REVIEWS

The diverse roles of DNA methylation

in mammalian development and
disease
Maxim V. C. Greenberg and Deborah Bourc’his*
Abstract | DNA methylation is of paramount importance for mammalian embryonic development.
DNA methylation has numerous functions: it is implicated in the repression of transposons and
genes, but is also associated with actively transcribed gene bodies and, in some cases, with gene
activation per se. In recent years, sensitive technologies have been developed that allow the
interrogation of DNA methylation patterns from a small number of cells. The use of these technol­
ogies has greatly improved our knowledge of DNA methylation dynamics and heterogeneity in
embryos and in specific tissues. Combined with genetic analyses, it is increasingly apparent that
regulation of DNA methylation erasure and (re-)establishment varies considerably between
different developmental stages. In this Review, we discuss the mechanisms and functions of DNA
methylation and demethylation in both mice and humans at CpG-rich promoters, gene bodies
and transposable elements. We highlight the dynamic erasure and re-establishment of DNA
methyl­ation in embryonic, germline and somatic cell development. Finally, we provide insights
into DNA methylation gained from studying genetic diseases.

Pericentromeric satellite
The methylation of the fifth carbon of cytosines revealed to introduce toxic 3-methylcytosine lesions into
repeats (5-methylcytosine (5mC)) originated in bacteria and DNA20. The systematic co-​evolution of DNMTs with a
Tandem repeats enriched in was present in the first eukaryote1. As eukaryotic 5mC specific alkylation repair enzyme (ALKBH2) may have
heterochromatin modifications is mostly found in the context of symmetrical CpG allowed eukaryotes to tolerate DNA methylation.
such as DNA methylation and
dinucleo­tides1–3, it was anticipated decades ago that a Nevertheless, mammalian genomes exhibit par-
histone H3 Lys9
trimethylation. mecha­nism was in place to recognize the hemimethylated ticularly high CpG methylation levels; although there
CpG site after DNA replication and faithfully methyl­ are some tissue-​specific differences, 70–80% of CpGs are
ate the daughter strand4; elegant experiments have methylated21. Moreover, DNMT-​deficient mice exhibit
sub­sequently validated this hypothesis5. Early studies severe developmental abnormalities, culminating in
in vitro and in vivo indicated that 5mC was associated early embryonic lethality22,23. Deregulation of DNA
with transcriptional repression6–8. Accordingly, DNA methylation is also a defining feature of virtually all
methylation has since been implicated in the classical cancer types24. In addition to XCI and genomic imprint-
epigenetic pheno­mena of genomic imprinting9–12 and ing, DNA methylation has a major role in repressing
X-​chromosome inactivation (XCI)13,14 (see below). transposons25,26 and germline-​specific genes27. DNA
Despite its ancient origins, DNA cytosine methyla- methylation is also highly enriched in pericentromeric
tion has been lost in several eukaryotic lineages, includ- satellite repeats28 and in the bodies of transcribed genes29,
ing in many animals15,16; common model organisms although the precise function of 5mC in both of these
such as Drosophila melanogaster, Caenorhabditis ele- contexts is unclear.
gans, fission yeasts and bakers’ yeasts exhibit virtually Remarkably, the mammalian genome undergoes two
Genetics and Developmental no 5mC. In fact, cytosine methylation comes at a cost: extensive waves of reprogramming of CpG methylation
Biology Department, Institut 5mC is inherently mutagenic because it can spontane- patterns during embryogenesis — following fertilization
Curie, Paris Sciences Lettres ously undergo deamination, leading to C → T transi- and after germline cell specification30,31. Much progress
University, INSERM, CNRS,
Paris, France.
tions17. Thus, organisms with CpG methylation also have has been made recently in understanding the genetic
reduced CpG content18,19. For example, mammals have requirements for these epigenome reprogramming pro-
*e-​mail: deborah.bourchis@
curie.fr roughly 5-fold fewer CpG dinucleotides than expected cesses. Genome-​wide methods with base-​pair resolution
https://doi.org/10.1038/ from the nucleotide composition of their genome. have been developed to elucidate the nuances of DNA
s41580-019-0159-6 Furthermore, DNA methyltransferases (DNMTs) were methylation dynamics during embryonic development,

590 | OCTOBER 2019 | volume 20 www.nature.com/nrm

 M E T H Y L AT I O N I N B I O L O G Y

and precision epigenome editing tools are increasingly is maintained upon DNA replication. This depends on
being used to ascertain the function of DNA methylation the activity of the methylation maintenance enzyme
on a locus-​specific basis. DNMT1 in concert with another multidomain protein,
In this Review, we discuss the latest major advances E3 ubiquitin-​protein ligase UHRF1 (Table 1). UHRF1
in our understanding of the functions of DNA methyl­ specifically binds hemimethylated CpG dinucleo­
ation and its establishment, maintenance and erasure tides at replication forks through its SET- and RING-
during mammalian development. We also discuss new associated (SRA) domain 45,46, and H3K9me2 and
insights gained into the patterning of DNA methylation H3K9me3 through its tandem TUDOR-PHD (TTD-
during development. Finally, we discuss the most cur- PHD) domain47–50 (Fig. 1c). By itself, DNMT1 exists in
rent views of the diverse functions of DNA methylation an auto-​inhibitory configuration, as its repli­cation foci
in genetic diseases. targeting sequence (RFTS) is buried in the catalytic
MTase domain, akin to the ADD domain of the DNMT3
Cellular functions of DNA methylation enzymes51–53 (Fig. 1c). UHRF1 recruits DNMT1 through
Cytosine methylation is pervasive throughout mammal­ its ubiquitin-​like (UBL) domain, thereby releasing its
ian genomes, but likely carries out distinct functions in auto-​inhibition and allowing RFTS binding to his-
different genomic regions. Moreover, even when DNA tone H3 tails that were previously ubiquitylated by the
methylation is associated with transcription silencing, RING finger domain of UHRF1 (refs54,55). DNMT1 then
the underlying mechanisms are not necessarily identical methyl­ates the daughter DNA strand53. Accordingly,
at gene promoters, gene bodies or repeated sequences. mouse ESCs expressing a mutated Uhrf1 recapitulate the
In this section, we discuss the latest evidence for the DNA methylation phenotype of the Dnmt1 mutant45,46.
genomic effects and functional mechanisms of DNA Active DNA demethylation is carried out by the
methylation. TET methylcytosine dioxygenases, which progres­
sively oxidize 5mC to 5-hydroxymethylcytosine
The writers and erasers of methylation (5hmC), 5-formyl­c ytosine (5fC) and 5-carboxylcy-
There are three phases of DNA methylation: estab- tosine (5caC)56–60 (Table 1). All the oxidized forms can
lishment (de novo DNA methylation), maintenance promote DNA demethyl­ation during replication61,62;
and demethylation. In mammals, there are two major in the case of 5fC and 5caC, demethylation can also
de novo DNA methylation enzymes, DNMT3A and occur through base removal by thymine DNA gly-
DNMT3B32,33, which contain a highly conserved DNMT cosylase (TDG) followed by the activity of the base
domain (the MTase domain) in the carboxy terminus excision repair pathway60,63,64. A more detailed descrip-
and two chromatin reading domains, ATRX-​DNMT3- tion of DNA methylation and demethylation mech-
DNMT3L (ADD) and PWWP (Fig. 1; Table 1). There is anisms can be found in Supplementary Box 1 and
also a catalytically inactive DNMT, DNMT3L, which Supplementary Figure 1.
interacts with and stimulates the activity of DNMT3A
and DNMT3B specifically in the germline34,35 (Table 1). DNA methylation represses transcription
DNA methylation is usually excluded from CpG-​rich The repressive role of DNA methylation in transcription
promoters of actively transcribed genes, which are has long been recognized with a correlation between
typically enriched in trimethylated histone H3 Lys4 DNA methylation and gene silencing that increases
(H3K4me3)36. The ADD domain, which binds to the with the density of CpG dinucleotides at promoters65.
K4 residue of H3 tails, is repelled by increasing num- However, how this leads to transcription inhibition is
bers of methyl moieties at K4, with H3K4me3 being the still not entirely resolved, as the methyl mark per se does
most repelling of ADD domains34,37,38 (Fig. 1a). When not not seem to confer silencing. Regions of accessible chro-
binding H3K4, the ADD binds the MTase domain and matin are frequently lowly methylated or unmethylated,
auto-​inhibits the activity of the DNMT3 enzymes; bind- indicating that binding of transcription factors and DNA
ing of the ADD domain to unmethylated H3K4 releases methylation are mutually exclusive66. Certain transcrip-
the MTase domain and enables DNA methylation39 tion factors are sensitive to CpG methylation: a recent
(Fig. 1a,b). In contrast to active promoters, the bodies survey of 542 human transcription factors found that
of actively transcribed genes are enriched with DNA 117 (22%) exhibited decreased binding to their motifs
methylation1,2,29. As RNA polymerase II (Pol II) tran- when methylated compared with unmethylated67. By
scribes, the histone methyltransferase SETD2 con- preventing the binding of such transcription factors,
comitantly trimethylates H3K36 (refs40,41); the PWWP DNA methylation can therefore impede transcription
domain binds to H3K36me3 in vitro42, which strongly activation of CGI promoters that contain their sequence-​
suggests there is a mechanistic link between tran­ recognition motifs. Using the same principle, methyl-
scription and DNA methylation at gene bodies (Fig. 1b). ated cytosines can also serve as binding modules for
Indeed, a DNMT3B protein with a PWWP mutation transcription activators (Box 1).
loses affinity for H3K36me3-marked gene bodies in DNA methylation can also contribute to heterochro-
mouse embryonic stem cells (ESCs), and DNMT3B is matin formation through the recruitment to chroma-
lost from gene bodies in Setd2 mutants43. Additionally, tin of chromatin remodellers and modifiers by DNMT
mouse Setd2-mutant oocytes exhibit grossly deregulated proteins: de novo DNMTs function in complex with
targeting of DNA methylation44. the chromatin remodeller lymphocyte-​specific helicase
Whereas de novo DNA methylation can occur in any (LSH) and with H3K9 methyltransferases and histone
sequence context, only symmetrical CpG methylation deacetylases68–75. Protein recruitment can also occur

NATuRE REvIEWS | MolECulAr CEll BIoloGy volume 20 | OCTOBER 2019 | 591

REVIEWS

MTase

a ADD DNMT3A, B b

P
W
H3K4me3 PW K4 ADD K4 ADD

PW
W
P H3K36me3
MTase MTase

P
PWW
Active or poised Inactive Gene body
unmethylated promoter methylated promoter

Replisome

RING RFTS
Ub
SRA
H3K9me2 TTD K9 MTase
RING
UHRF1
UBL SRA
TTD
MTase

RFTS
DNMT1
UBL

CpG 5mCpG

Fig. 1 | DNA methylation machinery and mechanisms. a | Mechanism of DNA methylation at promoters. Left:
trimethylated histone H3 Lys4 (H3K4me3), which marks active and poised promoters, prevents binding to chromatin of the
ATRX-​DNMT3-DNMT3L (ADD) domain of DNA (cytosine-5)-methyltransferase 3A (DNMT3A) and DNMT3B (and also of
DNMT3L), thereby causing it to bind to the methyltransferase (MTase) domain and auto-​inhibit the DNMT3 enzymes.
Right: in the absence of H3K4 methylation, the ADD domain binds to H3K4 and the auto-​inhibition is relieved, thereby
allowing the MTase domain to methylate the DNA. b | DNA methylation at gene bodies. In gene bodies, the ADD domain
binds unmethylated H3K4, thereby releasing the auto-​inhibition of the DNMT3 enzymes. H3K36me3 is deposited in gene
bodies of actively transcribed genes and serves as a recruitment module for the DNMT3 PWWP domain. c | Maintenance
of DNA methylation. Left: E3 ubiquitin-​protein ligase UHRF1 is recruited to replicating DNA through its SET- and RING-​
associated (SRA) domain, which binds hemimethylated CpG sites, and through its TUDOR (TTD) domain, which binds
H3K9me2. The RING domain of UHRF1 ubiquitylates the histone H3 tails (Ub). The replication foci targeting sequence
(RFTS) of DNMT1 folds into the MTase domain, thereby preventing its catalytic activity. UHRF1 recruits DNMT1 through an
interaction between its ubiquitin-​like (UBL) domain and the DNMT1 RFTS. Right: the auto-​inhibition of DNMT1 is released
when the RFTS binds to ubiquitylated H3 tails, which enables the maintenance of symmetrical DNA methylation at CpG
sites. Some domains of UHRF1 and DNMT1 were omitted from the figure for simplicity. 5m, methylation of the fifth carbon.

through 5mC and the readers, methyl-​CpG-binding CpG-​island promoters

domain (MBD) proteins76,77. Mammals have five MBD The mammalian genome is generally CpG poor, with
proteins: MBD1–MBD4 and methyl-​CpG-binding the exception of CpG islands (CGIs), which are rela-
protein 2 (MeCP2). Four of the MBD proteins exhibit a tively small genomic regions of roughly 1 kb on average.
linear relationship between an increase in CpG binding Over two-​thirds of mammalian promoters are CGIs83,84:
and increased CpG methylation78, but MBD3 does not virtually all housekeeping genes have CGI promoters,
have preference for methylated cytosines79. All MBDs and so do several developmentally regulated genes85.
interact with nucleosome remodelling and histone CGIs are very rarely methylated86; they are particu-
deacetylase complexes, which leads to gene silencing80,81. larly unmethylated in the dividing male germline cells,
It should be noted that genetic evidence in vivo for MBD explaining why they are not subject to CpG erosion by
function is complicated by redundancy between the deamination during evolution and their remarkably
proteins. Finally, there are several zinc finger proteins high CpG content. Most inactive CGI promoters are
that recog­nize and bind DNA methylated sequences silenced by Polycomb repressive complex 2-mediated
(reviewed elsewhere82). Such factors could contribute to H3K27 methylation, which is a more plastic mode
DNA methylation-​based silencing independently of, or of silencing than DNA methylation and therefore
redundantly with, MBDs. more amenable to gene (re)activation in response to

592 | OCTOBER 2019 | volume 20 www.nature.com/nrm

 M E T H Y L AT I O N I N B I O L O G Y

Table 1 | DNA methylation and demethylation factors and their functions

Factor Function Mouse loss-​of-function Human diseases associated with genetic
phenotype mutations
DNMT1 Maintenance DNA • Low global DNA methylation • Hereditary sensory autonomic neuropathy 1E
methyltransferase • Derepression of IAP transposons (HSAN1E; OMIM 614116)
• Early embryonic lethality • Autosomal-​dominant cerebellar ataxia,
deafness and narcolepsy (ADAC-​DN; OMIM
604121)
UHRF1 DNMT1 cofactor • Low global DNA methylation
• Early embryonic lethality
DNMT3A De novo DNA • Constitutive knockouts die • Microcephalic dwarfism
methyltransferase ~4 weeks after birtha • Tatton-​Brown–Rahman syndrome (TBRS;
• Sterility in both males and females OMIM 602729)
in germline-​specific knockouts • Acute myeloid leukaemia (AML; OMIM 601626)
DNMT3B De novo DNA Constitutive knockouts die mid-​ Immunodeficiency, centromeric instability and
methyltransferase gestationa. More important for facial anomalies syndrome (ICF; OMIM 602900)
embryonic DNA methylation than
for germline DNA methylation
DNMT3C De novo DNA Males are infertile likely owing to
methyltransferase defect in methylating transposon
(Muroidea specific) promoters during spermatogenesis
DNMT3L De novo DNA • Male germline cells unable to
methyltransferase undergo meiosis
cofactor • Females unable to establish
maternal imprinting, leading to
mid-​gestation lethality of progeny
TET1 DNA demethylation Loss has subtle effects and the
via oxidation of embryos are viableb,c
methylcytosine
TET2 DNA demethylation Increased self-​renewal of • AML (OMIM 601626)
via oxidation of haematopoetic stem cellsb,c • Chronic myelomonocytic leukaemia
methylcytosine • Lymphomas
• Myeloproliferative neoplasms
TET3 DNA demethylation Germline conditional knockout leads
via oxidation of to impaired paternal demethylation
methylcytosine and reduced fecundityc
DNMT, DNA (cytosine-5)-methyltransferase; IAP, intracisternal A particle. aDnmt3a–/–;Dnmt3b–/– double mutants exhibit early
embryonic lethality; bTet1–/–;Tet2–/– double mutants display a range of developmental defects and partial lethality; cTet1–/–;Tet2–/–;Tet3–/–
triple mutant embryos exhibit gastrulation failure.

developmental or environmental cues (reviewed else- of DNMT3B recruitment during XCI remains unclear,
where87). Nevertheless, there are three major classes of and so does the reason why SMCHD1 evolved such an
genes, in which stable, lifelong DNA-​methylation-based important XCI-​specific function. Perhaps the hierar-
silencing in somatic tissues is very important: genes chical and highly regulated heterochromatization of the
on the inactive X chromosome, imprinted genes and X chromosome provides a unique chromatin environment
germline-​specific genes. Here, we discuss the processes that facilitates DNA methylation of X-linked CGIs.
by which DNA methylation can be targeted to these
specific classes of CGI promoters. Genomic imprinting. In genomic imprinting, DNA
methylation is differentially established in the two
X-​chromosome inactivation. In female mammals, one parental germlines; these epigenetically imprinted pat-
X chromosome in each cell is randomly silenced by the terns withstand the genomic reprogramming of DNA
activity in cis of the non-​coding RNA X-​inactive specific methylation that takes place during early embryogen-
transcript (XIST). In this process of XCI, DNA methyl- esis. In both mouse and human, around 20 genomic
ation of X-​linked CGIs appears to occur relatively late regions known as imprinting control regions (ICRs)
and to function as a final lock added after the genes withstand this reprogramming and force mono-​allelic
have already been silenced14,88–90. In mice, X-​linked CGI-​ expression of neighbouring genes 95–101. The major-
promoter silencing mostly depends on DNMT3B — the ity of ICRs are methylated in the oocyte and these all
other DNMT3s are dispensable91 — and in a subset of coincide with extremely CpG-​rich CGIs101, whereas
X-linked CGIs it also requires structural maintenance the three paternal ICRs map to CpG-​poor intergenic
of chromosomes flexible hinge domain-​containing 1 sequences. During oocyte growth, DNMT3A assisted
(SMCHD1)91–93 (Fig. 2a). In humans, SMCHD1 is also by DNMT3L methyl­ates oocyte-​expressed gene bod-
involved in XCI, but its link with DNA methylation ies, including their intragenic CGIs in a transcription-​
has not been demonstrated94. The precise mechanism dependent manner, whereas the rest of the genome

NATuRE REvIEWS | MolECulAr CEll BIoloGy volume 20 | OCTOBER 2019 | 593

REVIEWS

Retrotransposons remains hypomethylated35,102–106 (Fig. 2b). Mammalian zinc-​finger protein 57 (ZFP57)110–112 (Fig. 2b). ZFP57
Transposable elements that oocyte transcription typically emanates from alternative recruits KRAB-​a ssociated protein 1 (KAP1; also
propagate in the genome promoters that are located upstream from canonical CGI known as TIF1β) and other silencing factors, includ-
through RNA intermediates promoters (which are used after fertilization) and often ing DNMTs111,113,114. This selective DNA binding of
and reverse transcription.
coincide with sequences of retrotransposons106–109. ZFP57–KAP1 in oocytes allows ICRs to maintain
What distinguishes maternal (and also paternal) allele-​specific methylation in post-​fertilization embryos,
ICRs from other sequences that are methylated in the which undergo global DNA methylation erasure and
gametes is their enrichment in specific genetic motifs re-​establishment. It should be noted that mutations in
(TGCCGC), which, when methylated, are recognized Zfp57 are not fully penetrant in mice: a recent study
by the Krüppel-​associated box (KRAB)-containing demonstrated that a similar protein, ZFP445, cooper-
ates with ZFP57 at nearly all ICRs to maintain methyl-
ation imprints in mice, and perhaps has an even more
Box 1 | DNA methylation-mediated transcription activation important role than ZFP57 in humans115. In summary,
the ability of maternally imprinted CGIs to undergo
DNA methylation can be read by methyl CpG binding domain (MBD)-containing
transcriptional repressors (see the figure, part a), but methylated DNA motifs can also DNA methylation relies on a combination of the unusual
be specifically recognized by transcription activators. Various methods, including transcriptional landscape of the oocyte and on specific
stable isotope labelling with amino acids238, microarrays246 and systematic evolution genetic sequences.
of ligands by exponential enrichment67 — all of which are designed to discriminate
between methylated and unmethylated DNA motifs recognized by DNA binding factors Germline-​specific genes. CGI promoters of germline-​
— collectively revealed that humans and mice express dozens of transcription specific genes are silenced by DNMT3B-​mediated
factors with binding preferences of specific methylated sequences. These include the DNA methylation with the onset of somatic differenti-
cell pluripotency factors KLF4 (refs238,246–248) and OCT4 (also known as POU5F1)67, ation during embryo implantation27,116. Germline genes
the homeobox proteins HOXB13 (ref.67) and the NKX neural patterning factors67. C/EBPα
are acut­ely sensitive to loss of DNA methylation and are
also uses a methyl-specific binding motif, which is important for keratinocyte differenti­
derepressed in Dnmt triple-​knockout mouse ESCs117,
ation249. It is intriguing that several transcription factors (TFs) involved in cell-type
transitions exhibit methylcytosine binding specificity and can function at chromatin DNMT1-depleted human fibroblasts118, Dnmt3B-​mutant
that is otherwise refractory to transcription activation in order to facilitate these mouse embryos27,119 and human diseases linked with
transitions. A recent study showed that C/EBPα and KLF4 (and another transcription DNMT3B mutations120. What makes CGI promoters
factor) recruit the methylcytosine dioxygenase TET2 to enhancers for demethylation of germline genes, which account for only 5% of the
during cell-type reprogramming201 (see the figure, part a). total number of CGIs, dependent on DNA methyla-
DNA methylation can have a counterintuitive role in activating genes. The genomic tion, whereas the great majority of autosomal CpG-​rich
distribution of DNA methylation is largely mutually exclusive from that of histone H3 promoters remain unmethylated? The key may be a
Lys27 trimethylation (H3K27me3), which is a gene-repressive histone modification non-​canonical Polycomb repressive complex 1 (PRC1)
that is catalysed by Polycomb repressive complex 2 (PRC2)250–253. Whereas gene
known as PRC1.6, which is a regulator of germline-​gene
silencing is usually preserved when replacing one repressive modification by another,
repression in mouse ESCs121,122. The complex contains
displacement of PRC2 by DNA methylation and, consequently, loss of H3K27me3
occasionally correlates with gene activation, in normal physiological conditions DNA binding proteins123 that provide sequence-​specific
and in cancer223,254. During neurogenesis, DNA (cytosine-5)-methyltransferase 3A targeting, such as MAX 121,124, MGA 121,122 and E2F6
(DNMT3A)-mediated de novo DNA methylation evicts PRC2 and, consequently, (ref.125). Furthermore, PRC1.6 associates with L3MBTL2
H3K27me3 from the regulatory regions of neural genes208,228 (see the figure, part b). (refs 126,127) , which interacts with the H3K9 methyl-
The human FOXA2 gene appears to be regulated in a similar fashion during endoderm transferase G9A (also known as EHMT2)126. Mouse
development255. Moreover, during mouse embryonic de novo DNA methylation, the G9a-​mutant embryos fail to acquire DNA methyla-
imprinted Zdbf2 gene is activated by DNA methylation upstream of its promoter, tion specifically at a large subset of germline-​specific
which disrupts H3K27 trimethylation183. In the absence of the Polycomb-to-DNA genes119 (Fig. 2c). Many aspects of the role of PRC1.6 in
methylation switch, Zdbf2 remains silenced throughout life, resulting in reduced
repressing germline-​specific genes remain unclear, such
body size.
as the means by which CGIs of germline genes are tar-
a b H3K27me3 geted by DNMT3B but not by DNMT3A. Nevertheless,
MBD
contrary to most CGI promoters, which resist de novo
PRC2 DNA methylation at implantation, germline CGIs have
evolved sequence-​specific means of recruiting the DNA
methylation machinery to ensure lifelong somatic
silencing.
TET2
TF Harnessing the repetitive genome
Genome defence against transposable elements has
been proposed as a major driver of the evolution of
Reprogramming DNA methylation128,129. Indeed, the main targets of DNA
methyl­ation in mammalian genomes are not genes but
TF TF transposable elements, and in particular retrotrans­
DNMT3 posons. There are millions of copies of retrotransposons
in the mouse and human genomes, which occupy roughly
CpG 5mCpG 5hmCpG half of the genomic space130. The expression of the most
active retrotransposons is controlled by CpG-​rich pro-
5hm, hydroxymethylation of the fifth carbon; 5m, methylation of the fifth carbon. moters, and DNA methylation is important for their

594 | OCTOBER 2019 | volume 20 www.nature.com/nrm

 M E T H Y L AT I O N I N B I O L O G Y

a X-chromosome inactivation b Maternal genomic imprinting

Growing oocyte
Imprinted
Oocyte-specific Canonical
promoter
promoter promoter
DNMT3B MaLR TGCmCGC
SMCHD1

SMCHD1 DNMT3A
DNMT3L

UHRF1 SETDB1
KAP1
Blastocyst DNMT1 ZFP57

TGCmCGC

Active Constitutively
inactive
c Germline-specific gene promoters
PRC1.6

L3MBTL2
PCGF6
GLP RING1A/B

G9a E2F6 RYBP

H3K9me2
MGA
MAX

DNMT3B
CpG 5mCpG

Fig. 2 | Targeting DNA methylation to CGI promoters. a | Structural maintenance of chromosome flexible hinge
domain-​containing 1 (SMCHD1)-dependent de novo DNA methylation at CpG islands (CGIs) during X-​chromosome
inactivation. SMCHD1 forms homodimers and may be involved in condensing the chromatin of the X chromosome
that is undergoing inactivation. At a subset of CGIs, SMCHD1 is required for DNA (cytosine-5)-methyltransferase 3B
(DNMT3B)-mediated DNA methylation through an unclear mechanism. b | Establishment of maternal imprinting. In the
growing oocyte, DNA methylation is strongly correlated with transcription elongation in a DNMT3A–DNMT3L-​dependent
manner. In mice, many of these transcripts arise from mammalian apparent long terminal repeat retrotransposons (MaLR).
After fertilization, imprinted promoters withstand the early embryonic global DNA methylation erasure through the
binding to methylated TGCCGC motifs of the zinc-​finger protein 57 (ZFP57)–KRAB-​associated protein 1 (KAP1) complex,
which recruits the histone methyltransferase SETDB1 and the maintenance DNA methylation factors DNMT1 and UHRF1,
thereby maintaining gene silencing. c | Model for the establishment of DNA methylation at CGI promoters of germline-​
specific genes. A subset of germline genes is bound by the non-​canonical Polycomb repressive complex 1 (PRC1), PRC1.6.
The complex also consists of L3MBTL2, which can recruit the heterodimeric H3K9 methyltransferase complex, G9a–GLP.
G9a is required for DNMT3B-​mediated deposition of DNA methylation specifically at germline genes. 5m, methylation of
the fifth carbon.

transcriptional silencing. This is exemplified in mouse its ADD and its de novo methylation activity. Unlike
Dnmt1 knockout embryos, where intracisternal A par- DNMT3A and DNMT3B, which are broadly expressed in
ticle (IAP) retrotransposons, which constitute a class various developmental contexts in both sexes, DNMT3C
of evolutionarily young retrotransposons that can still expression is confined to male fetal germ cells. Although
actively mobilize in rodent genomes, are massively homozygous Dnmt3C-​mutant mice are viable, males have
derepressed131 (Fig. 3a). small testes and are infertile (Fig. 3a).
Two independent mouse genetic screens have recently Whole-​genome methylation analysis revealed that
discovered DNMT3C — a previously unknown de novo DNMT3C selectively methylates and represses the pro-
DNMT with a specific role in controlling retrotrans- moters of evolutionarily young transposable elements,
posons129,132 (Table 1). Originally annotated as a pseudo­ which account for only 1% of the mouse genome129.
gene, Dnmt3C originates from a tandem duplication of The molecular phenotype of the Dnmt3C mutation is
the Dnmt3B gene in the Muroidea lineage — the rodent identical to that of mutants of the germline methylation
superfamily that contains mice and rats. During evolu- cofactor DNMT3L35 and also to that of mutants of the
tion, DNMT3C lost its PWWP domain, but maintained piwi-​interacting RNA pathway; piwi-​interacting RNAs

NATuRE REvIEWS | MolECulAr CEll BIoloGy volume 20 | OCTOBER 2019 | 595

REVIEWS

constitute a highly conserved class of small RNAs spe- is the tip of the small-​RNA-directed DNA methylation
cifically dedicated to the silencing of transposable ele- spear (Fig. 3b). Given the sterility of the mouse Dnmt3C
ments in the germline133–139. Thus, although mechanistic mutant, one wonders how the human male germline
confirmation awaits, it is highly likely that DNMT3C copes with transposable elements given that the human

a Wild-type embryo (E9.5) Wild-type testis b

Transposon MIWI2 piRNA
RNA

K4 DNMT3L H3K4me3
?
DNMT3C

Pol II

De novo methylation of
IAP transposons IAP and LINE1 transposon promoters
transposons

Dnmt1–/– embryo Dnmt3C–/– or

Dnmt3L–/– testis
• Developmental
delay • Meiotic failure
• Lethality • Infertility

IAP transposons IAP and LINE1

transposons
d
Mouse Dnmt1–/– ESCs
Wild-type placenta
c High DNA methylation Low DNA methylation SETDB1
Category A
(e.g. LINE1, MMERGLN) K9
H3K9me3
UHRF1 UHRF1 UHRF1
K9 Transposon K9 K27

IAP
Category B (e.g. IAP)
Mouse Uhrf1–/– ESCs
and placenta
K9 K9

Category C (e.g. MERVL)

K9 K27

Regulatory Gene
region body CpG 5mCpG

Fig. 3 | DNA methylation-​based regulation of transposons. a | In vivo consequences of DNA methylation loss at
transposons. In the mouse embryo, loss of DNA (cytosine-5)-methyltransferase 1 (DNMT1) leads to drastic upregulation of
intracisternal A particle (IAP) retrotransposons. This is associated with severe developmental defects and prenatal lethality
in embryonic day 9.5 (E9.5). Dnmt3C or Dnmt3L mutants have small testes and exhibit specific loss of DNA methylation at
IAP and at long interspersed nuclear element 1 (LINE1) retrotransposons in the male germline, which leads to meiotic
catastrophe and infertility. b | Model of piwi-​interacting RNA (piRNA)-directed DNA methylation in mice. piRNAs generated
from transposon transcripts that are loaded onto MIWI2 (also known as PIWIL4) localize to the nucleus and recruit
DNMT3C–DNMT3L through an unknown mechanism. As DNMT3C contains a chromatin-​reading ATRX-​DNMT3-DNMT3L
(ADD) domain, there likely is a mechanism to circumvent the gene-​activating effects of trimethylated histone H3 Lys4
(H3K4me3), such as its removal by a lysine demethylase. c | Compensation for loss of DNA methylation in mouse embryonic
stem cells (ESCs). During conversion from culturing mouse ESCs in media supporting high global DNA methylation to
media supporting low global DNA methylation, transposons can be generally divided into three classes. In class A, H3K9
methylation remains unperturbed, but H3K27me3 invades the transposon body; in class B, H3K9 methylation remains
unchanged; in class C, there is an epigenetic switch from H3K9-methylation-​based silencing to H3K27me3-based silencing.
d | UHRF1 prevents SETDB1-mediated silencing of IAPs at hemimethylated DNA. In Dnmt1 conditional knockouts, UHRF1
remains bound to post-​replication hemimethylated DNA and prevents the H3K9 methyltransferase SETDB1 from silencing
IAP transposons. This is physiologically relevant in wild-​type mouse placenta, where IAP is derepressed owing to UHRF1
binding to hemimethylated DNA. In Uhrf1 mutant ESCs and placenta, SETDB1 catalyses the trimethylation of H3K9 and
consequently IAPs are repressed. 5m, methylation of the fifth carbon; MERVL, mouse endogenous retrovirus with leucine
tRNA primer; MMERGLN, mouse endogenous retrovirus with glutamine tRNA primer; Pol II, RNA polymerase II.

596 | OCTOBER 2019 | volume 20 www.nature.com/nrm

 M E T H Y L AT I O N I N B I O L O G Y

genome lacks the DNMT3C gene. Perhaps human cytoplasm and cannot counteract SETDB1-mediated
PWWP-​less DNMT3B isoforms perform a similar repression through H3K9 methylation146.
function to rodent DNMT3C; alternatively, small-​RNA-
directed DNA methylation may simply not exist in the The puzzling case of gene bodies
same manner in humans, and instead we have evolved The enrichment of DNA methylation in gene bodies pre-
a different mechanism for controlling transposable sents a paradox: on the one hand, gene-​body methyla-
elements. tion is highly conserved across eukaryotes — more than
In addition to the short-​term effects of repressing it is conserved at transposable elements, for example —
transposable elements, DNA methylation also pro- indicating it has an important function1,2. On the other
motes their irreversible genetic inactivation through hand, DNA methylation is mutagenic, so why is it so
mutagenic deamination. Finally, DNA methylation prominent in coding sequences? Importantly, gene-​body
may support genome stability by limiting recombina- DNA methylation is positively correlated with transcrip-
tion between non-​allelic copies of transposable elements tion29,148,149, hence, its role is not linked to gene silencing.
with high sequence similarity. This possibility is sup- Two hypotheses have been proposed for the function of
ported by positive correlations between elevated levels of DNA methylation in gene bodies: that it facilitates tran-
chromosomal rearrangements and genome-​wide DNA scription elongation and/or co-​transcriptional splicing,
hypomethylation — mostly involving transposable and that it represses intragenic cryptic promoters.
elements— in various human cancers140. Also, during DNA methylation is enriched at exons relative to
meiosis, DNA methylation may limit transposable ele- introns29,150 and may affect the processivity of Pol II and,
ments from engaging in homology-​dependent search through this, splicing. Nucleosomes are also enriched
and recombination141. It should be mentioned that at exons, and de novo DNA methyltransferase activ-
DNA methylation may exert genomic stability protec- ity requires histone binding, which would explain the
tive functions not only through targeting interspersed enrichment of DNA methylation at exons. Several
transposon repeats, but also through targeting tandem studies indicate that DNA methylation affects splicing
satellite repeats found in telomeric, centromeric and and gene expression, not vice versa151–153. At the human
pericentromeric regions (reviewed elsewhere142). CD45 gene, CCCTC-​binding factor (CTCF) slows
In striking contrast to in vivo situations, transposable Pol II elongation rates at exon 5, thereby facilitating its
elements are not reactivated in mouse ESC models of inclusion; DNA methylation prevents CTCF binding,
constitutive DNA methylation deficiency, except IAPs, thereby leading to exon exclusion154. By contrast, at other
and their upregulation is only modest in Dnmt1 knock- loci, DNA methylation can facilitate exon inclusion
out or Dnmt triple-​knockout cells117,143,144. However, by by recrui­ting MeCP2 (ref.155), in line with the obser-
conditionally triggering the ablation of DNA methyla- vation that alternative exons exhibit lower levels of
tion, independent studies have recently demonstrated DNA methylation on average than constitutive ones151.
that several classes of transposable elements — including Additionally, heterochromatin protein 1 can regulate
IAPs — are indeed reactivated upon withdrawal of DNA exon inclusion by recruiting splicing factors to H3K9me3-
methylation145–147. Following this initial response, histone modified nucleosomes wrapped in methylated DNA156.
methylation restores long-​term transcriptional silenc- However, these mechanisms only account for a fraction
ing in hypomethylated genomes. Depending on the of alternative splicing events.
class of transposable element, silencing involves depo- The second hypothesis — that DNA methylation
sition of H3K9me3, H3K27me3 or a combination of the inhibits intragenic promoters — is attractive for numer-
two145 (Fig. 3c). ous reasons. First, it is consistent with DNA methylation
Paradoxically, the initial burst of IAP reactivation being a transcriptional repressor. Moreover, as discussed
that follows conditional Dnmt1 deletion is not observed above (Fig. 1b), the de novo DNA methylation machin-
upon removal of its DNA-​targeting cofactor UHRF1 ery is recruited to DNA through binding to H3K36me3,
(ref.146). The basis of this discrepancy has been geneti- which is known to prevent the use of cryptic promot-
cally dissected: suppression of post-​replicative mainte- ers in baker’s yeast157. Indeed, methylation of intra-
nance of DNA methylation in Dnmt1 mutants initially genic CGIs prevents promoter activity, and differential
generates an excess of hemimethylated DNA, which methylation can regulate transcription initiation in a
transiently prolongs UHRF1 residency on DNA, thereby tissue-​specific and cell-​type-specific manner in mam-
inhibiting the recruitment of the histone methyltrans- mals158. However, intragenic CGIs are often conserved
ferase histone-​lysine N-​methyltransferase SETDB1, its from mouse to human, and are more likely alternative
catalysis of H3K9 trimethylation and silencing of IAPs promoters rather than illegitimate, cryptic promoters.
(Fig. 3d). Following longer culture and progressive dilu- A recent study in mouse ESCs showed that DNMT3B-​
tion of hemimethylated DNA, UHRF1-based inhibition mediated gene-​body methylation restricts the activity of
of histone methylation is relieved and SETDB1-mediated cryptic promoters downstream of H3K36me3 (ref.159).
silencing is enabled. Interestingly, accumulation of Although the finding is appealing, the observed effect,
hemimethylated DNA naturally happens in vivo, in although statistically significant, occurs in a very small
specific developmental contexts146. Whether this leads to proportion of cells within a cell population. Moreover,
IAP reactivation depends on the nuclear availability of Dnmt triple-​k nockout ESCs do not exhibit inhibi-
UHRF1, for example in the mouse placenta. This does tion of intragenic cryptic promoters as the Dnmt3B
not occur in the early embryo or in migrating primordial single-​mutant cells do117,160. More functional analyses
germ cells (PGCs), where UHRF1 is sequestered in the are needed to confirm the role of DNA methylation in

NATuRE REvIEWS | MolECulAr CEll BIoloGy volume 20 | OCTOBER 2019 | 597

REVIEWS

repressing cryptic promoters, particularly in cell types dilution of DNA methylation, as the maintenance
other than mouse ESCs, in which DNA methylation is enzyme DNMT1 provided by the oocyte is excluded
not required for cell viability and other safeguards may from the nucleus during subsequent cell divisions170
be in place in gene bodies. (Fig. 4a) . This model was challenged by three recent
In summary, both of the above hypotheses are logical whole-​genome bisulfite sequencing analyses of embryos
and sound, and not necessarily mutually exclusive. In from hybrid crosses between two mouse strains (to infer
both cases, DNA methylation appears to be at most a parental specificity with available single-​nucleotide poly­
fine-​tuner, as loss of gene-​body methylation does not morphism) in conjunction with genome-​wide 5hmC
lead to drastic molecular phenotypes. Future work may mapping, which revealed that the maternal methylome
reveal hidden mechanisms and/or functions that will also undergoes limited TET3-dependent oxidation in
help explain the highly conserved prevalence of DNA the zygote171–173. Other studies also indicated that most
methylation in gene bodies. zygotic DNA methylation is lost independently of repli-
cation174 and that, at the paternal genome, this may occur
Methylation patterning in development in a TET3-independent manner through an unclear
The discovery of DNA methylation reprogramming mechanism174,175. Although human data are more limited,
in mammalian development was made over three they indicate that generally similar dynamics take place,
decades ago. However, it was not until the advent of with some species-​specific differences. An initial rapid
whole-​genome bisulfite sequencing that DNA methyla- demethylation also primarily affects the paternal genome
tion patterns could be assessed with single-​nucleotide of human embryos, but it occurs from fertilization to
resolution, and, more recently, using a small number of the two-​cell stage, whereas it is completed at the one-​cell
cells (Supplementary Box 2). Subsequently, techniques stage in the mouse176–178. There is then a gradual genome-​
have been developed to assess the oxidized forms of 5mC wide loss of DNA methylation until the blastocyst stage,
with the same precision. In this section, we describe the although the maternal genome retains higher DNA
recent nuanced findings and revised insights these tech- methylation levels in humans compared with mice179.
niques have provided into the role of DNA methylation Single-​c ell whole-​genome bisulfite sequencing
in development. revealed prevalent occurrences of de novo DNA methyl­
ation in the human embryonic genome while it under-
Early embryos and the germlines goes global demethylation: first in the paternal genome at
The traditional view of mammalian epigenetic repro- the one-​cell stage, and then genome-​wide at the eight-​cell
gramming predicates an extensive expunction of DNA stage, coincidently with embryonic genome activation179,
methylation during both early embryonic and germline mainly targeting transposable elements. During mouse
development. This may be required for acquiring epi­ development, some de novo DNA methylation may also
genetic plasticity at important developmental stages, but occur concurrently with global demethylation of the
also for limiting the inheritance of acquired epimuta- embryonic or the germline genomes. Indeed, 5mC is lost
Whole-​genome bisulfite tions. However, the mechanisms of DNA methylation before 5hmC is gained in the paternal zygotic genome174;
sequencing reprogramming in early embryogenesis and in germline therefore, de novo DNA methylation must occur in this
Sodium bisulfite treatment
development have key differences, although in both time window to provide 5mC for TET enzymes to oxi-
converts unmodified cytosines
— but not (hydroxy)methylated cases it is now clear that the process is neither complete dize. Similarly, during the active germline demethylation,
cytosines — into uracils nor totally linear: a limited number of sequences resist TET1 may help to maintain the hypomethylated state
(thymines following PCR). DNA methylation loss, and de novo DNA methylation against spurious events of de novo DNA methylation180.
Paired with next-​generation may be occurring during this process. In summary, during global DNA methylation erasure
sequencing, this technique
generates genome-​wide, single-​
both in embryos and in the germline, the major role of
nucleotide resolution maps of Global passive and active demethylation and local TET may be protection from ectopic DNA methylation
DNA methylation. de novo methylation. Genome-​wide DNA demethyla- rather than driving active demethylation per se.
tion has been a contentious area of research. In germline
Intergenerational epigenetic
reprogramming, a two-​step demethylation accompanies Methylation reprogramming and the biological implica-
inheritance
Epigenetic information that is the acquisition of germ cell identity in PGCs161,162. The tions of resisting it. Although embryonic and germline
inherited from the parents (for first phase consists of passive demethylation of the bulk demethylation is global, a substantial amount of DNA
example, genomic imprinting). of the genome as a default route163–165. This is followed methylation persists at the end of both processes. This
by TET1-dependent and TET2-dependent demethyla- observation has raised the interesting possibility of the
Transgenerational
epigenetic inheritance
tion, which mostly affects imprinted loci and germline-​ existence of intergenerational epigenetic inheritance and
Epigenetic information that is specific genes (Fig. 4a) and coincides with the appearance even transgenerational epigenetic inheritance (reviewed
inherited from generations that of 5hmC in PGC nuclei in the mouse164–166. elsewhere181).
were not exposed to the initial During post-​f ertilization reprogramming, the In the inner cell mass of preimplantation embryos,
cue that caused the epigenetic
embryo loses gamete-​specific DNA methylation patterns approximately 20% of CpGs retain gamete-​inherited
change.
inherited from the oocyte and the sperm as it progresses methylation in both mice171 and humans179 (Fig. 4a).
Inner cell mass towards pluripotency; this also occurs in two phases. In These notably map to ICRs, as expected from the inter-
Refers to the pluripotent cells mouse one-​cell-stage embryos (zygotes), joint appear- generational nature of genomic imprinting, which is
in the blastocyst of ance of TET3-mediated 5hmC and 5mC loss indicates linked to the sequence-​specific DNA demethylation
preimplantation embryos,
which can be derived and
that the paternal genome is initially actively demethyl- resistance through KRAB–ZFP recruitment of KAP1
cultured as embryonic stem ated by TET3 (refs167–169); the two parental genomes then (refs111,113,114). However, collectively, the ICRs comprise a
cells. undergo rounds of passive, DNA replication-​dependent negligible proportion of the genome. In addition to these

598 | OCTOBER 2019 | volume 20 www.nature.com/nrm

 M E T H Y L AT I O N I N B I O L O G Y

a Somatic tissue
DNMT3A
differentiation TET1
DNMT3B TET2
DNMT3C TET3
DNMT3L

Fertilization Birth
Epiblast
Sperm
100 Mouse stem cells
ZGA
Muroidea
DNMT1 excluded
75
from nucleus
CpG methylation (%)

Human Germline
50
ZGA
Oocyte Epiblast
25
extra-embryonic
tissues PGCs

0 Blastocyst

Mouse: E0.5 E3.5 E6.5 E13.5 Ovulation

Human: E0.5 E6 E14 W7-10 Ovulation

b c d
Transient Lifelong
Pre-implantation imprint imprint Tissue-specfic enhancer Postnatal neurons
Enhancer
Maternal activation
TET2
TF
DNMT3A DNMT3A
Paternal
Enhancer
Somatic decommissioning
Adult neurons
Maternal
DNMT3A
MeCP2 MeCP2 MeCP2
Paternal
Liz Zdbf2

H3K27me3 CpG 5mCpG 5hmCpG mCAC

Fig. 4 | DNA methylation reprogramming in development. a | Embryonic and germline DNA methylation erasure and
establishment. The methylcytosine dioxygenase TET3 is active in the fertilized zygote, leading to hydroxymethylation
and active DNA demethylation. Following passive demethylation (dashed line), DNA methylation reaches a low point at
the blastocyst stage, which is followed by DNA (cytosine-5)-methyltransferase 3A (DNMT3A)-mediated and DNMT3B-​
mediated de novo DNA methylation after blastocyst implantation. DNMT3L is also expressed in this time window, but is
not absolutely required to methylate the embryonic genome. In extra-​embryonic tissues, DNMT3A, DNMT3B and DNMT3L
are expressed, but to a lesser extent than in the embryo proper, which correlates with relative DNA hypomethylation. Post
implantation, in the epiblast, a subset of stem cells is specified for the germline, where they undergo two waves of DNA
demethylation: one passive and one mediated by TET1 and TET2. Male gametes become highly methylated before birth
through the activity of DNMT3A and DNMT3L, and, in the case of Muroidea (the superfamily that includes mice and rats),
DNMT3C. The oocyte gains methylation after meiosis and prior to ovulation through the activity of DNMT3A and DNMT3L
in mice, and likely through DNMT3A in humans. b | Transient imprinting at the Zdbf2 locus. During pre-​implantation
development, the expression of long isoform of Zdbf2 (Liz) is imprinted: the maternal allele is methylated and silenced, and
the paternal allele is unmethylated and expressed. The Zdbf2 promoters on both alleles are silenced through Polycomb-​
mediated trimethylation of histone H3 Lys27 (H3K27me3). During implantation, the expression of paternal Liz leads to
DNA methylation upstream of the canonical Zdbf2 promoter and eviction of H3K27me3, whereas the maternal allele
remains repressed by H3K27me3. Thus, although the imprinting of Liz is only transient, it causes permanent imprinting of
Zdbf2, because the post-​implantation methylation of the paternal locus enables lifelong expression of paternal Zdbf2 by
antagonizing Polycomb-​mediated repression. Failure to activate Zdbf2 in the embryo leads to a postnatal growth defect.
c | DNA methylation turnover at tissue-​specific enhancers. Dynamic DNA methylation in somatic tissues is frequently
found at regulatory elements. TET enzymes in collaboration with transcription factors (TF) lead to DNA demethylation and
enhancer activation. Inactive enhancers are bound less by transcription factors and are more accessible to de novo DNA
methyltransferases. d | The role of CpA methylation in neurons. After birth, DNMT3A deposits DNA methylation in gene
bodies. Methylated CAC sequences (mCAC) are recognized by methyl-​CpG-binding protein 2 (MeCP2), the binding
of which establishes a lifelong epigenetic memory of gene silencing. 5hm, hydroxymethylation of the fifth carbon;
5m, methylation of the fifth carbon; E, embryonic day; PGC, primordial germ cell; W, weeks.

NATuRE REvIEWS | MolECulAr CEll BIoloGy volume 20 | OCTOBER 2019 | 599

REVIEWS

classical ICRs, there are hundreds of ‘transient’ methyl­ of elements. Therefore, whether young transpos-
ation imprints: these are mainly inherited from the able elements contribute to epigenetic inheritance is
oocyte, maintain maternal-​specific DNA methylation unclear. However, although very rare, some single-​
until the blastocyst stage, probably through KRAB–ZFP-​ copy sequences also escape germline DNA methyl­
dependent protection, similarly to classical ICRs101,115, ation reprogramming in mice and humans; intriguingly,
and lose it through DNA demethylation or remethyl- these sequences exhibit very low levels of sequence and
ation after implantation101,176,179. This phenomenon may syntenic conservation between the two species161,166,190.
be particularly pervasive in humans, where the mater- Even more intriguingly, the DNA methylation of human
nal genome exhibits substantially higher preservation germline escapee sequences was generally not erased
of DNA methylation than the paternal genome prior to also during embryonic methylation reprogramming.
implantation, but also after implantation, exclusively in Whether these sparse sequences could serve as vectors
placental tissues130,179. Investigation of the Zdbf2 locus, of epigenetic inheritance is an alluring possibility.
whose transient imprinting is conserved in mice and
humans, demonstrated that transient hypomethylation The de novo DNA methylation programme
of the paternal allele can cause long-​lasting imprinting Following reprogramming in PGCs, sex-​specific pat-
through a cascade of downstream epigenetic changes terns of DNA methylation are established in the male
that affect postnatal size182,183 (Fig. 4b). Future work will be and female germlines. At the end of gametogenesis, the
needed to assess whether other transient imprints have sperm genome exhibits ~80% CpG methylation, with a
developmental roles. genomic distribution roughly similar to that of somatic
The bulk of the residual gametic methylation in the cells, although with a slightly higher DNA methylation
blastocyst is not related to single-​copy sequences but content. By contrast, the oocyte genome is only ~50%
instead to transposable elements. Young retrotrans- methylated, and nearly exclusively in gene bodies105,171,192
posons of the IAP class are particularly resistant to pre- (Fig. 4a). The relative hypomethylation of the oocyte is
implantation demethylation in mice171,184,185. In human associated with cytoplasmic retention of DNMT1,
blastocysts, the highest retention of DNA methylation itself secondary to the sequestration of UHRF1 to the
is also observed at evolutionarily young and potentially cytoplasm by the protein STELLA193. In Stella-​mutant
active transposable elements, in particular the hominid-​ oocytes, DNMT1 enters the nucleus, resulting in ectopic
specific SINE-​VNRT-Alu elements178. Indications are DNA methylation deposition, a twofold increase in DNA
that DNA methylation is retained through KRAB–ZFP-​ methylation and female infertility. Puzzlingly, this sug-
mediated sequence-​specific recruitment of KAP1 (ref.186) gests that DNMT1 is capable of de novo DNA methyl­
in a manner analogous to the selective DNA binding of ation in the oocyte, although this is not the case in
KAP1 in ICRs. KRAB–ZFPs require time to develop a embryonic cells, for example33. The gene-body specificity
match for recently emerged transposable elements187, and of oocyte DNA methylation is linked to a tight coupling of
thus the most recent IAP elements, which are often poly- de novo DNA methylation with gene transcription, in
morphic between mouse strains, are not resistant to DNA both mice and humans106,177. However, notable discrep-
methylation reprogramming during preimplantation ancies exist between the two species. First, whereas gene-​
development188. body methylation strictly requires DNMT3L-​mediated
Compared with preimplantation development, the stimulation of DNMT3A in mouse oocytes192, DNMT3L
germline wave of demethylation is more extensive. By is not expressed in human oocytes177. Second, thousands
day 13.5 of mouse embryogenesis, PGCs exhibit only of syntenic regions show different DNA methylation pat-
6–8% of methylated CpGs171 (Fig. 4a). Similar basal levels terns in mouse and in human oocytes. These differences
are retained in human fetal germ cells after 9 weeks of were recently linked to species-​specific insertions of long
pregnancy189,190. Contrary to embryonic development, terminal repeat retrotransposons, whose transcription is
ICRs undergo demethylation in developing PGCs, as a particularly high during oogenesis109. By defining new
prerequisite for subsequent acquisition of sex-​specific transcription units during oocyte maturation, these
DNA methylation patterns during male and female elements appear to be essential contributors of oocyte
germline differentiation. Similarly to methylation repro- methylation during development and mammalian evo-
gramming in preimplantation development, germline lution, and potential drivers of the emergence of new
retention of DNA methylation is mostly observed at maternally imprinted loci (Fig. 2b).
young and potentially harmful retrotransposons. In At the end of embryonic reprogramming — at
the mouse, this includes IAPs and long terminal repeat embryonic day 4.5 in the mouse — the blastocyst con-
elements of the endogenous retrovirus 1 class of retro- sists of the inner cell mass, which will go on to form
transposons (ERV1)161,191. In humans, SINE-​VNRT-Alu the epiblast and then embryonic tissues, and of the
(SVA) elements and human-​specific elements of the long trophoectoderm and the primitive endoderm, which
interspersed nuclear element (LINE) family remain par- contribute to the formation of extra-​embryonic tissues:
tially methylated compared with the rest of the genome the extra-​embryonic endoderm (ExE) and the visceral
and comparatively to older LINE types with a wider and endoderm, respectively. Two recent genome-​wide pro-
therefore older distribution in the primate lineage190. filing studies have confirmed that the ExE and the vis-
Owing to difficulties in mapping repetitive elements ceral endoderm are both hypomethylated relative to the
in the genome, it is not clear whether the same retro- epiblast194,195 (Fig. 4a). DNA hypomethylation persists
transposon copies resist both embryonic and germline in the differentiated placenta, which is largely derived
demethylation or whether they represent distinct sets from ExE cells196,197. Hypomethylation is correlated with

600 | OCTOBER 2019 | volume 20 www.nature.com/nrm

 M E T H Y L AT I O N I N B I O L O G Y

Box 2 | Experimentally editing DNA methylation decreased expression of the DNMT3 enzymes compared
with the epiblast, and even retrotranposons are relatively
There is no universal mode of gene regulation by DNA methylation, as the hypomethylated194. By contrast, a specific class of CGI
methylation modification is associated with genomic regions harbouring inactive promoters are methylated in the ExE and the visceral
and active genes with protein-coding genes and repetitive elements alike; there endoderm but not in the epiblast194,195. Genes associated
are also broad genomic regions that are methylated but where the modification serves
with this set of promoters are developmentally impor-
no obvious function. When analysing the effects of changes in DNA methylation in
DNA methylation mutants (or in conditions of induced hypomethylation), there is tant: their increased methylation either reflects a need
always the risk of confusing primary effects with confounding and/or secondary to keep them repressed in extra-​embryonic tissues or
effects. Methods exist for analysing the role of DNA methylation at specific loci. indicates that embryonic tissues have more stringent
For example, CpG sequences can be inserted into endogenous promoters to trigger mechanisms to prevent aberrant DNA methylation and
de novo DNA methylation256. However, the most promising avenue for altering long-​term silencing.
the methylation state of a given genomic locus is the emerging approach of The remethylation of the epiblast genome after
epigenome editing. implantation is very rapid: somatic-​tissue levels of DNA
Epigenome editing tools preceded the CRISPR–Cas era; indeed, modifying DNA methylation are attained already by mouse embryonic
methylation has been successfully achieved by targeting chromatin modifiers to DNA
day 6.5 (refs171,198) (Fig. 4a). This is notable, as the stem
through zinc-finger proteins257 or transcription activator-like effectors (TALEs)258,259.
cells in the epiblast are still pluripotent, so DNA methyl-
However, zinc-finger constructs are cumbersome to generate, and although TALEs
are relatively straightforward to assemble, they are sensitive to DNA methylation260 ation patterns established in epiblast stem cells have the
and thus not ideal for manipulating DNA methylation. Given its ease of use, the tool potential to be propagated through life in all tissues and
of choice for targeting DNA methylation to specific loci is now CRISPR–dCas9, maintain epigenetic memory of early embryogenesis.
in which catalytically inactive Cas9 (dCas9) is fused with the catalytic domain of a Nevertheless, whole-​genome surveys in both mice101,197
DNA methyltransferase (DNMT) or a TET methylcytosine dioxygenase and targeted and humans149,199,200 revealed widespread yet focal dis-
to genomic sequences by single guide RNAs (sgRNAs). Successful targeting of crepancies in DNA methylation patterns between vari­
methyltransferases at mammalian genomes was achieved by fusions of dCas9 with ous adult tissues, indicative of microwaves of de novo
the catalytic domain of mammalian DNMT3A261,262 (DNMT3A CD; see the figure, part a) DNA methylation and demethylation occurring during
and with the MQ1 CpG DNA methylase of the bacterium Mollicutes spiroplasma263 tissue differentiation. In mice, the majority (75%) of
(see the figure, part b). MQ1 is highly active and can generate considerable off-target
tissue-​specific differential DNA methylation regions are
methylation; this was mitigated by generating a Q147L mutation, which reduces the
intrinsic DNA binding capacity of MQ1. at enhancers or other regulatory elements, in line with
A caveat of dCas9–methyltransferase fusion proteins is the limited range of DNA studies showing that enhancers undergo active turnover
methylation from the targeted site, which can only be mitigated using multiple sgRNAs of DNA methylation, likely resulting from the combined
for targeting. In an interesting attempt to utilize DNMT biology, a dCas9 fused to activities of TET2 and tissue-​specific transcription fac-
chimeric DNMT3A catalytic domain–DNMT3L was demonstrated to effectively tors201,202 (Fig. 4c). In humans, a minority of DNA methyl-
methylate DNA at a greater distance from the target sequence, likely through the ation regions (<50%) map to regulatory elements; a large
formation of multimeric dCas9–DNMT3A–DNMT3L complexes264. Perhaps the most subset is found in undefined regions, where the func-
successful DNA methylome editing strategy thus far was derived from the dCas9–SunTag tional relevance of differences in DNA methylation is
system, in which an array of GCN4 epitopes is fused to dCas9 (refs265,266). By expressing unclear. Some of the differences between the mouse and
the DNMT3A catalytic domain linked to a single-chain antibody that recognizes the
human datasets may be due to the greater sequencing
epitope (scFv–DNMT3A), dCas9–SunTag recruits multiple enzymes to one locus (see the
figure, part c). Furthermore, modular recruitment of DNMT3A can decrease off-target depth of the human study199, which may have allowed
DNA methylation levels266. Similarly, using dCas9–SunTag to recruit scFv–TET1 has more nuanced extrapolations from the data.
proved efficient for targeting demethylation activity267. The brain exhibits a peculiar methylome among
Several studies are already putting epigenome editing to the test. Both DNMT3A somatic tissues. In both mice and humans, there is a
and TET1 have been successfully targeted in vivo to alter gene expression and alter postnatal spike in CpA methylation, which is generated
genome architecture262, and targeted DNA methylation was used to assess the role by DNMT3A203–205. In fact, total CpA methylation levels
of DNA methylation at neuronal genes during differentiation223. An exceptional study in neural tissues are roughly equivalent to those of CpG
that set out to determine the extent of epigenetic memory of transient epigenome methylation, although because CpA is a much more prev-
editing demonstrated that using a combination of dCas9–KRAB (a transcription alent dinucleotide, mCpA is found at a lower frequency
repression domain), dCas9–DNMT3A and dCas9–DNMT3L proffered long-term gene
than mCpG. Recent work indicates that the MBD protein
silencing268. This approach was notably employed to study the role of deregulated genes
in breast cancer269. We expect a torrent of forthcoming research using these precision MeCP2 binds mCAC sequences in gene bodies of young
tools to assess locus-specific functions of DNA methylation. mice neurons, and that MeCP2 binding maintains the
silencing of these genes in later stages of life206,207 (Fig. 4d).
a b This mechanism could help us understand the role of
sgRNA DNA methylation, and in particular non-​CpG methyl-
Q147L ation, in regulating neurological and behavioural traits.
Future studies will hopefully further clarify how defec-
DNMT3A MQ1
dCas9 CD tive DNA methylation early in life can lead to neuro­
nal abnormalities. Epigenome editing may prove to be
a useful avenue to explore these questions208 (Box 2).
c
3A 3A 3A 3A 3A
Genetic diseases, epigenetic outcomes
scFv antibody Disease-​associated mutations have been characterized
in DNMT and TET genes. Their effects on DNA methyl­
GCN4 epitope
ation patterns are very diverse, as are their tissue and
SunTag pathological manifestations, ranging from congenital

NATuRE REvIEWS | MolECulAr CEll BIoloGy volume 20 | OCTOBER 2019 | 601

REVIEWS

syndromes of immunodeficiency, growth phenotypes ICF — only 70 cases have been reported — it has been
or neurodegeneration to haematological cancers. Other, linked to germline mutations in three other genes, whose
phenotypically overlapping, syndromes have been linked products are postulated to facilitate DNA methylation:
to genes with known but also to those with unknown zinc finger and BTB domain-​containing protein 24
functions in DNA methylation and demethylation. (ZBTB24; defining ICF2), cell division cycle-​associated
The characterization of genetic mutations in the DNA protein 7 (CDCA7; ICF3) and HELLS (which encodes
methyl­ation pathways has greatly enriched our under- LSH; ICF4)216,217.
standing of the complexity and specificity of DNA Interestingly, although hypomethylation of peri­
methylation-​based gene regulation. centromeric satellite repeats is a hallmark of ICF, there
are also ICF-​type-specific defects that point towards
DNMT1 and neurological disorders differential genomic-​context-dependent interactions of
Heterozygous mutations in DNMT1 have been identified DNMT3B with ZBTB14, CDCA7 and LSH218. Notably,
in a spectrum of autosomal-​dominant forms of progres- individuals with ICF1 uniquely show hypomethylation
sive cognitive and behavioural deterioration, including of CGI promoters of germline genes and X-​linked genes,
hereditary sensory autonomic neuropathy 1E with in agreement with DNMT3B targeting these sequences
dementia and hearing loss (HSNA1E; OMIM 614116)209 during mouse development91,116 and indicating that this
and autosomal-​dominant cerebellar ataxia, deafness and targeting may occur independently of ZBTB14, CDCA7
narcolepsy (ADAC-​DN; OMIM 604121)210 (Table 1). The and LSH. By contrast, hypomethylation of centromeric
mean age of onset of HSNA1E is around 37 years, with α-​satellite repeats and neural gene clusters distinguishes
hearing and sensory loss being the initial symptoms fol- ICF2, ICF3 and ICF4 from ICF1. This may implicate
lowed by decline in cognitive function, ataxia and brain ZBTB14, CDCA7 and LSH in the recruitment of a
atrophy211. The average life expectancy is 50 years and DNMT other than DNMT3B to these sequences. Finally,
dementia is the main driver of morbidity. how whole-​body hypomethylation of pericentromeric
Strikingly, all DNMT1 mutations occur at the RFTS repeats specifically results in a defective immune system
domain, with HSNA1E mutations clustering in the remains unclear. Modelling the ICF syndrome in mice is
amino-​terminal and middle part of the domain and poorly informative owing to embryonic lethality and/or
ADAC-​DN mutations distinctively segregating to the lack of immunoglobulin deficiency219,220. However, a
carboxy terminus. As discussed above, proper folding of recent study of zbtb24-mutant zebrafish reported that
the RFTS is crucial for DNMT1 function52,53. Genome-​ pericentromeric hypomethylation during embryonic
wide analysis of the peripheral blood of individuals with development primarily activates the innate immune
HSNA1E or ADAC-​DN indicated that disturbances to system through the derepression of pericentro-
DNA methylation were very moderate compared with meric transcripts221. Such an animal model may help
non-​affected siblings, with hypomethylation of inter- refine our mechanistic understanding of ICF pathology
genic regions and hypermethylation of some CGIs212,213. in humans.
These DNA methylation signatures may be used as
pathological biomarkers in accessible cell types, but DNMT3A, cell growth and malignancies
they may not necessarily relate to the brain-​specific Heterozygous, germline mutations in DNMT3A are
manifestation of the diseases. Interestingly, DNMT1 associated with growth disorders of early prenatal
proteins with mutated RFTS domains form cytoplasmic onset, with contrasting growth phenotypes depending
aggregates in cell-​based overexpression assays, whereas on the nature of the mutations (Table 1). Missense, gain-​
wild-​type DNMT1 is normally nuclear211. The induction of-function mutations in the PWWP domain are found
of cellular stress by protein aggregation could result in in individuals with microcephalic dwarfism222 — a group
toxicity and age-​dependent degeneration of the central of pathologies that manifest as a profound yet propor-
and peripheral nervous systems. tionate reduction in body size and head size. These are
gain-​of-function mutations, which disrupt interactions
DNMT3B and immunodeficiency of DNMT3A with H3K36me3 moieties and alter its
The immunodeficiency, centromeric instability and chromatin binding specificity: in fibroblasts and blood
facial anomalies (ICF) syndrome (OMIM 602900) was cells of individuals carrying the mutations, ectopic
the first genetic disease to be linked to congenital DNA gain of DNA methylation is observed in large regions
methylation defects214 (Table 1). Individuals with ICF suf- known as DNA methylation valleys or canyons223,224, at
fer from atypical immunoglobulin deficiency that causes the expense of the bivalent H3K27me3 and H3K4me3
recurrent and often life-​threatening infections. The diag- modi­fications that normally characterize these regions
nosis of ICF typically includes a karyotype with multi- and regulate important developmental genes, such as
radiated chromosomes 1, 9 and 16, in association with Hox transcription factors and morphogen genes. The
specific hypomethylation of pericentromeric satellite introduction of a PWWP gain-​of-function mutation in
repeats II and III. Recessive mutations in the DNMT3B mice remarkably recapitulated the reductions in body
gene define ICF1, which accounts for the majority of size and brain weight in microcephalic dwarfism and
ICF cases215. As expected given the lethality of mouse the ectopic methylation phenotypes222,225.
models of Dnmt3B loss-​of-function mutations33, most By contrast, heterozygous DNMT3A haploinsuf-
ICF1 mutations cause single amino acid substitutions ficiency mutations characterize the Tatton-Brown–
or deletions and are thought to result in reduced rather Rahman syndrome (TBRS; also known as DNMT3A-
than abolished DNMT3B activity. Despite the scarcity of overgrowth syndrome; OMIM 602729), which combines

602 | OCTOBER 2019 | volume 20 www.nature.com/nrm

 M E T H Y L AT I O N I N B I O L O G Y

Neomorphic mutations
macrocephalic overgrowth with moderate intellectual availability of 5mC for TET2. The 5hmC modification
Typically, dominant mutations, disability226 (Table 1). TBRS mutations occur across may therefore be the culprit of malignancy, by affecting
which confer altered the DNMT3A gene, including in the PWWP domain gene regulation and/or other cellular processes through
expression or novel function for (upstream from the gain-of-function mutations), the recruitment of 5hmC readers238. A recent study of
the protein product.
ADD domain and MTase domain. The effects of these mice with mutations in both Dnmt3A and Tet2 illustrates
mutations on DNA methylation patterns have not yet the complex epistasis between the two enzymes, which
been determined, although the large genomic domains involves a combination of cooperative, competitive and
that are hypermethylated in the DNMT3A gain-​of- independent activities239.
function mutants are seemingly not affected in individ- TET2 is also indirectly impaired in multiple cancers
uals with TBRS222. Given that DNMT3A loss of function due to genetic mutations in enzymes that regulate the
results in stem cell hyperproliferation in various tissues production of metabolites necessary for its oxidative
in mice227,228, DNMT3A haploinsufficiency may pro- activity. TET2 requires 2-oxoglutarate as a co-​substrate
mote overgrowth in TBRS by increasing the number of for catalysing the three-​step 5mC hydroxylation reaction.
cells in tissues and, consequently, body size. Inversely, Isocitrate dehydrogenase 1 (IDH1) and IDH2 promote
DNMT3A gain of function could favour differentiation TET2 activity by producing 2-oxoglutarate. In the vast
over stem cell proliferation, and, in turn, reduce organ majority of lower-​grade gliomas, neomorphic mutations
and body size. The DNMT3A locus is strongly linked in IDH1 and IDH2 render them capable of producing
with natural height variation in genome-wide associ- 2-hydroxyglutarate. By competing with 2-oxoglutarate
ation studies229, pointing towards an important role of as a TET2 co-​substrate, 2-hydroxyglutarate induces
DNMT3A in human size physiology. extensive gain of 5mC and transcriptomic changes that
Finally, somatic DNMT3A mutations are linked to are relevant to the oncogenic phenotype, thereby clas-
enhanced proliferation of immature myeloid cells and sifying it as a potential oncometabolite240. In addition,
the development of adult haematological malignan- as part of the Krebs cycle in mitochondria, succinate
cies. The mutations are found in ~15–35% of cases of dehydrogenases (SDHs) convert the TET2 inhibitor
acute myeloid leukaemia (AML; OMIM 601626), the succinate. Somatic mutations in SDH genes are found
vast majority being missense mutations in arginine 882 in gastrointestinal stromal tumours and neuroendocrine
(R882), which is in the MTase domain. R882 mutations tumours (paragangliomas and pheochromocytomas), in
have dominant negative effects on the formation of association with increased intracellular levels of succi-
wild-​type DNMT3A tetramers, thereby reducing the nate, inhibition of TET2 activity and — similarly to IDH
methylation activity of the enzyme by 80% in vitro230. mutations — widespread hypermethylation within and
Consequently, very focal yet genome-​wide hypomethyl- outside CGIs241–243.
ation is observed at specific sites in CpG-​dense regions, In summary, the diseases associated with mutations
which precedes the onset of AML231. Hypermethylation in DNA methylation and demethylation factors offer a
of bivalent CGI promoters was also reported, but is a nuanced understanding of the functions of these pro-
consequence rather than a cause of AML progression, teins, beyond what can be achieved in loss-​of-function
being secondary to rapid cellular proliferation of malig- studies. Continued research in these areas will hopefully
nant myeloid cells. Interestingly, germline R882 muta- lead to therapeutic breakthroughs for both cancer and
tions also exist and cause early-​onset AML in individuals genetic diseases.
with TBRS232.
Conclusions and future perspective
Hydroxymethylation by TET2 and cancer As Timothy Bestor — one of the fathers of the mammal­
Whereas somatic loss of DNMT3A activity is fairly com- ian DNA methylation field — likes to note, the number
mon in AML, TET2 mutations are even more frequent of studies mentioning DNMT1 has increased linearly
causes of haematological malignancies, including AML, since its discovery in the 1980s, and the scientific interest
chronic myelomonocytic leukaemia, lymphomas and in DNA methylation has continued to grow. However,
myeloproliferative neoplasms (Table 1). Acquisition of despite the major inroads made into understanding DNA
(mostly heterozygous) TET2 mutations confers a pro- methylation mechanisms and patterns, much remains
liferative advantage to haematopoietic stem cells and unknown. For example, why do DNA methylation-
likely is an early driver of leukaemogenesis233. Moreover, deficient embryos die at such an early stage of develop-
restoration of TET2 function can reverse leukaemic ment? It seems like a foregone conclusion that lack of
progression, further supporting a causative role for the DNA methylation would be lethal, but it is not clear what
enzyme in the aetiology of the disease234. In contrast confers the embryonic failure: aberrant expression of
to the DNMT3A mutations that primarily decrease protein-​coding genes; massive derepression of transpos-
DNA methylation levels, TET2 mutations promote able elements; genomic instability; a combination or all
hypermethylation in myeloid malignancies, mostly at of the above or something less obvious? Furthermore, in
enhancer regions235,236. How haploinsufficiency muta- cancer cells, DNA methylation is commonly deregulated
tions in TET2 and DNMT3A — two enzymes with and DNMT3 and TET genes are frequently mutated. But
opposing functions — can cause similar malignancy which methylation changes are relevant for oncogenesis
phenotypes is puzzling and compels explanation. TET2- and which are only incidental?
mutant cancer cells exhibit the expected reduction in We are now entering an era of unprecedented genetic
5hmC content235,237; in DNMT3A-​mutant cancer cells, tools, sensitive and highly quantitative sequencing
decreased 5hmC is also expected owing to the lower techno­logies, and the ability to alter DNA methylation

NATuRE REvIEWS | MolECulAr CEll BIoloGy volume 20 | OCTOBER 2019 | 603

REVIEWS

with surgical precision (Box 2). Major questions that this process to assess the developmental effects of DNA
have been traditionally difficult to address now seem methylation.
possible to answer. Yet challenges remain. For exam- As a final note, although the focus of this review is
ple, when assessing DNA methylation patterns asso- mice and humans, our understanding of mammalian
ciated with a specific cell type or developmental stage, DNA methylation has greatly benefited from studies in
we can only infer steady-​state levels of methylation that distant species such as flowering plants and filamentous
result from the opposing activities of the enzymes fungi (reviewed elsewhere244). Comparative analyses
that generate and remove DNA methylation. There is of non-​traditional model organisms provide valuable
a need for sequence-​specific quantification of DNA insights into conserved mechanisms and functions,
methylation turnover. Combining experimental and and also surprises: in the milkweed bug, DNA methyl­
theoretical approaches would allow modelling of the ation is important for fertility, but this may occur inde-
principles of local DNA methylation dynamics and pendently of the traditional role of DNA methylation in
help us to understand where and why imbalances may transcription control245. With the torrid pace of techno-
occur during ageing and tumorigenesis. Furthermore, logical advancement and knowledge gleaned from stud-
a major conundrum in mammalian DNA methylation ies in diverse systems, the future of mammalian DNA
biology is the function of the extensive demethylation methylation research promises exciting and impactful
that occurs in the early embryo and germline, in parti­ discoveries ahead.
cular whether it is required for the establishment of
pluripotency. At present, it appears impossible to stymie Published online 9 August 2019

1. Zemach, A., McDaniel, I. E., Silva, P. & Zilberman, D. 19. Cooper, D. N. & Krawczak, M. Cytosine methylation 36. Piunti, A. & Shilatifard, A. Epigenetic balance of gene
Genome-​wide evolutionary analysis of eukaryotic DNA and the fate of CpG dinucleotides in vertebrate expression by Polycomb and COMPASS families.
methylation. Science 328, 916–919 (2010). genomes. Hum. Genet. 83, 181–188 (1989). Science 352, aad9780 (2016).
2. Feng, S. et al. Conservation and divergence of 20. Rošić, S. et al. Evolutionary analysis indicates that 37. Otani, J. et al. Structural basis for recognition
methylation patterning in plants and animals. Proc. DNA alkylation damage is a byproduct of cytosine of H3K4 methylation status by the DNA
Natl Acad. Sci. USA 107, 8689–8694 (2010). DNA methyltransferase activity. Nat. Genet. 50, methyltransferase 3A ATRX-​DNMT3-DNMT3L
3. Doskočil, J. & Šorm, F. Distribution of 5-methylcytosine 452–459 (2018). domain. EMBO Rep. 10, 1235–1241 (2009).
in pyrimidine sequences of deoxyribonucleic acids. 21. Li, E. & Zhang, Y. DNA methylation in mammals. 38. Zhang, Y. et al. Chromatin methylation activity of
Biochim. Biophys. Acta 55, 953–959 (1962). Cold Spring Harb. Perspect. Biol. 6, a019133 Dnmt3a and Dnmt3a/3L is guided by interaction of
4. Riggs, A. D. X inactivation, differentiation, and DNA (2014). the ADD domain with the histone H3 tail. Nucleic
methylation. Cytogenet. Cell Genet. 14, 9–25 (1975). 22. Li, E., Bestor, T. H. & Jaenisch, R. Targeted mutation of Acids Res. 38, 4246–4253 (2010).
5. Bird, A. P. Use of restriction enzymes to study eukaryotic the DNA methyltransferase gene results in embryonic 39. Guo, X. et al. Structural insight into autoinhibition and
DNA methylation. II. The symmetry of methylated sites lethality. Cell 69, 915–926 (1992). histone H3-induced activation of DNMT3A. Nature
supports semi-​conservative copying of the methylation 23. Okano, M., Bell, D. W., Haber, D. A. & Li, E. 517, 640–644 (2015).
pattern. J. Mol. Biol. 118, 49–60 (1978). DNA methyltransferases Dnmt3a and Dnmt3b are 40. Krogan, N. J. et al. Methylation of histone H3 by Set2
6. Ben-​Hattar, J. & Jiricny, J. Methylation of single CpG essential for de novo methylation and mammalian in Saccharomyces cerevisiae is linked to
dinucleotides within a promoter element of the development. Cell 99, 247–257 (1999). transcriptional elongation by RNA polymerase II.
Herpes simplex virus tk gene reduces its transcription 24. Baylin, S. B. & Jones, P. A. Epigenetic determinants of Mol. Cell. Biol. 23, 4207–4218 (2003).
in vivo. Gene 65, 219–227 (1988). cancer. Cold Spring Harb. Perspect. Biol. 8, a019505 41. Sun, X. J. et al. Identification and characterization of
7. Watt, F. & Molloy, P. L. Cytosine methylation prevents (2016). a novel human histone H3 lysine 36-specific
binding to DNA of a HeLa cell transcription factor 25. Arand, J. et al. In vivo control of CpG and non-​CpG methyltransferase. J. Biol. Chem. 280, 35261–35271
required for optimal expression of the adenovirus major DNA methylation by DNA methyltransferases. (2005).
late promoter. Genes Dev. 2, 1136–1143 (1988). PLOS Genet. 8, e1002750 (2012). 42. Dhayalan, A. et al. The Dnmt3a PWWP domain reads
8. Iguchi-​Ariga, S. M. & Schaffner, W. CpG methylation 26. Walsh, C. P., Chaillet, J. R. & Bestor, T. H. Transcription histone 3 lysine 36 trimethylation and guides DNA
of the cAMP-​responsive enhancer/promoter sequence of IAP endogenous retroviruses is constrained by methylation. J. Biol. Chem. 285, 26114–26120 (2010).
TGACGTCA abolishes specific factor binding as well cytosine methylation. Nat. Genet. 20, 116–117 43. Baubec, T. et al. Genomic profiling of DNA
as transcriptional activation. Genes Dev. 3, 612–619 (1998). methyltransferases reveals a role for DNMT3B in
(1989). 27. Borgel, J. et al. Targets and dynamics of promoter genic methylation. Nature 520, 243–247 (2015).
9. Ferguson-​Smith, A. C., Sasaki, H., Cattanach, B. M. DNA methylation during early mouse development. 44. Xu, Q. et al. SETD2 regulates the maternal
& Surani, M. A. Parental-​origin-specific epigenetic Nat. Genet. 42, 1093–1100 (2010). epigenome, genomic imprinting and embryonic
modification of the mouse H19 gene. Nature 362, 28. Lewis, J. D. et al. Purification, sequence, and cellular development. Nat. Genet. 51, 844–856 (2019).
751–755 (1993). localization of a novel chromosomal protein that 45. Bostick, M. et al. UHRF1 plays a role in maintaining
10. Li, E., Beard, C. & Jaenisch, R. Role for DNA binds to methylated DNA. Cell 69, 905–914 (1992). DNA methylation in mammalian cells. Science 317,
methylation in genomic imprinting. Nature 366, 29. Lister, R. et al. Human DNA methylomes at base 1760–1764 (2007).
362–365 (1993). resolution show widespread epigenomic differences. 46. Sharif, J. et al. The SRA protein Np95 mediates
11. Bartolomei, M. S., Webber, A. L., Brunkow, M. E. & Nature 462, 315–322 (2009). epigenetic inheritance by recruiting Dnmt1 to
Tilghman, S. M. Epigenetic mechanisms underlying 30. Monk, M., Boubelik, M. & Lehnert, S. Temporal and methylated DNA. Nature 450, 908–912 (2007).
the imprinting of the mouse H19 gene. Genes Dev. 7, regional changes in DNA methylation in the 47. Nady, N. et al. Recognition of multivalent histone
1663–1673 (1993). embryonic, extraembryonic and germ cell lineages states associated with heterochromatin by UHRF1
12. Stöger, R. et al. Maternal-​specific methylation of the during mouse embryo development. Development 99, protein. J. Biol. Chem. 286, 24300–24311 (2011).
imprinted mouse Igf2r locus identifies the expressed 371–382 (1987). 48. Arita, K. et al. Recognition of modification status on
locus as carrying the imprinting signal. Cell 73, 61–71 31. Sanford, J. P., Clark, H. J., Chapman, V. M. & Rossant, J. a histone H3 tail by linked histone reader modules of
(1993). Differences in DNA methylation during oogenesis the epigenetic regulator UHRF1. Proc. Natl Acad. Sci.
13. Mohandas, T., Sparkes, R. S. & Shapiro, L. J. and spermatogenesis and their persistence during USA 109, 12950–12955 (2012).
Reactivation of an inactive human X chromosome: early embryogenesis in the mouse. Genes Dev. 1, 49. Rothbart, S. B. et al. Association of UHRF1 with
evidence for X inactivation by DNA methylation. 1039–1046 (1987). methylated H3K9 directs the maintenance of DNA
Science 211, 393–396 (1981). 32. Okano, M., Xie, S. & Li, E. Cloning and methylation. Nat. Struct. Mol. Biol. 19, 1155–1160
14. Lock, L. F., Takagi, N. & Martin, G. R. Methylation characterization of a family of novel mammalian DNA (2012).
of the Hprt gene on the inactive X occurs after (cytosine-5) methyltransferases. Nat. Genet. 19, 50. Rothbart, S. B. et al. Multivalent histone engagement
chromosome inactivation. Cell 48, 39–46 (1987). 219–220 (1998). by the linked tandem tudor and PHD domains of
15. Zemach, A. & Zilberman, D. Evolution of eukaryotic 33. Okano, M., Bell, D. W., Haber, D. A. & Li, E. DNA UHRF1 is required for the epigenetic inheritance
DNA methylation and the pursuit of safer sex. methyltransferases Dnmt3a and Dnmt3b are essential of DNA methylation. Genes Dev. 27, 1288–1298
Curr. Biol. 20, R780–R785 (2010). for de novo methylation and mammalian (2013).
16. Raddatz, G. et al. Dnmt2-dependent methylomes lack development. Cell 99, 247–257 (1999). 51. Song, J., Rechkoblit, O., Bestor, T. H. & Patel, D. J.
defined DNA methylation patterns. Proc. Natl Acad. 34. Ooi, S. K. et al. DNMT3L connects unmethylated Structure of DNMT1–DNA complex reveals a role for
Sci. USA 110, 8627–8631 (2013). lysine 4 of histone H3 to de novo methylation of DNA. autoinhibition in maintenance DNA methylation.
17. Holliday, R. & Grigg, G. W. DNA methylation and Nature 448, 714–717 (2007). Science 331, 1036–1040 (2011).
mutation. Mutat. Res. 285, 61–67 (1993). 35. Bourc’his, D., Xu, G. L., Lin, C. S., Bollman, B. & 52. Takeshita, K. et al. Structural insight into maintenance
18. Bird, A. P. & Taggart, M. H. Variable patterns of total Bestor, T. H. Dnmt3L and the establishment of methylation by mouse DNA methyltransferase 1
DNA and rDNA methylation in animals. Nucleic Acids maternal genomic imprints. Science 294, (Dnmt1). Proc. Natl Acad. Sci. USA 108, 9055–9059
Res. 8, 1485–1497 (1980). 2536–2539 (2001). (2011).

604 | OCTOBER 2019 | volume 20 www.nature.com/nrm

 M E T H Y L AT I O N I N B I O L O G Y
53. Ishiyama, S. et al. Structure of the Dnmt1 reader 79. Saito, M. & Ishikawa, F. The mCpG-​binding domain of 104. Smith, E. Y., Futtner, C. R., Chamberlain, S. J.,
module complexed with a unique two-​mono-ubiquitin human MBD3 does not bind to mCpG but interacts Johnstone, K. A. & Resnick, J. L. Transcription is
mark on Histone H3 reveals the basis for DNA with NuRD/Mi2 components HDAC1 and MTA2. required to establish maternal imprinting at the
methylation maintenance. Mol. Cell 68, 350–360 J. Biol. Chem. 277, 35434–35439 (2002). Prader–Willi syndrome and Angelman syndrome
(2017). 80. Nan, X. et al. Transcriptional repression by the locus. PLOS Genet. 7, e1002422 (2011).
54. Nishiyama, A. et al. Uhrf1-dependent H3K23 methyl-CpG-binding protein MeCP2 involves a histone 105. Smallwood, S. A. et al. Dynamic CpG island
ubiquitylation couples maintenance DNA methylation deacetylase complex. Nature 393, 386–389 (1998). methylation landscape in oocytes and preimplantation
and replication. Nature 502, 249–253 (2013). 81. Ng, H. H. et al. MBD2 is a transcriptional repressor embryos. Nat. Genet. 43, 811–814 (2011).
55. Qin, W. et al. DNA methylation requires a DNMT1 belonging to the MeCP1 histone deacetylase complex. 106. Veselovska, L. et al. Deep sequencing and de novo
ubiquitin interacting motif (UIM) and histone Nat. Genet. 23, 58–61 (1999). assembly of the mouse oocyte transcriptome define
ubiquitination. Cell Res. 25, 911–929 (2015). 82. Ren, R., Horton, J. R., Zhang, X., Blumenthal, R. M. & the contribution of transcription to the DNA
56. Kriaucionis, S. & Heintz, N. The nuclear DNA base Cheng, X. Detecting and interpreting DNA methylation methylation landscape. Genome Biol. 16, 209 (2015).
5-hydroxymethylcytosine is present in purkinje neurons marks. Curr. Opin. Struct. Biol. 53, 88–99 (2018). 107. Peaston, A. E. et al. Retrotransposons regulate host
and the brain. Science 324, 929–930 (2009). 83. Gardiner-​Garden, M. & Frommer, M. CpG Islands in genes in mouse oocytes and preimplantation embryos.
57. Tahiliani, M. et al. Conversion of 5-methylcytosine to vertebrate genomes. J. Mol. Biol. 196, 261–282 Dev. Cell 7, 597–606 (2004).
5-hydroxymethylcytosine in mammalian DNA by MLL (1987). 108. Franke, V. et al. Long terminal repeats power evolution
partner TET1. Science 324, 930–935 (2009). 84. Larsen, F., Gundersen, G., Lopez, R. & Prydz, H. of genes and gene expression programs in mammalian
58. Ito, S. et al. Role of tet proteins in 5mC to 5hmC CpG islands as gene markers in the human genome. oocytes and zygotes. Genome Res. 27, 1384–1394
conversion, ES-​cell self-​renewal and inner cell mass Genomics 13, 1095–1107 (1992). (2017).
specification. Nature 466, 1129–1133 (2010). 85. Ku, M. et al. Genomewide analysis of PRC1 and PRC2 109. Brind’Amour, J. et al. LTR retrotransposons transcribed
59. Ito, S. et al. Tet proteins can convert 5-methylcytosine occupancy identifies two classes of bivalent domains. in oocytes drive species-​specific and heritable changes
to 5-formylcytosine and 5-carboxylcytosine. Science PLOS Genet. 4, e1000242 (2008). in DNA methylation. Nat. Commun. 9, 3331 (2018).
333, 1300–1303 (2011). 86. Bird, A., Taggart, M., Frommer, M., Miller, O. J. & 110. Li, X. et al. A maternal-​zygotic effect gene, Zfp57,
60. He, Y. F. et al. Tet-​mediated formation of Macleod, D. A fraction of the mouse genome that is maintains both maternal and paternal imprints.
5-carboxylcytosine and its excision by TDG in derived from islands of nonmethylated, CpG-​rich DNA. Dev. Cell 15, 547–557 (2008).
mammalian DNA. Science 333, 1303–1307 (2011). Cell 40, 91–99 (1985). 111. Quenneville, S. et al. In embryonic stem cells, ZFP57/
61. Hashimoto, H. et al. Recognition and potential 87. Marasca, F., Bodega, B. & Orlando, V. How polycomb-​ KAP1 recognize a methylated hexanucleotide to affect
mechanisms for replication and erasure of cytosine mediated cell memory deals with a changing chromatin and DNA methylation of imprinting control
hydroxymethylation. Nucleic Acids Res. 40, environment. BioEssays 40, e1700137 (2018). regions. Mol. Cell 44, 361–372 (2011).
4841–4849 (2012). 88. Singer-​Sam, J. et al. Use of a HpaII-​polymerase chain 112. Strogantsev, R. et al. Allele-​specific binding of ZFP57
62. Otani, J. et al. Cell cycle-​dependent turnover of reaction assay to study DNA methylation in the Pgk-1 in the epigenetic regulation of imprinted and non-​
5-hydroxymethyl cytosine in mouse embryonic stem CpG island of mouse embryos at the time of X-​ imprinted monoallelic expression. Genome Biol. 16,
cells. PLOS ONE 8, e82961 (2013). chromosome inactivation. Mol. Cell. Biol. 10, 112 (2015).
63. Maiti, A. & Drohat, A. C. Thymine DNA glycosylase 4987–4989 (1990). 113. Messerschmidt, D. M. et al. Trim28 is required for
can rapidly excise 5-formylcytosine and 89. Grant, M., Zuccotti, M. & Monk, M. Methylation epigenetic stability during mouse oocyte to embryo
5-carboxylcytosine: Potential implications for active of CpG sites of two X-​linked genes coincides with transition. Science 335, 1499–1502 (2012).
demethylation of CpG sites. J. Biol. Chem. 286, X-inactivation in the female mouse embryo but not 114. Galonska, C., Ziller, M. J., Karnik, R. & Meissner, A.
35334–35338 (2011). in the germ line. Nat. Genet. 2, 161–166 (1992). Ground State Conditions Induce Rapid Reorganization
64. Weber, A. R. et al. Biochemical reconstitution of 90. Keohane, A. M., O’Neill, L. P., Belyaev, N. D., of Core Pluripotency Factor Binding before Global
TET1-TDG-​BER-dependent active DNA demethylation Lavender, J. S. & Turner, B. M. X-​inactivation and Epigenetic Reprogramming. Cell Stem Cell 17,
reveals a highly coordinated mechanism. Nat. Commun. histone H4 acetylation in embryonic stem cells. 462–470 (2015).
7, 10806 (2016). Dev. Biol. 180, 618–630 (1996). 115. Takahashi, N. et al. ZNF445 is a primary regulator of
65. Weber, M. et al. Distribution, silencing potential and 91. Gendrel, A.-V. et al. Smchd1-dependent and genomic imprinting. Genes Dev. 33, 49–54 (2019).
evolutionary impact of promoter DNA methylation in -independent pathways determine developmental 116. Auclair, G., Guibert, S., Bender, A. & Weber, M.
the human genome. Nat. Genet. 39, 457–466 (2007). dynamics of CpG Island methylation on the inactive Ontogeny of CpG island methylation and specificity
66. Stadler, M. B. et al. DNA-​binding factors shape the X chromosome. Dev. Cell 23, 265–279 (2012). of DNMT3 methyltransferases during embryonic
mouse methylome at distal regulatory regions. Nature 92. Blewitt, M. E. et al. SmcHD1, containing a structural-​ development in the mouse. Genome Biol. 15, 545
480, 490–495 (2011). maintenance-of-​chromosomes hinge domain, has a (2014).
67. Yin, Y. et al. Impact of cytosine methylation on DNA critical role in X inactivation. Nat. Genet. 40, 117. Karimi, M. M. et al. DNA methylation and SETDB1/
binding specificities of human transcription factors. 663–669 (2008). H3K9me3 regulate predominantly distinct sets of
Science 356, eaaj2239 (2017). 93. Gdula, M. R. et al. The non-​canonical SMC protein genes, retroelements, and chimeric transcripts in
68. Dennis, K., Fan, T., Geiman, T., Yan, Q. & Muegge, K. SmcHD1 antagonises TAD formation and mESCs. Cell Stem Cell 8, 676–687 (2011).
Lsh, a member of the SNF2 family, is required for compartmentalisation on the inactive X chromosome. 118. O’Neill, K. M. et al. Depletion of DNMT1 in
genome-​wide methylation. Genes Dev. 15, Nat. Commun. 10, 30 (2019). differentiated human cells highlights key classes of
2940–2944 (2001). 94. Nozawa, R. S. et al. Human inactive X chromosome is sensitive genes and an interplay with polycomb
69. Myant, K. et al. LSH and G9a/GLP complex are compacted through a PRC2-independent SMCHD1- repression. Epigenetics Chromatin 11, 12 (2018).
required for developmentally programmed DNA HBiX1 pathway. Nat. Struct. Mol. Biol. 20, 566–573 119. Auclair, G. et al. EHMT2 directs DNA methylation for
methylation. Genome Res. 21, 83–94 (2011). (2013). efficient gene silencing in mouse embryos. Genome
70. Tao, Y. et al. Lsh, chromatin remodeling family member, 95. Wutz, A. et al. Imprinted expression of the Igf2r gene Res. 26, 192–202 (2016).
modulates genome-​wide cytosine methylation patterns depends on an intronic CpG island. Nature 389, 120. Velasco, G. & Francastel, C. Genetics meets DNA
at nonrepeat sequences. Proc. Natl Acad. Sci. USA 745–749 (1997). methylation in rare diseases. Clin. Genet. 95,
108, 5626–5631 (2011). 96. Thorvaldsen, J. L., Duran, K. L. & Bartolomei, M. S. 210–220 (2019).
71. Esteve, P. O. et al. Direct interaction between DNMT1 Deletion of the H19 differentially methylated domain 121. Endoh, M. et al. PCGF6-PRC1 suppresses premature
and G9a coordinates DNA and histone methylation results in loss of imprinted expression of H19 and differentiation of mouse embryonic stem cells by
during replication. Genes Dev. 20, 3089–3103 (2006). Igf2. Genes Dev. 12, 3693–3702 (1998). regulating germ cell-​related genes. eLife 6, e21064
72. Epsztejn-​Litman, S. et al. De novo DNA methylation 97. Yang, T. et al. A mouse model for Prader–Willi (2017).
promoted by G9a prevents reprogramming of syndrome imprinting-​centre mutations. Nat. Genet. 122. Stielow, B., Finkernagel, F., Stiewe, T., Nist, A.
embryonically silenced genes. Nat. Struct. Mol. Biol. 19, 25–31 (1998). & Suske, G. MGA, L3MBTL2 and E2F6 determine
15, 1176–1183 (2008). 98. Fitzpatrick, G. V., Soloway, P. D. & Higgins, M. J. genomic binding of the non-​canonical Polycomb
73. Fuks, F., Burgers, W. A., Brehm, A., Hughes-​Davies, L. Regional loss of imprinting and growth deficiency in repressive complex PRC1.6. PLOS Genet. 14,
& Kouzarides, T. DNA methyltransferase Dnmt1 mice with a targeted deletion of KvDMR1. Nat. Genet. e1007193 (2018).
associates with histone deacetylase activity. Nat. Genet. 32, 426–431 (2002). 123. Gao, Z. et al. PCGF homologs, CBX proteins, and RYBP
24, 88–91 (2000). 99. Lin, S. P. et al. Asymmetric regulation of imprinting on define functionally distinct PRC1 family complexes.
74. Fuks, F., Burgers, W. A., Godin, N., Kasai, M. & the maternal and paternal chromosomes at the Dlk1- Mol. Cell 45, 344–356 (2012).
Kouzarides, T. Dnmt3a binds deacetylases and is Gtl2 imprinted cluster on mouse chromosome 12. 124. Maeda, I. et al. Max is a repressor of germ cell-​related
recruited by a sequence‐specific repressor to silence Nat. Genet. 35, 97–102 (2003). gene expression in mouse embryonic stem cells.
transcription. EMBO J. 20, 2536–2544 (2001). 100. Williamson, C. M. et al. Identification of an imprinting Nat. Commun. 4, 1754 (2013).
75. Deplus, R. et al. Dnmt3L is a transcriptional repressor control region affecting the expression of all 125. Velasco, G. et al. Dnmt3b recruitment through E2F6
that recruits histone deacetylase. Nucleic Acids Res. transcripts in the Gnas cluster. Nat. Genet. 38, transcriptional repressor mediates germ-​line gene
30, 3831–3838 (2002). 350–355 (2006). silencing in murine somatic tissues. Proc. Natl Acad.
76. Meehan, R. R., Lewis, J. D., McKay, S., Kleiner, E. L. 101. Proudhon, C. et al. Protection against de novo Sci. USA 107, 9281–9286 (2010).
& Bird, A. P. Identification of a mammalian protein methylation is instrumental in maintaining parent-​ 126. Qin, J. et al. The polycomb group protein L3mbtl2
that binds specifically to DNA containing methylated of-origin methylation inherited from the gametes. assembles an atypical PRC1-family complex that is
CpGs. Cell 58, 499–507 (1989). Mol. Cell 47, 909–920 (2012). essential in pluripotent stem cells and early
77. Hendrich, B. & Bird, A. Identification and 102. Kaneda, M. et al. Genetic evidence for Dnmt3a-​ development. Cell Stem Cell 11, 319–332 (2012).
characterization of a family of mammalian methyl-​CpG dependent imprinting during oocyte growth obtained 127. Trojer, P. et al. L3MBTL2 protein acts in concert with
binding proteins. Mol Cell Biol 18, 6538–6547 by conditional knockout with Zp3-Cre and complete PcG protein-​mediated monoubiquitination of H2A to
(1998). exclusion of Dnmt3b by chimera formation. Genes establish a repressive chromatin structure. Mol. Cell
78. Baubec, T., Ivánek, R., Lienert, F. & Schübeler, D. Cells 15, 169–179 (2010). 42, 438–450 (2011).
methylation-​dependent and -independent genomic 103. Chotalia, M. et al. Transcription is required for 128. Yoder, J. A., Walsh, C. P. & Bestor, T. H. Cytosine
targeting principles of the MBD protein family. Cell establishment of germline methylation marks at methylation and the ecology of intragenomic
153, 480–492 (2013). imprinted genes. Genes Dev. 23, 105–117 (2009). parasites. Trends Genet 13, 335–340 (1997).

NATuRE REvIEWS | MolECulAr CEll BIoloGy volume 20 | OCTOBER 2019 | 605

REVIEWS
129. Barau, J. et al. The DNA methyltransferase DNMT3C 155. Maunakea, A. K., Chepelev, I., Cui, K. & Zhao, K. 183. Greenberg, M. V. C. et al. Transient transcription in the
protects male germ cells from transposon activity. Intragenic DNA methylation modulates alternative early embryo sets an epigenetic state that programs
Science 354, 909–912 (2016). splicing by recruiting MeCP2 to promote exon postnatal growth. Nat. Genet. 49, 110–118 (2017).
130. Sanchez-​Delgado, M. et al. Human oocyte-​derived recognition. Cell Res. 23, 1256–1269 (2013). 184. Lane, N. et al. Resistance of IAPs to methylation
methylation differences persist in the placenta revealing 156. Yearim, A. et al. HP1 is involved in regulating the reprogramming may provide a mechanism for
widespread transient imprinting. PLOS Genet. 12, global impact of DNA methylation on alternative epigenetic inheritance in the mouse. Genesis 35,
e1006427 (2016). splicing. Cell Rep. 10, 1122–1134 (2015). 88–93 (2003).
131. Walsh, C. P., Chaillet, J. R. & Bestor, T. H. Transcription 157. Carrozza, M. J. et al. Histone H3 methylation by Set2 185. Smith, Z. D. et al. A unique regulatory phase of DNA
of IAP endogenous retroviruses is constrained by directs deacetylation of coding regions by Rpd3S to methylation in the early mammalian embryo. Nature
cytosine methylation. Nat. Genet. 20, 116–117 suppress spurious intragenic transcription. Cell 123, 484, 339–344 (2012).
(1998). 581–592 (2005). 186. Rowe, H. M. et al. KAP1 controls endogenous
132. Jain, D. et al. rahu is a mutant allele of Dnmt3c, 158. Maunakea, A. K. et al. Conserved role of intragenic retroviruses in embryonic stem cells. Nature 463,
encoding a DNA methyltransferase homolog required DNA methylation in regulating alternative promoters. 237–240 (2010).
for meiosis and transposon repression in the mouse Nature 466, 253–257 (2010). 187. Ecco, G., Imbeault, M. & Trono, D. KRAB zinc finger
male germline. PLOS Genet. 13, e1006964 (2017). 159. Neri, F. et al. Intragenic DNA methylation prevents proteins. Development 144, 2719–2729 (2017).
133. Kuramochi-​Miyagawa, S. et al. DNA methylation of spurious transcription initiation. Nature 543, 72–77 188. Kazachenka, A. et al. Identification, characterization,
retrotransposon genes is regulated by Piwi family (2017). and heritability of murine metastable epialleles:
members MILI and MIWI2 in murine fetal testes. 160. Teissandier, A. & Bourc’his, D. Gene body DNA implications for non-​genetic inheritance. Cell 175,
Genes Dev. 22, 908–917 (2008). methylation conspires with H3K36me3 to preclude 1259–1271 (2018).
134. Aravin, A. A. et al. A piRNA pathway primed by aberrant transcription. EMBO J. 36, 1471–1473 189. Gkountela, S. et al. DNA demethylation dynamics in
individual transposons is linked to de novo DNA (2017). the human prenatal germline. Cell 161, 1425–1436
methylation in mice. Mol Cell 31, 785–799 (2008). 161. Guibert, S., Forné, T. & Weber, M. Global profiling of (2015).
135. Aravin, A. A., Hannon, G. J. & Brennecke, J. The Piwi– DNA methylation erasure in mouse primordial germ 190. Tang, W. W. C. et al. A unique gene regulatory network
piRNA pathway provides an adaptive defense in the cells. Genome Res. 22, 633–641 (2012). resets the human germline epigenome for development.
transposon arms race. Science 318, 761–764 (2007). 162. Vincent, J. J. et al. Stage-​specific roles for Tet1 and Cell 161, 1453–1467 (2015).
136. Carmell, M. A. et al. MIWI2 is essential for Tet2 in DNA demethylation in primordial germ cells. 191. Hajkova, P. et al. Epigenetic reprogramming in mouse
spermatogenesis and repression of transposons in the Cell Stem Cell 12, 470–478 (2013). primordial germ cells. Mech. Dev. 117, 15–23
mouse male germline. Dev. Cell 12, 503–514 (2007). 163. Yamaguchi, S. et al. Dynamics of 5-methylcytosine (2002).
137. Molaro, A. et al. Two waves of de novo methylation and 5-hydroxymethylcytosine during germ cell 192. Kobayashi, H. et al. Contribution of intragenic DNA
during mouse germ cell development. Genes Dev. 28, reprogramming. Cell Res. 23, 329–339 (2013). methylation in mouse gametic DNA methylomes to
1544–1549 (2014). 164. Yamaguchi, S. et al. Tet1 controls meiosis by establish oocyte-​specific heritable marks. PLOS Genet.
138. Manakov, S. A. et al. MIWI2 and MILI have differential regulating meiotic gene expression. Nature 492, 8, e1002440 (2012).
effects on piRNA biogenesis and DNA methylation. 443–447 (2012). 193. Li, Y. et al. Stella safeguards the oocyte methylome by
Cell Rep. 12, 1234–1243 (2015). 165. Yamaguchi, S., Shen, L., Liu, Y., Sendler, D. & Zhang, Y. preventing de novo methylation mediated by DNMT1.
139. Barau, J. et al. The novel DNA methyltransferase Role of Tet1 in erasure of genomic imprinting. Nature Nature 564, 136–140 (2018).
DNMT3C protects male germ cells from transposon 504, 460–464 (2013). 194. Smith, Z. D. et al. Epigenetic restriction of
activity. Science 354, 909–912 (2016). 166. Hackett, J. A. et al. Germline DNA demethylation extraembryonic lineages mirrors the somatic
140. Jordà, M. et al. The epigenetic landscape of Alu dynamics and imprint erasure through transition to cancer. Nature 549, 543–547 (2017).
repeats delineates the structural and functional 5-hydroxymethylcytosine. Science 339, 448–452 195. Zhang, Y. et al. Dynamic epigenomic landscapes
genomic architecture of colon cancer cells. (2013). during early lineage specification in mouse embryos.
Genome Res. 27, 118–132 (2017). 167. Gu, T.-P. et al. The role of Tet3 DNA dioxygenase in Nat. Genet. 50, 96–105 (2018).
141. Zamudio, N. et al. DNA methylation restrains epigenetic reprogramming by oocytes. Nature 477, 196. Gama-​Sosa, M. A. et al. Tissue-​specific differences in
transposons from adopting a chromatin signature 606–610 (2011). DNA methylation in various mammals. Biochim.
permissive for meiotic recombination. Genes Dev. 29, 168. Wossidlo, M. et al. 5-Hydroxymethylcytosine in the Biophys. Acta 740, 212–219 (1983).
1256–1270 (2015). mammalian zygote is linked with epigenetic 197. Hon, G. C. et al. Epigenetic memory at embryonic
142. Nishibuchi, G. & Déjardin, J. The molecular basis of the reprogramming. Nat. Commun. 2, 241 (2011). enhancers identified in DNA methylation maps from
organization of repetitive DNA-​containing constitutive 169. Iqbal, K., Jin, S.-G., Pfeifer, G. P. & Szabo, P. E. adult mouse tissues. Nat. Genet. 45, 1198–1206
heterochromatin in mammals. Chromosom. Res. 25, Reprogramming of the paternal genome upon (2013).
77–87 (2017). fertilization involves genome-​wide oxidation of 198. Seisenberger, S. et al. The dynamics of genome-​wide
143. Hutnick, L. K., Huang, X., Loo, T.-C., Ma, Z. & Fan, G. 5-methylcytosine. Proc. Natl Acad. Sci. USA 108, DNA methylation reprogramming in mouse primordial
Repression of retrotransposal elements in mouse 3642–3647 (2011). germ cells. Mol. Cell 48, 849–862 (2012).
embryonic stem cells is primarily mediated by a DNA 170. Howell, C. Y. et al. Genomic imprinting disrupted by a 199. Schultz, M. D. et al. Human body epigenome maps
methylation-​independent mechanism. J. Biol. Chem. maternal effect mutation in the Dnmt1 gene. Cell 104, reveal noncanonical DNA methylation variation.
285, 21082–21091 (2010). 829–838 (2001). Nature 523, 212–216 (2015).
144. Matsui, T. et al. Proviral silencing in embryonic stem 171. Wang, L. et al. Programming and inheritance of 200. Ziller, M. J. et al. Charting a dynamic DNA methylation
cells requires the histone methyltransferase ESET. parental DNA methylomes in mammals. Cell 157, landscape of the human genome. Nature 500,
Nature 464, 927 (2010). 979–991 (2014). 477–481 (2013).
145. Walter, M. et al. An epigenetic switch ensures 172. Guo, F. et al. Active and passive demethylation of male 201. Sardina, J. L. et al. Transcription factors drive Tet2-
transposon repression upon dynamic loss of DNA and female pronuclear DNA in the mammalian zygote. mediated enhancer demethylation to reprogram cell
methylation in embryonic stem cells. eLife 5, e11418 Cell Stem Cell 15, 447–459 (2014). fate. Cell Stem Cell 23, 727–741 (2018).
(2016). 173. Shen, L. et al. Tet3 and DNA replication mediate 202. Rasmussen, K. D. et al. TET2 binding to enhancers
146. Sharif, J. et al. Activation of endogenous retroviruses demethylation of both the maternal and paternal facilitates transcription factor recruitment in
in Dnmt1−/− ESCs involves disruption of SETDB1- genomes in mouse zygotes. Cell Stem Cell 15, hematopoietic cells. Genome Res. 29, 564–575
mediated repression by NP95 binding to 459–470 (2014). (2019).
hemimethylated DNA. Cell Stem Cell 19, 81–94 174. Amouroux, R. et al. De novo DNA methylation drives 203. Lister, R. et al. Global epigenomic reconfiguration
(2016). 5hmC accumulation in mouse zygotes. Nat. Cell Biol. during mammalian brain development. Science 341,
147. Berrens, R. V. et al. An endosiRNA-​based repression 18, 225–233 (2016). 6146 (2013).
mechanism counteracts transposon activation during 175. Santos, F. et al. Active demethylation in mouse 204. Guo, J. U. et al. Distribution, recognition and
global DNA demethylation in embryonic stem cells. zygotes involves cytosine deamination and base regulation of non-​CpG methylation in the adult
Cell Stem Cell 21, 694–703 (2017). excision repair. Epigenetics Chromatin 6, 39 (2013). mammalian brain. Nat. Neurosci. 17, 215–222
148. Bender, C. M. et al. Roles of cell division and gene 176. Smith, Z. D. et al. DNA methylation dynamics of the (2014).
transcription in the methylation of CpG islands. human preimplantation embryo. Nature 511, 205. Gabel, H. W. et al. Disruption of DNA-​methylation-
Mol. Cell. Biol. 19, 6690–6698 (1999). 611–615 (2014). dependent long gene repression in Rett syndrome.
149. Varley, K. E. et al. Dynamic DNA methylation across 177. Okae, H. et al. Genome-​wide analysis of DNA Nature 522, 89–93 (2015).
diverse human cell lines and tissues. Genome Res. 23, methylation dynamics during early human 206. Lagger, S. et al. MeCP2 recognizes cytosine
555–567 (2013). development. PLOS Genet. 10, e1004868 (2014). methylated tri-​nucleotide and di-​nucleotide sequences
150. Laurent, L. et al. Dynamic changes in the human 178. Guo, H. et al. The DNA methylation landscape of to tune transcription in the mammalian brain.
methylome during differentiation. Genome Res. 20, human early embryos. Nature 511, 606 (2014). PLOS Genet. 13, e1006793 (2017).
320–331 (2010). 179. Zhu, P. et al. Single-​cell DNA methylome sequencing 207. Stroud, H. et al. Early-​life gene expression in neurons
151. Gelfman, S., Cohen, N., Yearim, A. & Ast, G. of human preimplantation embryos. Nat. Genet. 50, modulates lasting epigenetic states. Cell 171,
DNA-methylation effect on cotranscriptional splicing 12–19 (2018). 1151–1164 (2017).
is dependent on GC architecture of the exon-​intron 180. Hill, S. et al. Epigenetic reprogramming enables the 208. Ziller, M. J. et al. Dissecting the functional
structure. Genome Res. 23, 789–799 (2013). transition from primordial germ cell to gonocyte. consequences of de novo DNA methylation dynamics
152. Shayevitch, R., Askayo, D., Keydar, I. & Ast, G. Nature 555, 392–396 (2018). in human motor neuron differentiation and physiology.
The importance of DNA methylation of exons on 181. Skvortsova, K., Iovino, N. & Bogdanović, O. Cell Stem Cell 22, 559–574 (2018).
alternative splicing. RNA 24, 1351–1362 (2018). Functions and mechanisms of epigenetic inheritance 209. Klein, C. J. et al. Mutations in DNMT1 cause
153. Yang, X. et al. Gene body methylation can alter gene in animals. Nat. Rev. Mol. Cell Biol. 19, 774–790 hereditary sensory neuropathy with dementia and
expression and is a therapeutic target in cancer. (2018). hearing loss. Nat. Genet. 43, 595 (2011).
Cancer Cell 26, 577–590 (2014). 182. Duffié, R. et al. The Gpr1/Zdbf2 locus provides new 210. Winkelmann, J. et al. Mutations in DNMT1 cause
154. Shukla, S. et al. CTCF-​promoted RNA polymerase II paradigms for transient and dynamic genomic autosomal dominant cerebellar ataxia, deafness and
pausing links DNA methylation to splicing. Nature imprinting in mammals. Genes Dev. 28, 463–478 narcolepsy. Hum. Mol. Genet. 21, 2205–2210
479, 74–79 (2011). (2014). (2012).

606 | OCTOBER 2019 | volume 20 www.nature.com/nrm

 M E T H Y L AT I O N I N B I O L O G Y
211. Baets, J. et al. Defects of mutant DNMT1 are linked mutation. Am. J. Med. Genet. Part A 173, 250–253 silencing in breast cancer. Genome Res. 22, 246–258
to a spectrum of neurological disorders. Brain 138, (2017). (2012).
845–861 (2015). 233. Langemeijer, S. M. C. et al. Acquired mutations in 255. Bahar Halpern, K., Vana, T. & Walker, M. D.
212. Sun, Z. et al. Aberrant signature methylome by TET2 are common in myelodysplastic syndromes. Paradoxical role of DNA methylation in activation
DNMT1 hot spot mutation in hereditary sensory Nat. Genet. 41, 838–842 (2009). of FoxA2 gene expression during endoderm
and autonomic neuropathy 1E AU. Epigenetics 9, 234. Cimmino, L. et al. Restoration of TET2 function blocks development. J. Biol. Chem. 289, 23882–23892
1184–1193 (2014). aberrant self-​renewal and leukemia progression. Cell (2014).
213. Kernohan, K. D. et al. Identification of a methylation 170, 1079–1095 (2017). 256. Takahashi, Y. et al. Integration of CpG-​free DNA
profile for DNMT1-associated autosomal dominant 235. Ko, M. et al. Impaired hydroxylation of induces de novo methylation of CpG islands in
cerebellar ataxia, deafness, and narcolepsy. 5-methylcytosine in myeloid cancers with mutant pluripotent stem cells. Science 356, 503–508 (2017).
Clin. Epigenetics 8, 91 (2016). TET2. Nature 468, 839–843 (2010). 257. Li, F. et al. Chimeric DNA methyltransferases target
214. Jeanpierre, M. et al. An embryonic-​like methylation 236. Rasmussen, K. D. et al. Loss of TET2 in hematopoietic DNA methylation to specific DNA sequences and
pattern of classical satellite DNA is observed in ICF cells leads to DNA hypermethylation of active repress expression of target genes. Nucleic Acids Res.
syndrome. Hum. Mol. Genet. 2, 731–735 (1993). enhancers and induction of leukemogenesis. Genes Dev. 35, 100–112 (2007).
215. Xu, G. L. et al. Chromosome instability and 29, 910–922 (2015). 258. Bernstein, D. L., Le Lay, J. E., Ruano, E. G. &
immunodeficiency syndrome caused by mutations in a 237. Madzo, J. et al. Hydroxymethylation at gene Kaestner, K. H. TALE-​mediated epigenetic suppression
DNA methyltransferase gene. Nature 402, 187–191 regulatory regions directs stem/early progenitor cell of CDKN2A increases replication in human fibroblasts.
(1999). commitment during erythropoiesis. Cell Rep. 6, J. Clin. Invest. 125, 1998–2006 (2015).
216. Thijssen, P. E. et al. Mutations in CDCA7 and HELLS 231–244 (2014). 259. Maeder, M. L. et al. Targeted DNA demethylation and
cause immunodeficiency–centromeric instability–facial 238. Spruijt, C. G. et al. Dynamic readers for 5-(hydroxy) activation of endogenous genes using programmable
anomalies syndrome. Nat. Commun. 6, 7870 (2015). methylcytosine and its oxidized derivatives. Cell 152, TALE-​TET1 fusion proteins. Nat. Biotechnol. 31,
217. de Greef, J. C. et al. Mutations in ZBTB24 are 1146–1159 (2013). 1137–1142 (2013).
associated with immunodeficiency, centromeric 239. Zhang, X. et al. DNMT3A and TET2 compete and 260. Valton, J. et al. Overcoming transcription activator-​
instability, and facial anomalies syndrome type 2. cooperate to repress lineage-​specific transcription like effector (TALE) DNA binding domain sensitivity
Am. J. Hum. Genet. 88, 796–804 (2011). factors in hematopoietic stem cells. Nat. Genet. 48, to cytosine methylation. J. Biol. Chem. 287,
218. Velasco, G. et al. Comparative methylome analysis 1014–1023 (2016). 38427–38432 (2012).
of ICF patients identifies heterochromatin loci that 240. Turcan, S. et al. IDH1 mutation is sufficient to establish 261. Vojta, A. et al. Repurposing the CRISPR-​Cas9 system
require ZBTB24, CDCA7 and HELLS for their the glioma hypermethylator phenotype. Nature 483, for targeted DNA methylation. Nucleic Acids Res 44,
methylated state. Hum. Mol. Genet. 27, 2409–2424 479–483 (2012). 5615–28 (2016).
(2018). 241. Xiao, M. et al. Inhibition of α-​KG-dependent histone 262. Liu, X. S. et al. Editing DNA methylation in the
219. Ueda, Y. et al. Roles for Dnmt3b in mammalian and DNA demethylases by fumarate and succinate that mammalian genome. Cell 167, 233–247 (2016).
development: a mouse model for the ICF syndrome. are accumulated in mutations of FH and SDH tumor 263. Lei, Y. et al. Targeted DNA methylation in vivo using an
Development 133, 1183–1192 (2006). suppressors. Genes Dev. 26, 1326–1338 (2012). engineered dCas9-MQ1 fusion protein. Nat. Commun.
220. Wu, H. et al. Converging disease genes in ICF 242. Letouzé, E. et al. SDH mutations establish a 8, 16026 (2017).
syndrome: ZBTB24 controls expression of CDCA7 hypermethylator phenotype in paraganglioma. 264. Stepper, P. et al. Efficient targeted DNA methylation
in mammals. Hum. Mol. Genet. 25, 4041–4051 Cancer Cell 23, 739–752 (2013). with chimeric dCas9-Dnmt3a-​Dnmt3L
(2016). 243. Killian, J. K. et al. Succinate dehydrogenase methyltransferase. Nucleic Acids Res. 45,
221. Rajshekar, S. et al. Pericentromeric hypomethylation mutation underlies global epigenomic divergence in 1703–1713 (2017).
elicits an interferon response in an animal model of gastrointestinal stromal tumor. Cancer Discov. 3, 265. Huang, Y.-H. et al. DNA epigenome editing using
ICF syndrome. eLife 7, e39658 (2018). 648–657 (2013). CRISPR-​Cas SunTag-​directed DNMT3A. Genome Biol.
222. Heyn, P. et al. Gain-​of-function DNMT3A mutations 244. Zhang, H., Lang, Z. & Zhu, J.-K. Dynamics and function 18, 176 (2017).
cause microcephalic dwarfism and hypermethylation of DNA methylation in plants. Nat. Rev. Mol. Cell Biol. 266. Pflueger, C. et al. A modular dCas9-SunTag DNMT3A
of Polycomb-​regulated regions. Nat. Genet. 51, 19, 489–506 (2018). epigenome editing system overcomes pervasive
96–105 (2019). 245. Bewick, A. J. et al. Dnmt1 is essential for egg off-target activity of direct fusion dCas9-DNMT3A
223. Li, Y. et al. Genome-​wide analyses reveal a role of production and embryo viability in the large milkweed constructs. Genome Res. 28, 1193–1206 (2018).
Polycomb in promoting hypomethylation of DNA bug, Oncopeltus fasciatus. Epigenetics Chromatin 12, 267. Morita, S. et al. Targeted DNA demethylation in vivo
methylation valleys. Genome Biol. 19, 18 (2018). 6 (2019). using dCas9-peptide repeat and scFv-​TET1 catalytic
224. Jeong, M. et al. Large conserved domains of low DNA 246. Hu, S. et al. DNA methylation presents distinct domain fusions. Nat. Biotechnol. 34, 1060–1065
methylation maintained by Dnmt3a. Nat. Genet. 46, binding sites for human transcription factors. eLife 2, (2016).
17–23 (2014). e00726 (2013). 268. Amabile, A. et al. Inheritable silencing of endogenous
225. Sendžikaitė, G., Hanna, C. W., Stewart-​Morgan, K. R., 247. Liu, Y. et al. Structural basis for Klf4 recognition of genes by hit-​and-run targeted epigenetic editing. Cell
Ivanova, E. & Kelsey, G. A DNMT3A PWWP mutation methylated DNA. Nucleic Acids Res. 42, 4859–4867 167, 219–232 (2016).
leads to methylation of bivalent chromatin and growth (2014). 269. Saunderson, E. A. et al. Hit-​and-run epigenetic editing
retardation in mice. Nat. Commun. 10, 1884 (2019). 248. Hashimoto, H. et al. Distinctive Klf4 mutants prevents senescence entry in primary breast cells from
226. Tatton-​Brown, K. et al. Mutations in the DNA determine preference for DNA methylation status. healthy donors. Nat. Commun. 8, 1450 (2017).
methyltransferase gene DNMT3A cause an Nucleic Acids Res. 44, 10177–10185 (2016).
overgrowth syndrome with intellectual disability. 249. Rishi, V. et al. CpG methylation of half-​CRE sequences Acknowledgements
Nat. Genet. 46, 385–388 (2014). creates C/EBPα binding sites that activate some The laboratory of D.B. is part of the Laboratoire d’Excellence
227. Challen, G. A. et al. Dnmt3a is essential for tissue-​specific genes. Proc. Natl Acad. Sci. USA 107, (LABEX) entitled DEEP (11-LBX0044) and is supported by the
hematopoietic stem cell differentiation. Nat. Genet. 20311–20316 (2010). European Research Council (ERC) (grant ERC-​Cog EpiRepro).
44, 23–31 (2012). 250. Tanay, A., O’Donnell, A. H., Damelin, M. & Bestor, T. H.
228. Wu, H. et al. Dnmt3a-​dependent nonpromoter DNA Hyperconserved CpG domains underlie Polycomb-​ Author contributions
methylation facilitates transcription of neurogenic binding sites. Proc. Natl Acad. Sci. USA 104, The authors contributed equally to all aspects of the article.
genes. Science 329, 444–448 (2010). 5521–5526 (2007).
229. Tatton-​Brown, K. et al. Mutations in epigenetic 251. Brinkman, A. B. et al. Sequential ChIP-​bisulfite Competing interests
regulation genes are a major cause of overgrowth sequencing enables direct genome-​scale investigation The authors declare no competing interests.
with intellectual disability. Am. J. Hum. Genet. 100, of chromatin and DNA methylation cross-​talk. Genome
725–736 (2017). Res. 22, 1128–1138 (2012). Peer review information
230. Russler-​Germain, D. A. et al. The R882H DNMT3A 252. Statham, A. L. et al. Bisulfite sequencing of chromatin Nature Reviews Molecular Cell Biology thanks G. Kelsey,
mutation associated with AML dominantly inhibits immunoprecipitated DNA (BisChIP-​seq) directly R. Lister and T. Nakano for their contribution to the peer
wild-​type DNMT3A by blocking its ability to form informs methylation status of histone-​modified DNA. review of this work.
active tetramers. Cancer Cell 25, 442–454 (2014). Genome Res. 22, 1120–1127 (2012).
231. Spencer, D. H. et al. CpG island hypermethylation 253. Jermann, P., Hoerner, L., Burger, L. & Schubeler, D. Publisher’s note
mediated by DNMT3A is a consequence of AML Short sequences can efficiently recruit histone H3 Springer Nature remains neutral with regard to jurisdictional
progression. Cell 168, 801–816 (2017). lysine 27 trimethylation in the absence of enhancer claims in published maps and institutional affiliations.
232. Kosaki, R., Terashima, H., Kubota, M. & Kosaki, K. activity and DNA methylation. Proc. Natl Acad.
Acute myeloid leukemia-​associated DNMT3A p. Sci. USA 111, E3415–E3421 (2014). Supplementary information
Arg882His mutation in a patient with Tatton-​Brown– 254. Hon, G. C. et al. Global DNA hypomethylation coupled Supplementary information is available for this paper at
Rahman overgrowth syndrome as a constitutional to repressive chromatin domain formation and gene https://doi.org/10.1038/s41580-019-0159-6.

NATuRE REvIEWS | MolECulAr CEll BIoloGy volume 20 | OCTOBER 2019 | 607