (2000). J. Biol. Chem., 275: 24893-24899. MEDLINE

Mizuno S, Chijiwa T, Okamura T, Akashi K, Fukumaki Y, Niho Y, Sasaki H. (2001). Blood, 97: 1172-1179. Article MEDLINE

Molloy PL, Watt F. (1990). Philos. Trans. R. Soc. Lond. B. Biol. Sci., 326: 267-275. MEDLINE

Monk M. (1990). Philos. Trans. R. Soc. Lond. [Biol.] 326: 299-312.

Monk M, Boubelik M, Lehnert S. (1987). Development, 99: 371-382. MEDLINE

Morgan HD, Sutherland HG, Martin DI, Whitelaw E. (1999). Nature Genet., 23: 314-318. Article MEDLINE

Nan XS, Ng HH, Johnson CA, Laherty CD, Turner BM, Eisenman RN, Bird A. (1998). Nature, 393: 386-389. Article MEDLINE

Noble JR, Rogan EM, Neumann AA, Maclean K, Bryan TM, Reddel RR. (1996). Oncogene, 13: 1259-1268. MEDLINE

Okano M, Bell DW, Haber DA, Li E. (1999). Cell, 99: 247-257. MEDLINE

Okano M, Xie S, Li E. (1998). Nature Genet., 19: 219-220. Article MEDLINE

Pfeifer GP, Tanguay RL, Steigerwald SD, Riggs AD. (1990). Genes Dev., 4: 1277-1287. MEDLINE

Pogribny IP, Miller BJ, James SJ. (1997). Cancer Lett., 115: 31-38. MEDLINE

Reik W, Dean W, Walter J. (2001). Science, 293: 1089-1093. MEDLINE

Riggs AD. (1975). Cytogenet. Cell. Genet., 14: 9-25. MEDLINE

Robertson KD, Uzvolgyi E, Liang G, Talmadge C, Sumegi J, Gonzales FA, Jones PA. (1999). Nucleic Acids Res., 27: 2291-2298. Article MEDLINE

Schmutte C, Yang AS, Nguyen TT, Beart RW, Jones PA. (1996). Cancer Res., 56: 2375-2381. MEDLINE

Schwarz S, Bourgeois C, Soussaline F, Homsy C, Podesta A, Jost JP. (2000). Eur. J. Cell Biol., 79: 488-494. MEDLINE

Song J, Stirzaker C, Harrison J, Melki J, Clark S. (2002). Oncogene, 21: 1048-1061. Article MEDLINE

Stern LL, Mason JB, Selhub J, Choi SW. (2000). Cancer Epidemiol. Biomarkers Prev., 9: 849-853. MEDLINE

Stirzaker C, Millar DS, Paul CL, Warnecke PM, Harrison J, Vincent PC, Frommer M, Clark SJ. (1997). Cancer Res., 57: 2229-2237. MEDLINE

Thayer RE, Singer MF, Fanning TG. (1993). Gene, 133: 273-277. MEDLINE

Vachtenheim J, Horakova I, Novotna H. (1994). Cancer Res., 54: 1145-1148. MEDLINE

Warnecke P. (1998). PhD thesis, DNA Methylation in Early Development, University of Sydney, Sydney.

Warnecke P, Mann J, Frommer M, Clark S. (1998). Genomics, 51: 182-190. Article MEDLINE

Warnecke PM, Clark SJ. (1999). Mol. Cell. Biol., 19: 164-172. MEDLINE

Wu JJ, Issa JP, Herman J, Bassett DE, Nelkin BD, Baylin SB. (1993). Proc. Natl. Acad. Sci. USA, 90: 8891-8895. MEDLINE

Xu GL, Bestor TH, Bourchis D, Hsieh CL, Tommerup N, Bugge M, Hulten M, Qu XY, Russo JJ, ViegasPequignot E. (1999). Nature, 402: 187-191. Article MEDLINE

Yoder JA, Walsh CP, Bestor TH. (1997). Trends Genet., 13: 335-340. Article MEDLINE

Figure 1 Global and gene specific DNA methylation changes in the early embryo. During early development the methylation profile of the embryo changes with an initial wave of demethylation from fertilization up until the eight cell stage, followed by a wave of de novo methylation from the morulae to blastocyst stage (implantation). The global DNA methylation profile shown in the top panel is based on restriction enzyme analysis (Monk et al., 1987) and therefore principally reflects the repetitive fraction of the genome. The lower panel depicts the gene specific methylation profile, with each class of genes or DNA regions being represented as an oval on a contiguous DNA strand. The level of methylation shown is based on genomic sequencing individual candidate genes from these various regions (Warnecke, 1998; Warnecke et al., 1998; Warnecke and Clark, 1999). The intensity of methylation at each stage during development for both the global methylation profile and the gene specific methylation profile is indicated by the degree of shading

Figure 2 Global DNA methylation changes that occur in a cancer cell. In a normal cell, the methylation at individual CpG sites (^mCpG) is established in a regional specific manner, with some regions that remain unmethylated and others that are hypermethylated. This defined methylation pattern is drastically altered in a cancer cell, with some DNA regions undergoing methylation resulting in hypomethylated areas and some DNA regions undergoing de novo methylation resulting in hypermethylated areas. Hypomethylation in a cancer cell is linked with DNA instability and gene activation, whereas hypermethylation in a cancer cell is linked with gene mutation and gene silencing. ^mCpG depicted as shaded circles; hypomethylated areas depicted as a line; hypermethylated areas depicted as ovals

Figure 3 Cancer-specific methylation pattern. The proportion of AML and prostate cancer patients methylated in eight different tumour suppressor or tumour related genes is plotted. Methylation profile for AML was measured by direct bisulphite sequencing DNA isolated from bone marrow from 12 normal patients (N) or from 20 patients with cancer (C) (Melki et al., 1999). Methylation profile for prostate cancer was measured by bisulphite sequencing DNA isolated from the prostate from 18 patients with cancer (C) or from matched normal (N) prostate cells (Clark et al. submitted)

Figure 4 Hypermethylation model (seeding and silencing). It has always been a conundrum as to what comes first the chicken or the egg, that is 'Does methylation cause gene silencing or does gene silencing cause methylation?' We propose that in cancer, gene silencing comes first and triggers hypermethylation of the CpG island associated gene. In a normal cell, CpG island-associated genes are actively transcribed and remain essentially unmethylated, whereas in a cancer cell the genes are silent and associated CpG island is hypermethylated. Our model suggests that in a normal cell, CpG islands are subjected to a steady, but light 'snowfall' or 'seeding' of de novo methylation, but remain methylation free due to demethylation promoted by active transcription of the gene. Whereas in a cancer cell, gene silencing is a critical precursor and that in combination with 'seeding' methylation shifts the balance in the weather conditions and triggers a 'snow storm' of methylation