The History of Cancer Epigenetics

7/27/2019 The History of Cancer Epigenetics

1/11

PERSPECTIVES

abnormal during fetal development, therebypredisposing to paediatric cancers,and theycan change during normal ageing and con-tribute to common cancer risk in adults.Theycan also support clonal evolution in humancancers,contributing to tumour progression.

But how was this key role for epigeneticsin cancer development discovered,how has itcome to rival genetics and what else do weneed to know?

Hypomethylation and gene ac tivation

Loss of DNA methylation at CpG dinu-cleotides was the first epigenetic abnormalityto be identi fied in cancer cells. At asymposium at Johns Hopkins in 1982 ontumour-cell heterogeneity,Andy Feinberg andBert Vogelstein wondered what mechanismaccounted for high-frequency mutations,adaptation to tumour microenvironment andplasticity in some cancers.The conference wasorganized by Donald Coffey,who had intro-duced the two investi gators. At the ti me,many groups were excited by observationsthat DNA methylation might be linked to tis-sue-specifi c gene silencing,so Feinberg and

Vogelstein searched for di fferences betweencancers and normal t issues. They usedSouthern blott ing to analyse DNA that hadbeen digested wi th methylation-sensit iverestriction enzymes and found that a sub-stantial proport ion of CpGs that weremethylated in normal tissues were unmethy-lated in cancer cells1.Ehrlich and colleaguesthen carried out similar investigations usinghigh-performance liquid chromatography toshow that the 5-methylcytosine content wasglobally reduced2 (seeTIMELINE).The loss ofmethylation involved every tumour type

studi ed, both benign and mali gnant; fur -thermore, pre-mali gnant adenomas alsouniversally had altered DNA methylation3,4.

Hypomethylation of DNA has mechanis-tic implications.First, it can lead to gene acti-vation. It has been found recently that manyCpG islands are normally methylated insomatic tissues5.These methylated islands can

become hypomethylated in cancer and nearbygenes become activated. Examples of genesthat are affected by hypomethylation includeoncogenes such asHRAS6 and the CT genes those that are expressed normally in thetesti s and aberrantl y in tumours. Theirhypomethylation leads, for example, toMAGEexpression in melanoma a promis-ing target of immunotherapy7. The relatedcancer/testi s antigen CAGEwas also shownto be activated by hypomethylati on, whichwas confi rmed using 5-aza-2-deoxycytidine(5-azaCdR), an inhibitor of DNA methyla-tion,as well as promoter reporter transfectionexperiments;hypomethylation of CAGE wasfound to precede the development of stomachand liver cancer at high frequency8.Althoughhypomethylation was the originally identi fiedepigenetic change in cancer, it was overlookedin preference of hypermethylation for manyyears and is only now undergoing a renais-sance.This is, in part,because of previous biasin experimental design; if one looks for alteredmethylation only at sites that are normallyunmethylated, then one wil l only observehypermethylation. The frequency ofhypomethylated sites might be quite high,as

indicated by high-throughput genomic-methylation analysis of tumours9,10,includingcancers of thestomach,kidney,colon,pan-creas, liver,uterus, lungandcervix1018.Strongsupport for hypomethylation leading to acti-vati on of genes that are impor tant in cancerincludes promoter CpG demethylation in theoverexpression ofcycli n D211andmaspiningastric carcinoma12,MN/CA9overexpressionin human renal-cell carcinoma13, S100A4metastasis-associated gene in colon cancer14

and human papillomavirus 16 (HPV16)expression in cervi cal cancer15,16. Extensive

Since its discovery in 1983, the epigenetics

of human cancer has been in the shadows

of human cancer genetics. But this area

has become increasingly visible with a

growing understanding of specific

epigenetic mechanisms and their role in

cancer, including hypomethylation,

hypermethylation, loss of imprinting and

chromatin modification. This timeline traces

the field from its conception to the present

day. It also addresses the genetic basis of

epigenetic changes an emerging area

that promises to unite cancer genetics and

epigenetics, and might serve as a model for

understanding the epigenetic basis of

human disease more generally.

Epigenetic inheritance is defined as cellularinformation,other than the DNA sequenceitself, that is heri table during cell division.There are three main, inter-related types ofepigeneti c inheri tance: DNA methylation,genomic imprinting and histone modifi ca-tion (BOX 1).Epigenetic inheri tance accountsfor unusual phenomena such as position-effect vari egati on in f li es, telomere and

mating-type silencing in yeast, and transgene-induced gene silencing in plants and animals.However, it has become increasingly apparentthat epigeneti c inheritance is impor tant inmany physiological and pathophysiologicalconditions. It is key to our understanding ofthe di fferences between growing, senescentand immortal cells, tumour and normal cells,various differentiated cell s,and ageing cells.Epigenetic templates that control gene expres-sion are transmitted to daughter cells inde-pendently of the DNA sequence. Thesemetastable patterns can sometimes become

NATURE REVI EWS | CANCER VOLUME 4 | FEBRUARY 2004 | 1

The history of cancer epigenetics

Andrew P. Feinberg and Benjamin Tycko

T IMEL INE


2/11

2 | FEBRUARY 2004 | VOLUME 4 www.nature.com/reviews/cancer

P E R S P E C T I V E S

ICF syndrome (immunodeficiency,chro-mosomal instabil ity and facial anomalies) was found by several investigators to be causedby loss-of-function mutations in the cytosineDNA methyltransferaseDNMT3B2527.ICFsyndrome has as its cardinal features loss ofmethylation in classical satellite DNA andmitogen-inducible formation of bizarre multi-radial chromosomes that contain arms fromchromosomes 1 and 16 (REF.26).Indeed,this iswhy DNMT3B was successfully scrutinized as acandidate gene for the disorder by Viegas-Pequignot and Bestor,Gartler,Li and others.Nevertheless, ICF syndrome does not lead tocancer (discussed below).An even more directlink between hypomethylation and chromoso-mal instabil ity was made by Jaenischs group,who found that neurofibromatosis 1(Nf1)+/Trp53+/ mice showed a 2.2-foldincrease in frequency of LOH when a hypo-morphic Dnmt1allele was introduced28.

Another potential connection betweenhypomethylation and chromosomal instabilityis the hypomethylation of L1 retrotransposonsin colorectal cancer, which might promotechromosomal rearrangement29.

Fourth,hypomethylation is a mechanismof drug, toxin and viral effects in cancer. Inaddition to gene ampli fication,hypomethyla-tion of the multidrug-resistance geneMDR1correlates with increased expression and drugresistance inacute myelogenous leukaemia30.Toxic carcinogens might also act throughmethylati on alterations. For example, cad-mium inhibits DNA methyltransferase activ-ity and leads to acute hypomethylation,which i s followed by hypermethylation ofDNA after chronic exposure to this epige-netic carcinogen31.Similarly,arsenic inducesRashypomethylation in mice32. Finally,

Thi rd, Ehrli ch and colleagues recentlylinked tumour hypomethylation in cancer tochromosomal instabil ity.Hypomethylation

is particularly severe in pericentromericsatellite sequences, and several cancers(Wilms tumour,ovarian and breast carcino-mas) frequently contain unbalanced chro-mosomal translocations with breakpoints inthe pericentromeric DNA of chromosomes 1and 16 (REF. 23). These rearrangements arespecifi c, and not due to global genomicinstabili ty in Wil ms tumours, t(1;16)translocations are sometimes the onlydetectable abnormality.These unbalancedtranslocations produce loss of heterozygosity(LOH) for markers on chromosome 16,which, in turn, strongly correlates withtumour anaplasia24.Demethylation of satel-lite sequences might predispose to theirbreakage and recombination. The presumedcausal relationship in these cancers has notbeen proven, but a developmental di sorder

recent studies of pancreati c cancer byGoggins and colleagues showed widespreadhypomethylati on associated with prol ifera-

tion-linked genes, including14-3-310,17.Second,a cellular methylator phenotype

has been linked to mismatch repair, fi rst byLengauer and colleagues, who showed thatcancer cells that are deficient in DNA mis-match repair silenced retroviral constructpromoters by DNA methylation19. Thi sobservation was challenged by Jones andcolleagues20and the methylator phenotypeconcept (about which more below) willprobably be debated until its genetic basis iselucidated. Nevertheless, the idea makessense,as hypermethylation of the mismatch-repair geneMLH1is commonly found inmismatch-repair-defective tumours,as firstdescribed by Kolodners group21. In addition,abnormal imprinting (discussed below) isalso more commonly found in mismatch-repair-defective colorectal cancers22.

Box 1 | The three main types of epigenetic information

Cytosine DNA methylation is a covalent modification of DNA, in which a methyl group is

transferred from S-adenosylmethionine to the C-5 position of cytosine by a family of cytosine

(DNA-5)-methyltransferases. DNA methylation occurs almost exclusively at CpG nucleotides

and has an important contributing role in the regulation of gene expression and the silencing of

repeat elements in the genome.

Genomic imprinting is parent-of-origin-specific allele silencing, or relative silencing of one

parental allele compared with the other parental allele. It is maintained, in part, bydifferentially methylated regions within or near imprinted genes, and it is normally

reprogrammed in the germline.

Histone modifications including acetylation,methylation and phosphorylation are

important in transcriptional regulation and many are stably maintained during cell division,

although the mechanism for this epigenetic inheritance is not yet well understood. Proteins

that mediate these modifications are often associated within the same complexes as those that

regulate DNA methylation.

Al tered g ene

methylat ion

in cancer

Parent-of-or igin-

spec i fi c loss of

heterozygosity

in tumours

Hypermethylat ion

of calci tonin

Hypermethylat ion

of a tumo ur -

suppressor gene

Promoter methylat ion

enforces gene s i lencing on

the inact ive X chromos ome

M ouse m ode l fo r

canc er epigenet ics

Frequent tumou r-

suppressor gene

methylat ion in cancer

Ret inoblastoma

tumour -suppressor

gene se ts epigenet i c

states through the

S W I / S N F com p lex

Universal i ty of

epigenet i c changes

i n hum an cance r

CpG -cluster methylat ion

marks inact ive

promoters on the

inact ive X chromosom e

Human tr i thorax

homo logue i s a

p r o t o - oncogene

Histone deacetylat ion in

X inac t ivat ion

Gatekeeper ro le for

epigenet ic

al terat ions in cancer

Chromat in- remodel l i ng

com plex mutated in

hum an cance r s

D N A

hypomethylat ion

l inked to

ch r om osom a l

instability

Funct ional

associat ion betwe en

epigenet i cs and

tumour suppression

1983 1985 1986 1987 1989 1990 1992 1993 1994 1995 1997 1998 1999

T imeline | Milestones in human cancer epigenetics Insul in- l ike growth fac tor-2 lossof imprint ing (LOI) in canc er

Several epigenet ic

defects l i nked to

Beckwi thWiedemann

synd r om e


3/11


and Horsthemke laborator ies,were the fi rstto indicate that tumour-suppressor silencingmight occur by an epigenetic pathway43,44. Inthe Horsthemke study in 1989,hypermethy-lation was specifi cally linked toRB,which ledhis group to suggest it might have a directrole in tumour-suppressor gene silencing43.In 1991, Dryjas group showed that thehypermethylation was confined to one allele,again indicating specificity.He argued explic-it ly that it l eads to gene silencing44. Directconfirmation of epigenetic silencing of atumour-suppressor gene was provided bySakais group in 1993,who showed a 92%reduction ofRBexpression in tumours withpromoter hypermethylation45 and byHorsthemkes group in 1994(REF.46).

Two years later, beginning in 1995,severalgroups, including the Baylin, Jones andSidransky laborator ies,confirmed promoterhypermethylation at numerous other loci in

cancer cells,supporting the principle of epige-netic gene inactivation in cancer.Key tumour-suppressor proteins including the INK4A(also known as p16;encoded byCDKN2A)cyclin-dependent kinase inhibitor, the mis-match-repair enzyme MLH1, the vonHippelLindau (VHL) tumour suppressorand E-cadherin were all shown to be elimi-nated both in cell lines and in primary cancersby an epigenetic pathway that correlates withdense CpG methylation of their gene promot-ers.The primary publi cati ons that describethese correlations for theCDKN2A andVHLgenes appeared between 1994 and 1995 (REFS4750) and were extended to includeMLH1between 1997 and 1998 (REFS 21,51,52). I t isimportant to note that the basic relationshipbetween CpG-island methylati on and geneinactivation,and the identi fication of CpGislands themselves,came from early studies ofX chromosome inactivation (BOX 2).

After leaving the Baylin laboratory,J.-P.Issa(and colleagues) then showed methylationprofili ng data that indicated a dichotomousclassif ication of human carcinomas into fre-quent promoter methylation and infrequentmethylation groups,and led him to promote

the idea of a CpG-island methylator pheno-type termed CIMP in human cancer53.This attractive,but stil l controversial,concepthas stimulated many follow-up studies and,aswe discuss below,there are now several candi-date biochemical mechanisms that couldconceivably account for CIMP.The fi rst func-tional report showing a relationship betweentumour-suppressor activi ty and DNA methy-lation was performed by West and Barrett in1993,in which they examined a model of pro-gressive loss of tumour-suppressor activity inSyrian hamster cells54.

cervical cancer latency seems to be caused, inpart , by hypermethylation of the HPV16genome, and latent EpsteinBarr virus inhuman lymphoma cells uses a similar strat-egy to enforce sil encing of a subset of itsgenes16,33. In cervical cancer,activation of theHPV genome and progression occur withprogressive hypomethylation of the virus inprecursor lesions.

An exciting recent development in cancerhypomethylation involves a link to diet. Acommon polymorphism of methylenetetrahy-drofolate reductase (MTHFR), which isinvolved in biosynthesis of the methylati onprecursor S-adenosylmethionine,was associ-ated wi th i ncreased colorectal cancer preva-lence in a population-based study,and cancerincidence was lower in patients with highdietary methionine,which increases methyla-tion content34. Reduced MTHFR was alsolinked to alcohol consumption34,35,and colonic

hypomethylation was found in patients withcolorectal cancer36.These results are consistentwith studies in rodents showing that choline-or choline- and methionine-deficient diets leadto hepatocellular carcinoma, without anyadded carcinogen, fi rst shown by Poir ier37andconfirmed by many groups.

The mechanism behind global hypo-methylation in cancers remains unknown,butan indirect link was indicated by two differentdiscoveries that might turn out to be con-nected.First,ATP-dependent DNA helicases ofthe SNF2 family the catalytic componentsof SWI/SNF chromatin-remodell ing com-plexes are essential for maintaining normalDNA methylation.Individuals with the devel-opmental disorder ATRX (-thalassaemia,myelodysplasia) have mutations in theATRXgene,which encodes a SNF2-family heli case.

In ATRX cells, the ribosomal DNA repeats arehypomethylated38.Although ATRX is associ-ated with a pre-malignant myelodysplasia,but not cancer per se, a second discoveryshows that the connection of SWI/SNF com-plexes with human cancer is unambiguous.As uncovered by Delattre and coll eagues in1998, germli ne and somati c mutations inSNF5(also known asINI1) which encodesa SWI/SNF complex component causerare,but lethal, cancer malignant rhabdoidtumour39. It is not yet known if,as might bepredicted, this class of neoplasm has moreextensive global demethylation than mostcancers.The more general question, whetherSWI/SNF function is altered in more com-mon types of cancers, is also unanswered,butresults from research on the retinoblastoma(RB) tumour suppressor (see below) indicatethat this is indeed the case.

A second chromatin protein that has

been linked to hypomethylation and cancerwas recently identified by Muegge and col-leagues,who found that Lsh,a SNF2-familymember, is required for maintenance ofnormal methylation. Gene knockout leadsto a global defect in genomic methylati on,as well as a severe prol i ferative defect andchromosomal instability40. Further supportfor a link between hypomethylati on andtumorigenesis was provided by Hirohashisgroup, who identi fied a common spli cevari ant ofDNMT3Bin patients with l ivercancer, which is associated wi th hypo-methylation and causes hypomethylation ofpericentromeric satellite sequences whentransfected into cells41.

Hypermethylation and gene silencing

Of course,hypomethylation is not the onlyway in which methylation can contribute tocancer. Steve Bayli n and Barry Nelkin, inconversations at Johns Hopkins withFeinberg and Vogelstein,decided to examinecalcitonin, which is a marker of small -celllung cancer and was their main interest atthe time.In 1986, they were surprised to findsite-specifi c hypermethylation of calcitonin,

with relative silencing of calcitonin expres-sion42.However,calcitonin i s not a tumour-suppressor gene,and the first l ink betweenhypermethylation and tumour-suppressorgenes was made, fi tt ingly enough,on the firstknown tumour-suppressor gene theretinoblastoma geneRB. Thi s gene mightnot come to mind as a locus that is frequentlyinactivated by the epigenetic pathway,but, infact,theRBpromoter is methylated in a sig-nifi cant subset of sporadic and even heredi-tary retinoblastomas.The papers reporti ngthis phenomenon,publi shed by the Dryja

NATURE REVIEWS | CANCER VOLUME 4 | FEBRUARY 2004 | 3

Epigenotype

pheno t ype

analys is show s

gatekeeper role

of LOI in a hum an

cance r synd r om e

Altered histone

lysine

methylat ion at

si lenced

tumour -

suppressor loc i

Polycomb p rote ins hypothesized to accou nt for

widespread gen e s i lencing in cancer

Clinical trials of

com b ined 5 - aza-

2 -deoxycyt id ine

and tr ichostat in A

therapy in pat ients

wi th cancer

DNA h ypom ethylat ion

l inked to en vironmental

tox ins and d iet

2001 2002 2003

Renaissance of

hypom ethylat ion and

gene ac t ivat ion in c ancer


4/11



allele is more abundantly (or exclusively)expressed in the offspring was discoveredin embryological studies published in themid-1980s63,64.But even earlier, cancer cytoge-neticists studying two human neoplasms,hydati diform moles and ovarian teratomas,had produced data that presaged these find-ings. Moles which are placenta-derived(that is, extraembryonic) tumours werefound to contain two complete sets of pater-nal chromosomes with no maternal contribu-tion65, whereas fi ndings in both mice andhumans indicated that ovarian teratomas,which contain many tissue types,but neverplacental trophoblast, carried a bi-maternalchromosome complement66.

Evidence for a role for human imprintedgenes in development then came from twoneurodevelopmental disorders,PraderWill isyndrome (PWS) and Angelman syndrome(AS).These were found to be caused by uni -

parental chromosomal disomies of the longarm of chromosome 15,which were mater-nal in PWS and paternal in AS.More perti-nent to this timeline, in the late 1980s,severalindependent studies reported a strong ( infact, absolute) parent-of-origin bias in LOHfor chromosome 11p15 alleles in Wilmstumours and embryonal rhabdomyosarco-mas, with invari able loss of maternal andduplication of paternal alleles6770.This strik-ing finding was hard to explain without pos-tulating a role for imprinted gene(s) in thesetumours,as was suggested first by Sapienzaand colleagues69 in 1989.An important cluewas also provided by studies of the disorderBeckwithWiedemann syndrome (BWS),which causes prenatal overgrowth, birthdefects (including a large tongue,ear creasesand abdominal-wall closure defects) and pre-disposition to various embryonal tumours ofchildhood including Wilms tumour. Rarefamilial cases of BWSalso indicated a parent-of-origin effect,as the overgrowth phenotypewas only seen after maternal transmission71.More direct evidence came from mapping ofBWS to 11p15 (REFS72,73), followed by obser-vations from Mannenset al. that chromoso-

mal rearrangements in 11p15 in patientswith BWSwere all of maternal origin74.

The discovery of bona fidehumanimprinted genes was made independentlyin the early 1990s by Tycko,Ohlsson,FeinbergPolychronakos and Reeve (and colleagues)7579,following from mouse studies8082 of thesesame genes IGF2and H19. Additionalimprinted human genes were uncovered in thePWS/ASregion of chromosome 15(REFS83,84).Studies of the mechanism of imprinting werein ful l swing by the mid-1990s, with somelandmarks being the direct demonstration of a

and the same is as true for hypomethylati onas for hypermethylati on. Indeed,a signif i-cant challenge to the causal role of hyperme-thylation in i nit iating the process of genesilencing comes from a recent report show-ing that methylation of histone H3 lysine 9 that is,chromatin modification occurredin conjunction with re-silencing ofCDKN2Ain the absence of DNA methylation, in cells inwhich CDKN2A had previously been acti-vated by DNA methyltransferase knockout59.Consistent with this observation, Clarkand Melki also point out cogently thatCDKN2A is silenced in prol iferating coloniesof mammary epitheli al cell s that escapesenescence,even in the absence of promotermethylation, suggesting that hypermethyla-tion is not responsible for silencing,but helpsto maintain silencing60.

Clearly,one must look at methylation incancer as an example of epigenetic dysregula-tion,with both hypomethylation and hyper-methylation having significant roles.Nicelysummarizing this situation,a recent study of

Wilms tumours found different unique-geneloci that were affected by hypomethylation orhypermethylation in the same tumour61.Moreover, the fi nal epigeneti c programmevaries strongly by tumour type;an interestingexample is a counteri ntui tive report ofCDKN2A hypomethylation in some breastcancers compared with normal breast tissue62.

Loss of imprinting in cancer

Imprinting which refers to conditioning ofthe maternal and paternal genomes duringgametogenesis,such that a specifi c parental

That DNA methylation is causal in main-taining the silent epigenetic state has beenshown by the potency of demethylating drugsin reactivating gene expression and by recentstudies that use somatic-cell knockout proce-dures55,or antisense and RNA interference56,to eliminate DNA methyltransferases fromcancer cells.These manipulations resulted intumour-suppressor gene reactivation,but thedetails differed depending on the experimen-tal system: acute elimination ofDNMT1inHCT116 cells by antisense or RNA i nterfer-ence was sufficient to reactivateCDKN2A,whereas in this same cell line,a double somaticknockout of bothDNMT1andDNMT3Bwasrequired to demethylate and reactivate thisgene.A possible explanation is thatDNMT3Bcan ultimately replace the function ofDNMT1during cell selection, as occurs inknockout experiments,but DNMT3Bdoesnot completely substi tute forDNMT1acutely.

What has not been as clear is that themechanism of initial silencing of these genesis hypermethylati on, and this has been the

subject of some debate: see, for example,Bestor57. Indeed, Modr ich and colleaguesshowed that activation of MHL1by5-azaCdR is rapidly reversed sponta-neously52. So, methylation changes couldarise secondarily to other epigenetic changes,such as chromati n modi fi cati on, but thenhelp to maintain the silenced state.Consistent with this idea,although tumour-suppressor gene silencing per secan be adominant trait in somatic-cell genetic exper-iments58,atrans-acting defect in methylationhas not been demonstrated in tumour cell s

Box 2 | An X connection

In the background of the discoveries that link CpG-island methylation and gene inactivation is a

large body of important correlative and mechanistic data that are related to X chromosome

inactivation.For example,an early clue to a role in gene silencing came from studies of

Mohandas et al., who showed in 1981 that 5-aza-2-deoxycytidine (5-azaCdR), an inhibitor of

DNA methylation176 could reactivate the inactive X chromosome176. The discovery of the

functional significance of what are now termed CpG islands also came from studies of the X

chromosome. Wolf, Migeon and colleagues showed in 1984 that clusters of CpG dinucleotides

are specifically methylated on the inactive X chromosome and reactivated with 5-azaCdR177.

These were later extensively characterized and termed CpG islands by Adrian Bird and

colleagues,who found that they were common in the promoters of autosomal genes178. Studies

from the Gartler laboratory showed that gene reactivation on the inactive X chromosome is

associated with large regions of promoter demethylation after 5-azaCdR treatment, indicating a

causal relationship between methylation and gene silencing on the inactive X chromosome179.

Incidentally, a peculiarity of the DNA methylation literature is the term CpG: what else connects

the two sugars but a phosphate? After all,we dont say TpApTpA box!

The connection between epigenetic gene silencing and chromatin modifications, another

theme that is increasingly important in the progress of cancer epigenetics,was also highlighted

early on in studies of X inactivation with memorable images of 45 human chromosomes

intensely stained by an antibody specific for acetylated histone H4, with the lone inactive X

chromosome globally deacetylated and unstained180.


5/11


PEG3might lie outside the region of LOH99.These results are provocative,as they seem togo against the paternal allele activegrowthpromoter paradigm.In fact,data fromPeg3-knockout mice100 indicate the expectedgrowth-promoting function of this paternallyexpressed gene (the knockout mice are small);but human PEG3has tumour-suppressoractivity in transfected cell s101 and the gene isepigenetically silenced in gliomas102.Selectiveloss of alleles has also been reported in neu-roblastomas103, but in these tumours theparent-of-origin dependence is by no meansabsolute and an unequivocally imprintedlocus on chromosome 1 has not materiali zed.Recently,Morison, Reeve and colleagues havefound selective loss of maternal alleles onchromosome 9p in childhood acute lym-phoblastic leukaemias104.There are some datato indicate that LOI can also lead to tumour-suppressor gene silencing; for example,ARHI

a candidate breast tumour gene that wasfound by the Yu laboratory shows aberrantallele-specific silencing105. In addition,LIT1an untranslated RNA found by the Feinberg,Oshimura and Higgins groups to undergoLOI in about half of patients with BWS106,107

might cause downregulation ofCDKN1C(which encodes KIP2,also known as p57).

Chromatin and met hylation

The third epigenetic mechanism histonemodification has been the last to be linkedto cancer research.A link between chromatinand DNA methylation, however, dates backto the 1980s, in the elegant observation by theCedar and Graessmann laboratories thatnaked DNA templates, pre-methylatedin vitroand then transfected or microinjectedinto cell s, only became transcript ionallysilenced after packaging into a repressiveform of chromatin108,109.Proteins that bind tomethylated CpGs were soon identified byAdrian Birds group110,111,and work from thatlaboratory,along with Steve Baylins and AlanWolfes groups, showed that these proteins(MECP2 and MBD2), as well as DNAmethyltransferases themselves (DNMT1,

DNMT3A and DNMT3B),physically associ-ate with histone deacetylases112115.DNMT1maintains patterns of methylation duringrepli cati on, and DNMT3A and DNMT3Bcan add methylation to previously unmethy-lated templates.A recurring theme of DNAmethylation is that it s machinery co-opts amore fundamental system of chromati nmodification.For example,whereas DNMTsassociate with chromatin proteins in mam-mals, theDrosophilaorthologues of methyl-binding proteins Mbd2 and Mbd3 do notbind methyl-C, but stil l show conserved

role for DNA methylation in maintaining

allele-specific gene expression85and the discov-ery of species-conserved imprinted chromoso-mal domains, containing several imprintedgenes,found by chromosomal-walking experi-ments done for the chromosome 11p15 BWSregion by the Feinberg and Tycko groups,aswell as others8689.

Loss of imprinting (LOI), leading topathological biallelic expression ofIGF2(REFS 78,79) in Wilms tumours, was discov-ered by the Feinberg and Reeve laboratoriesin 1993,and in 1994 the Feinberg and Tyckolaboratories showed that this abnormality in

embryonal tumours is invariably li nked to again of DNA methylation that is localized tothe 5 sequences and transcribed region ofthe closely l inked and reciprocally imprintedH19gene,which i s thereby transcriptionallysilenced90,91. The presence ofH19hyperme-thylati on (a somatic gain of an epigeneti cmark on the previously expressed maternalallele) was found not only in the tumourDNA, but also in the non-neoplasti c kidneyparenchyma surrounding some of thetumours90. These observations were the firstto indicate a gatekeeper rol e for epigeneticalterations in cancer, as they are the earliestobservable geneti c change(FIG. 1). In 1997,mosaicism for H19hypermethylation inpatients with Wilms tumour was confi rmedin an independent series of cases92. The neteffect of thi s epigenetic lesion is silencing ofH19,a gene that encodes an abundant splicedbut non-tr anslated RNA, and a reciprocal

increase in expression ofIGF2(REFS 90,91).The roughly twofold increase in effectiveIGF2gene dosage is now considered the mostlikely explanation for the associated tumoursusceptibil ity, althoughH19RNA is growthsuppressive in some cancer cell lines93.Themechanism by which IGF2 promotestumour formation might be by inhibiti ngapoptosis,as Hanahan and colleagues foundthat knockout of one allele (preventing bial-lelic progression) arrests tumour progressionin an activated T-antigen transgenic tumourmodel94, and this arrested progressioninvolves increased apoptosis95.

Epidemiological support for the gate-keeper role of altered methylation and LOI inWilms tumour came from the discovery, in2000, that Knudsons hypothesis did notexplain the bimodal age distribution of mostWilms tumours.Tumours that are late ari s-ing, that is, not in infancy, were found toinvolve epigenetic rather than genetic alter-ations,and early-arising tumours had classicalgenetic changes involving Wilms tumour 1(WT1) and LOH96. Definit ive clinical evi-dence for a gatekeeper function of the gain ofDNA methylation upstream ofH19and LOI

ofIGF2in Wilms tumours has since comefrom studies of tumour susceptibil ity in thepre-neoplastic disorder BWS(see below).

Indicating generality of a role forimprinted genes in cancer, a recent publica-tion reported selective loss of paternal alleles(6/6 cases of LOH) on chromosome 19q inoligodendrogliomas97.As indicated by datafrom the Oshimura laboratory98, the putativetumour-suppressor gene identified by thesedata might bePEG3 a paternally expressedimprinted gene that maps to band 19q13.4,orperhaps another nearby imprinted gene,as


LO I

a Somaticmosaicism

b BeckwithWiedemannsyndrome

c Wilms tumour d Multiple Wilmstumours

Secondgeneticevent

Figure 1 | Gatekeeper role for loss of

imprinting ofIGF2in Wilms tumour.a | Loss

of imprinting (LO I, dark nuclei) has been shown to

arise sporadically as a somatic mosaic epigenetic

alteration, because LO I has been found in

parenchymal kidney tissue of patients with Wilms

tumour, as well as in pre-malignant nephrogenicrests (pink circles). b | A lternatively, LO I can arise

in the germline or very early in development in

BeckwithWiedemann syndrome, causing

nephromegaly (overgrowth of the whole kidney).

In both cases, overgrowth is caused by a double

dose ofIGF2expression and possibly silencing of

H19. c | A second, presumably genetic, event

can then lead to Wilms-tumour formation (red

circles). d | T his will be more common in those

with BeckwithWiedemann syndrome, so

multiple tumours arise.


6/11



Chinnaiyan and colleagues for EZH2, anorthologue of the Drosophilachromatin-repressor protein enhancer of zeste.Increasedexpression ofEZH2is linked to generali zedhypermethylation and gene silencing inmetastatic prostate cancer131.As we discuss inmore detail below,another methylation andchromati n connection involves RB,whichassociates both with DNA methyltransferasesand wi th the SUV39H1 histone methyltrans-ferase and HP1,providing a mechanism forrepressing cell-cycle genes,such as cyclin E.

BWS: epigenetic casualty in cancer

A key barrier to the acceptance of epigeneticalterations as a cause rather than a consequenceof cancer,has been the lack,until recently,ofwell-defined human pre-neoplastic disordersthat are caused by epigenetic mutations.Thereare known disorders involving genes thatencode the methylation machinery of the cell

(for example,DNMT3Bin ICF syndrome andMECP2in Rett syndrome),but these disordersdo not predispose to cancer.By contrast,manytumour-suppressor genes,when mutated in thegermline, cause cancer predispositi on syn-dromes (TP53in LiFraumeni syndrome beingthe original).

However, the discovery of the mechanismsof BWSprovides the genetic smoking gun forat least one epigenetic mechanism in cancer,genomic imprinting.The generali zed over-growth characteristic of BWS sometimesincludes kidney enlargement,and the affectedkidneys can contain persistent nephrogenicblastema the precursor of Wilms tumour.By no means do all children with BWSdevelop Wi lms tumours,but the relative ri skis 816 (REF. 132). Potentially explaini ng thisclinical heterogeneity,in the years between1993 and 2000,BWS was shown to have vari -ous molecular causes, including LOI ofIGF2(REFS 90,91,133) or,alternatively,point muta-tions in theCDKN1Cgene134 or epigeneticlesions in the nearby antisense RNA LIT1(REFS106,107),which,together, lie within a sep-arate imprinted subdomain of chromosome11p15.Furthermore,a review of BWScases in

the li terature up to 1999 indicated that cancerpredisposition might be specifically associatedwith LOI ofIGF2and hypermethylation ofH19(REF.135).To test this idea,several groups,including Mannens, Weksberg, Reik andMaher, found association of cancer in BWSwith hypermethylation ofH19,although inthose studies, because of patient numbers,one could not distinguish statistically an asso-ciation specific to H19from uniparentaldisomy includingH19and other genes136138.In the fi rst epigenotypephenotype studyfor any disease,DeBaun and Feinberg found

line was followed over prolonged passage intissue culture, lysine 9 methylation accompa-nied the cytosine methylati on-i ndependentre-sil encing of demethylated CDKN2Aalleles59.The histone methylation/ HP1-bind-ing cycle,which is present in organisms asdiverse as yeast,Drosophilaand humans, is anancient mechanism for propagating epige-neti c states, whereas the analogous CpGmethylation/histone-deacetylase-bindingcycle is evidently a later addition.This fail-safemechanism,which stabil izes silent chromatinin mammals,keeps parasiti c retroelementsrepressed and might have co-evolved withthese invasive sequences120.

Another very recent advance that mightprove relevant comes from observations byHenikoff and Ahmad that histones can beselectively replaced at transcriptionally activeloci, in a manner that i s independent of DNAreplication121. This process entails the tran-

scription-dependent accumulation of a highlyconserved histone variant H3.3 whichsubstitutes for the canonical H3 histone.Hi stone replacement, which presumablyoccurs after invasion of promoters by stronglyactivating transcription factors,offers an expla-nation for how the cell might reactivate genesthat were previously silenced via histonemethylation.This is an attractive idea,giventhat there are no known histone demethy-lases122.Whether histone replacement might beperturbed in cancer cells is an open question.

Very recently,an intr iguing relationshipbetween genomic imprinting,DNA methyla-tion and chromatin has been established bythe discovery and functional analysis ofCTCF,an insulator protein that establisheschromati n boundaries and the binding ofwhich is blocked by DNA methylation,shownby the Lobanenkov,Ohlsson, Felsenfeld andTilghman laboratories123126. Cui, Feinbergand colleagues have found that LOI in Wilmstumour depends on hypermethylation ofCTCF binding sites127, which reside in theDNA upstream ofH19,but hypomethylationof IGF2, not apparently involving CTCF,seems to occur in some colorectal cancers128,

whereas de la Chapelle and colleagues havefound hypermethylati on ofCTCFin othercolorectal cancers129.

A paralogue ofCTCF, termedBORIS,wasrecently discovered by Lobanenkov,Ohlssonand colleagues to be amplified and overex-pressed within the 20q amplicon in breastcancer, and i ts overexpression i s thought toimpede normal CTCF binding130. It is intrigu-ing that BORISis itself a cancer/testis geneand, therefore,its hypomethylation might beli nked to hypermethylati on at other sites.Asimilar mechanism has been suggested by

transcriptional repressor function/histonedeacetylase association116.

Methylation at lysine residues in histoneshas been known for many years, but thismodification was only recently recognized byJenuwein and colleagues and Allis and col-leagues as crucially important for normalgene regulation117,118.Histone methylation is aparsimonious explanation for the perpetua-ti on of silent epigeneti c states through celldivisions.The silent state can be maintainedby a cycle of histone methylati on, which iscatalysed by theSUV39H1histone methyl-transferase an orthologue of aDrosophilaprotein that is involved in suppression of

position-effect variegation followed byrecruitment of the binding protein hete-rochromatin protein-1 (HP1) to the lysine-9-methylated histone,which perpetuates thecycle by recruiting SUV39H1 (FIG. 2). Thiswork was influenced by the knowledge thatposition-effect variegation is mediated by het-erochromati n formation. Indeed, Jones hasfound that methylation of lysine 9 in histoneH3 correlates with silencing of theCDKN2Atumour suppressor in cancer cells119.Moreover,Vogelstein and colleagues foundthat when a DNA-methyltransferase-null cell

M BD

D NMT

K 9 K 9 K 9C pG

K 9 K 9

M e M e M eM e

M e

C pG

M e

Ac

HP1 HP1 HP1

HDAC

M BD

HM T

Figure 2 | Co-operative and self-reinforcing

organization of the chromatin and DNA-

modifying machinery responsible for gene

silencing in normal and malignant cells.

Histone (H 3) modifications include lysine (K )

acetylation (Ac) and lysine methylation (M e).

Lysines at other positions are also modified. The

HP1 protein recognizes M eK9 and, as this protein

also binds the histone methyltransferase (H M T),

heterochromatin can spread. H istone

deacetylases (HD AC ) deacetylate lysine residues

as a prerequisite for their subsequent methylation.

DNA methyltransferases (DN M T) participate in

multiprotein complexes that contain HD ACs and

HM Ts, and methyl-C binding proteins (M BD ) can

be loaded onto methylated DN A through their

interactions with both HDAC s and HM Ts. M uch of

the evidence comes from studies of constitutive

heterochromatin, but recent studies indicate

similar interactions of genes silenced de novoin

cancer cells.


7/11


binding of these methyltransferases to theoncogenic PMLRAR fusion protein inpromyelocytic leukaemia156.This secondarymethylation scenario was questi oned basedon the finding ofRARB2methylation inmany leukaemias that lack PMLRAR157,but these findings can be reconciled i f othertranscripti onal repressors, present i n thePMLRAR-negative cases, also recrui tmethyltransferases.A second type of repres-sor that might shut down gene promotersbefore DNA methylation is typified by theSLUGtranscription factor, which Fearon andcolleagues found binds and silences theE-cadherin promoter a known target ofde novomethylation in breast cancer celllines158, although, to our knowledge, therehave not yet been reports of methyltransferaserecruitment via SLUG.

In an important li ne of work that wasinitiated by the Goff laboratory in 1994 and

substantially extended by Doug Deans labo-rator y,RB has been shown to function as abrake on the cell cycle at least in part byestablishing and enforcing stable epigeneticsilencing of its target genes. It does this bypart icipating in a multiprotein complex thatincludes chromatin-remodell ing enzymes ofthe SWI/SNF class159, as well as histonedeacetylases160,161 and epigenetic silencingproteins of the polycomb class (geneti cantagonists of tr ithorax-group proteins)162.Furthermore,based on at least two reports,the DNA methyltransferase DNMT1 alsobinds to RB163,164. Interestingly, the Riz1gene,which encodes a histone methyl trans-ferase that can associate with Rb,was foundby Huang and colleagues to sometimes beinactivated epigenetically and act as atumour suppressor in mice165. This findingindicates a hypothetical sequence of eventsin which a cancer-associated epigeneticlesion (including silencing at theRiz1locus)exerts downstream effects that abrogate thefunction of a tumour-suppressor protein(Rb), and this, in turn, erases stable epige-netic silencing at the cyclin E gene and otherRb targets.Involvement of polycomb-group

proteins in human cancer is not restr icted totheir interaction with RB: overexpression ofEZH2in metastatic prostate cancer, alreadymentioned above, is another example131.

Epigenetic chemotherapy

The demethylating drug 5-azaCdR whichinactivates methyltransferases was firstshown to transform cultured cells byWeinstein and colleagues in 1984 (REF. 166),and these studies were stimulated by theearlier finding by Taylor and Jones that5-azaCdR has reproducible effects on cell

that in a large registry of patients with BWSthat was studi ed prospectively, gain ofmethylation at H19, presumably resulting inbiallelic expression ofIGF2,was specificall yand statistically associated with cancer risk,whereas loss of methylation at LIT1wasspecifically associated with birth defects(macrosomia and midl ine abdominal-walldefects)139. This specifi city for cancer r iskindicates a gatekeeper role of LOI in BWS(FIG.1). The mechanisms accounting for gainof methylati on at H19or loss of methyla-tion at LIT1are not known,but recent f ind-ings by DeBaun, Feinberg, Maher andothers140,141, which associate BWS wi thin vitroferti li zation, together with the sim-ple fact that the epigenetic abnormality iswidespread in various somatic tissues, indi-cates that these epigenetic aberrations occurvery early in development, before the devel-opment of malignancy. In other words,epi-

geneti c lesions in BWS are a cause, not aconsequence,of cancer.

LOI might have a causal role in commoncancer as well.A recent study indicates that,although BWS is rare,epigenetic alterationsaffecting IGF2might be common in the gen-eral population and associated with moreprevalent malignancies. LOI of IGF2wasfound by Cui,Cruz-Correa and Feinberg innormal lymphocytes and colonic mucosa in10% of healthy adults,and the odds ratio forLOI was 5.15 for patients with a positive fam-ily history,and 21.7 for patients with a pasthistory,of colorectal cancer142.Here too,theepigeneti c abnormality is found in normalcells,and so is not an epiphenomenon of thecancer phenotype.Interestingly, the mecha-nism of LOI in colorectal cancer i nvolveshypomethylation ofIGF2, rather than hyper-methylation ofH19that is seen in embryonaltumours128,which is consistent with the ideathat cancer is li nked to epigenetic disequilib-rium rather than hypomethylation or hyper-methylation per se.The timing of epigeneticlesions in cancer is a topic that cannot be over-emphasized. Promoter hypermethylationmight also often be an early event.Ominously,

in smokers with bronchial epithelial atypiathat a pathologist would classify as pre-neoplasti c, Herman and colleagues foundalready substantial hypermethylation of theCDKN2Apromoter region143.In an intriguingrecent study,Sapienza has found famil ial clus-tering of epigenetic alterations involvingH19,indicating that methylation might represent anepigenetic cancer-associated polymorphism inthe population144.

An alternative genetic argument can bederived from mouse models,with the caveatthat mouse and human carcinogenesis

might di ffer. Three mouse models poten-tially li nk DNA methylation to cancer.Thefi rst of these was the demonstrati on byJaenisch and colleagues that aDnmt1hypo-morphic mutation reduces the frequencyof intesti nal neoplasia when crossed toApcMi nmice145. These results indicate thathypomethylation might abate the risk ofcancer, but more recent studies fromJaenisch and colleagues indicate just theopposite, with a high frequency of lym-phomas in mice with a hypomorphicDnmt1allele146. Together, the data indicate that adisruption in the balance of methylation isassociated wi th cancer risk, an idea that isconsistent wi th observations of human can-cer. In a third model heterozygousknockout of theHic1gene,which is hyper-methylated in human colorectal cancer Bayli n and coll eagues found a modest, butsignif icant, increase in cancer frequency of

very late onset, with hypermethylati on ofthe wild-type allele147.

Tumour suppressors and chromatin

The counter-argument from human geneticstudies is that the many tumour-suppressorgenes in cancer are not modif iers of DNAmethylation,even though they are involved invirtually every other potential growth or reg-ulatory pathway. On the other hand, manytumour-suppressor genes are involved insome aspect of chromatin structure.The firstof these involved the cloning by Canaani ,Rowley and Cleary of theALL1gene (alsoknown asMLL or HRX) in 1991. ALL1which is altered by chromosomal transloca-tions involving band 11q23 to produce fusiongenes in human leukaemias148150 is thehuman homologue ofDrosophila tri thorax,which functions to stably maintain activeexpression states of homeobox genes in grow-ing tissues.The protein encoded byALL1par-ti cipates in a megadalton-size multi proteincomplex that has chromatin remodell ing,his-tone acetylation/deacetylati on and histonemethylation activities151.

A suggested candidate for a mechanism of

methylation modification was the DNAmethyltransferaseDNMT1,which was origi-nally reported to be overexpressed in cancercells152,but these studies have been controver-sial153,154,and a well-controlled real-time PCRanalysis did not show a significant increase155.An alternative hypothesis involves expressionof transcriptional repressors or,equivalently,the loss of activators,as the primary event.Arecent example,put forward by Peli cci andcolleagues to support the fi rst possibi li ty, isthe recruitment of DNMT1 and DNMT3A tothe retinoic-acid receptor (RARB2) locus by



8/11



Benjami n Tycko is at the Insti tu te for Cancer

Genetics and Department of Pathology, Columbia

Uni versit y College of Physicians and Surgeons,

New York 10032,USA.

Corr espondence to A.P.F. or B.T.

e-mail : afeinberg@jhu .edu; bt12@columbi a.edu

doi10.1038/nrc1279

1. Feinberg, A. P. & Vogelstein, B. Hypomethylationdistinguishes genes of some human cancers from their

normal counterparts. Nature301, 8992 (1983).2. G ama-Sosa, M . A. et al. The 5-methylcytosine content of

DN A from human tumors. Nucleic Acids Res. 11,68836894 (1983).

3. Goelz, S. E., Vogelstein, B., Hamilton, S. R. & Feinberg, A. P.

Hypomethylation of DNA from benign and malignant

human colon neoplasms. Science228, 187190 (1985).4. Feinberg, A. P., Gehrke, C. W., Kuo, K. C. & Ehrlich, M .

Reduced genomic 5-methylcytosine content in human

colonic neoplasia. Cancer Res. 48, 11591161 (1988).5. Strichman-Almashanu, L. Z. et al. A genome-wide screen

for normally methylated human C gG islands that can

identify novel imprinted genes. Genome Res. 12,543554 (2002).

6. Feinberg, A. P. & Vogelstein, B. Hypomethylation ofras

oncogenes in primary human cancers. Biochem.

Biophys. Res. Commun. 111, 4754 (1983).7. De Smet, C. et al. The activation of human gene MAGE-1in

tumor cells is correlated with genome-wide demethylation.

Proc. Natl Acad. Sci. USA 93, 71497153 (1996).8. Cho, B. et al. Promoter hypomethylation of a novel

cancer/testis antigen gene CAGEis correlated with its

aberrant expression and is seen in premalignant stage of

gastric carcinoma. Biochem. Biophys. Res. Commun.

307, 5263 (2003)9. Adorjan, P. et al. Tumour class prediction and discovery

by microarray-based D NA methylation analysis. Nucleic

Acids Res. 30, e21 (2002).10. Iacobuzio-Donahue, C. A. et al. Exploration of global gene

expression patterns in pancreatic adenocarcinoma using

cDN A microarrays. Am. J. Pathol. 162, 11511162 (2003).11. O shimo, Y. et al. Promoter methylation of cyclin D2 gene

in gastric carcinoma. Int. J. Oncol. 6, 16631670 (2003).12. Ak iyama, Y., M aesawa, C. , Ogasawara, S., Terashima,

M . & M asuda, T. C ell-type-specific repression of the

maspin gene is disrupted frequently by demethylation at

the promoter region in gastric intestinal metaplasia and

cancer cells. Am. J. Pathol. 163, 19111919 (2003).13. Cho, M . et al. Hypomethylation of the MN/CA9promoter

and upregulated MN/CA9expression in human renal cellcarcinoma. Br. J. Cancer85, 563567 (2001).

14. Nak amura, N. & Takenaga, K. Hypomethylation of the

metastasis-associated S100A4gene correlates with

gene activation in human colon adenocarcinoma cell

lines. Clin. Exp. Metastasis16, 471479 (1998).15. Badal, V. et al. C pG methylation of human papillomavirus

type 16 DN A in cervical cancer cell lines and in clinical

specimens: genomic hypomethylation correlates with

carcinogenic progression. J. Virol. 77, 62276234 (2003).16. De Capoa, A. et al. DN A demethylation is directly related

to tumour progression: evidence in normal, pre-malignant

and malignant cells from uterine cervix samples. Oncol.

Rep. 10, 545549 (2003).17. Sato, N. et al. Frequent hypomethylation of multiple

genes overexpressed in pancreatic ductal

adenocarcinoma. Cancer Res. 63, 41584166 (2003).18. Piyathilake, C. J. et al. Race- and age-dependent

alterations in global methylation of D NA in squamous cell

carcinoma of the lung (United States). Cancer Causes

Control14, 3742 (2003).19. Lengauer, C. , K inzler, K . W. & Vogelstein, B. D NA

methylation and genetic instability in colorectal cancer

cells. Proc. Natl Acad. Sci. USA 94, 25452550 (1997).20. Pao, M . M . et al. DN A methylator and mismatch repair

phenotypes are not mutually exclusive in colorectal

cancer cell lines. Oncogene19, 943952 (2000).21. K ane, M . F. et al. M ethylation of the hMLH 1 promoter

correlates with lack of expression of hM LH 1 in sporadic

colon tumors and m ismatch repair- defective human

tumor cell lines. Cancer Res. 57, 808811 (1997).22. Cui, H., H oron, I. L., O hlsson, R., Hamilton, S. R. &

Feinberg, A . P. Loss of imprinting in normal tissue of

colorectal cancer patients with microsatellite instabi lity.

Nature Med. 4, 12761280 (1998).23. Q u, G. Z. , G rundy, P. E., Narayan, A. & Ehrlich, M .

Frequent hypomethylation in Wilms tumors of

pericentromeric DNA in chromosomes 1 and 16. Cancer

Genet. Cytogenet. 109, 3439 (1999).

in molecular genetics is the mode of propa-gation of the histone code, through disassem-bly and reassembly during cell divi sion,andthe mechanism for this process.Cancer epi-genetics will probably advance substantiallywhen that process is better understood.Second, it is remarkable that tumour-suppressor genes are drawn from so manyaspects of cell biology except DNA methyla-ti on. This indicates that the methylationchanges themselves are secondary to otherimportant causal elements, although thatneed not necessarily be the case. It mightsimply be that the known mediators of DNAmethylation are factors that are essential formammalian life,but that accessory factors formethylati on propagati on that are mutatedare not known or that their role in methyla-tion is not known. Third, the most com-pell ing evidence for a causal role of epigeneticchanges comes from the study of well-defined

human genetic and epigenetic syndromes.The main problem is that so few familial dis-orders seem to involve the epigenetic machin-ery.One notable exception i s BWS,which iscaused by epigenetic defects,and those alter-ations are specifically li nked to cancer risk inaffected patients. The future might revealpopulation epigenetic polymorphisms thatcontr ibute to cancer ri sk, but that do notcause a stark definable syndrome; the recentidenti fication of a methylation variant thatis linked to colorectal cancer might be sucha polymorphism.

Finally,we would argue that age is centralto our understanding of cancer epigenetics,an idea that goes all the way back to Hollidaysobservati ons of methylation erosion duringageing.The single leading risk factor for can-cer is age.Although that has often been attrib-uted to the accumulation of mutations overtime, an alternati ve and complementaryinterpretation is that age it self disrupts theepigenetic programme,increasing cancer risk.This relationship might be even more true fornon-malignant disease than for cancer, as epi-genetics might explain why most commondisorders that involve complex genetics are of

adult onset; after all , the genome has beenthere since birth, so the genetic factors arepresumably there as well.We feel that sub-stantially more attention must be paid to epi-genetic variation in the population,epigeneticchanges during ageing and the relationshipbetween these changes and common diseasesincluding cancer. For these studies, the bestmodel organisms are humans themselves.

Andrew P. Feinberg is at the Departments of

Medicine,Oncology,and Molecular Biology &

Geneti cs, Johns Hopki ns Universit y School of

Medicine,Baltimore, Maryland 21205,USA.

differentiation in tissue culture167.This drugis now used in some clinical situations,notably as part of combined chemotherapyregimens for myelodysplastic syndrome andleukaemias168.Part of its activity in patientsmight be due to its ability to reactivategrowth suppressors,such as the INK4B (alsoknown as p15) cyclin-dependent kinaseinhibitor168. However, data that wereobtained withDnmt1-knockout cells thatare resistant to 5-AzaCdR indi cate that theincorporation of the drug into DNA, andthe resulti ng formation of covalentDNADnmt adducts,might contr ibute toits cytotoxic effects169.A necessity for intra-venous administrati on has limi ted the use-ful ness of 5-AzaCdR, but an orally acti veinhibi tor,zebulari ne170, is enteri ng clinicaltrials. 5-AzaCdR can also restore a normalpattern of imprinting to cells171.The histonedeacetylase inhibitor trichostatin A is

already in use and seems to have efficacyagainst leukaemias. Potential synergybetween trichostatin A and 5-azaCdR172 isnow being tested in this clinical setting173.

These promising therapies seek to reacti-vate tumour-suppressor genes that have beensilenced epigenetically,and they are justifiedin patients with cancer in which other treat-ments have failed or are expected to fail.Nevertheless,a recent exchange inSciencehashighlighted the concern that genomic insta-bil ity,because of hypomethylation,might bean adverse long-term consequence174,175.

Conclusions

In the past 20 years, cancer epigenetics hascome full circle,with a renaissance of interestin hypomethylation and its role inactivating oncogenes and chromosomalrearrangement, as well as hypermethylationaffecting tumour-suppressor genes. In thepast 10 years, the discovery of imprintedgenes and their role in cancer has added anew dimension to the field, and the impact ofthe role of chromati n modif icati ons is justbeginning to be felt.One of the most intrigu-ing recent advances is the convergence of

mechanistic studies linking DNA methyla-tion,genomic imprint ing and histone modi-fi cati on.Although cancer epigenetics is nowconsidered to be well within the mainstream,there are several remaining questions thatcontinue to limit its complete acceptance andstill stimulate debate in assigning causal rela-tionships.First, the mechanism of epigeneticinheri tance other than DNA methylation issti ll largely unknown,yet i t must be impor-tant as nonmethylated species handle epige-neti c modif ication quite well . One couldargue that the most important open question


9/11


10/11

1 0 | FEBRUARY 2004 | VOLUME 4 www.nature.com/reviews/cancer


cells and progression stages of colon cancer. Proc. Natl

Acad. Sci. USA 88, 34703474 (1991).153. Lee, P. J. et al. Limited upregulation of DNA

methyltransferase in human colon cancer reflecting

increased cell proliferation. Proc. Natl Acad. Sci. USA 93,1036610370 (1996).

154. De M arzo, A. M . et al. Ab normal regulation of DN A

methyltransferase expression during colorectal

carcinogenesis. Cancer Res. 59, 38553860(1999).

155. Eads, C . A. et al. C pG island hypermethylation in human

colorectal tumors is not associated with DN Amethyltransferase overexpression. Cancer Res. 59,23022306 (1999).

156. Di Croce, L. et al. M ethyltransferase recruitment and

DN A hypermethylation of target promoters by an

oncogenic transcription factor. Science295, 10791082(2002).

157. Esteller, M . et al. C ancer epigenetics and methylation.

Science297, 18071808 (2002).158. H ajra, K . M ., C hen, D. Y. & Fearon, E. R . The SLUG zinc-

finger protein represses E-cadherin in breast cancer.

Cancer Res. 62, 16131618 (2002).159. Dunaief, J. L. et al. The retinoblastoma protein and BR G 1

form a complex and cooperate to induce cell cycle arrest.

Cell79, 119130 (1994).160. Luo, R. X., Postigo, A. A. & Dean, D. C . R b interacts with

histone deacetylase to repress transcription. Cell92,463473 (1998).

161. M agnaghi-Jaulin, L . et al. Retinoblastoma protein

represses transcription by recruiting a histone

deacetylase. Nature391, 601605 (1998).162. Dahiya, A., Wong, S., G onzalo, S., G avin, M . & Dean, D. C.

Link ing the Rb and polycomb pathways. Mol. Cell8,557569 (2001).

163. Pradhan, S. & K im, G . D. T he retinoblastoma gene

product interacts with maintenance human DN A

(cytosine-5) methyltransferase and modulates its activity.

EMBO J. 21, 779788 (2002).164. R obertson, K . D. et al. DN M T1 forms a complex with Rb,

E2F1 and HDA C 1 and represses transcription from E2F-

responsive promoters. Nature Genet. 25, 338342 (2000).165. Steele-Perkins, G . et al. Tumor formation and inactivation

of RIZ1, an Rb-binding member of a nuclear protein-

methyltransferase superfamily. Genes Dev. 15,22502262 (2001).

166. Hsiao, W.-L., G attoni-C elli, S. & Weinstein, I. B. Effects

of 5-azacytidine on the progressive nature of cell

transformation. Mol. Cell. Biol. 5, 18001803 (1985).167. Taylor, S . M . & Jones, P. A. M ultiple new phenotypes

induced in 10T 1/2 and 3T 3 cells treated with 5-

azacytidine. Cell17, 771779 (1979).

168. Daskalakis, M . et al. Demethylation of a hypermethylated

P15/INK4Bgene in patients with myelodysplastic

syndrome by 5-aza-2-deoxycytidine (decitabine)

treatment. Blood100, 29572964 (2002).169. Juttermann, R ., Li, E. & Jaenisch, R . Toxicity of 5-aza-2-

deoxycytidine to mamm alian cells is mediated primarily

by covalent trapping of D NA methyltransferase rather

than DNA demethylation. Proc. Natl Acad. Sci. USA 91,1179711801 (1994).

170. Cheng, J. C. et al. Inhibition of DN A methylation and

reactivation of silenced genes by zebularine. J. Natl

Cancer Inst. 95, 399409 (2003).171. Barletta, J. M ., Rainier, S . & Feinberg, A. P. R eversal of

loss of imprinting in tumor cells by 5-aza-2-

deoxycytidine. Cancer Res. 57, 4850 (1997).172. Cameron, E. E., B achman, K . E., M yohanen, S. ,

Herman, J. G . & Baylin, S . B . S ynergy of demethylation

and histone deacetylase inhibition in the re-expression of

genes silenced in cancer. Nature Genet. 21, 103107(1999).

173. Shaker, S. , Bernstein, M ., M omparler, L. F. & M omparler, R . L.Preclinical evaluation of antineoplastic activity of inhibitors

of DN A methylation (5-aza-2-deoxycytidine) and histone

deacetylation (trichostatin A, depsipeptide) in

combination against myeloid leukemic cells. Leuk. Res.

27, 437444 (2003).174. Eden, A ., G audet, F. & Jaenisch, R . R esponse to

comment on C hromosomal instability and tumors

promoted by dna hypomethylation and Induction of

tumors in mice by genomic hypomethylation. Science

302, 1153 (2003).175. Yang, A . S., Estecio, M . R ., G arciaM anero, G. ,

K antarjian, H . M . & Issa, J. P. C omment on

C hromosomal instability and tumors promoted by

DN A hypomethylation and Induction of tumors in nice

by genomic hypomethylation. Science302, 1153(2003).

176. M ohandas, T., Sparkes, R . S. & Shapiro, L. J.

Reactivation of an inactive human X chromosome:

128. C ui, H. et al. Loss of im printing in colorectal cancer linked

to hypomethylation ofH19and IGF2. Cancer Res. 62,64426446 (2002).

129. N akagawa, H . et al. Loss of im printing of the insulin-like

growth factor II gene occurs by biallelic methylation in a

core region of H19-associated CT C F-binding sites in

colorectal cancer. Proc. Natl Acad. Sci. USA 98,591596 (2001).

130. Loukinov, D. I. et al. BO R IS, a novel male germ-line-

specific protein associated with epigenetic

reprogramming events, shares the same 11-zinc-finger

domain with C TC F, the insulator protein involved inreading imprinting marks in the soma. Proc. Natl Acad.

Sci. USA 99, 68066811 (2002).131. Varambally, S . et al. The polycomb group protein EZH 2 is

involved in progression of prostate cancer. Nature419,624629 (2002).

132. D eBaun, M . R. & Tucker, M . A. R isk of cancer during the

first four years of life in children from T he

BeckwithWiedemann Syndrome R egistry. J. Pediatr.

132,398400 (1998).133. Weksberg, R ., Shen, D. R. , Fei, Y. L.,Song, Q . L., &

Squire, J. D isruption of insulin-like growth factor 2

imprinting in BeckwithWiedemann syndrome. Nature

Genet. 5,143150 (1993).134. Hatada, I. et al. An imprinted gene p57KIP2 is mutated in

BeckwithWiedemann syndrome. Nature Genet. 14,1711733 (1996).

135. Tycko, B . G enomic imprinting and cancer. Results Probl.

Cell. Differ. 25, 133169 (1999).136. Engel, J. R . et al. Epigenotype-phenotype correlations in

BeckwithWiedemann syndrome. J. Med. Genet. 37,921926 (2000).

137. Bliek, J. et al. Increased tumour risk for BWS patients

correlates with aberrant H19and not KCNQ1OT1

methylation: occurrence ofKCNQ1OT1hypomethylation

in familial cases of BWS. Hum. Mol. Genet. 10, 467476(2001).

138. Weksberg, R . et al. Tumor development in the

BeckwithWiedemann syndrome is associated with a

variety of constitutional molecular 11p15 alterations

including imprinting defects of KC NQ 1OT 1. Hum. Mol.

Genet. 10, 29893000 (2001).139. DeBaun, M. R . et al. Epigenetic alterations of H19 and

LIT 1 distinguish patients with BeckwithWiedemann

syndrome with cancer and birth defects. Am. J. Hum.

Genet. 70, 604611 (2002).140. D eBaun, M . R. , N iemitz, E. L. & Feinberg, A. P.

Association ofin vitrofertilization with

BeckwithWiedemann syndrome and epigenetic

alterations of LIT1 and H19. Am. J. Hum. Genet. 72,156160 (2002).

141. M aher, E. R. et al. BeckwithWiedemann syndrome and

assisted reproduction technology (ART) . J. Med. Genet.

40, 6264 (2003).142. Cui, H . et al. Loss ofIGF2imprinting: a potential marker of

colorectal cancer risk. Science299, 64426446 (2003).143. Belinsky, S . A . et al. Aberrant methylation ofp16(INK4a) is

an early event in lung cancer and a potential biomark er

for early diagnosis. Proc. Natl Acad. Sci. USA 95,1189111896 (1998).

144. Sandovici, I. et al. Familial aggregation of abnormal

methylation of parental alleles at the IGF2/H19and IGF2R

differentially methylated regions. Hum. Mol. Genet. 12,15691578 (2003).

145. Laird, P. W. et al. Suppression of intestinal neoplasia by

DN A hypomethylation. Cell81, 197205 (1995).146. G audet, F. et al. Induction of tumors in mice by genomic

hypomethylation. Science300, 489492 (2003).147. Chen, W. Y. et al. Heterozygous disruption ofHic1

predisposes mice to a g ender-dependent spectrum of

malignant tumors. Nature Genet. 33, 197202

(2003).148. Ziemin-van der Poel, S. et al. Identification of a gene,

MLL , that spans the breakpoint in 11q23 translocations

associated with human leukemias. Proc. Natl Acad. Sci.

USA 88, 1073510739 (1991).149. T kachuk, D. C. , K ohler, S . & Cleary, M . L. Involvement of

a homolog ofDrosophilatrithorax by 11q23

chromosomal translocations in acute leukemias. Cell71,691700 (1992).

150. G u, Y. et al. The t(4;11) chromosome translocation of

human acute leukemias fuses theALL-1gene, related to

Drosophilatrithorax, to the AF-4gene. Cell71, 701708(1992).

151. Nakamura, T. et al. ALL-1 is a histone methyltransferase

that assembles a supercomplex of proteins involved in

transcriptional regulation. Mol. Cell10, 11191128(2002).

152. El-Deiry, W. S . et al. High expression of the DN A

methyltransferase gene characterizes human neoplastic

103. C aron, H. et al. C hromosome bands 1p35-36 contain

two distinct neuroblastoma tumor suppressor loci, one of

which is imprinted. Genes Chromosomes Cancer30,168174 (2001).

104. M orison, I. M ., Ellis, L. M ., Teague, L. R . & Reeve, A. E.

P referential loss of maternal 9p alleles in childhood acute

lymphoblastic leukemia. Blood99, 375377 (2002).105. Yuan, J. et al. Aberrant methylation and silencing ofARHI,

an imprinted tumor suppressor gene in which the

function is lost in breast cancers. Cancer Res. 63,41744180 (2003).

106. Lee, M . P. et al. Loss of imprinting of a paternallyexpressed transcript, with antisense orientation to

KVLQT1, occurs frequently in Beck withWiedemann

syndrome and is independent of insulin-like growth factor

II imprinting. Proc. Natl Acad. Sci. USA 96, 52035208(1999).

107. Smilinich, N. J. et al. A maternally methylated CpG island

in KvLQT1is associated with an antisense paternal

transcript and loss of imprinting in B eckwithWiedemann

syndrome. Proc. Natl Acad. Sci. USA 96, 80648069(1999).

108. B uschhausen, G ., W ittig, B ., G raessmann, M . &

G raessmann, A. C hromatin structure is required to block

transcription of the methylated herpes simplex virus

thymidine k inase gene. Proc. Natl Acad. Sci. USA 84,11771181 (1987).

109. K eshet, I., Lieman-Hurwitz, J. & C edar, H . DN A

methylation affects the formation of active chromatin. Cell

44, 535543 (1986).110. Lewis, J. D. et al. Purification, sequence, and cellular

localization of a novel chromosomal protein that binds to

methylated D NA . Cell69, 905914 (1992).111. M eehan, R. R ., L ewis, J. D. & Bird, A. P. C haracterization

of M eCP 2, a vertebrate DN A binding protein with affinity

for methylated DN A. Nucleic Acids Res. 20, 50855092(1992).

112. Jones, P. L . et al. M ethylated DNA and M eCP 2 recruit

histone deacetylase to repress transcription. Nature

Genet. 19, 187191 (1998).113. N an, X. et al. Transcriptional repression by the methyl-

C pG -binding protein M eCP 2 involves a histone

deacetylase complex. Nature393, 386389 (1998).114. Rountree, M. R ., Bachman, K . E. & Baylin, S. B. DN M T1

binds HDA C 2 and a new co-repressor, D M AP 1, to form a

complex at replication foci. Nature Genet. 25, 269277(2000).

115. Wade, P. A. et al. M i-2 complex couples DN A methylation

to chromatin remodelling and histone deacetylation.

Nature Genet. 23, 6266 (1999).116. R oder, K . et al. Transcriptional repression byDrosophila

methyl-CpG -binding proteins. Mol. Cell Biol. 20,

74017409 (2000).

117. R ea, S. et al. Regulation of chromatin structure by site-

specific histone H3 methyltransferases. Nature406,593599 (2000).

118. Strahl, B. D., Ohba, R., C ook, R. G . & Allis, C . D.

M ethylation of histone H3 at lysine 4 is highly conserved

and correlates with transcriptionally active nuclei in

Tetrahymena. Proc. Natl Acad. Sci. USA 96,1496714972 (1999).

119. Nguyen, C . T. et al. Histone H3-lysine 9 methylation is

associated with aberrant gene silencing in cancer cells

and is rapidly reversed by 5-aza-2-deoxycytidine.

Cancer Res. 62, 64566461 (2002).120. Yoder, J. A ., Walsh, C . P. & Bestor, T. H . C ytosine

methylation and the ecology of intragenomic parasites.

Trends Genet. 13, 335340 (1997).121. Ahmad, K . & Henikoff, S. Histone H3 variants specify

modes of chromatin assembly. Proc. Natl Acad. Sci. USA

99, S16477S16484 (2002).122. Bannister, A . J., Schneider, R . & K ouzarides, T. H istone

methylation: dynamic or static?Cell109, 801806(2002).

123. Lobanenkov, V. V., N icolas, R. H., Plumb, M . A. , Wright, C . A.

& G oodwin, G . H . Sequence-specific DNA -binding

proteins which interact with (G + C )-rich sequences

flanking the chicken c-mycgene. Eur. J. Biochem. 159,181188 (1986).

124. H olmgren, C. et al. C pG methylation regulates the

Igf2/H 19 insulator. Curr. Biol. 11, 11281130 (2001).125. H ark, A . T. et al. CTC F mediates methylation-sensitive

enhancer-blocking activity at the H19/Igf2locus. Nature

405, 486489 (2000).126. B ell, A. C ., West, A. G . & Felsenfeld, G . T he protein CTC F

is required for the enhancer blocking activity of vertebrate

insulators. Cell98, 387396 (1999).127. C ui, H. et al. Loss of im printing of insulin-like growth

factor-II in Wilms tumor commonly involves altered

methylation but not mutations of CT CF or its binding site.

Cancer Res. 61, 49474950 (2001).


11/11


evidence for X inactivation by DN A methylation. Science

211, 393396 (1981).177. Wolf, S. F., Jolly, D . J., L unnen, K . D. , Friedmann, T., &

M igeon, B. R. M ethylation of the hypoxanthine

phosphoribosyltransferase locus on the human X

chromosome: implications for X-chromosome

inactivation. Proc. Natl Acad. Sci. USA 81, 28062810(1984).

178. Antequera, F., M acleod, D. & B ird, A. P. Specific

protection of methylated CpG s in mammalian nuclei. Cell

58, 509517 (1989).

179. Hansen, R. S. & G artler, S . M . 5-azacytidine-inducedreactivation of the human X chromosome-linked PGK1

gene is associated with a large region of cytosine

demethylation in the 5 C pG island. Proc. Natl Acad. Sci.

USA 87, 41744178 (1990).

180. Jeppesen, P. & Turner, B . M . T he inactive X chromosome

in female mammals is distinguished by a lack of histone

H4 acetylation, a cytogenetic marker for gene

expression. Cell74, 281289 (1993).

AcknowledgementsWe thank E . Fearon, B . Vogelstein, R . O hlsson and G . K lein for

reading the manuscript. This work was supported by grants

from the National Institutes of Health. We apologize in advance

for any errors of omission or commission.

C ompeting interests statementThe authors declare that theyhave no competing financial interests.

Online links

DATABASESThe following terms in this article are linked online to:

Cancer.gov:http://cancer.gov/acute lymphoblastic leukaemias | acute myelogenous

leukaemia | cervical cancer | colon cancer | liver cancer | lung

cancer | malignant rhabdoid tumour | pancreatic cancer |

stomach cancer | uterine cancer | Wilms tumour

LocusLink:http://www.ncbi.nlm.nih.gov/LocusLink/ALL1| ARHI| ATRX| BORIS| C AG E | calcitonin |CDKN1C|

CTCF | DNM T1 | DNMT3A | DNMT3B | EZH2| H19| HP1 |

HRAS | IGF2| LIT1| Lsh | MBD2 | MD R1 | MECP 2 |MLH1|

Nf1| PEG3 |PM L | RAR | RARB2| RB |SLUG | SNF5 |SUV39H1 | Trp53| WT1

FURTHER INFORMATIONDNA Methylation Society:http://dnamethsoc.server101.com/

Imprinted Gene Catalogue:http://cancer.otago.ac.nz/IGC /Web/home.html

Imprinting: http://www.geneimprint.comAccess to this interacti ve links box is free online.

NATURE REVIEWS | CANCER VOLUME 4 | FEBRUARY 2004 | 1 1

Documents

The History of Cancer Epigenetics