NATUREIVol 447 24 May 2007 do:10.1038/nature05918 INSIGHT REVIEW Stability and flexibility of epigenetic gene regulation in mammalian development Wolf Reik During development cells start in a pluripotent state from which they can differentiate into many cell types, and progressively develop a narrower potential. their gene-expression programmes become more defined, restricted and potentially, 'locked in. Pluripotent stem cells express genes that encode a set of core transcription factors while genes that are required later in development are repressed by histone marks, which confer short-term, and therefore flexible, epigenetic silencing. by contrast the methylation of DNa confers long-term epigenetic silencing of particular sequences-transposons, imprinted genes and pluripotency-associated genes-in somatic cells. Long-term silencing can be reprogrammed by demethylation of DNA and this process might involve dNa repair. It is not known whether any of the epigenetic marks has a primary role in determining cell and lineage commitment during development. F%832Orht is, by definition, epigenetic Differences in the pro- epigenetic marks(which can be removed before a cell divides or within Develo f gene expression that result in the development of different very few cell divisions)with the long-term stability and heritability of organs and tissues occur without changes to the sequence of our DNA other marks(which can be maintained for many divisions)(Fig. 1) one or two exceptions). There is nothing mysterious in this con- During the early stages of development, genes that are required later in ubsets of the -30,000 genes in our genome are active in different development are transiently held in a repressed state by histone modifi tissues and organs, depending on their regulation by different sets or cations, which are highly flexible and easily reversed when expression of combinations of transcription factors. This implies that if we were to these genes is needed. During differentiation that are crucial for take all of the transcription factors that activate genes in a liver cell and pluripotency are silenced by histone modifications, as well as by dnA transfer them to a brain cell (while inactivating all brain-specific tran- methylation. Some of these genes are also silent in mature germ cell scription factors), then the brain cell would turn into a liver cell. meaning that epigenetic marks probably need to be reversed rapidly after A recent study provides tantalizing insight into this concept of fertilization to allow re-expression of pluripotency-associated genes i epigenetic control of development. Takahashi and Yamanaka identi- the next generation. By contrast, long-term silencing of transposons fied four transcriptional regulators that when expressed in fibroblasts, and imprinted genes-which is based on DNA methylation-need resulted in these cells being reprogrammed to become embryonic stem to be stably maintained from the gametes into the early embryo and the (ES)-like cells Extending this concept a little further, in somatic-cell adult organism. Methylation of imprinted genes can only be erased nuclear transfer, the nucleus of a somatic cell from an adult individual is primordial germ cells(PGCs), the cells that ultimately give rise to the transplanted into an oocyte from which the nucleus has been removed, germ line. Probably because there is a requirement for both removing resulting in reprogramming of the adult nucleus and therefore successful epigenetic marks and retaining epigenetic marks between generations, development of the cloned animal Cloning, however, is inefficient, because most(if not all) cloned erations. In this review, I address how the fascinating interplay betwee animals have epigenetic defects, particularly in DNA methylation. transcription factors and epigenetic factors is beginning to provide an Therefore, our lack of understanding ofhow epigenetic marks are repro- explanation for how pluripotency and development are regulated. grammed is a key obstacle to cloning. Similarly, the reprogramming of fibroblasts to become ES-like cells is a rare event in vitro, and epigenetic Flexibility for developmental gene regulation defects such as lack of demethylation of the Oct4(also known as Pou5f1) In this section, three issues are addressed. First, are differentiation promoter, affecting expression of the encoded transcription factor, have specific genes held in an epigenetically silenced manner in pluripotent been noted in these ES-like cell cell types, in order to be activated later? And is the removal of epigenetic These observations highlight that, in addition to transcription fac- marks from these genes needed for their activation? Second,are tors, changes in gene expression during development are accompanied pluripotency-associated genes epigenetically inactivated in differentiated or caused by epigenetic modifications", such as methylation of DNA at cell types? This inactivation could, in principle, be irreversible, because CpG sequences(in vertebrates), modification of histone tails and the somatic cell types are not required to give rise to pluripotent cells. One presence of non-nucleosomal chromatin-associated proteins. Therefore, exception is the germ line, where reactivation of pluripotency-associated as development and differentiation proceed, differentiated cells accumu- genes is needed at the initial stages of development; however, later, the late epigenetic marks that differ from those of pluripotent cells, and dif- silencing of these genes is essential for the differentiation of mature germ ferentiated cells of different lineages also accumulate different marks. cells. And therefore, third, is the removal of permanent silencing marks In this review, I focus on the role of epigenetic regulation in devel- from the gametic genomes after fertilization crucial to activate essential opment, particularly comparing the short-term flexibility of certain genes, such as pluripotency-associated genes, early in development? Laborato ory of Developmen ntal Genetics and Imprinting aham institute, Cambridge CB22 3AT, UK. @2007 Nature Publishing Group
Development is, by definition, epigenetic. Differences in the programmes of gene expression that result in the development of different organs and tissues occur without changes to the sequence of our DNA (with one or two exceptions). There is nothing mysterious in this concept; subsets of the ~30,000 genes in our genome are active in different tissues and organs, depending on their regulation by different sets or combinations of transcription factors. This implies that if we were to take all of the transcription factors that activate genes in a liver cell and transfer them to a brain cell (while inactivating all brain-specific transcription factors), then the brain cell would turn into a liver cell. A recent study provides tantalizing insight into this concept of epigenetic control of development. Takahashi and Yamanaka identified four transcriptional regulators that when expressed in fibroblasts, resulted in these cells being reprogrammed to become embryonic stem (ES)-like cells1 . Extending this concept a little further, in somatic-cell nuclear transfer, the nucleus of a somatic cell from an adult individual is transplanted into an oocyte from which the nucleus has been removed, resulting in reprogramming of the adult nucleus and therefore successful development of the cloned animal. Cloning, however, is inefficient, because most (if not all) cloned animals have epigenetic defects, particularly in DNA methylation. Therefore, our lack of understanding of how epigenetic marks are reprogrammed is a key obstacle to cloning2 . Similarly, the reprogramming of fibroblasts to become ES-like cells is a rare event in vitro, and epigenetic defects such as lack of demethylation of the Oct4 (also known as Pou5f1) promoter, affecting expression of the encoded transcription factor, have been noted in these ES-like cells1 . These observations highlight that, in addition to transcription factors, changes in gene expression during development are accompanied or caused by epigenetic modifications2–7, such as methylation of DNA at CpG sequences (in vertebrates4,5), modification of histone tails6 and the presence of non-nucleosomal chromatin-associated proteins7 . Therefore, as development and differentiation proceed, differentiated cells accumulate epigenetic marks that differ from those of pluripotent cells, and differentiated cells of different lineages also accumulate different marks. In this review, I focus on the role of epigenetic regulation in development, particularly comparing the short-term flexibility of certain epigenetic marks (which can be removed before a cell divides or within very few cell divisions) with the long-term stability and heritability of other marks (which can be maintained for many divisions) (Fig. 1). During the early stages of development, genes that are required later in development are transiently held in a repressed state by histone modifications, which are highly flexible and easily reversed when expression of these genes is needed. During differentiation, genes that are crucial for pluripotency are silenced by histone modifications, as well as by DNA methylation. Some of these genes are also silent in mature germ cells, meaning that epigenetic marks probably need to be reversed rapidly after fertilization to allow re-expression of pluripotency-associated genes in the next generation. By contrast, long-term silencing of transposons and imprinted genes — which is based on DNA methylation — needs to be stably maintained from the gametes into the early embryo and the adult organism. Methylation of imprinted genes can only be erased in primordial germ cells (PGCs), the cells that ultimately give rise to the germ line. Probably because there is a requirement for both removing epigenetic marks and retaining epigenetic marks between generations, epigenetic information can sometimes be inherited across multiple generations. In this review, I address how the fascinating interplay between transcription factors and epigenetic factors is beginning to provide an explanation for how pluripotency and development are regulated. Flexibility for developmental gene regulation In this section, three issues are addressed. First, are differentiationspecific genes held in an epigenetically silenced manner in pluripotent cell types, in order to be activated later? And is the removal of epigenetic marks from these genes needed for their activation? Second, are pluripotency-associated genes epigenetically inactivated in differentiated cell types? This inactivation could, in principle, be irreversible, because somatic cell types are not required to give rise to pluripotent cells. One exception is the germ line, where reactivation of pluripotency-associated genes is needed at the initial stages of development; however, later, the silencing of these genes is essential for the differentiation of mature germ cells. And therefore, third, is the removal of ‘permanent’ silencing marks from the gametic genomes after fertilization crucial to activate essential genes, such as pluripotency-associated genes, early in development? Stability and flexibility of epigenetic gene regulation in mammalian development Wolf Reik1 During development, cells start in a pluripotent state, from which they can differentiate into many cell types, and progressively develop a narrower potential. Their gene-expression programmes become more defined, restricted and, potentially, ‘locked in’. Pluripotent stem cells express genes that encode a set of core transcription factors, while genes that are required later in development are repressed by histone marks, which confer short-term, and therefore flexible, epigenetic silencing. By contrast, the methylation of DNA confers long-term epigenetic silencing of particular sequences — transposons, imprinted genes and pluripotency-associated genes — in somatic cells. Long-term silencing can be reprogrammed by demethylation of DNA, and this process might involve DNA repair. It is not known whether any of the epigenetic marks has a primary role in determining cell and lineage commitment during development. 1 Laboratory of Developmental Genetics and Imprinting, The Babraham Institute, Cambridge CB22 3AT, UK. 425 NATURE|Vol 447|24 May 2007|doi:10.1038/nature05918 INSIGHT REVIEW
INSIGHT REVIEW NATURE Vol 447 24 May 2007 There is recent evidence for the first type of epigenetic regulation: dentified demethylase)"2. The H3K27 methylation mark occurs mostly that is, the temporary inactivation of differentiation-specific genes in outside the context of DNA methylation. In contrast to the terminal pluripotent cell types(Fig 2a). Genes that are required during develop- silencing achieved by DNA methylation(discussed later), developmental ment and differentiation -for example, those in the homeobox(Hox), genes that are silenced by PRCs in pluripotent tissues require repressive distal-less homeobox(Dlx), paired box(Pax)and sine-oculis-related marks to be rapidly and flexibly removed when differentiation begins homeobox(Six) gene families-are held repressed in pluripotent Es Strikingly, in cancer cells, the genes targeted by PRCs often become cells by the Polycomb group(PcG)-protein repressive system in mice DNA methylated, which might result in a more permanent locking- and humans. This system marks the histones associated with these genes of a pluripotent state in cancer stem cells by inducing methylation of the lysine residue at position 27 of histone The second type of epigenetic regulation to be considered is whether H3(H3K27)-. ES cells that lack EED(embryonic ectoderm develop- pluripotency-associated genes are epigenetically inactivated in differen- ment), a component of the PcG-protein repressive complex(PRC), have tiated cell types. Several genes that are required for early development partly derepressed developmental genes and are prone to spontaneous or for germ-cell development only-for example, those that encode differentiation o. Interestingly, some developmental genes are present pluripotency-sustaining transcription factors(such as OCT4 and within bivalent chromatin regions, which contain both inactivating NANOG)-are known to be expressed by ES cells but silenced on the marks(methylated H3K27) and activating marks(methylated H3K4). differentiation of these cells, with a defined kinetics of acquiring repres This could indicate that after the repressive marks have been removed sive histone modifications and DNA methylation"(Fig. 2b). Silencing (when expression of the components of PRCs are downregulated dur- by both histone modifications and DNA methylation in somatic tissues ing differentiation), these genes are automatically poised for transcrip- seems to be typical of this group of genes and of those that encode can tional activation through the H3K4 methylation mark. It is important cer-testis antigens, which are expressed during spermatogenesis>. It is to note that epigenetic silencing by PRCs might be mitotically heritable probable that this permanent type of epigenetic silencing safeguards se genes in differentiated cells, because be rapidly removed by enzymatic demethylation of H3K27(by an uni- that might lead to dedifferentiation and, perhaps, to a predisposition to DNA methylation 8E88 H3K27 methylation H3K4 methylation Pluripotency-associated genes Developmental genes Oocyte + Adult Germ cells cell mass DNA methylation H3K27 methylation Developmental genes Figure 1 Epigenetic gene regulation during mammalian development. hylation. During the early development of PGCs, DNA methylation and Key developmental events are shown together with global epigenetic repressive histone modifications(such as H3K9 methylation)are also erased. modifications and gene-expression patterns. Very early in development, Pluripotency-associated genes are re-expressed during a time window that DNA methylation is erased. In addition, pluripotency-associated genes begin onic germ cells to be derived in culture. Imprinted genes ar to be expressed, and developmental genes pressed by the PcG prote during this period, and developmental ge system and H3K27 methylation. During the differentiation of pluripotent lexible histone marks such as H3K27 methylation enable enes to be silenced for a short time in pluripotent cells. By otentially permanently, as a result of DNA methylation. At the same time, ethylation enables the stable silencing of imprinted g developmental genes begin to be expressed, and there is an increase in H3K4 transposons and some pluripotency-associated genes @2007 Nature Publishing Group
There is recent evidence for the first type of epigenetic regulation: that is, the temporary inactivation of differentiation-specific genes in pluripotent cell types (Fig. 2a). Genes that are required during development and differentiation — for example, those in the homeobox (Hox), distal-less homeobox (Dlx), paired box (Pax) and sine-oculis-related homeobox (Six) gene families — are held repressed in pluripotent ES cells by the Polycomb group (PcG)-protein repressive system in mice and humans. This system marks the histones associated with these genes by inducing methylation of the lysine residue at position 27 of histone H3 (H3K27)8–10. ES cells that lack EED (embryonic ectoderm development), a component of the PcG-protein repressive complex (PRC), have partly derepressed developmental genes and are prone to spontaneous differentiation8,10. Interestingly, some developmental genes are present within ‘bivalent’ chromatin regions, which contain both inactivating marks (methylated H3K27) and activating marks (methylated H3K4)9,11. This could indicate that after the repressive marks have been removed (when expression of the components of PRCs are downregulated during differentiation), these genes are automatically poised for transcriptional activation through the H3K4 methylation mark. It is important to note that epigenetic silencing by PRCs might be mitotically heritable (through an unknown mechanism)7 , but these marks could presumably be rapidly removed by enzymatic demethylation of H3K27 (by an unidentified demethylase)12. The H3K27 methylation mark occurs mostly outside the context of DNA methylation. In contrast to the terminal silencing achieved by DNA methylation (discussed later), developmental genes that are silenced by PRCs in pluripotent tissues require repressive marks to be rapidly and flexibly removed when differentiation begins. Stri kingly, in cancer cells, the genes targeted by PRCs often become DNA methylated, which might result in a more permanent locking-in of a ‘pluripotent’ state in cancer stem cells13. The second type of epigenetic regulation to be considered is whether pluripotency-associated genes are epigenetically inactivated in differentiated cell types. Several genes that are required for early development or for germ-cell development only — for example, those that encode pluripotency-sustaining transcription factors (such as OCT4 and NANOG) — are known to be expressed by ES cells but silenced on the differentiation of these cells, with a defined kinetics of acquiring repressive histone modifications and DNA methylation14 (Fig. 2b). Silencing by both histone modifications and DNA methylation in somatic tissues seems to be typical of this group of genes and of those that encode cancer–testis antigens, which are expressed during spermatogenesis15. It is probable that this permanent type of epigenetic silencing safeguards against accidental expression of these genes in differentiated cells, because that might lead to dedifferentiation and, perhaps, to a predisposition to Oocyte DNA methylation H3K27 methylation DNA methylation H3K27 methylation H3K4 methylation Pluripotency-associated genes Developmental genes Pluripotencyassociated genes Developmental genes Zygote Morula Blastocyst Embryo Adult Germ cells PGCs ES cells Embryonic germ cells Inner cell mass Embryo and somatic cells Germ cells Sperm Figure 1 | Epigenetic gene regulation during mammalian development. Key developmental events are shown together with global epigenetic modifications and gene-expression patterns. Very early in development, DNA methylation is erased. In addition, pluripotency-associated genes begin to be expressed, and developmental genes are repressed by the PcG protein system and H3K27 methylation. During the differentiation of pluripotent cells such as ES cells, pluripotency-associated genes are repressed, potentially permanently, as a result of DNA methylation. At the same time, developmental genes begin to be expressed, and there is an increase in H3K4 methylation. During the early development of PGCs, DNA methylation and repressive histone modifications (such as H3K9 methylation) are also erased. Pluripotency-associated genes are re-expressed during a time window that allows embryonic germ cells to be derived in culture. Imprinted genes are demethylated during this period, and developmental genes are expressed afterwards. Flexible histone marks such as H3K27 methylation enable developmental genes to be silenced for a short time in pluripotent cells. By contrast, DNA methylation enables the stable silencing of imprinted genes, transposons and some pluripotency-associated genes. 426 INSIGHT REVIEW NATURE|Vol 447|24 May 2007
NATUREIVol 447 24 May 2007 INSIGHT REVIEW cancer. Consequently, these genes are difficult to reactivate in cloned this involves demethylation of DNA PGCs at these stages have similar embryos because of inefficient reprogramming of repressive marks, par- properties to pluripotent cells, including the ability to form embryonic ularly of DNA methylation germ cells in culture". These studies are important because they are c Special epigenetic regulation needs to occur in PGCs developing in the first to show that in some developmental situations, removal of the early post-implantation embryo. Because these cells emerge from epigenetic marks( H3K27 methylation in the ES-cell study, and DNA ell types in the egg cylinder that are already on the way to lineage com- methylation in the PGC study) could be crucial for the activation of mitment and differentiation, the somatic gene-expression programme developmental genes. Whether DNA methylation in PGCs is erased needs to be suppressed. One of the key regulators of this process is by an active or a passive mechanism is unclear(discussed later). The BLIMPI(B-lymphocyte-induced maturation protein 1), which associ- promoters of the genes that undergo developmental demethylation tes with the arginine methyltransferase PRMT5 PRMT5 might partly (for example, Mvh, Dazl and Sycp3)contain CpG islands, as do the dif- repress Hox-family genes and other somatic genes in PGCs(Fig. 2c). ferentially methylated regions(DMRs)of imprinted genes, which als Pluripotency-associated genes and genes that have later roles in germ- undergo demethylation at these stages of PGC development. I am not cell development can also be repressed by DNA methylation(Fig 2b). aware of any reports of demethylation of CpG islands during develop So genes such as Mwh(also known as Ddx4), Dazl( deleted in azoo- ment other than in PGCs or in the zygote and pre- implantation embryo permia-like)and Sycp3(synaptonemal complex protein 3)are meth-(discussed later). Methylation of CpG islands might only be removable ylated in early PGCs and begin to be expressed after the erasure of DNA under exceptional circumstances. methylation which occurs between embryonic day(E)8.0 and E12.5 Some key pluripotency-associated genes(such as Oct4 and Nanog) in PGCs. Interestingly, pluripotency-associated genes such as Nanog are epigenetically inactivated at later stages of gametogenesis and in also begin to be reactivated at these stages, but it is not known whether the mature gametes, including by DNA methylation. Therefore, after Temporary repression of developmental genes by the peG protein system Histone demethylase? Repression of pluripotency-associated genes by histone methylation and dNA methylation Pluripotent Differentiated c Maintenance of silencing of somatic genes in early germ cells rly PGC Histone demethylase? Histone methyltransferase tion of b, Pluripotency-associated genes are stably silenced during genes during the differentiation of somatic cells and differentiation, through histone methylation and DNA methylation germ cells. The expression or repr h as oct and d developmental genes is indicated, and the associated modifications ES-cell differentiation, and this process can involve both histone of the histone tails and/or DNA are represented by different colours. methylation(such as methylation of H3k9 mediated by G9A; also a, In pluripotent cells, the repression of genes that are needed later in known as EHMT2)(green) and DNA methylation(red). Whethera development is flexible and can involve the PcG-protein repressive histone demethylase is required for the removal of H3K4 methylation system. Silent developmental genes an be marked by both H3K27 is unknown. c, For germ-cell development, the repression of somatic methylation (yellow)and H3K4 methylation(blue), possibly allowing genes needs to be maintained in early germ cells, and this process might rapid gene activation after loss of repression by PcG-protein-containi involve histone arginine methylation(pink). Hox-family genes and other developmental genes remain silent in early germ cells; some of this involves a histone demethylase is unknown. Further increases in H3K4 silencing might require histone arginine methylation brought about by methylation might be required for proper developmental gene expression. PRMT5 27 @2007 Nature Publishing Group
cancer16. Consequently, these genes are difficult to reactivate in cloned embryos because of inefficient reprogramming of repressive marks, particularly of DNA methylation17. Special epigenetic regulation needs to occur in PGCs developing in the early post-implantation embryo18. Because these cells emerge from cell types in the egg cylinder that are already on the way to lineage commitment and differentiation, the somatic gene-expression programme needs to be suppressed. One of the key regulators of this process is BLIMP1 (B-lymphocyte-induced maturation protein 1), which associates with the arginine methyltransferase PRMT5. PRMT5 might partly repress Hox-family genes and other somatic genes in PGCs19 (Fig. 2c). Pluripotency-associated genes and genes that have later roles in germcell development can also be repressed by DNA methylation (Fig. 2b). So genes such as Mvh (also known as Ddx4), Dazl (deleted in azoospermia-like) and Sycp3 (synaptonemal complex protein 3) are methylated in early PGCs and begin to be expressed after the erasure of DNA methylation20, which occurs between embryonic day (E) 8.0 and E12.5 in PGCs. Interestingly, pluripotency-associated genes such as Nanog also begin to be reactivated at these stages, but it is not known whether this involves demethylation of DNA. PGCs at these stages have similar properties to pluripotent cells, including the ability to form embryonic germ cells in culture21. These studies are important because they are the first to show that in some developmental situations, removal of epigenetic marks (H3K27 methylation in the ES-cell study, and DNA methylation in the PGC study) could be crucial for the activation of developmental genes. Whether DNA methylation in PGCs is erased by an active or a passive mechanism is unclear (discussed later). The promoters of the genes that undergo ‘developmental’ demethylation (for example, Mvh, Dazl and Sycp3) contain CpG islands, as do the differentially methylated regions (DMRs) of imprinted genes, which also undergo demethylation at these stages of PGC development. I am not aware of any reports of demethylation of CpG islands during development other than in PGCs or in the zygote and pre-implantation embryo (discussed later). Methylation of CpG islands might only be removable under exceptional circumstances. Some key pluripotency-associated genes (such as Oct4 and Nanog) are epigenetically inactivated at later stages of gametogenesis and in the mature gametes, including by DNA methylation. Therefore, after a Temporary repression of developmental genes by the PcG protein system b c Pluripotent cell Histone demethylase? Histone methyltransferase Histone demethylase? Histone methyltransferase DNMT Histone demethylase? Histone methyltransferase Pluripotent cell Pluripotent cell Differentiated cell Repression of pluripotency-associated genes by histone methylation and DNA methylation Maintenance of silencing of somatic genes in early germ cells Differentiated cell Early PGC PRC PRC PRMT5 Figure 2 | Epigenetic regulation of pluripotency-associated genes and developmental genes during the differentiation of somatic cells and germ cells. The expression or repression of pluripotency-associated genes and developmental genes is indicated, and the associated modifications of the histone tails and/or DNA are represented by different colours. a, In pluripotent cells, the repression of genes that are needed later in development is flexible and can involve the PcG-protein repressive system. Silent developmental genes can be marked by both H3K27 methylation (yellow) and H3K4 methylation (blue), possibly allowing rapid gene activation after loss of repression by PcG-protein-containing repressive complexes (PRCs). Whether the loss of H3K27 methylation involves a histone demethylase is unknown. Further increases in H3K4 methylation might be required for proper developmental gene expression. b, Pluripotency-associated genes are stably silenced during differentiation, through histone methylation and DNA methylation. For example, genes such as Oct4 and Nanog are silenced during ES-cell differentiation, and this process can involve both histone methylation (such as methylation of H3K9 mediated by G9A; also known as EHMT2) (green) and DNA methylation (red). Whether a histone demethylase is required for the removal of H3K4 methylation is unknown. c, For germ-cell development, the repression of somatic genes needs to be maintained in early germ cells, and this process might involve histone arginine methylation (pink). Hox-family genes and other developmental genes remain silent in early germ cells; some of this silencing might require histone arginine methylation brought about by PRMT5. 427 NATURE|Vol 447|24 May 2007 INSIGHT REVIEW
INSIGHT REVIEW NATURE Vol 447 24 May 2007 fertilization, the repressive epigenetic marks might need to be removed involve pre-existing histone marks". After fertilization, the methylation for transcriptional activation of these genes and correct early lineage of imprinted-gene DMRs is maintained by DNmTlo( the oocyte form development to take place(discussed later) of DNMTI) for one division cycle during very early pre-implantation developmentand then by DNMTls(the somatic form of DNMTI)in Stability for transposon silencing and imprinting embryonic and adult tissues In contrast to developmental genes, which need to be epigenetically Imprinted genes can be directly silenced by methylation of DMrs gulated with flexibility, transposons(if possible)need to be silenced (which often contain CpG islands )that overlap the promoter. More fre- rent them from moving around in the genome and potentially causing a single dMr that is methylated in the germ line and is responsible for utations". Therefore, many transposon families are both methyl- regulating gene silencing in the rest of the cluster. So far, there are two ted themselves and marked by repressive histone modifications distinct models for how, after fertilization, imprinted genes are silenced uch as H3K9 methylation), and these marks are important for the through the action of nearby unmethylated DMRs. First, the DRover heritable silencing of transposons. Some transposon families(such as laps the promoter of a long, non-coding, unspliced, nuclear RNA ntracisternal A particles; IAPs)are also resistant to the erasure of Dna The presence of the unmethylated and expressed copy of the non-coding methylation in the zygote and in PGCs, possibly resulting in epigenetic RNA results in the silencing of linked genes, a process that involves inheritance across generations(discussed later) repressive histone modifications. It is unclear how the presence of the Imprinted genes are a class of mammalian genes with possible non-coding RNA leads to gene silencing in cis. In one model, repressive mechanistic relationships to transposons in that CpG islands in their complexes(for example, PRCs)might be targeted during transcription promoters become methylated and in that silencing relies on long- Alternatively, the RNA might coat the region to be inactivated, sim term epigenetic stability In imprinted genes(and transposons), DNA larly to how Xist RNA(inactive X-specific transcripts)coats the inactive methylation is introduced during either oogenesis or spermatogenesis, Xchromosome. This might establish a physical structure from which by the de novo methyltransferase DNA methyltransferase 3A(DNMT3A) RNA polymerase II( Pol II)is excluded, resulting in transcriptional and its cofactor DNMT3-like(DNMT3L)(Fig 3a). How particular silencing(Fig. 3b). In one case of silencing mediated by an imprinted imprinted genes are selected for de novo methylation during oogenesis non-coding RNA, the developmental kinetics of inactivation are mark or spermatogenesis is not understood, although this targeting could edly similar to those of imprinted X-chromosome inactivation. Both a Acquisition of dna methylation in germ cells Immature gamete DNMT b Silencing of the X chromosome and im Embryonic lineage d methyltransferase? Differentiated 想8 Extra-embryonic lineage Figure 3 Developmental regulation of imprinting and X-chromosome adjacent genes as a consequence of the physical exclusion of pol ll and the inactivation. a, During germ-cell development, selected imprinted genes acquisition of histone modifications and/or DNA methylation, depending and transposons become methylated. This process depends on de novo the embryonic lineage. DNA methylation stabilizes gene silencing methyltransferases such as DNMT3A and its cofactor DNMT3L. It in embryonic tissues but is less important in extra-embryonic tissues, is possible that the targeting of DNA methylation requires arginine where PRC-mediated silencing might predominate. This mechanism of arried out by PRMT7. Mature male germ cells postzygotic gene silencing occurs in X-chromosome inactivation and in have chromatin that is largely based on non-histone proteins known as some forms of autosomal gene imprinting H3K9 methylation is shown tamines(dark pink); this alters the packaging of the DNA. b, Expression in green, H3K27 methylation in yellow, histone arginine methylation in of non-coding RNAs(wavy blackline) in cis can result in the silencing of pink and DNA methylation in red. 28 @2007 Nature Publishing Group
fertilization, the repressive epigenetic marks might need to be removed for transcriptional activation of these genes and correct early lineage development to take place (discussed later). Stability for transposon silencing and imprinting In contrast to developmental genes, which need to be epigenetically regulated with flexibility, transposons (if possible) need to be silenced completely and stably (at least from the perspective of the host) to prevent them from moving around in the genome and potentially causing mutations22. Therefore, many transposon families are both methylated themselves and marked by repressive histone modifications (such as H3K9 methylation), and these marks are important for the heritable silencing of transposons. Some transposon families (such as intracisternal A particles; IAPs) are also resistant to the erasure of DNA methylation in the zygote and in PGCs, possibly resulting in epigenetic inheritance across generations (discussed later). Imprinted genes are a class of mammalian genes with possible mechanistic relationships to transposons23, in that CpG islands in their promoters become methylated and in that silencing relies on longterm epigenetic stability. In imprinted genes (and transposons), DNA methylation is introduced during either oogenesis or spermatogenesis, by the de novo methyltransferase DNA methyltransferase 3A (DNMT3A) and its cofactor DNMT3-like DNMT3L)24,25 (Fig. 3a). How particular imprinted genes are selected for de novo methylation during oogenesis or spermatogenesis is not understood, although this targeting could involve pre-existing histone marks26. After fertilization, the methylation of imprinted-gene DMRs is maintained by DNMT1o (the oocyte form of DNMT1) for one division cycle during very early pre-implantation development27 and then by DNMT1s (the somatic form of DNMT1) in embryonic and adult tissues28. Imprinted genes can be directly silenced by methylation of DMRs (which often containCpG islands) that overlap the promoter. More frequently, however, imprinted genes occur in clusters, and there is usually a single DMR that is methylated in the germ line and is responsible for regulating gene silencing in the rest of the cluster. So far, there are two distinct models for how, after fertilization, imprinted genes are silenced through the action of nearby unmethylated DMRs. First, the DMR overlaps the promoter of a long, non-coding, unspliced, nuclear RNA29,30. The presence of the unmethylated and expressed copy of the non-coding RNA results in the silencing of linked genes, a process that involves repressive histone modifications31,32. It is unclear how the presence of the non-coding RNA leads to gene silencing in cis. In one model, repressive complexes (for example, PRCs) might be targeted during transcription33. Alternatively, the RNA might ‘coat’ the region to be inactivated, similarly to how Xist RNA (inactive X-specific transcripts) coats the inactive X chromosome31,34. This might establish a physical structure from which RNA polymerase II (Pol II) is excluded, resulting in transcriptional silencing35 (Fig. 3b). In one case of silencing mediated by an imprinted non-coding RNA, the developmental kinetics of inactivation are markedly similar to those of imprinted X-chromosome inactivation. Both Acquisition of DNA methylation in germ cells Silencing of the X chromosome and imprinted genes a b Immature gamete Mature gamete DNMT DNMT Histone methyltransferase? Pluripotent cell PRC Differentiated cells Embryonic lineage Extra-embryonic lineage PRMT7? Pol II Pol II Pol II Figure 3 | Developmental regulation of imprinting and X-chromosome inactivation. a, During germ-cell development, selected imprinted genes and transposons become methylated. This process depends on de novo methyltransferases such as DNMT3A and its cofactor DNMT3L. It is possible that the targeting of DNA methylation requires arginine methylation of histones, carried out by PRMT7. Mature male germ cells have chromatin that is largely based on non-histone proteins known as protamines (dark pink); this alters the packaging of the DNA. b, Expression of non-coding RNAs (wavy black line) in cis can result in the silencing of adjacent genes as a consequence of the physical exclusion of Pol II and the acquisition of histone modifications and/or DNA methylation, depending on the embryonic lineage. DNA methylation stabilizes gene silencing in embryonic tissues but is less important in extra-embryonic tissues, where PRC-mediated silencing might predominate. This mechanism of postzygotic gene silencing occurs in X-chromosome inactivation and in some forms of autosomal gene imprinting. H3K9 methylation is shown in green, H3K27 methylation in yellow, histone arginine methylation in pink and DNA methylation in red. 428 INSIGHT REVIEW NATURE|Vol 447|24 May 2007
NATUREIVol 447 24 May 2007 INSIGHT REVIEW non-coding RNAs(Kcnqlotl and Xist)begin to be expressed from the sequences. For example, the Nanog promoter becomes highly methyl paternal allele in the two-cell embryo, and gene silencing in cis and the ated in mature sperm quisition of histone modifications follow during the next few cleavage Distinct genome-wide reprogramming events also occur immedi- divisions and are largely complete by the blastocyst stage(Fig. 3b). ately after fertilization and during early pre-implantation development The second model of how imprinted genes are silenced involves an(Fig. 4b). Many sequences in the paternal genome become suddenl epigenetically regulated chromatin insulator. In this model, tissue- demethylated shortly after fertilization. This demethylation occurs specific enhancers are located on one side of the DMR overlapping with after the removal of protamines(basic proteins that are associated with he insulator, whereas the silenced genes are on the other side". Silencing DNA in sperm)and the acquisition of histones by the paternal genome occurs when the DMR is unmethylated and binds chromatin-organizing during the long Gl phase, before DNA replication. Methylation can proteins such as CTCF(CCCTC-binding factor), resulting in a higher- be observed by staining cells with an immunofluorescently labelled order chromatin structure that prevents interactions between remote antibody specific for 5-methylcytosine. Judged by the substantial loss enhancers and promoters of immunofluorescence signal, together with the considerable loss of X-chromosome inactivation is another example of a relatively stable methylation of Linel elements as determined by bisulphite sequencing epigenetic silencing event; in this case, large regions of a whole chro- the paternal genome loses a significant amount of methylation, although mosome are involved In mice, imprinted X-chromosome inactivation more precise measurements and more information about which is probably largely initiated by expression of Xist from the paternal sequences are affected and unaffected would be valuable Sequences chromosome at the two-cell stage". (The nature of the imprinting that are known not to be affected include IAPs and paternally meth leading to paternal expression is still unknown, but it is unlikely to ylated DMRs in imprinted genes(Fig. 4c). A recent study provides be DNa methylation ) Imprinted X-chromosome inactivation is then intriguing insight into a protein that might protect the genome from stable(even in the absence of DNA methylation")in the extra-embry- demethylation. The protein stella(also known as DPPA3)binds to onic tissues. Although the PcG protein system(which confers H3K27 DNA and was originally identified because expression of the encoding methylation marks) has some influence on gene silencing, these mod gene is upregulated during early PGC development. Stella is present in fications do not seem to confer heritable silencing" Random X-chro- large amounts in oocytes and, after fertilization, translocates to both mosome inactivation is initiated in the epiblast after reprogramming of pronuclei. Deletion of the gene from the oocyte(and therefore removal printed inactivation" This rep ing might be initiated by of the protein from the zygote)results in early pre-implantation lethality the silencing of Xist expression, and if this is the case, it is possible that of embryos, as well as loss of methylation of the following sequences the mitotic memory' for inactivation simply resides in the expression the maternally methylated genes PegI (also known as Mest), Pegs (alse of Xist. The subsequent upregulation of Xist expression during the dif- known as Nnat)and Peg10; the paternally methylated genes H19 and ferentiation of epiblast cells is again followed by coating, gene silencing Rasgrfl(Ras protein-specific guanine-nucleotide-releasing factor 1); and acquisition of histone marks". However, in contrast to imprinted and IAPs". So stella might, either directly or indirectly, protect specific X-chromosome inactivation, CpG islands in inactivated genes on the sequences from demethylation in the zygote, but it is unknown how X chromosome become methylated and, although it has not been tested other sequences are protected ( Fig. 4c) genetically, this might constitute long-term memory for inactivation The mechanism of active demethylation in the zygote is still during embryonic and adult life(Fig. 3b). It is important to note that unknown. However, the DNA deaminases AID and APOBECI have this methylation of CpG islands seems to be a dead end in that it does been shown in vitro to deaminate 5-methylcytosine in DNA to thym not need to be reprogrammed during the normal life cycle. (In the germ ine; this results in TG mismatches, which can be repaired by the line, the inactivated X chromosome does not become methylated. base-excision repair pathway. Interestingly, Aid and Apobecl are located in a cluster of genes with Stella, growth differentiation factor 3(Gdf Breaking stability by epigenetic reprogramming and Nanog. Stella, Gdf3 and Nanog are all expressed in pluripotent tis- DNA-methylation patterns that have been acquired during develop- sues, and Gdf3 and Nanog have important roles in conferring stem-cell ment are stable in somatic cells and during adult life. DNA-methylation identity on ES cells. Indeed, Aid and Apobecl are also expressed by patterns are somatically heritable essentially through the action of oocytes, stem cells and germ cells, and recent work shows that in vi DNMTI, the maintenance methyltransferase. At most CPG sites, targeting of Aid to the methylated H19 DMR in the zygote results the error rate of maintaining methylation (-1% per division)is low in efficient and substantial demethylation of this region(C. F Chan, in relation to the number of cell divisions that are needed to pro- H. Morgan, F. Santos, D. Lucifero, S. Petersen- Mahrt, W. Dean and duce a mammalian organism(44 for humans). Indeed, methylation W.R., unpublished observations). Although it is unclear whether of CpG islands is never erased during normal development. By con- AID and/or APOBECI are responsible for the demethylation of trast, methylation of CpG islands in imprinted-gene DMRs needs to the paternal genome in the zygote, the evidence suggests that base be erased in the germ line so that gender-specific methylation can be excision or mismatch repair might have a role in this process. I think posed subsequently, during germ-cell development. This erasure that this suggestion is supported by the recent identification of a dNA takes place in a defined period-from E10.5 to E12.5 in PGCs-in glycosylase-lyase-DEMETER-that preferentially excises 5-methyl all imprinted genes that have been tested,, and it could occur by cytosine from DNA in Arabidopsis thaliana?/. DEMETER is required active demethylation of DNA by an unknown mechanism, possibly for the demethylation and activation of the imprinted gene MEDEA involving DNA repair(discussed later). This mechanism for erasure (see page 418). Another DNA-damage-responsive gene, the mouse gene ight also underlie the demethylation and activation of non-imprinted Gadd45(growth arrest and DNA-damage-inducible 45), might also such as Mvh, Dazl and Sycp3, which takes place at about the same have a role in demethylation stage(Fig 4a) Although there have been suggestions that the methyl group could pigenetic reprogramming in PGCs entails widespread loss of dna be directly removed from dNa by hydrolytic attack or by oxidation, methylation, as well as H3K9 methylation". In addition to the erasure these mechanisms have not been substantiated. The relative flexibility of genomic imprints, this epigenetic reprogramming might also help of histone methylation might be brought about by the attachment of to return PGCs to a pluripotent state(because at these stages of PGc the methyl group through a carbon-nitrogen bond, together with the development, pluripotent embryonic germ cells can be established existence of enzymes that can directly remove the methyl group, leav- in culture), through the reactivation of genes such as Nanog. Not all ing the rest of the histone molecule intact. By contrast, the current wa.ch as IAPs remain fairly highly methylated. Later in oogenesis and evidence suggests that methyl groups attach through a carbon-carbon genomic methylation is lost, however, at these stages; some transpose bond to the cytosine base and therefore might not be able to be directly rmatogenesis, de novo methylation occurs not only sex-specifically removed, so demethylation inevitably has to proceed by pathways that in imprinted genes but also in transposons and in single-copy gene involve base-excision or mismatch repair 429 @2007 Nature Publishing Group
non-coding RNAs (Kcnq1ot1 and Xist) begin to be expressed from the paternal allele in the two-cell embryo, and gene silencing in cis and the acquisition of histone modifications follow during the next few cleavage divisions and are largely complete by the blastocyst stage34 (Fig. 3b). The second model of how imprinted genes are silenced involves an epigenetically regulated chromatin insulator. In this model, tissuespecific enhancers are located on one side of the DMR overlapping with the insulator, whereas the silenced genes are on the other side36. Silencing occurs when the DMR is unmethylated and binds chromatin-organizing proteins such as CTCF (CCCTC-binding factor), resulting in a higherorder chromatin structure that prevents interactions between remote enhancers and promoters37. X-chromosome inactivation is another example of a relatively stable epigenetic silencing event; in this case, large regions of a whole chromosome are involved. In mice, imprinted X-chromosome inactivation is probably largely initiated by expression of Xist from the paternal chromosome at the two-cell stage38. (The nature of the imprinting leading to paternal expression is still unknown, but it is unlikely to be DNA methylation.) Imprinted X-chromosome inactivation is then stable (even in the absence of DNA methylation39) in the extra-embryonic tissues. Although the PcG protein system (which confers H3K27 methylation marks) has some influence on gene silencing, these modifications do not seem to confer heritable silencing40. Random X-chromosome inactivation is initiated in the epiblast after reprogramming of imprinted inactivation41,42. This reprogramming might be initiated by the silencing of Xist expression, and if this is the case, it is possible that the mitotic ‘memory’ for inactivation simply resides in the expression of Xist. The subsequent upregulation of Xist expression during the differentiation of epiblast cells is again followed by coating, gene silencing and acquisition of histone marks43. However, in contrast to imprinted X-chromosome inactivation, CpG islands in inactivated genes on the X chromosome become methylated and, although it has not been tested genetically, this might constitute long-term memory for inactivation during embryonic and adult life43 (Fig. 3b). It is important to note that this methylation of CpG islands seems to be a dead end in that it does not need to be reprogrammed during the normal life cycle. (In the germ line, the inactivated X chromosome does not become methylated.) Breaking stability by epigenetic reprogramming DNA-methylation patterns that have been acquired during development are stable in somatic cells and during adult life. DNA-methylation patterns are somatically heritable essentially through the action of DNMT1, the maintenance methyltransferase44. At most CpG sites, the error rate of maintaining methylation (~1% per division) is low in relation to the number of cell divisions that are needed to produce a mammalian organism (44 for humans). Indeed, methylation of CpG islands is never erased during normal development. By contrast, methylation of CpG islands in imprinted-gene DMRs needs to be erased in the germ line so that gender-specific methylation can be imposed subsequently, during germ-cell development. This erasure takes place in a defined period — from E10.5 to E12.5 in PGCs — in all imprinted genes that have been tested45,46, and it could occur by active demethylation of DNA by an unknown mechanism, possibly involving DNA repair (discussed later). This mechanism for erasure might also underlie the demethylation and activation of non-imprinted genes such as Mvh, Dazl and Sycp3, which takes place at about the same stage20 (Fig. 4a). Epigenetic reprogramming in PGCs entails widespread loss of DNA methylation, as well as H3K9 methylation47. In addition to the erasure of genomic imprints, this epigenetic reprogramming might also help to return PGCs to a pluripotent state (because at these stages of PGC development, pluripotent embryonic germ cells can be established in culture), through the reactivation of genes such as Nanog. Not all genomic methylation is lost, however, at these stages; some transposons such as IAPs remain fairly highly methylated48. Later in oogenesis and spermatogenesis, de novo methylation occurs not only sex-specifically in imprinted genes but also in transposons and in single-copy gene sequences. For example, the Nanog promoter becomes highly methylated in mature sperm49. Distinct genome-wide reprogramming events also occur immediately after fertilization and during early pre-implantation development (Fig. 4b). Many sequences in the paternal genome become suddenly demethylated shortly after fertilization50–53. This demethylation occurs after the removal of protamines (basic proteins that are associated with DNA in sperm) and the acquisition of histones by the paternal genome during the long G1 phase, before DNA replication. Methylation can be observed by staining cells with an immunofluorescently labelled antibody specific for 5-methylcytosine. Judged by the substantial loss of immunofluorescence signal, together with the considerable loss of methylation of Line1 elements as determined by bisulphite sequencing48, the paternal genome loses a significant amount of methylation, although more precise measurements and more information about which sequences are affected and unaffected would be valuable. Sequences that are known not to be affected include IAPs and paternally methylated DMRs in imprinted genes (Fig. 4c). A recent study provides intriguing insight into a protein that might protect the genome from demethylation. The protein stella (also known as DPPA3) binds to DNA and was originally identified because expression of the encoding gene is upregulated during early PGC development. Stella is present in large amounts in oocytes and, after fertilization, translocates to both pronuclei. Deletion of the gene from the oocyte (and therefore removal of the protein from the zygote) results in early pre-implantation lethality of embryos, as well as loss of methylation of the following sequences: the maternally methylated genes Peg1 (also known as Mest), Peg5 (also known as Nnat) and Peg10; the paternally methylated genes H19 and Rasgrf1 (Ras protein-specific guanine-nucleotide-releasing factor 1); and IAPs54. So stella might, either directly or indirectly, protect specific sequences from demethylation in the zygote, but it is unknown how other sequences are protected (Fig. 4c). The mechanism of active demethylation in the zygote is still unknown. However, the DNA deaminases AID and APOBEC1 have been shown in vitro to deaminate 5-methylcytosine in DNA to thymine55; this results in T•G mismatches, which can be repaired by the base-excision repair pathway. Interestingly, Aid and Apobec1 are located in a cluster of genes with Stella, growth differentiation factor 3 (Gdf3) and Nanog. Stella, Gdf3 and Nanog are all expressed in pluripotent tissues, and Gdf3 and Nanog have important roles in conferring stem-cell identity on ES cells. Indeed, Aid and Apobec1 are also expressed by oocytes, stem cells and germ cells55, and recent work shows that in vivo targeting of AID to the methylated H19 DMR in the zygote results in efficient and substantial demethylation of this region (C. F. Chan, H. Morgan, F. Santos, D. Lucifero, S. Petersen-Mahrt, W. Dean and W.R., unpublished observations). Although it is unclear whether AID and/or APOBEC1 are responsible for the demethylation of the paternal genome in the zygote, the evidence suggests that baseexcision or mismatch repair might have a role in this process. I think that this suggest ion is supported by the recent identification of a DNA glycosylase–lyase — DEMETER — that preferentially excises 5-methylcytosine from DNA in Arabidopsis thaliana56,57. DEMETER is required for the demethylation and activation of the imprinted gene MEDEA (see page 418). Another DNA-damage-responsive gene, the mouse gene Gadd45 (growth arrest and DNA-damage-inducible 45), might also have a role in demethylation58. Although there have been suggestions that the methyl group could be directly removed from DNA by hydrolytic attack or by oxidation, these mechanisms have not been substantiated2 . The relative flexibility of histone methylation might be brought about by the attachment of the methyl group through a carbon–nitrogen bond, together with the existence of enzymes that can directly remove the methyl group, leaving the rest of the histone molecule intact12. By contrast, the current evidence suggests that methyl groups attach through a carbon–carbon bond to the cytosine base and therefore might not be able to be directly removed, so demethylation inevitably has to proceed by pathways that involve base-excision or mismatch repair55–57. 429 NATURE|Vol 447|24 May 2007 INSIGHT REVIEW